Neurogliaform cortical interneurons derive from cells in the preoptic area

Delineating the basic cellular components of cortical inhibitory circuits remains a fundamental issue in order to understand their specific contributions to microcircuit function. It is still unclear how current classifications of cortical interneuron subtypes relate to biological processes such as their developmental specification. Here we identified the developmental trajectory of neurogliaform cells (NGCs), the main effectors of a powerful inhibitory motif recruited by long-range connections. Using in vivo genetic lineage-tracing in mice, we report that NGCs originate from a specific pool of 5-HT3AR-expressing Hmx3+ cells located in the preoptic area (POA). Hmx3-derived 5-HT3AR+ cortical interneurons (INs) expressed the transcription factors PROX1, NR2F2, the marker reelin but not VIP and exhibited the molecular, morphological and electrophysiological profile of NGCs. Overall, these results indicate that NGCs are a distinct class of INs with a unique developmental trajectory and open the possibility to study their specific functional contribution to cortical inhibitory microcircuit motifs.

Among cortical INs, neurogliaform cells (NGCs) display unique characteristics. They represent the main source of 'slow' cortical inhibition by acting on metabotropic GABA B receptors (Tamás et al., 2003), and are thought to be key effectors of a powerful inhibitory circuit recruited by long-range connections such as interhemispheric and thalamic projections (Craig and McBain, 2014;Palmer et al., 2012;De Marco García et al., 2015). Whether the current description of NGCs captures an IN subtype related to a distinct developmental specification process is unclear. Here we used in vivo genetic lineage-tracing to follow the developmental origin and trajectory of NGCs. We found that they originate from a distinct pool of 5-HT 3A R-expressing Hmx3+ cells located in the rostral POA region, ventrally to the anterior commissure. In the embryonic POA, Htr3a-GFP+ INs in the Hmx3+ domain expressed CGE-enriched TFs such as PROX1 and NR2F2, but only rarely, if not, MGE-related TFs such as NKX2.1 or LHX6. In the cortex, Hmx3-derived Htr3a-GFP+ INs expressed markers of CGE-derived INs such as NPY and/or reelin, as well as CGE-enriched TFs such as SP8, NR2F2 and PROX1, but neither MGE-specific markers such as PV or SST nor TFs such as LHX6 or SOX6. Finally, single-molecule in situ hybridization and electrophysiological recordings followed by post hoc reconstructions indicated that Hmx3-derived Htr3a-GFP+ cells exhibited the molecular, electrophysiological and morphological profile of NGCs. Taken together, these results demonstrate that cortical NGCs have a precise developmental trajectory that is linked to the expression of the transcription factor (TF) Hmx3 in a discrete embryonic subpallial region. eLife digest Our brain contains over a 100 billion nerve cells or neurons, and each of them is thought to connect to over 1,000 other neurons. Together, these cells form a complex network to convey information from our surroundings or transmit messages to designated destinations. This circuitry forms the basis of our unique cognitive abilities.
In the cerebral cortex -the largest region of the brain -two main types of neurons can be found: projection neurons, which transfer information to other regions in the brain, and interneurons, which connect locally to different neurons and harmonize this information by inhibiting specific messages. The over 20 different types of known interneurons come in different shapes and properties and are thought to play a key role in powerful computations such as learning and memory.
Since interneurons are hard to track, it is still unclear when and how they start to form and mature as the brain of an embryo develops. For example, one type of interneurons called the neurogliaform cells, have a very distinct shape and properties. But, until now, the origin of this cell type had been unknown.
To find out how neurogliaform cells develop, Niquille, Limoni, Markopoulos et al. used a specific gene called Hmx3 to track these cells over time. With this strategy, the shapes and properties of the cells could be analyzed. The results showed that neurogliaform cells originate from a region outside of the cerebral cortex called the preoptic area, and later travel over long distances to reach their final location. The cells reach the cortex a few days after their birth and take several weeks to mature.
These results suggest that the traits of a specific type of neuron is determined very early in life. By labeling this unique subset of interneurons, researchers will now be able to identify the specific molecular mechanisms that help the neurogliaform cells to develop. Furthermore, it will provide a new strategy to fully understand what role these cells play in processing information and guiding behavior.
To determine whether Hmx3; tdTOM+/Htr3a-GFP+ cells in the POA eventually give rise to a specific subpopulation of cortical INs, we examined brains at various postnatal ages. From P5 to P21, Hmx3; tdTOM+/Htr3a-GFP+ INs were found distributed preferentially in superficial cortical layers and in a variety of other brain regions including in the hippocampus ( Overall, these findings indicate that the cortical Hmx3-derived lineage observed in the POA differentiate into INs that are Htr3a-GFP+, glial cells that are Htr3a-GFP negative and, for a small fraction, to NeuN+ neurons negative for Htr3a-GFP. A second distinct region in the POA expressing Dbx1 was previously reported to give rise to subsets of cortical INs (Gelman et al., 2011). To determine whether a fraction of Htr3a-GFP+ INs also originate from Dbx1-expressing cells, we examined Dbx1-Cre::Htr3a-GFP; R26R-tdTOM fl/fl brains at postnatal periods. While the overall contribution of Hmx3-derived cells to the Htr3a-GFP IN population in the cortex increased with postnatal maturation from P5 (6.8 ± 0.2%; 77/1118 cells) to P21 (16.0 ± 0.3%; 863/5405 cells) ( Figure Figure 3F). Overall, these results indicate that Hmx3-derived 5-HT 3A R+ cortical INs largely originate from Hmx3-expressing cells but not from the Dbx1+ domain, which gives rise to INs expressing MGE-related markers.
We next examined whether Hmx3; tdTOM+/Htr3a-GFP+ INs expressed classical CGE markers such as reelin, NPY and VIP (Lee et al., 2010;Murthy et al., 2014;Vucurovic et al., 2010).    Figure 6D). With regard to their first action potential (AP) at rheobase, NGCs are reported to display a type 1 profile with only an after-hyperpolarization potential (AHP), whereas SBCs show a type 2 profile consisting of an AHP followed by an after-depolarization potential (ADP) (Cadwell et al., 2016;Jiang et al., 2015). Strikingly, all (29 out of 29) recorded Hmx3; tdTOM+/Htr3a-GFP+ INs were of type 1, thus confirming their NGC identity. Moreover, the vast majority of them (23 out of 29) were of type 1A with a deep and wide AHP and only a few (6 out of 29) were of type 1B with a shallow and narrow AHP ( Figure 6E, left, Figure 6-source data 2). Htr3a-GFP+ INs negative for Hmx3; tdTOM had more variable profiles, but the majority of them (24 out of 28) were displaying a type two profile with an average ADP amplitude of 2.30 ± 0.51 mV, suggesting that they were SBCs. Most of them (20 out of 28) were of type 2B with a small ADP below the spike threshold, a few others (4 out of 28) were of type 2A with a big ADP above the spike threshold ( Figure 6E, right, Figure 6-source data 2) and another few of them (4 out of 28) had not measurable ADP (not shown). Hmx3; tdTOM+/ Htr3a-GFP+ INs showed also a higher tendency to late-spiking when compared to Htr3a-GFP+ INs ( Figure 6F, Figure 6-source data 2). Hmx3; tdTOM+/Htr3a-GFP+ INs had bigger AP delay average, but not significantly different from Htr3a-GFP+ INs negative for Hmx3; tdTOM ( Figure 6-figure supplement 2C, Figure 6-source data 2). However, the variability of individual cell values was higher for Hmx3; tdTOM+/Htr3a-GFP+ INs, indicating that these cells tend to be more late-spiking, a characteristic of NGCs (Cadwell et al., 2016;Jiang et al., 2015). Alignement of the first APs at rheobase revealed other putative differences between the two groups ( Figure 6G Figure 6-source data 1). Morphological recovery of Hmx3; tdTOM+/Htr3a-GFP+ INs located in L2-6 revealed that all cells (7 out of 7) had also the characteristic morphology of NGCs ( Figure 6-figure supplement 4). Furthermore, characteristic properties of NGC like the tendency to late spiking, the depth of AHP, and the level of Vthr were significantly more pronounced in these cells compared to L1 cells ( Figure 6-figure supplement 3C,E). Overall, these data indicate that Htr3a-GFP+ INs displaying the molecular, morphological and electrophysiological properties of NGC INs originate from Hmx3-expressing cells in the embryonic POA (Figure 7, orange), whereas SBCs in layer 1, as well as VIP +INs, are more likely to originate from the CGE (Figure 7, green).

Discussion
Distinct subtypes of local GABAergic INs are required to regulate microcircuit function (Cardin et al., 2009;Fu et al., 2014;Kepecs and Fishell, 2014;Pfeffer et al., 2013;Pi et al., 2013;Pinto and Dan, 2015;Sohal et al., 2009;Zhang et al., 2014). Whether current classifications of cortical IN subtypes relate to intrinsic biological processes such as their developmental specification is a key question in the field (Huang, 2014;Taniguchi et al., 2013). Among cortical INs, NGCs are considered as belonging to a distinct subtype acting as a main effector of a powerful inhibitory motif recruited by long-range connections (Tamás et al., 2003;Craig and McBain, 2014;Palmer et al., 2012). Here we aimed to track the developmental trajectory of NGCs using genetic fate mapping strategies. We find that NGCs derive from Htr3a-GFP+/Hmx3+ cells located in the embryonic POA, but not from Dbx1+ cells. Strikingly, L1 Hmx3-derived Htr3a-GFP+ INs display the distinct molecular, morphological and electrophysiological properties of NGCs, whereas Htr3a-GFP + INs negative for Hmx3 have the profile of SBCs (Cadwell et al., 2016). Hmx3-derived Htr3a-GFP + NGCs represent about a third of reelin-expressing Htr3a-GFP+ INs. At a molecular level, Hmx3; tdTOM+/Htr3a-GFP+ NGCs express CGE-related TFs such as NR2F2 and PROX1, indicating that they share common features with CGE-derived INs. Overall, these results indicate that cortical NGCs derive from a discrete embryonic area located in the subpallial POA and that their specification is linked to the expression of the TF Hmx3.

Htr3a-GFP+ INs derived from Hmx3+ cells express CGE-enriched transcription factors
Here, we find that a fraction (about 15%) of Htr3a-GFP+ cortical INs originate from Hmx3+ but not Dbx1+ cells in the POA. The overall fraction of Hmx3; tdTOM+/Htr3a-GFP+ INs to the total Htr3a-GFP+ IN population in the cortex almost doubled from P9 to P21, a period during which neural migration is largely achieved. Given that about 40% of developing cortical INs undergo apoptosis during early postnatal life (Southwell et al., 2012) higher levels of programmed cell death in Htr3a-GFP+ INs negative for tdTOM could thus account for the relative postnatal increase in the cortical Hmx3; tdTOM+/Htr3a-GFP+ cell population. Overall, our data support the general view that the differential expression of TFs in progenitor cells originating from distinct subpallial germinal zones controls the specification of cortical IN subtypes (Huang, 2014;Kessaris et al., 2014;Anastasiades and Butt, 2011;Flames et al., 2007;Gelman et al., 2009). A striking example in the field relates to chandelier INs, which have been shown to derive from Nkx2.1+ cells produced specifically at late embryonic time-points in a restricted region of the MGE (Taniguchi et al., 2013). Three major germinal zones contribute to the generation of cortical INs, including the MGE, the CGE and the POA (Kessaris et al., 2014). The majority of cortical IN subtypes (about 60-70%) originates from Nkx2.1+ progenitors in the MGE and includes fast-spiking PV+ basket INs, chandelier cells and SST + Martinotti cells. In addition to NKX2.1, sequential expression of the TFs such as LHX6 (Anastasiades and Butt, 2011;Du et al., 2008;Liodis et al., 2007) and SOX6 (Azim et al., 2009;Batista-Brito et al., 2009) controls the specification and migration of MGE-derived IN subtypes.
Here we find that Htr3a-GFP+ cortical INs originating from Hmx3+ cells in the POA do not express MGE-enriched TFs such as LHX6 or SOX6. In the embryonic POA, we observe that only a small fraction of Hmx3; tdTOM+/Htr3a-GFP+ cells expresses the TF NKX2.1, which has been shown to be strongly expressed in the ventricular zone of the POA (Flames et al., 2007). This could be due to either down-regulation of NKX2.1 in postmitotic Hmx3; tdTOM+/Htr3a-GFP+ cells as previously observed in migrating MGE-derived INs (Nó brega-Pereira et al., 2008) or to the fact that the majority of Hmx3; tdTOM+/Htr3a-GFP+ cells do not originate from NKX2.1 progenitors. In line with this second possibility, recent genetic fate-mapping experiments using a Nkx2.1-ires-Flpo knock-in mouse line did not appear to label INs in L1 (He et al., 2016). Overall, further work needs to be done to clarify the precise origin of mitotic cells giving rise to the pool of Hmx3; tdTOM+/Htr3a-GFP+ cells observed in the embryonic POA. In contrast to the absence of co-localization with MGEenriched TFs, we find that Hmx3; tdTOM+/Htr3a-GFP+ INs express TFs such as PROX1 and NR2F2 in the embryonic POA and in the postnatal cortex. PROX1 and NR2F2 have been shown to be expressed in CGE cells and these TFs are maintained in subsets of cortical INs as they mature in the developing cortex (Cai et al., 2013;Ma et al., 2012;Rubin and Kessaris, 2013;Murthy et al., 2014;Kanatani et al., 2008). Our results thus indicate that the specification of Hmx3-derived and CGE-derived Htr3a-GFP+ INs shares common transcriptional controls and that the expression of Hmx3 in a fraction of Htr3a-GFP+ defines the distinct subtype of NGCs. To gain insights on the requirement of Hmx3 in the specification of Hmx3-derived Htr3a-GFP+ NGCs, cell-type specific genetic deletion strategies are needed. Finally, the molecular pathways specifically controlled by Hmx3 in NGCs remain to be identified.

Htr3a-GFP+ INs derived from Hmx3+ cells express CGE but not MGEenriched neurochemical markers
MGE-derived INs express the neurochemical markers PV or SST and are preferentially distributed in lower cortical layers, whereas CGE-derived INs specifically express the 5-HT 3A R, but not PV or SST, and populate more superficial cortical layers Huang, 2014;Rudy et al., 2011). Using in situ hybridization, we confirmed that Hmx3+ lineage give rise to superficial cortical Htr3a-GFP+ INs expressing the Htr3a transcript. Reelin, VIP and NPY have been used as neurochemical markers to identify different subtypes of Htr3a-GFP+ cortical INs (Lee et al., 2010;Murthy et al., 2014;Vucurovic et al., 2010). Expressions of reelin and VIP are mutually exclusive in Htr3a-GFP+ INs, whereas a fraction of them is found to co-express reelin and NPY (Lee et al., 2010). Using these markers, we find that Hmx3; tdTOM+/Htr3a-GFP+ INs express reelin and/or NPY, but not VIP, PV or SST. This is in line with previous results showing that Hmx3+ INs express NPY and not VIP, PV or SST (Gelman et al., 2009). Finally, we find that cortical INs from the Dbx1 + domain express the MGE-enriched markers PV or SST and only rarely co-label with Htr3a-GFP + INs. In addition, Dbx1-derived cortical INs express the MGE-related TFs SOX6 and LHX6 but not the CGE-enriched TF PROX1. Taken together, our findings thus indicate that Hmx3+ but not Dbx1 + cells give rise to a subpopulation of cortical Htr3a-GFP+ INs, which share molecular similarities with CGE but not MGE-derived INs. However, given that both Hmx3+ and Hmx3-Htr3a-GFP+ INs express reelin and/or NPY, these classical neurochemical markers are not sufficient to segregate Hmx3-and non-Hmx3-derived Htr3a-GFP+ IN subtypes.
Hmx3-derived Htr3a-GFP+ INs display the molecular, morphological and electrophysiological properties of NGCs Electrophysiological recordings obtained from Htr3a-GFP+ INs revealed the existence of many different subtypes of INs (Lee et al., 2010). Recently, electrophysiological and morphological characterization of L1 INs combined to single-cell transcriptomics delineated two main types of INs, namely NGCs and SBCs (Cadwell et al., 2016). Our findings support this observation and indicate that Hmx3-derived Htr3a-GFP+ INs exhibit the morphological and electrophysiological signature of NGCs and strongly express Car4, a transcript present at high level in NGCs, but not in SBCs. In contrast, Htr3a-GFP+ INs in L1 that do not derive from Hmx3+ cells, have low levels of Car4 and display the electrophysiological profile of SBCs. These results indicate that Htr3a-GFP+ cortical INs in L1 can be subdivided in two major groups characterized by distinct intrinsic properties and that these subgroups are determined by their sites of origin and the differential expression of the TF Hmx3. Finally, we show that all Hmx3; tdTOM+/Htr3a-GFP+ INs analysed in deeper cortical layers also display molecular, morphological and electrophysiological profiles of NGCs, indicating that the Hmx3-Cre line labels NGCs across neocortical layers. In vivo studies of the canonical cortical microcircuit have mainly relied on the use of the mutually exclusive SST-, PV-and VIP-Cre driver lines (Cardin et al., 2009;Fu et al., 2014;Kepecs and Fishell, 2014;Pfeffer et al., 2013;Pi et al., 2013;Pinto and Dan, 2015;Sohal et al., 2009;Zhang et al., 2014) but they do not give access to NGCs. These cells are the main source of 'slow' GABA B -receptor mediated inhibition in the neocortex (Tamás et al., 2003) and are thought to constitute the core cellular component of a canonical inhibitory circuit in L1 (Craig and McBain, 2014). NGCs acts through GABA B -receptors to inhibit the activity of projection neurons and halt ongoing network activity through dendritic calcium channels (Craig and McBain, 2014). Long-range interhemispheric inhibition has been shown to be mediated through a GABA B -receptor dependent mechanism and it has been proposed that this process requires the recruitment of L1 cortical INs, possibly of the neurogliaform-type (Craig and McBain, 2014;Palmer et al., 2012). However, given the diversity of L1 cortical INs (Cadwell et al., 2016;Jiang et al., 2013) and the lack of molecular tools to specifically target NGCs in vivo, it has so far not been possible to manipulate and interrogate exclusively NGCs in cortical networks. Our findings redefine the Hmx3-Cre mice as a valuable tool to specifically investigate the functional contribution of NGCs in the cortical microcircuit motif.

Imaging and quantification of interneuron identity and distribution
Images were acquired using confocal microscopes (Nikon A1R or Axio Imager.Z2 Basis LSM 800) equipped with oil-immersion 40x, 60x and 63x objectives (CFI Plan Fluor 40x/1.3 and CFI Plan Apo VC H 60x/1.4, Nikon or Plan-APO (UV) VIS-IR 40x/1.4 and Plan-Apochromat f/ELYRA 63x/1.4, LSM). For widefield illustrations (Figure 2-figure supplement 1), images were taken with Axioscan.Z1 slidescanner (Zeiss), equipped with Plan-Apochromat 10x/0.45 objective (Zeiss). Images were lightly treated (gamma, brightness and and despeckle filter only) for visual purpose with Photoshop CC and manual counts were achieved with Fiji. Data are presented as brain averages calculated from at least three slices at different rostro-caudal levels per brain (except for P5 Dbx1 brain 3). A detailed description of the counts, cells and brains in the different experiments is given in Supplementary file 1.

Statistical analysis and prediction model
Animals were used regardless of their sex and statistical analysis was done with R programming language and GraphPad Prism. No statistics were used to determine optimal group sample size; however, sample sizes were similar to those used in previous publications from our group and others. Normality of the samples was assessed with D'Agostino-Pearson test and when distribution was not normal, non-parametric tests were applied. Using bmrm (v3.3) package for L1-regularized logistic regression model, data were standardized, and a L1-regularized logistic regression model was trained to distinguish between Htr3a-GFP+ INs that were Hmx3-derived and those which were not. This model assigned a linear weight that reflects the power of each feature in the model logistic regression and computed a probability that a given cell is Hmx3-derived in such a way that the misclassification error on the training data was minimized ( Figure 6I-J). Classification performance of the L1-regularized logistic regression algorithm was assessed by leave-one-out-cross-validation (LOOCV). It consists in training a model on all but one cell, feeding the model with this isolated cell to predict its origin and finally assessing if the prediction is correct. Looping it over all cells, yields a prediction value for each cell, which is used to estimate the generalization error of the classifier. Finally, in order to determine if the prediction made by the logistic regression model improved over the signal contained into each feature taken individually, receiver operating characteristic (ROC) curves were drawn to visualize the sensitivity/specificity ratio for each feature and for the leave-oneout predictions. Areas under the curves (AUC) were analyzed to determine the strongest signals ( Figure 6-figure supplement 2, Figure 6-source data 2).