Transcriptional regulation of MGE progenitor proliferation by PRDM16 controls cortical GABAergic interneuron production

The complex circuitry of the mammalian neocortex comprises two major types of neuronal cells, excitatory pyramidal neurons and inhibitory interneurons, with different functions and origins (Campbell, 2003; Wilson and Rubenstein, 2000). In the mouse, both types of neurons are generated between embryonic day (E)10 and E16, from neural progenitors residing in the proliferative zones located along the lateral ventricles of the telencephalon (Anthony et al., 2004; Marin and Muller, 2014). Excitatory neurons are produced from dorsal (pallial) telencephalic proliferative zones (Govindan and Jabaudon, 2017), whereas the majority of inhibitory interneurons are produced from two transient proliferative structures in the ventral (subpallial) telencephalon, known as the medial (MGE) and caudal (CGE) ganglionic eminences (Miyoshi et al., 2010; Nery et al., 2002; Wonders and Anderson, 2006; Xu et al., 2004). Progenitors in the MGE give rise to two major groups of inhibitory interneurons, fast-spiking and non-fast-spiking, which can be identified by their expression of the markers parvalbumin (PV) and somatostatin (SST), respectively (Xu et al., 2004). The majority of SST-expressing interneurons are born in the early stages of neurogenesis, whereas PV-expressing cells are generated later (Hu et al., 2017; Miyoshi et al., 2007). Both populations undergo a process of tangential migration into the developing cortex, following a spatiotemporal pattern in which earlier born cells typically occupy the deep layers of the cortex, whereas later born cells occupy more superficial layers (Lopez-Bendito et al., 2004). During neurogenesis, neural progenitors in the MGE transition through a series of competence states as they produce these two types of interneurons: radial glia (RG) divide at the surface of the ventricle, self-renewing while giving rise to either neurons (direct neurogenesis) or more committed transit-amplifying progenitors with limited self-renewal capacity, that usually divide once to generate two neurons (indirect neurogenesis) (Harwell et al., 2015; Turrero Garcia and Harwell, 2017). In order to understand how the balance of excitation and inhibition is achieved during development and maintained in the mature cortex, it is necessary to know the mechanisms by which MGE progenitors regulate their proliferation and consequently their neuronal output (Lim et al., 2018; Petryniak et al., 2007). One crucial factor that ensures correct neural progenitor amplification and cell lineage progression is PRDM16, a transcriptional regulator that is specifically expressed in radial glia of the developing telencephalon (Baizabal et al., 2018; Chuikov et al., 2010; Shimada et al., 2017). In the cortex, Prdm16 controls indirect neurogenic divisions, radial glia lineage progression and the production of late born upper layer pyramidal neurons by regulating the epigenetic state of developmental enhancers (Baizabal et al., 2018). Prdm16 is also expressed in radial glia of the MGE, but it is unknown whether it regulates lineage progression programs controlling the production of cortical interneurons. Here we show that Prdm16 expression in MGE progenitors is necessary in order to maintain their proliferative potential and ensure the generation of sufficient numbers of cortical interneurons and the maintenance of correct inhibitory input onto pyramidal cells. PRDM16 controls the expression of both generic and MGE-specific neuronal differentiation programs through its binding to sets of cis-regulatory elements that are either common between this area and the developing neocortex or exclusive to the MGE.

We generated a Cre-lox system-based conditional knockout mouse model, in which Prdm16 was deleted in cells with a developmental history of Nkx2.1 expression (i.e. neural progenitors in the MGE and their progeny) (Cohen et al., 2014; Xu et al., 2008). Additionally, these mice carried a Cre-dependent fluorescent reporter, tdTomato (tdT), to allow identification of cortical interneurons derived from Nkx2.1-expressing MGE progenitors in the adult brain after the expression of this gene is shut down (Fig. 1A,C) (Madisen et al., 2010). To verify the specific ablation of Prdm16 in Nkx2.1-expressing cells, we examined the brains of wild-type (WT) and conditional knockout (cKO) mice at E13, a developmental stage midway through cortical interneuron neurogenesis (Turrero Garcia and Harwell, 2017), confirming the absence of PRDM16 in the ventricular zone of the MGE by immunostaining (Fig. 1B). We analyzed the brains of postnatal day (P)30 mice in order to understand the consequences of Prdm16 deletion in the mature cortex. We performed immunofluorescence staining to detect tdT, SST and PV, markers of the two major subgroups of MGE-derived forebrain interneurons (Fig. 1C). We found that the number of tdT-positive (tdT⁺) interneurons in cKO cortices was 27.27% lower than in WT controls (Fig. 1D). Decreased numbers of tdT⁺ cells were observed across all cortical layers except for layer VI (Fig. 1E), and the proportion of tdT⁺ cells that were positive for either SST or PV was the same in WT (27.39% SST⁺ and 29.96% PV⁺) and cKO cortices (26.60% SST⁺ and 27.30% PV⁺). The total numbers of tdT⁺ cortical interneurons expressing SST (Fig. 1F) or PV (Fig. 1H) were decreased by similar proportions as the overall tdT⁺ cells (30.43% in SST⁺ and 35.03% in PV⁺ cells). The loss of SST⁺ (Fig. 1G) and PV⁺ (Fig. 1I) in tdT⁺ cortical interneurons was also observed across all cortical layers except for layer VI. We observed a similar phenotype in the hippocampus (Fig. S1A), where there was a decrease in the number of tdT⁺ Nkx2.1-lineage interneurons (Fig. S1B), which was consistent across all regions except the subiculum (Fig. S1C), and across all layers (Fig. S1D). In the hippocampus, the loss was cell type-specific: the number of SST⁺, tdT⁺ cells was decreased (Fig. S1E), whereas, unexpectedly, PV⁺, tdT⁺ cells were not affected (Fig. S1F). As in the cortex, the distribution of tdT⁺ cells in the hippocampus of cKO mice closely recapitulated that of the WT in all areas (Fig. S1G,H). There was a trend toward decreased numbers of Nkx2.1-lineage cells in the striatum of cKO animals (Fig. S1I) and a significant reduction in the nucleus accumbens (Fig. S1J). Together, these observations indicate that Prdm16 regulates the number of Nkx2.1-lineage cells throughout the forebrain; this effect was particularly clear for MGE-derived interneurons in the adult cortex and hippocampus.

The proper complement of inhibitory interneurons is essential for maintaining the excitatory and inhibitory network balance; however, cKO mice do not display obvious seizure activity despite the decrease in PV⁺ and SST⁺ cortical interneurons (Neves et al., 2013; Southwell et al., 2014). We hypothesized that a decrease in inhibitory tone due to the loss of MGE-derived interneurons in the cKO cortex could be compensated by populations derived from another major embryonic source of cortical inhibitory neurons, the caudal ganglionic eminence (CGE) (Denaxa et al., 2018). To test this, we performed immunofluorescence staining in P30 WT and cKO brains to detect reelin and vasointestinal peptide (VIP), two markers of CGE-derived cortical interneurons (Fig. 2A) (Lee et al., 2010; Miyoshi et al., 2010). We did not detect a significant difference in the total number of VIP⁺ (Fig. 2B) or reelin⁺ (Fig. 2C) cortical interneurons. Their distribution across cortical layers I-VI was largely similar between WT and cKO animals (Fig. S2A,B); however, there was a significant increase in the number of reelin⁺ cells in layer I of the mutant cortex (Fig. S2B). Reelin⁺ interneurons originate from either the CGE or the MGE (Miyoshi et al., 2010); the latter population is largely SST⁺ (Pesold et al., 1999). To distinguish between these two reelin⁺ subpopulations, we quantified the proportion of cortical reelin⁺ cells that were tdT⁺, as cells labeled with the Nkx2.1-Cre driver line originate in the MGE. We found that approximately two-thirds (67.16%) of reelin⁺ interneurons in WT cortices were MGE-derived, but this proportion was significantly lower (52.41%) in cKO mice (Fig. 2D). In line with this, cKO cortices contained an increased number of CGE-derived (i.e. tdT^–) reelin⁺ interneurons (Fig. 2E), concentrated in cortical layers I and II-III (Fig. 2F). We did not observe an analogous increase in the VIP⁺ population (Fig. S2A). These results suggest that there is an increase in the number of CGE-derived interneurons in the upper layers of cKO cortices, which could partially compensate for the decrease in the number of MGE-derived cortical interneurons. This effect was specific to reelin⁺ interneurons, which express a common marker even though they derive from different developmental origins. No increase was observed in VIP⁺ interneurons, which is consistent with evidence that survival and maturation of this subgroup is not dependent upon activity (Fig. S2C) (De Marco García et al., 2011; Priya et al., 2018).

The observed increase in the number of non-MGE-derived upper-layer cortical interneurons does not necessarily imply that these cells can fully compensate for the physiological function of the missing population. We decided to investigate whether the inhibitory input onto excitatory pyramidal cells was altered in the cKO cortex. We performed whole cell electrophysiological recordings in pyramidal neurons located in layers II-III, where we had observed the greatest increase in CGE-derived reelin⁺ interneurons. We recorded from cells in the cortex of 3-week-old WT and cKO mice, analyzing their inhibitory input by comparing the frequency and amplitude of spontaneous miniature inhibitory postsynaptic currents (mIPSC; Fig. 3A). We found that the frequency of mIPSCs was 33.54% lower in the cKO cortex (Fig. 3B), but their amplitudes were not significantly different (Fig. 3C), consistent with an overall decrease in the number of inhibitory GABAergic inputs. From this, we conclude that the loss of MGE-derived interneurons in the cortex of cKO mice results in a decrease in the inhibitory input to pyramidal neurons in the upper layers that cannot be fully rescued by the increase in the number of upper layer CGE-derived reelin⁺ interneurons.

We sought to understand the causes leading to the decreased numbers of tdT⁺ cortical interneurons in the cortex of cKO mice. One possibility is that defects in the migratory capacity of Nkx2.1-lineage interneurons could impair their ability to reach their final destination in the cortex through tangential migration (Anderson et al., 1997; Corbin et al., 2001). We examined the position of tdT⁺ interneurons in the developing brains of WT and cKO animals (Fig. S3A), and found no significant differences between them: at E13, tdT⁺ cells migrating into the cortex had covered a similar proportion in animals of both genotypes (Fig. S3B), and they had largely occupied the entire cortex by E15 (Fig. S3C). Another possibility is that loss of Prdm16 function in MGE-derived postmitotic neurons could induce cell death (Close et al., 2012; Denaxa et al., 2012). However, immunostaining for the apoptosis marker cleaved caspase-3 did not differ between WT and cKO cortices at any of the developmental stages analyzed (Fig. S3D). We conclude that Prdm16 must exert its role at the progenitor level, guiding the generation of cortical interneurons rather than their survival or long-distance migration.

We analyzed the Nkx2.1-expressing progenitors that give rise to these cells. First, we compared the number of progenitors in the MGE of WT and cKO mice at E13, a mid-neurogenic stage (Turrero Garcia and Harwell, 2017). We performed immunofluorescence staining for phosphorylated histone 3 (pH3), a marker of late G2/M phase in cycling cells (Fig. 4A). We found a significant reduction in the overall density of dividing cells in the MGE of cKO mice (Fig. 4B); this effect could be observed both in the ventricular zone (VZ), where radial glia reside (Fig. 4C), and in the subventricular zone (SVZ) (Fig. 4D), which harbors other types of transit-amplifying progenitors (Turrero Garcia and Harwell, 2017). To assess the proliferative capacity of MGE progenitors, we infected embryonic mice with GFP-expressing retrovirus, which allowed us to identify and analyze the progeny (or ‘clones’) derived from single radial glial cells 24 h after infection (Fig. 4E) (Baizabal et al., 2018; Harwell et al., 2015). Although the majority of clones in the MGE of control mice were composed of two cells, with a smaller proportion of one-cell clones and progressively smaller proportions of bigger clones, in the MGE of cKO mice this distribution was shifted towards smaller clone sizes: almost two-thirds of the clones were composed of a single cell and about one-third of the total were two-cell clones (Fig. 4F). Correspondingly, the average clone size was smaller in the MGE of cKO mice (Fig. 4G). These results indicate that the changes in cortical interneuron output that we observed are the result of defects in the number and proliferative capacity of MGE progenitors.

We sought to understand better what molecular changes lead to the decreased number and proliferative capacity of progenitors in the MGE of cKO embryos. For this purpose, we performed RNA sequencing (RNA-Seq) on the dissected MGEs of WT and cKO embryos at E14 (Fig. 5A), comparing three samples of either genotype (Fig. 5B). We found 510 differentially regulated genes (absolute fold change >2, adjusted P-value <0.05; see Table S1). Of these, 326 were downregulated in the cKO samples and 184 upregulated (Fig. 5C). Among the downregulated genes (Fig. 5C,D), there are several with known functional roles at various stages of cortical interneuron development, such as Egfr (Namba et al., 2017) or Maf (Pai et al., 2019); migration, Erbb4 (Li et al., 2012); and maturation, Adamts15 (Levy et al., 2015). There are also a number of differentially expressed genes encoding neurotransmitter receptors, such as Npy1r, Htr1a or Drd1. Several of the upregulated genes (Fig. 5C,D) are expressed in MGE radial glia and control their proliferative capacity, as is the case for Hes5 (Wang et al., 2013) and Ccnd1 (Glickstein et al., 2007), whereas others control MGE regionalization, such as Shh (Xu et al., 2005), Fgf15 (Danjo et al., 2011) and Otx2 (Hoch et al., 2015). Overall, these results suggest that the defects we found in cKO mice (Fig. 4) might result from an inability of MGE radial glia to progress correctly through their differentiation program due to the lack of Prdm16.

In the cortex, Prdm16 regulates neurogenesis by modulating the expression of genes involved in the generation of intermediate progenitors and neuronal differentiation (Baizabal et al., 2018). To understand the molecular mechanisms by which Prdm16 regulates the proliferation and neuronal output of MGE progenitors, we performed chromatin immunoprecipitation followed by sequencing (ChIP-Seq) for PRDM16 on MGE samples isolated from E14 WT mice. We detected 3517 statistically reproducible (IDR P<0.05) ChIP-Seq peaks across two experimental replicates, representing PRDM16 binding sites (Table S2). We compared the enrichment of PRDM16 binding sites with those previously characterized in PRDM16 ChIP-Seq experiments performed in E15 cortex (Baizabal et al., 2018). Analysis of all the potential PRDM16 binding sites (Table S3) showed a large overlap between the MGE and cortex datasets (Table S4). To understand more about the epigenetic state of those sites, we compared our list with a published dataset studying histone modifications in the MGE at E13 (Sandberg et al., 2016). Potential PRDM16 binding sites were enriched in open chromatin histone marks H3K4me3, H3K4me1 and H3K27ac, but not in the repressive H3K27me3 modification (Fig. 6A,B). We analyzed the distribution of PRDM16 peaks across the genome and found that the majority of peaks were located in intergenic and intronic genomic regions, whereas about 10% of PRDM16 binding peaks were associated with transcription start sites (TSS) (Fig. 6C). This distribution across genomic regions was enriched in open chromatin marks and similar to that of the E15 cortex dataset (Fig. S4C-E, Fig. S5A) (Baizabal et al., 2018). We then examined the genes closest to MGE PRDM16 peaks by performing gene ontology (GO) analysis. The nearest genes in every annotated genomic region were associated to similar GO terms, with a relative enrichment of terms related to DNA synthesis and repair in genes close to TSS-associated peaks (Fig. S4B). All of the top 30 GO terms for genes were related to developmental processes, including the generation and migration of neurons (Fig. 6D). We compared the MGE and cortex ChIP-Seq datasets to find out which PRDM16 peaks were exclusive to MGE progenitors. We generated a list of highly enriched peaks (greater than threefold enrichment over input) for both datasets and obtained a list of 3206 peaks. When we analyzed the regional specificity of those peaks (Fig. 6E), we found that about one-third of them were unique to the MGE (Table S5), two-thirds were common to both datasets (Table S4) and less than 10% were cortex specific (Table S6). GO terms of nearest genes for both subsets of peaks had overall similar representation, with a slightly lower representation of synaptic signaling-related terms in MGE-exclusive peaks (Fig. S5B).

It is possible that PRDM16 exhibits lineage-specific genomic binding in order to direct GABAergic versus glutamatergic cell fate specification. To explore this possibility, we analyzed the sequence of PRDM16 peaks, searching for de novo or known binding motifs either in all MGE peaks (Fig. 6F) or in the MGE-specific subset (Fig. 6G). MGE peaks that overlapped with cortical peaks were enriched in binding motifs for transcriptional regulators associated with neurogenesis, including LHX2 and SOX10 (Chou and Tole, 2019; Weider and Wegner, 2017). MGE-exclusive peaks showed motif enrichment for TCF12 and ASCL1, which are basic helix-loop-helix (bHLH) transcription factors involved in progenitor proliferation and neuronal cell fate determination (Dennis et al., 2019). We compared the binding sites of PRDM16 in the MGE that we obtained in this study and those of ASCL1 obtained from an in vitro model (Castro et al., 2011) and found 127 overlapping peaks, several of which were associated with genes that control progenitor identity (Table S7). Together, these data suggest that PRDM16 forms MGE-specific complexes with transcription factors known to regulate genetic programs guiding the specification of cortical interneurons (Sandberg et al., 2018). We cross-compared the list of 510 differentially regulated genes from our RNA-Seq experiment (Fig. 5) with the lists of PRDM16 ChIP-Seq peaks in the cortex (Baizabal et al., 2018) and the MGE in order to identify genes that could be potential direct targets of PRDM16. We identified a total of 75 genes that were associated with PRDM16 binding sites; 50 of them contained peaks that were common to the MGE and cortex, 21 contained only MGE-exclusive peaks and 4 displayed peaks unique to the cortex (Fig. 6H). This suggests that PRDM16 exerts its gene regulatory functions through both common and unique sites between the MGE and the cortex.

To understand the function of PRDM16 in vivo, we selected two genes that contained PRDM16 ChIP-Seq peaks: Pdzrn3 and Lmo1. Several ChIP-Seq peaks in the Pdzrn3 genomic locus were common to the MGE and cortex datasets (Fig. 7A). We previously showed that Pdzrn3 repression by PRDM16 controls the migration of pyramidal neurons in the cortex (Baizabal et al., 2018). A ChIP-Seq peak in the Lmo1 genomic locus was exclusive to the MGE (Fig. 7B). Lmo1 is a direct target of Arx, a gene essential for basal forebrain development (Fulp et al., 2008), and has been proposed to play a role in cortical interneuron development (Friocourt and Parnavelas, 2011) but its specific function in this context is still unknown. To understand how PRDM16 binding controls cortical interneuron specification through Pdzrn3 and Lmo1, we compared the expression levels of these genes in the MGE of WT and cKO mice at E13. We directly visualized the spatial pattern of Pdzrn3 and Lmo1 expression using single molecule fluorescent in situ hybridization (RNAscope) on embryonic tissue sections of WT and cKO embryos. We detected RNA puncta representing single mRNA transcripts of either Pdzrn3 (Fig. 7C) or Lmo1 (Fig. 7E). Automated quantification of these puncta in the VZ and subventricular/mantle zone (SVZ/MZ) revealed that there was a significant increase in the expression levels of both genes in cKO animals that was specific to the VZ in both cases (Fig. 7D,F). This suggests that PRDM16 primarily acts in VZ radial glia to repress the expression of genes involved in neuronal differentiation, thus allowing these cells to maintain their proliferative capacity and transition through transit-amplifying stages in order to generate sufficient numbers of cortical interneurons.

Prdm16 is expressed in radial glia in all telencephalic proliferative zones, suggesting that it might play a key role in the specification of most forebrain neuron lineages (Baizabal et al., 2018; Chuikov et al., 2010; Inoue et al., 2017; Shimada et al., 2017). In the cortex, PRDM16 binds to developmental enhancers involved in the specification and migration of upper layer pyramidal neurons, and promotes indirect neurogenic divisions of RG and production of transit-amplifying intermediate progenitors (Baizabal et al., 2018). Here we show that loss of Prdm16 in MGE progenitors leads to a decrease in their proliferation and to a reduction in interneuron numbers in the cortex and hippocampus; however, in contrast to the cortical Prdm16 mutant phenotype, this reduction is not layer- or neuronal subtype-specific (Fig. 1; Fig. S1). The decrease in MGE-derived interneurons is not caused by an increased rate of developmental cell death or to an altered pattern of tangential migration within this population (Fig. S3), but rather to defects in the proliferation of Nkx2.1-expressing progenitors (Fig. 4). We observed both a reduction in the number of dividing progenitors and a decrease in their proliferative capacity, as evidenced by the smaller size of retrovirus-labeled clones in the mutant MGE. Because the retroviral genome is only transmitted to one of the two daughter cells in radial glia divisions (Cepko et al., 1995), and given the fact that the viral genome can be silenced (Halliday and Cepko, 1992), our clone analysis experiments probably underestimate the actual size of clones derived from the infected progenitors. However, this should be the case for both the control and experimental samples, so we believe the interpretation of the data is clear: MGE progenitors have a lower proliferative capacity in the brain of cKO mice. This might be due to a number of reasons, including alterations in their cell cycle dynamics (such as cell cycle length), changes in the division mode of radial glia and/or the presence of different types of progenitors in the mutant MGE (Glickstein et al., 2009; Petros et al., 2015; Pilz et al., 2013; Ross, 2011). The loss of both early- and late-born neuronal lineages from the MGE could reflect key differences in the general role that transit-amplifying progenitors might play in different proliferative regions: in the cortex they have been shown to be the main source of upper-layer pyramidal neurons (Mihalas et al., 2016), but their contribution to the neuronal output of the MGE is still largely unexplored. Different subtypes of transit-amplifying MGE progenitors are biased towards the generation of PV⁺ and SST⁺ cortical interneurons (Petros et al., 2015), but it is unknown whether transit-amplifying cells are required throughout the entire neurogenic period. Based on our results, we propose that the majority of MGE-derived interneurons are generated through a transit-amplifying progenitor; the decreased ability of radial glia to transition into this type of progenitor in Prdm16 mutants is reflected by the uniform loss of interneurons across cortical layers and between both histological subgroups. In the future, it will be important to determine the fate potential and diversity of transit-amplifying progenitors throughout the neurogenic period for GABAergic cortical interneurons (Kelly et al., 2019 preprint).

The vast majority of cortical interneurons are derived from two distinct ventral telencephalic sources, the MGE and CGE (Wonders and Anderson, 2006). In our study, the loss of MGE-derived interneurons in mutant cortices appears to be partially compensated by an increase in the number of interneurons from a CGE-derived population (Fig. 2). This is in line with previous research (Denaxa et al., 2018), suggesting a homeostatic mechanism for setting interneuron numbers in response to network activity within the developing cortex. Our electrophysiological recordings demonstrate that a compensatory increase in the number of CGE-derived interneurons was not enough to restore inhibitory inputs onto cortical pyramidal cells to wild-type levels (Fig. 3). This is probably due to the specific circuit and functional features of reelin⁺ interneurons, which cannot fill in for the inhibitory circuit functions of PV and SST subgroups derived from the MGE (Olah et al., 2007; Pesold et al., 1999). This suggests that maintaining the proper number and diverse complement of cortical interneuron subgroups derived from different developmental sources is necessary to maintain inhibitory balance (Denaxa et al., 2018).

Numerous studies have investigated the roles of different gene expression programs in the specification of cortical interneurons from MGE progenitors (Nord et al., 2015). Our work adds to this growing body of knowledge by identifying a series of genes whose expression is controlled by Prdm16 (Fig. 5; Tables S1, S8). Overall, the functions of these genes support the idea that Prdm16 promotes the generation of cortical interneurons by guiding MGE radial glia through progressively more differentiative divisions, including the production of transit-amplifying progenitors. In the future, it will be important to determine how Prdm16-regulated genes in the MGE control the proliferative capacity and fate potential of different progenitor subtypes to produce the right number and complement of cortical interneuron types.

In both the MGE and the developing cortex, PRDM16 binds to cis-regulatory elements in the genome and prevents premature neuronal differentiation by repressing neuronal maturation genes (Figs 6, 7; Figs S4, S5) (Baizabal et al., 2018). This is consistent with the observation that PRDM16 binds to a largely overlapping set of distal regulatory elements associated with marks of open chromatin in both MGE and cortical progenitors (Fig. 6). We also found a number of overlapping genes containing PRDM16 binding sites in a recent ChIP-Seq dataset obtained from the entire heads of E13 mice (He et al., 2019 preprint) (Table S8). Just as in the cortex, Prdm16 repressor function in the MGE might serve to ensure proper timing of lineage-specific gene expression programs to ensure timely transitions from radial glia into transit-amplifying progenitors, which in turn would be responsible for the generation of sufficient numbers of cortical interneurons.

Although many common aspects of cortical and MGE progenitor function probably require overlapping networks of genes, there are fundamental differences in the migratory capacity and neurotransmitter identity of their neuronal progeny. How might PRDM16 contribute to lineage-specific differentiation programs? In the case of genes that are bound by PRDM16 in both cortical and MGE progenitors, such as Pdzrn3, which encodes an E3 ubiquitin ligase-RING domain containing protein, the molecular function of the encoded protein could be conserved, but the cellular response might be different, eliciting distinct effects on cell fate determination in each proliferative region. This would be consistent with the prominent role that this gene plays in the radial migration of pyramidal cells in the neocortex (Baizabal et al., 2018), although the migratory capacity of MGE-derived interneurons does not appear to be affected in Prdm16 cKO brains (Fig. S3A-C).

It is also very likely that PRDM16 regulates lineage-specific differentiation programs through associations to genomic loci that are exclusive to MGE progenitors (Fig. 6H). It is notable that MGE-specific PRDM16-bound genomic regions are enriched for sequence motifs of bHLH family transcription factors such as ASCL1, which is expressed exclusively in ventral telencephalic neural progenitors and is known to regulate GABAergic neuron specification (Long et al., 2009). We found that PRDM16 associated with genomic loci previously shown to be bound by ASCL1. Although it is important to note that the ASCL1 data was derived from an in vitro model, the significant overlap suggests the potential for in vivo interactions between PRDM16 and ASCL1 in the MGE (Table S7). Many bHLH transcription factors such as OLIG2, ASCL1 and NeuroD family members are expressed exclusively in either ventral or dorsal progenitors (Casarosa et al., 1999; Lu et al., 2000; Osorio et al., 2010). Prdm family members have been shown to interact with bHLH transcription factors (Hohenauer and Moore, 2012; Kinameri et al., 2008; Ross et al., 2012), which are in many cases crucial for proper neuronal production by neural progenitors, both in the dorsal and the ventral telencephalon. It has been proposed that interactions between Prdm family proteins and bHLH transcription factors is a conserved mechanism for the regulation of gene expression during neural development (Hohenauer and Moore, 2012; Ross et al., 2012). Such transcription factor complexes may function in the proliferative zones where they are expressed in order to direct differentiation and lineage progression programs towards region-specific neuronal subtype specification (Lindtner et al., 2019). Our genomic analysis of PRDM16 binding in MGE progenitors sheds light on the lineage-specific transcriptional programs regulated by PRDM16 and suggests that specialized PRDM16-associated protein complexes orchestrate these programs in different progenitor types.

In humans, the PRDM16 gene is located in a distal critical region of chromosome 1 that is deleted in 1p36 deletion syndrome (Jordan et al., 2015). Patients affected by this condition exhibit a spectrum of clinical features that includes intellectual disability, developmental delay and epileptic seizures. It is still unknown to what extent this phenotype is caused by the deletion of PRDM16, but it is likely that distinct mechanisms underlie each of its different clinical aspects. Our study begins to unravel the important question of how PRDM16 controls the fate of progenitors by regulating specific transcriptional programs necessary for neuronal fate determination and excitatory and inhibitory circuit development.

All animal procedures conducted in this study followed experimental protocols approved by the Institutional Animal Care and Use Committee of Harvard Medical School. Mouse lines are listed in Table S9. Mouse housing and husbandry conditions were performed in accordance with the standards of the Harvard Medical School Center of Comparative Medicine. Mice were group housed in a 12 h light/dark cycle, with access to food and water ad libitum. Samples were obtained from animals at embryonic days E13, E14 and E17 and postnatal days P0, P7, P14, P21 and P30, as indicated in the figure legends. All results reported in adult (P30) and late postnatal stages (P14, P21) include animals of both sexes; the sex of embryos and animals at early postnatal stages was not determined.

Postnatal animals were transcardially perfused with PBS followed by 4% paraformaldehyde (PFA) in 120 mM phosphate buffer; their brains were dissected out and post-fixed in 4% PFA for 2-4 h at room temperature or overnight at 4°C. Brains were sectioned into 75-100 µm sections on a vibratome (Leica Microsystems VT1200S) and either further processed for immunofluorescence staining or stored at 4°C in PBS containing 0.05% sodium azide. Embryonic brains were dissected out in ice-cold PBS and fixed in 4% PFA for 2-4 h at room temperature or overnight at 4°C. Their brains were cryoprotected in 30% sucrose in PBS overnight at 4°C, embedded in OCT compound (Sakura), frozen and stored at −20°C. Samples were sectioned at 20 µm on a cryostat (Thermo Fisher CryoStar NX70); sections were either stored at −20°C or further processed for immunofluorescence staining. For fresh frozen samples (used for fluorescent in situ hybridization; FISH), the brains were extracted in PBS and immediately embedded in OCT compound, then frozen on dry ice.

Samples were permeabilized with 0.5% Triton X-100 (Amresco) in PBS for 1-2 h and blocked in blocking buffer (10% goat serum, 0.1% Triton X-100 in PBS) for 1-2 h at room temperature. The sections were then incubated for 24-72 h with primary antibodies diluted in blocking buffer at 4°C. The samples were washed three times (10-30 min/wash) with PBS, counterstained with DAPI (4′,6-diamidino-phenylindole; Invitrogen) for 45 min (both steps at room temperature), and incubated with secondary antibodies diluted in blocking buffer for 2 h at room temperature or overnight at 4°C. Samples were then washed (three 10-30 min washes) and mounted on slides with ProLong Gold Antifade Mountant (Invitrogen).

Slides were allowed to reach room temperature and then washed three times with PBS. Sections were permeabilized with 0.5% Triton X-100 in PBS for 30 min and blocked with blocking buffer for 1 h at room temperature. Slides were incubated with primary antibodies diluted in blocking buffer overnight, in a humid chamber at 4°C. They were then washed with PBS (three 10-30 min washes), counterstained with DAPI (45 min), and incubated for 1-2 h with secondary antibodies diluted in blocking buffer, at room temperature. Slides were washed (three 10-30 min washes) with PBS and mounted with ProLong Gold Antifade Mountant.

Images were acquired using either a Leica DM5500B wide field microscope or a Leica SP8 laser point scanning confocal microscope. Objectives of 10×, 20× and 25× magnification were used, and the parameters of image acquisition (speed, resolution, averaging, zoom, z-stack, etc.) were adjusted for each set of samples. Images were further analyzed using ImageJ, both in its native and Fiji distributions, as described below. Brightness and contrast were adjusted as necessary for visualization, but the source images were kept unmodified.

Embryonic samples from medial regions of the neocortex, immunostained for tdT and counterstained with DAPI, were used. The length of the cortex was measured from the pallial/subpallial boundary (PSB) between the lateral ganglionic eminence (LGE) and the developing cortex to the start of the cortical hem, through the center of the tissue [set as the basal end of the SVZ at E13, and the SVZ-intermediate zone (IZ) boundary at E15]. The position of the cell soma closest to the cortical hem was taken as the furthest point of migration; its distance to the PSB was measured, divided by the total length of the cortex and presented as a percentage. Where available, this measurement was repeated across both hemispheres of each sample.

Mice (19-21 days old) were anesthetized by isoflurane inhalation and perfused transcardially with ice-cold artificial cerebrospinal fluid (ACSF) containing (in mM): 125 NaCl, 2.5 KCl, 25 NaHCO₃, 2 CaCl₂, 1 MgCl₂, 1.25 NaH₂PO₄ plus 25 glucose and 310 mOsm/kg. Cerebral hemispheres were removed and sliced in cold ACSF (300 µm coronal slices on a Leica VT1200S vibratome). Coronal slices containing somatosensory cortex were recovered for 15-20 min at 34°C in choline-based recovery solution (in mM): 110 choline chloride, 25 NaHCO₃, 2.5 KCl, 7 MgCl₂, 0.5 CaCl₂, 1.25 NaH₂PO₄, 25 glucose, 11.6 ascorbic acid and 3.1 pyruvic acid, and then transferred to a holding chamber with 34°C ACSF that progressively cooled down to room temperature (20-22°C). All recordings were obtained within 1-6 h after slicing and solutions were constantly bubbled with 95% O₂ and 5% CO₂. Individual slices were transferred to a recording chamber mounted on an upright microscope (Olympus BX51WI) and continuously perfused (1-2 ml/min) with ACSF at room temperature. Cells were visualized using a 40× water-immersion objective with infrared DIC optics. Whole-cell voltage-clamp recordings (room temperature) were made from pyramidal cells in L2/3 somatosensory cortex. Patch pipettes (2-4 MΩ) pulled from borosilicate glass (BF150-86-7.5, Sutter Instruments) were filled with a Cs⁺-based internal solution containing (in mM) 130 CsMeSO₄, 10 HEPES, 1.8 MgCl₂, 4 Na₂ATP, 0.3 NaGTP, 8 Na₂-phosphocreatine, 10 CsCl₂ and 3.3 QX-314 (Cl⁻ salt), pH 7.3 adjusted with CsOH, plus 295 mOsm/kg. For all voltage-clamp experiments, errors due to voltage drop across the series resistance (<20 MOhm) were left uncompensated. To isolate mIPSCs, cells were held at −0 mV and ACSF included 20 μM NBQX (Tocris), 10 μM (R)-CPP (Tocris) and 1 μM Tetrodotoxin (Sigma). Membrane currents were amplified and low-pass filtered at 3 kHz using a Multiclamp 700B amplifier (Molecular Devices), digitized at 10 kHz and acquired using National Instruments acquisition boards and a custom version of ScanImage written in MATLAB (Mathworks). Off-line analysis of mIPSC frequency was performed using custom routines written in MATLAB and Igor Pro (Wavemetrics). Statistical analyses were done in GraphPad Prism 7 software (GraphPad).

The retrovirus packaging cell line HEK293 gp NIT-GFP was grown to 90% confluence in DMEM supplemented with 10% fetal bovine serum. Cells were transfected with a pCMV-VSV-G vector using Lipofectamin 2000 (Invitrogen) and Optimem (Gibco). Two days after transfection, the cell supernatant was collected, filtered through a 0.45 µm filter (VWR International) and centrifuged at 25,000 rpm (∼80,000 g) for 90 min at 4°C. After discarding the supernatant, 100 µl of cold PBS+Ca²⁺ was added to the pellet, which was incubated overnight at 4°C. Viral particles were then resuspended, aliquoted and stored at −80°C. The titer of the viral preparation was determined to be 10⁶-10⁷ cfu. Timed pregnant mice were anesthetized using an isoflurane vaporizer and placed on a warming pad. An abdominal incision (approximately 2 cm wide) was made, and the uterine horns were exposed on top of a sterile gauze pad. The uterus was periodically moistened with sterile PBS prewarmed at 37°C during the entire procedure. Approximately 0.5-1 µl of retrovirus, mixed with 0.05% Fast Green for visualization, was injected into the lateral ventricles of each embryo using heat-pulled beveled glass micropipettes (Drummond). The abdominal cavity was sutured and stapled before administering buprenorphine (0.05-0.1 mg/kg) and ketoprofen (5-10 mg/kg). The pregnant dams were allowed to recover in a 35°C chamber for 1-2 h after surgery, then returned to usual housing conditions and sacrificed 24 h after surgery; the embryos were retrieved and processed as explained above.

The MGEs of E14 embryos were manually dissected out and flash-frozen with liquid nitrogen. After confirming the genotype of each embryo, three samples (i.e. six MGEs) were pooled for each experimental replicate (three WT and three KO pools). Tissue digestion, RNA extraction and sequencing and bioinformatic analyses were conducted by GENEWIZ (South Plainfield, NJ). After RNA extraction, sequencing reads (42-71 million reads/sample) were obtained in an Illumina HiSeq sequencer. Reads were trimmed to remove possible adapter sequences and nucleotides with poor quality using Trimmomatic v.0.36. The trimmed reads were mapped to the Mus musculus GRCm38 reference genome available on ENSEMBL using the STAR aligner v.2.5.2b. Unique gene hit counts were calculated using featureCounts from the Subread package v.1.5.2. The hit counts were summarized and reported using the gene_id feature in the annotation file. Only unique reads that fell within exon regions were counted. After extraction of gene hit counts, the gene hit counts table was used for downstream differential expression analysis. Using DESeq2, a comparison of gene expression between the defined groups of samples (WT versus KO) was performed. The Wald test was used to generate P-values and log₂ fold changes. Genes with an adjusted P-value <0.05 and absolute log₂ fold change >1 were considered differentially expressed genes for each comparison.

The MGE of 18 embryos from two litters of E14 WT mice were dissected out and finely minced. Cells were dual crosslinked by incubating in 1.5 mM EGS (ethylene glycol bis[succinimidyl succinate]) solution for 20 min with rotation, and then with 1% PFA and 1.5 mM EGS for an additional 10 min (both steps were performed at room temperature). Crosslinking was quenched by adding glycine to a final concentration of 125 mM and rotating for 5 min at room temperature. Cells were then washed twice with cold PBS containing EDTA-free protease inhibitor, centrifuged and stored at −80°C or freshly resuspended in lysis buffer (20 mM Tris-HCl pH 8.0, 85 mM KCl, 0.5% NP40) and incubated on ice for 30 min. Nuclei were pelleted by centrifugation at 1500 g for 5 min, resuspended in SDS buffer (0.2% SDS, 20 mM Tris-HCl pH 8.0, 1 mM EDTA) and incubated on ice for 10 min. Nuclei were then sonicated using a Covaris S2 ultrasonicator for shearing chromatin into fragments with a size range of 100-500 bp. After spinning chromatin at 18,000 g for 10 min, supernatant was transferred to a clean tube and one volume of 2× ChIP dilution buffer (0.1% sodium deoxycholate, 2% Triton X-100, 2 mM EDTA, 30 mM Tris-HCl pH 8.0, 300 mM NaCl) was added. At this step, a volume of supernatant containing around 0.5 million nuclei was set aside as input control and the remaining supernatant was incubated with 5 mg of anti-PRDM16 antibody overnight at 4°C with rotation. Next day, 50 ml of washed protein G beads (22.5 mg/ml; Novex) were added to the chromatin solution and incubated for 2 h at 4°C. After incubation, beads were washed twice with low salt wash buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8.0, 150 mM NaCl) followed by two washes with high salt wash buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8.0, 500 mM NaCl) then two washes with LiCl wash buffer (0.25 M LiCl, 0.5% NP40, 0.5% sodium deoxycholate, 1 mM EDTA, 10 mM Tris-HCl pH 8.0) and finally two washes with TE pH 8.0 (10 mM Tris-HCl, 1 mM EDTA). Beads were then resuspended in 90 ml of freshly prepared ChIP elution buffer (1% SDS, 0.1 M NaHCO₃) and incubated at 65°C for 30 min with rotation. The recovered supernatant was incubated in reverse crosslinking solution (250 mM Tris-HCl pH 6.5, 62.5 mM EDTA pH 8.0, 1.25 M NaCl, 5 mg/ml Proteinase K) at 65°C overnight. DNA was then extracted with phenol/chloroform/isoamyl alcohol, precipitated with 3 M sodium acetate pH 5.0 and resuspended in TE pH 8.0 low EDTA (10 mM Tris-HCl, 0.1 mM EDTA). Finally, samples were treated with RNase A (100 mg/ml) for 30 min at 37°C. For library preparation, genomic DNA was purified, end repaired, ligated with barcoded adapters, amplified for 14 PCR cycles and purified using the Ovation Ultralow System V2 (NuGEN) according to the manufacturer's instructions. Library fragments in the range of 100-800 bp were size-selected using agarose gel electrophoresis followed by DNA gel extraction (Qiagen). Recovered DNA was further cleaned and concentrated using a column (Zymo Research). Libraries were sequenced in an Illumina HiSeq 2500 sequencer to a sequencing depth of 30-40 million reads per sample.

Reference genomic sequence and annotations for the mouse genome (mm10) were obtained from the University of California Santa Cruz Genomics Institute (Casper et al., 2018). ChIP-Seq reads generated from PRDM16 ChIP and input libraries were aligned to the mouse genome using bowtie2 (Langmead and Salzberg, 2012) with default parameters except for ‘-I 50 -X 750’. The read densities were calculated with the Gaussian kernel (bandwidth of 35 bp) using the R package SPP (Kharchenko et al., 2008). Peaks were called using MACS2 (version 2.1.2) (Zhang et al., 2008) with ‘callpeak’ and a P-value threshold of 0.05 followed by irreproducible discovery rate (IDR) analysis, as described by Landt et al. (2012), with a threshold of ≤0.05. The MGE peak set was compared to the previously published cortical PRDM16 ChIP experiment (Baizabal et al., 2018), with a merged set of peaks defined by the union of the MGE and cortical peak sets. Peaks were considered to be present in both cortical and MGE samples if the ChIP/input ratio of library-size normalized read counts within the peak region was ≥3 for both experiments. Peaks were considered ‘cortical only’ (‘MGE only’) if the ChIP/input ratio was ≥3 for the cortical (MGE) experiment and ≤1.5 for the MGE (cortical) experiment. Gene ontology enrichment analysis was performed using the R package ClusterProfiler (Yu et al., 2012) with the parameters ont=‘BP’ and maxGSSize=20,000.

Motif analysis was performed on set peaks detected in the MGE using HOMER (Heinz et al., 2010). De novo motif analysis was conducted by parsing input sequences at ±100 bp from peak center. The FDR-corrected probability of that motif being over-represented amongst target sequences is presented as a q-value. Known motif enrichment was screened against the JASPAR database of previously determined high quality motifs.

Fresh frozen or PFA-fixed samples (see above) were submitted to the RNAscope protocol (Advanced Cell Diagnostics), following the manufacturer's instructions. RNAscope probes against each transcript (see Table S9) were purchased from ACD. In each experiment, probes against the candidate gene were used in combination with a tdTomato probe to confirm that all areas analyzed were within the portion of the MGE where Nkx2.1-Cre-mediated recombination had taken place.

The CellCounter tool in ImageJ or Fiji was used for all cell quantifications. In the mature cortex (Fig. 1C-J, Fig. 2A-J), 1 mm-wide sections were selected from images obtained at the level of the somatosensory cortex. All cells positive for the markers analyzed in each case were quantified within each cortical layer, as identified by the distribution of DAPI-counterstained nuclei. In the hippocampus (Fig. S1A-H), the entire structure was analyzed at equivalent rostro-caudal levels; layers and areas were determined by the distribution of DAPI-counterstained nuclei. In the striatum and nucleus accumbens (Fig. S1I,J), 500×500 µm squares were drawn in the same regions within these structures, at equivalent rostro-caudal levels, and numbers were normalized by square millimeter. In the developing cortex at different stages (Fig. 3A), the entire surface of the neocortex was measured on both hemispheres of each sample at equivalent rostro-caudal levels, and the number of cells positive for CC3 and tdT was quantified within the measured surface, then normalized by square millimeter. In E13 MGE samples (Fig. 3B), the entire surface of the MGE in sections at equivalent rostro-caudal locations was measured; cells positive for pH3 were quantified within the measured area and normalized by square millimeter. For the quantification of MGE clones (Fig. 3D-F), cells were considered as part of the same clone if they had similar fluorescence intensities and were located within 50 µm of the radial fiber in the case of RG-containing clones or within a 50 µm radius in the case of IP-containing clones. Clones were not analyzed if other GFP-labeled cells of different fluorescence intensity were present in their vicinity (within 100 µm of the radial fiber or within a 100 µm radius). Quantification of RNA puncta per area was performed using an automated data processing pipeline in MatLab, guided by MatBots (https://hms-idac.github.io/MatBots); each data point corresponds to the average values from the analysis of three fields (dimensions: 100×100 µm or 50×50 µm) selected within the VZ and SVZ/MZ of two sections or MGEs per sample (where available; in some cases, only one section could be analyzed). Cell and RNA puncta numbers were compiled in Microsoft Excel spreadsheets; GraphPad Prism 8 (GraphPad) was used to build graphs.

All statistical analyses were performed with GraphPad Prism 8, as detailed in the figure legends. All P-values were rounded to four decimal places, and are presented above each statistical comparison in the corresponding figures; those highlighted in bold are below 0.05, which was considered the cut-off for statistical significance (P-values deemed not statistically significant under this criterion are displayed in parenthesis above the corresponding comparisons in the figures).

The authors thank Patrick Seale and Jeremy Dasen for their generous gift of anti-PRDM16 antibodies; Bernardo Sabatini for providing equipment and resources for the electrophysiology experiments; Mahmoud El-Rifai and Aurélien Begué at the Neurobiology Imaging Facility at Harvard Medical School for performing RNAscope experiments; Steve Vu for technical support; all members of the Harwell, Goodrich, Lehtinen and Segal labs for discussions and comments; and A. Denise Garcia, Rebecca Ihrie and Debby Silver for their comments on the manuscript.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.187526.reviewer-comments.pdf