Distinct types of neurons in the neocortex are all derived from neural progenitor cells. Hence, the number of neural progenitor cells is critical to determine the size of the brain. The development of the neocortex begins with the generation and expansion of neural progenitor cells. In mammals, at least three types of neural progenitor cells are involved in the development of the neocortex: neuroepithelial cells (NECs), RGs, and intermediate progenitor cells (IPCs). NECs are the earliest type of neural progenitor cells that are highly polarized in a pseudostratified pattern (His, 1889; Ramon y Cajal and Azoulay, 1955). Because NECs are believed to generate all other types of cells in the neocortex, the size of the NEC pool is crucial to determine the numbers of progenitor cells and even the final number of cortical neurons (Malatesta et al., 2000; Noctor et al., 2001, 2002). To amplify their pool, NECs keep dividing symmetrically and exponentially before the onset of neurogenesis. NECs gradually transform into RGs for the onset of neurogenesis. Although RGs still maintain some NEC characteristics, such as bipolar morphology and apical–basal polarity (Rakic, 1972), they begin to lose tight junctions (Aaku-Saraste et al., 1996) and express specific RG proteins (Levitt and Rakic, 1980), such as glutamate/aspartate transporter (Shibata et al., 1997) and brain lipid-binding proteins (Feng et al., 1994). Although RGs could symmetrically divide to expand its pool as NECs, they can undergo asymmetrical division to produce neurons. In addition to their self-renewal and differentiation functions, the radial fiber of RGs guides neuronal migration. In this process, the overexpression of cleaved NICD fragments promote progenitor cells to express RG-specific markers (Gaiano et al., 2000). No differences in the number of NECs in the neural tube were observed between Hes1/5 double-knockout mice and control mice at the NEC stage E8.5, whereas the number of RGs decreased due to prematuration at later stages (E9.5–10.5) when NECs begin to transform to RGs (Hatakeyama et al., 2004). These studies suggest that the transition of NECs to RGs is dependent on Notch signaling, whereas the formation and expansion of NECs is independent of Notch signaling.

Neurogenesis from RGs to neurons can occur in a direct or an indirect manner. Direct neurogenesis is one RG divides to generate an RG and a neuron in the ventricular zone (VZ), and indirect neurogenesis is one RG may generate two RGs or two other types of progenitor cells, such as IPCs. Subsequently, IPCs symmetrically divide to generate two neurons. Indirect neurogenesis is beneficial for the increase in the final neuron pool and is more common in the mammalian neocortex compared with direct neurogenesis, which is the predominant neurogenesis manner in the developing cortex of other vertebrates, such as birds and reptiles (Englund et al., 2005; Guillemot, 2005; Cárdenas et al., 2018).

Prematuration is observed in animal models with a Notch signaling deficiency. The aforementioned studies have indicated that defects in the activation of Notch signaling inhibited the transition from NECs to RGs. Because Mind bomb 1 (Mib1), a RING-type E3 ubiquitin ligase, promotes the endocytosis of canonical Notch ligands, knocking out the Mib1 gene can impair Notch signaling. Conventional Mib1 knockout mice exhibited deficient Notch signaling that led to prematuration at E9.0–E9.5, resulting in the death of embryos before E12.5 (Koo et al., 2005). Furthermore, in animal models with Notch signaling deficiency, RGs transformed into IPCs early before differentiating into neurons. In Nestin-driven Mib1 knockout mice, the numbers of IPCs and mitotic cells outside the VZ region were increased at E13.5 (Yoon et al., 2008), resulting in an increase in the number of neurons from E14.0. Those findings suggest that Notch signaling activity is high in RGs but low in IPCs and neurons. To determine the activity of Notch signaling in RGs and IPCs separately, overexpression of NICD together with CBF1-EGFP, a reporter of Notch signaling, was utilized. The results revealed that NICD activated the CBF1-binding site in RGs but not in IPCs. Because NICD cannot activate Notch signaling in IPCs, Hes proteins can be overexpressed as an alternative method to activate Notch signaling. However, the numbers of IPCs decreased when Hes proteins were overexpressed (Mizutani et al., 2007; Ohtsuka and Kageyama, 2021b); this finding is in contrast to that of knockout experiments that indicated the attenuation of Notch activity. However, the reason underlying the inactivation of Notch signaling in IPCs remains to be elucidated. Because IPCs mediate indirect neurogenesis to effectively increase cell numbers and emergence of IPCs is crucial in the evolution of the mammalian neocortex (Cárdenas et al., 2018), the evolution of the mammalian neocortex should be examined by investigating the functional roles and molecular mechanisms of IPCs.

In gyrencephalic species, such as ferret and primates, a large population of proliferative cells can be noted in the basal region of the VZ. They are a subtype of RGs, called basal RGs (bRGs). These bRGs, unlike IPCs, have radial fibers but lose the apical attachment to the ventricular surface, unlike their apical cohorts, apical RGs (aRGs). bRGs can undergo self-renewal to expand the progenitor pool in the SVZ region. In the developing primate neocortex, the majority of bRGs are positioned in the outer SVZ (OSVZ), which is separated from the inner SVZ (ISVZ) by an inner fiber layer. During neocortical expansion, the thickness of the OSVZ gradually increases with the expansion of the VZ (Rakic, 1974; Smart et al., 2002; Lukaszewicz et al., 2005; Lui et al., 2011). Except for the similarity in morphological characteristics between bRG and aRGs, bRGs express some aRG genes, such as SOX2, PAX6, nestin, and GFAP, and undergo a Notch signaling–dependent pathway to self-renew or generate IPCs in the OSVZ (Fietz et al., 2010; Hansen et al., 2010). The induction of radial glial fiber divergence in the superficial neocortex by a large number of bRGs produced through the basal process combined with neuronal migration along the newly formed fibers can cause lateral dispersion and promote cortical folding in gyrencephalic species (Reillo et al., 2011; Gertz and Kriegstein, 2015; Llinares-Benadero and Borrell, 2019). Moreover, because of the abundant generation of bRGs and their daughter cells, the OSVZ was determined to be the predominant neurogenic zone at the mid-gestational stage that caused marked cortical neuronal expansion and an increase in brain size in humans, thus leading to the evolution of the cerebral cortex (Hansen et al., 2010; Lui et al., 2011; Llinares-Benadero and Borrell, 2019).

In the last decade, a group led by Professor Ryoichiro Kageyama in Japan published a series of discoveries describing several components in the Notch signaling pathway are expressed in a dynamic pattern called oscillation, which has been reported earlier and is essential in somitogenesis (Palmeirim et al., 1997). They found that the oscillation of Hes1 can maintain the pool of neural progenitor cells. Concurrently, the expression of Notch ligand Dll1 and the proneural gene Ngn2 were fluctuated in a manner which is coordinated but opposite to the oscillated expression pattern of Hes1. The fine balance of the oscillating gene expression pattern is orchestrated by several elaborate transcriptional regulatory mechanisms. The oscillating pattern of Hes1 expression can be regulated through a negative feedback loop. After the activation of Hes1 by the Notch ligand–receptor interaction, Hes1 protein cis-represses its own transcription by directly targeting its promoter. Another key is the short half-life of Hes1 mRNA and Hes1 protein. The half-life of Hes1 mRNA and Hes1 protein is as short as 20 min. As both Hes1 mRNA and Hes1 protein are degraded soon after their production, the Hes1 promoter can be released from autoinhibition. Also Hes protein represses proneural genes such as Mash1 (Chen et al., 1997) and the expression of Dll1 is directly regulated by Ngn2 and Mash1 through the regulation of enhancer regulatory elements (Castro et al., 2006), the oscillated pattern of Dll1 and proneuronal genes Ngn2 and Mash1 are similar to and follow that of Hes1 (Shimojo et al., 2008; Imayoshi et al., 2013). Nonetheless, the oscillating Ngn2 expression remains to be validated because previous findings have indicated that most cells, if not all, of Neurogenin2 CreER and R26R-CAG-loxPstop-EGFP mice had left the progenitor pool at 12 h after tamoxifen administration (Miyoshi and Fishell, 2012). Thus, the oscillating Dll1 expression pattern should be the most critical event in orchestrating Hes1 expression and Mash1 may be the upstream activator of Dll1 instead of Ngn2 (Imayoshi et al., 2013; Sueda et al., 2019). Interestingly, while the Hes genes oscillated in multiple tissues across species, the frequency varies. For instance, during somitogenesis when the oscillated Her/Hes expression regulated the formation of new somite, the frequency differs in different species: 30 min in zebrafish, 90 min in chick, 2 h in mouse (Cinquin, 2007), and 4–6 h in humans (Turnpenny et al., 2007; Kageyama et al., 2012; Hubaud and Pourquié, 2014; Matsuda et al., 2020). The period of Hes1 oscillation in mouse neural progenitor cells and fibroblasts is 2 h. However, the period is 3–5 h in mES cells (Kobayashi et al., 2009; Kobayashi and Kageyama, 2011), suggesting that the period may vary among cell types as well as the regulatory machinery. If the oscillation of Hes1 can maintain the pool of neural progenitor cells, the neuronal production step in neurogenesis indicates the escape of the oscillation cycle. Hence, neuronal differentiation can be induced by the sustained Ngn2 expression in the replacement of oscillatory Ngn2 expression (Shimojo et al., 2008). However, in this scheme, how Ngn2 and Dll1 expression escape the negative feedback loop controlled by Hes1 and changes from the oscillatory pattern to a sustained high expression pattern remain unclear.

The oscillatory Hes1 expression can be used to maintain neural progenitors in the cell cycle, whereas sustained Hes1 expression promotes cells to stay in a quiescent state (Sang et al., 2008; Sueda et al., 2019) that may contribute to boundary formation such as the boundary between the dorsal and ventral telencephalon (Baek et al., 2006). The sustained overexpression of Hes1 in mouse neural progenitor cells at E13.5 reduced the expression of Notch ligands (Dll1 and Jag1), proneural genes (Mash1 and Ngn2), and cell cycle regulators (cyclin D1 and cyclin E1) (Shimojo et al., 2008; Sueda et al., 2019). This result suggested that the sustained overexpression of Hes1 repressed both proliferation and differentiation. Thus, cells in the boundaries of the brain were not able to proliferate or differentiate. In Hes1-overexpressing transgenic mice, Pax6+/Hes1+ neural progenitor cells were maintained for a long time in the VZ even after birth. Nonetheless, compared with control mice, Hes1-overexpressing mice exhibited the suppressed proliferation of abnormal neural progenitor cells and a markedly elongated cell cycle length; this finding is in agreement with the previous study indicating that the sustained overexpression of Hes1 reduced the expression of cell cycle–related proteins such as cyclin D1 (Shimojo et al., 2008). Further investigation using transgenic mice to engineered wild-type Hes1 gene into the shortened or elongated form found both amplitude and frequency of oscillated Hes1 expression were impaired which resulted in neural prematuration and reduced brain size (Ochi et al., 2020) similar to the phenotype induced by engineered Dll1 gene (Shimojo et al., 2016). Notably, the shortened or elongated form of Dll1 gene would cause the deficiency in both neural development and somite formation. In addition to manipulating the pattern of oscillation, the basal level of Hes1 expression is also critical to its biological functions. Contrary to the mutant Hes1 mice expressing reduced as well as sustained levels of Hes1, overexpression of Hes1 prevented neural progenitor cells from self-renewal and differentiation, thus leading to a smaller brain size, a thinner cerebral cortex, the enlarged ventricles in Hes1-overexpressing mice and an apparent increase in the number of neural progenitor cells even in the late corticogenesis (Ohtsuka and Kageyama, 2021b). However, another interpretation has been raised by Borrell et al. (2012) proposing that Hes1 expression is crucial to maintaining the progenitor cell pool in the VZ by overexpressing Hes1 cDNA or downregulating Hes1 expression by using the RNA interference (RNAi) technique. Another study showed that activation of Notch signaling maintains the neural progenitor cell pool by overexpressing the NICD fragment (Mizutani et al., 2007). Thus, whether maintaining the neural progenitor pool is controlled by simply activation of Notch signaling or in the combination of the oscillated Hes1 expression remains to be clarified.

Robo signaling is a widely known pathway involved in neural development. Robos and Slits (ligands of Robo receptors) are responsible for regulating axon guidance which contributes to cortical circuits (Brose et al., 1999; Dickson and Gilestro, 2006). Moreover, Robo–slit signaling regulates neurogenesis in the central nervous system (CNS) of drosophila (Mehta and Bhat, 2001) and mice (Andrews et al., 2008). In the neocortex of Robo1/2 knockout mouse, neural progenitor cells in the VZ underwent a premature asymmetric division and increased the generation of IPCs, thus reducing the brain size. This deficiency was found to be mediated by Robo-mediated transcriptional activation of the Notch effector Hes1, which suggested the interplay between Robo and Notch signaling is crucial to regulate neurogenesis precisely (Borrell et al., 2012).

CNS evolution across species has been investigated for decades; however, it still remains largely unclear. The differential regulation of direct and indirect neurogenesis in different species is one of the most prominent hypotheses. Recently, a study examined the switch between Dll1–Notch and Robo–Slit signaling in corticogenesis to determine the predominant mode of indirect or direct neurogenesis along with its effects on the neuron number, brain size, and neural circuit complexity across amniotes. To elucidate the involvement of Notch–Dll1 and Robo–Slit signaling, the expression level in the neural progenitors of each representative species among diverse amniotes (snake, chick, mouse, and human) was analyzed. Data indicated a high Robo expression level and a low Dll1 expression level in brain structures including the dorsal telencephalon of snake, the medial dorsal telencephalon of chick, and the olfactory bulb, hippocampus, and spinal cord of mammals, but a high Dll1 expression level and a low Robo expression level in the advanced brain structures including the lateral dorsal telencephalon of chick and the neocortex of mammals (Cárdenas et al., 2018; Cárdenas and Borrell, 2020). In brief, Robo expression declined in the evolutionary process, whereas Dll1 expression increased during the evolution of amniotes. Furthermore, the combined gain-of-function of Dll1 and loss-of-function of Robo in the evolutionarily old region of the telencephalon in mouse, chick, and snake indicated the promotion of indirect neurogenesis. By contrast, the combined gain-of-function of Robo and loss-of-function of Dll1 in the evolutionarily young region of the telencephalon including the mouse neocortex and human cerebral organoids indicated the promotion of direct neurogenesis. This observation is correlated to the switch between direct and indirect neurogenesis. Progenitors in the snake dorsal cortex exhibit mostly direct neurogenesis with no indirect neurogenesis, as indicated by the absence of IPCs. By contrast, progenitors in the mammalian neocortex exhibit indirect neurogenesis most frequently. These findings were further confirmed in human organoids, indicating that the Robo–Dll reciprocal expression–based balance of direct/indirect neurogenesis is the key factor for evolution among amniotes (Cárdenas et al., 2018; Cárdenas and Borrell, 2020).

Sonic hedgehog (Shh) is a secreted protein encoded by Shh gene. Initially, hedgehog gene was identified from Drosophila melanogaster. Mutations in hedgehog gene lead to abnormal segmental patterning and polarity in flies (Nüsslein-Volhard and Wieschaus, 1980; Mohler, 1988). Shh signaling is essential for embryonic development in two stages. In the early stage, Shh is secreted from the notochord, located ventrally to the neural tube, and controls the neural axis by creating a concentration gradient (Echelard et al., 1993; Roelink et al., 1995). In the later stage, Shh regulates cell proliferation and differentiation during brain development by controlling cell cycle kinetics in various tissues and species such as the mouse neocortex (Bertrand and Dahmane, 2006; Komada et al., 2008; Komada, 2012) and chick spinal cord (Saade et al., 2013). Shh is essential to the development of IPCs (Shikata et al., 2011). Mutations in human SHH gene cause holoprosencephaly (HPE), which is an autosomal dominantly inherited disorder. Patients with HPE have intellectual disability, microcephaly, and epilepsy (Tekendo-Ngongang et al., 1993; Belloni et al., 1996; Roessler et al., 1996). Shh protein initiates signaling by binding to the transmembrane receptor Pathed (Ptch), which inhibits Smoothened (Smo) in the absence of Shh (Murone et al., 1999). When Smo is de-repressed, it causes Gli1-3 to move to the nucleus, thus inducing downstream gene expression (Wickström et al., 2013).

Ohtsuka and Kageyama (2021b) reported that the sustained overexpression of Hes1 in mice retained abnormal neural progenitors with both Pax6 and Hes1 expression in the VZ even after birth but still accompanied by smaller brains, thinner cerebral cortices, and enlarged ventricles due to defects in proliferation and neurogenesis. Later, Ohtsuka and Kageyama (2021a) observed that Hes1-overexpressing mice could be rescued from their defects by crossing them with transgenic mice expressing constitutively active Smo, an effector of Shh signaling. This result suggested that dysfunction in Notch signaling can be complemented by promoting Shh signaling (Ohtsuka and Kageyama, 2021a). However, as both the pathways are crucial during embryonic development, detailed molecular mechanisms through which they work together in parallel or complement remain to be elucidated.

The cortex of most of the reptiles such as alligators, geckos, and turtles shows a mixed pattern of the layered structure in the dorsal pallium dorsal to the ventricles and nuclear structures in the dorsal ventricular ridge ventral to the ventricles (Goffinet et al., 1986; Suzuki and Hirata, 2014; Briscoe and Ragsdale, 2018a,b; Nomura et al., 2020). In addition to different cytoarchitecture, neurons in the layered structure of reptiles migrate and integrate into the cortex roughly through an outside–in migration pattern (Suzuki and Hirata, 2014; Luzzati, 2015; Tosches et al., 2018; Nomura et al., 2020), opposite to the inside–out migration pattern in the developing mammalian neocortex. The pallium of birds is composed of four major subdivisions: hyperpallium, mesopallium, nidopallium, and arcopallium (Jarvis, 2009). The pallium in birds and some reptiles have a nuclear-type structure, in which neuronal cell bodies aggregate instead of layered laminated structures such as the neocortex in mammals. On the basis of trajectory tracing and in situ hybridization analyses, recent studies have identified that neurons with similar functions and molecular expression across the species have a nuclear or laminar structure, regardless of different cytoarchitectures (Zeier and Karten, 1971; Karten and Shimizu, 1989; Dugas-Ford et al., 2012; Suzuki et al., 2012; Chen et al., 2013). For example, neurons in the L2 field of the cortex in zebra finches receive signals from the thalamus and express genes such as Rorβ, similar to layer IV sensory neurons in the mammalian neocortex; neurons in the mesopallium and nidopallium and neurons in the arcopallium exhibit conserved projections and molecular expression similar to layer II–III and layer V–VI neurons in the mammalian neocortex, respectively (Chen et al., 2013). Considering conserved functions and the phylogenic tree, the laminated structure should be evolved from the nuclear type. A nuclear-to-layered hypothesis proposed by Karten indicated that the laminated pallium of the mammalian neocortex might be transformed from the nuclear type pallium in birds or reptiles (Karten, 1991).

Comparative analysis of neuroanatomical structures, gene expression profiles, and neural circuits is a common approach used to study pallium formation in teleosts (Wullimann and Mueller, 2004; Wullimann, 2009). The structure of the pallium shows distinct morphological features in different teleosts. However, mechanisms underlying the development of the teleost cortex and the gene expression profiles of neuronal connections remain largely unknown. The teleosts are close to land vertebrates such as amphibians and reptiles in evolution, and can be divided into ray-finned fishes and lobe-finned fishes. A ray-finned fish, called zebrafish, is the most common animal model used to study embryonic development, diseases, and neurological behaviors. Accumulating results of in situ hybridization, immunostaining, and neural circuit tracing indicate that molecular profiles and presumptive functions in the pallium and subpallium of the teleostean cortex are similar to those of other vertebrates. For example, Emx genes are enriched in the pallium and Dlx genes are enriched in the subpallium across species (Wullimann and Mueller, 2004; Wullimann, 2009). These genetic studies suggested that pallium formation from the neural tube in ray-finned fishes follows a special method called “eversion,” in which the neural tube bends outward to form two cerebral hemispheres, separated by an unpaired ventricle and covered with a thin roof plate. In contrast to ray-finned fishes, the pallium of other vertebrates, such as lobe-finned fishes, amphibians, reptiles, birds, and mammals, is generated during an evagination process, in which the roof of the neural tube is sunken down to separate two lateral ventricles (Huesa et al., 2009; Yamamoto et al., 2017). The two prominent differences between these processes in terms of morphological changes are the inverted mediolateral axis in the pallium and the position of ventricles. Because the lumen surface of the neural tube is critical to generating neural progenitor cells, changing the position of ventricles may cause alterations in the direction of the neuronal migration and orientation of neural fibers (Huesa et al., 2009; Wullimann, 2009; Yamamoto et al., 2017).

In fishes and mammals, considerable changes have been observed in the neuron number, pallium cytoarchitecture, and neural circuit complexity. Because components in Notch signaling are highly conserved, the activity of Notch signaling may be widely involved in multiple developmental events in the formation of the pallium such as the maintenance of neural and cortical progenitor cell pools, transition from aRGs to IPCs, and corticogenesis. Thus, we speculate that the evolutionary divergence in pallium formation may result from the dominant isoform switch, the distinct regulation mechanism, or the emergence of novel genes, which will be discussed in the following sections.

In zebrafish, Notch ligands in the Delta family include DeltaA, DeltaB, DeltaD, and DeltaC, whereas Delta-like ligands include Dll1, Dll3, and Dll4 in mammals. DeltaA–D are expressed in the developing zebrafish pallium (Smithers et al., 2000; Mueller and Wullimann, 2003; Takano et al., 2011), whereas Dll1 and Dll3 (Nelson et al., 2013) but not Dll4 are expressed in the developing mammalian pallium (Herman et al., 2018). Comparing the DNA sequences of these delta genes with mouse Dll1 and Dll3 revealed that the sequences of DeltaD, DeltaA, and DeltaB are similar to that of mouse Dll1, whereas the sequence of DeltaC is similar to that of mouse Dll3 (Figure 2A). As fewer studies have examined the roles of Delta proteins in the development of the zebrafish pallium, we would like to briefly introduce the functions of Delta genes during somitogenesis and elaborate their possible implications in pallial development.

In zebrafish, somitogenesis is controlled by segmentation clock, which is coordinated by several components in Notch signaling including a Notch ligand (DeltaC) and the downstream target hairy/E(spl) genes (her1 and her7) with an oscillatory expression pattern. Both DeltaD and DeltaC are essential for somitogenesis, and DeltaC, as one of the oscillators, is critical for proper somite segmentation (Mara et al., 2007). Unexpectedly, a study using DeltaD mutant embryos suggested that DeltaD is required for the oscillation of her1, the downstream target gene of Notch signaling, whereas the expression of DeltaD is maintained at a constant level (Holley et al., 2000). These findings suggest that DeltaC and DeltaD expressed in different patterns have distinct functions during somitogenesis, and the deficiency of any protein causes defects in somite formation (Holley et al., 2000; Mara et al., 2007). Further examination by Wright et al. (2011) revealed the puncta expression pattern of DeltaC and DeltaD in the retina and hindbrain, and, notably, DeltaC and DeltaD were colocalized in the retina but not in the hindbrain. In cellular level, DeltaD may be either expressed in the cytoplasm or on the plasma membrane depending on the expression level of DeltaC in the presomitic mesoderm (PSM) during the formation of new somite (Wright et al., 2011). Mechanically, DeltaC is expressed as a soluble form to physically attract DeltaD away from the cell membrane to switch off DeltaD-mediated Notch signaling in the DeltaC-enriched region (Wright et al., 2011). In the developing pallium, DeltaD and DeltaC are both expressed, as demonstrated by our in situ hybridization data (DeltaD and DeltaC, Figure 2B) and previous studies (Smithers et al., 2000; Takano et al., 2011) and, in protein level, DeltaC was expressed in a puncta pattern in the cytoplasm, as shown by the DeltaC–mCherry strategy (Figure 2C). Considering the interplay between DeltaC and DeltaD in somitogenesis, DeltaC and DeltaD may play similar roles in regulating Notch signaling to control pallial formation.

During mammalian neocortical development, Dll1 has been found to be expressed in neural progenitors with an oscillatory pattern (Shimojo et al., 2008). In situ hybridization in the E9.5 whole mount embryo indicated that Dll1 was expressed in the forebrain, whereas Dll3 was expressed only in the ventral region of the forebrain. During somite formation, Dll1 and Dll3 were differentially expressed in the posterior or anterior region of newly formed somite (Dunwoodie et al., 1997) and were both necessary for somitogenesis (Kusumi et al., 1998; Dunwoodie et al., 2002). Mutations in human DLL1 induce neurodevelopmental disorders with non-specific brain abnormalities (Fischer-Zirnsak et al., 2019), whereas mutations in DLL3 cause spondylocostal dysostosis with axial skeletal defects (Bulman et al., 2000). These pathological findings suggest the critical role of Dll1 in dorsal telencephalic development while Dll3 mainly functions in somitogenesis. The use of Dll3 cDNA to replace Dll1 gene resulted in embryonic lethality in transgenic mice, suggesting that at least some Dll1 functions cannot be replaced by Dll3 (Geffers et al., 2007). Besides, Dll1 was found on the plasma membrane, whereas Dll3 was observed in the cytosol with a puncta pattern in mouse PSM and cultured cell lines (Geffers et al., 2007; Chapman et al., 2011); this is similar to the distribution of DeltaC and DetlaD in zebrafish somitogenesis. Another reason to explain the interchangeable role of Dll1 by Dll3 is the absence of lysine within the intracellular domain of Dll3. As lysine in the intracellular domain would be ubiquitinated by the ubiquitin ligase, Mib1, to triggers the endocytosis to recycle the ligand on the signal sending cell and pulling Notch receptor on the signal sending cell to activate downstream signaling through exposing the S2 protease site (Ladi et al., 2005; Le Borgne et al., 2005; Sprinzak and Blacklow, 2021). Thus, Dll3 cannot be exhibited on the cell membrane to compensate the loss of Dll1. Besides the intracellular domain, Geffers et al. (2007) provided evidences showing N-terminal DSL domain and the first two EGF-like repeats of Dll1 were critical to activate Notch signaling and cannot be replaced by that of Dll3 using different forms of chimeric Dll1 and Dll3 fusion proteins. Komatsu et al. (2008) also reported that in addition to the DSL domain, the conserved DOS motif within the first two EGF-like repeats is vital for activating Notch signaling and suggested that the DOS motif may cooperate with the DSL domain in binding to the Notch receptor. Separate studies performing mutation and structural analysis have indicated the importance of the DOS motif in cell lines (Shimizu et al., 1999; Geffers et al., 2007; Cordle et al., 2008a) and C. elegans (Komatsu et al., 2008). However, mouse Dll3 and Dll4 and zebrafish DeltaC do not contain this DOS motif, which may explain why Dll3 is unable to activate Notch signaling in certain cell types (Ladi et al., 2005; Geffers et al., 2007). Although Komatsu et al. (2008) suggested that Notch ligands without the DOS motif, such as DeltaC and Dll3, may trigger non-canonical Notch signaling with non-canonical ligands with the DOS motif, the role of non-canonical Notch signal pathway in either neurogenesis or somitogenesis should be further confirmed.

Both zebrafish DeltaC and mouse Dll3 share some similar features such as the intracellular distribution and lack of a DOS motif. Mutation of either DeltaD or DeltaC in zebrafish would lead to defects in somite development (Holley et al., 2000; Mara et al., 2007), suggesting that DeltaD and DeltaC are both necessary for somitogenesis. Studies on human disorders have indicated that DLL1 is more crucial for the neocortical development (Fischer-Zirnsak et al., 2019), whereas DLL3 is more critical for somitogenesis (Bulman et al., 2000). Thus, during somite formation, the dominant isoform changes from DeltaD and DeltaC in zebrafish to DLL3 in humans. This may reflect an evolutionary change in dominant forms in distinct tissue development. As DeltaA–D are all expressed in the developing zebrafish pallium (Smithers et al., 2000; Mueller and Wullimann, 2003; Takano et al., 2011), the expression pattern of Delta-like genes in mice and clinical features of human diseases suggest that the compensation may occur in the developing zebrafish pallium but not in the developing human dorsal telencephalon. These findings imply that the dominant form regulating telencephalic development may switch during the course of evolution.

Cortical development involves multiple neural and cortical progenitors to produce cortical neurons at the right place and correct time. After the onset of corticogenesis, aRGs derived from RGs produce neurons either through the direct or indirect pathway (Figure 3A, black arrows: direct pathway; green arrows: indirect pathway). In the indirect pathway, aRGs generate to IPCs before producing neurons. Promotion of the indirect neurogenic pathway may be an evolutionary event (Cárdenas et al., 2018; Cárdenas and Borrell, 2020). A comparative approach using multiple species such as snakes and the mammalian pallium demonstrated the dominance of direct neurogenesis, whereas indirect neurogenesis gradually replaces direct neurogenesis in higher animals such as mammals (Figure 3A). Borrell’s team identified this evolutionary trend and suggested its relation to the gradient expression of Robo/Dll1 in the pallium across different species (Borrell and Reillo, 2012; Borrell et al., 2012; Cárdenas and Borrell, 2020). During indirect neurogenesis, RGs generate IPCs before becoming neurons, and one IPC symmetrically divides again to generate two neurons (Miyata et al., 2004; Noctor et al., 2004). IPCs act as a source of Notch ligands (Mizutani et al., 2007) to maintain the RG cell pool in a feedback loop of Notch signaling (Kawaguchi et al., 2008; Yoon et al., 2008; Lui et al., 2011; Nelson et al., 2013). Kawaguchi demonstrated that Dll1-positive cells in the VZ/SVZ of the E13.5 mouse neocortex were separated from those with active Notch signaling, and conditional Dll1 knockout mice driven by Nestin-Cre showed neuronal prematuration, suggesting that Dll1 can maintain neural progenitors in an undifferentiated state (Kawaguchi et al., 2008). Yoon et al. (2008) used Mib1 knockout mice in their study. Mib1 is a RING-type E3 ubiquitin ligase that promotes the endocytosis of canonical Notch ligands. They demonstrated that Mib1-positive cells may provide the Dll1 ligand to activate Notch signaling in adjacent cells in vitro. In addition, most Mib1-positive cells including IPCs and neurons can serve as Dll1 sources to activate Notch signaling in surrounding RGs (Yoon et al., 2008) which was supported by the asymmetric distribution of Dll1 and Mib1 during the asymmetric division of a neural progenitor to produce a progenitor and a neuron (Tozer et al., 2017).

Nelson et al. (2013) categorized the major expression of Dll1 into two clusters in the SVZ and VZ at the E14.5 mouse neocortex similar to that reported in the Allen Brain Atlas¹ (Figure 3A). The in situ hybridization results from the Allen Brain Atlas showed that Dll1 was expressed in the VZ and SVZ but not in the cortical plate in the cortices at various developmental stages. In addition, the number of Dll1-expressing cells gradually decreased over the course of development. Through multiphoton microscopy, Nelson discovered that Dll1-positive IPCs could contact Hes1/5-positive RGs through dynamic and transient elongate processes to maintain the RG cell pool. In addition, they suggested that other Notch ligands may participate in Notch signaling such as Dll3 in IPCs and Jag1 in RGs (Nelson et al., 2013). On the basis of the findings of these studies, we summarized that the oscillatory expression of Notch ligands such as Dll1 or Jag1 (Nelson et al., 2013) and receptors (Notch1/2) (Mase et al., 2021) can maintain RGs in the cell cycle (Figure 3). After differentiation into IPCs, Notch ligands are expressed at a constant level because Notch signaling cannot be activated to produce downstream Hes1, which represses Dll1 expression in a negative feedback loop (Figure 3B; Mizutani et al., 2007). With a constant expression of Dll1, IPCs can act as a Dll1 source to activate Notch signaling in RGs through ligand–receptor interaction. Oscillated Hes1 and Dll1 may occur only in RGs but not in IPCs. Although the detailed mechanism remains unclear, Notch signaling is believed to be inactivated in IPCs due to the lack of Notch receptors or the presence of molecules that inhibit the activation of Notch signaling. Finally, the expression of Dll1 is the lowest in neurons (Figure 3B). In the brains, such as reptiles and birds, which lack IPCs because they primarily rely on direct neurogenesis during pallium development, Notch ligands (Dll1 or Jag1) are produced solely by RGs. However, Notch ligands in RGs are oscillatory in response to the negative regulator Hes protein, which may explain the smaller RG pool in reptiles than in mammals due to the lack of an alternative source of Notch ligands (Figure 3A). During the development of the neocortex in humans, a significant increase in the number and types of neural/cortical progenitors contribute to a larger size of the dorsal telencephalon than in other vertebrates. The enriched bRG is the most prominent feature for massive cortical expansion. bRG is derived from aRGs similar to the derivation of IPCs from aRGs. Similar to direct and indirect neurogenic pathways, bRG may produce neurons directly or generate basal IPC before producing neurons. Notch signaling is activated in bRGs based on the expression of Hes1. In addition, according to Nelson’s study with Dll1d2YFP reporter, many neural/cortical progenitors, including RGs, bRGs and IPs, express Dll1, and basal IPCs can maintain bRG proliferation through physical contact with bRGs in the SVZ (Figure 3B; Nelson et al., 2013; Govindan and Jabaudon, 2017).

Nomura found that neural progenitors in the pallium of geckos required a long time to differentiate into neurons compared with other species of amniotes (Nomura et al., 2013). They applied pulse labeling to monitor the period from neural progenitors to neurons in the pallium of mouse (M. musculus), gecko (Poekilocerus pictus), turtle (Pelodiscus sinensis), and chick (Gallus gallus). Compared with the neural progenitors of other species, neural progenitors in geckos required a longer period to differentiate, twice as those required by mice and chicks. Although cortical progenitors stay in the progenitor stage for a longer period in geckos than in mice, the number of mitotic cells was lower in geckos, suggesting that the size of the neural progenitor pool may not be associated with the duration in the progenitor stage. Furthermore, they used a CBF1-driven reporter to monitor Notch activity in neural progenitors. Neural progenitors of geckos exhibited higher Notch activity than those of other species. Notably, the distribution of neural progenitors with active Notch signaling differed among species: mosaic in mice, turtles, and chicks but homogenous in geckos (Nomura et al., 2013). Two possibilities can explain the mosaic distribution of neural/cortical progenitors. First, this distribution may result from the oscillated expression pattern of Notch signaling components (Shimojo et al., 2008; Ohtsuka and Kageyama, 2021b). Second, this pattern may be due to scattered IPCs in the neural progenitor pool to deliver the Notch ligands (Mizutani et al., 2007). In either possibility, the cross-species study suggested that differences in the duration of neural progenitor differentiation and the number of mitotic cells within amniotes may be linked to the spatial distribution of neural/cortical progenitor cells, which may be uniform or mosaic. Thus, the emergence of the mosaic distribution of neural/cortical progenitor may be an evolutionary key to the expansion of the telencephalon.

Notch signaling is essential for self-renewal in RGs to maintain the progenitor cell pool during cortical development. Expansion of the neural progenitor pool and a prolonged neural progenitor self-renewal period are believed to be critical events in cerebral cortex evolution (Hansen et al., 2010; Lui et al., 2011; Borrell and Reillo, 2012; Geschwind and Rakic, 2013). Recently, a human-specific NOTCH2 partial duplicated paralog, NOTCH2NL gene, was found to be expressed in both human aRG and bRGs and to improve the expansion of cortical progenitors by activating NOTCH signaling through interrupting the cis-inhibition of DLL1 (Suzuki et al., 2018). Overexpression of NOTCH2NL in embryonic mice or human organoids prolonged the self-renewal stage and delayed neuronal differentiation, resulting in clonal expansion in neural progenitors (Fiddes et al., 2018; Suzuki et al., 2018). By contrast, NOTCH2NL knockout accelerated neuronal differentiation and reduced neurogenesis (Fiddes et al., 2018). Investigations on underlying mechanisms revealed that NOTCH2NLs would interact with NOTCH receptors and inhibit cell autonomous DLL1 (NOTCH ligand) function to enhance NOTCH activity during corticogenesis (Fiddes et al., 2018; Suzuki et al., 2018). In addition, the deletion or duplication of NOTCH2NL genes in humans induced microcephaly and megacephaly, respectively, suggesting the crucial role of NOTCH2NL in human neocortical development (Fiddes et al., 2018). Hence, enhancing NOTCH signaling at a proper level may contribute to cortical evolution.