Loss of Twist1 and balanced retinoic acid signaling from the meninges causes cortical folding in mice

Cortical folding is a stepwise process involving amplification of neuroprogenitor cell populations, heightened levels of neurogenesis and tangential dispersion of migrating neurons (Llinares-Benadero and Borrell, 2019; Del-Valle-Anton and Borrell, 2022). Cortical neurons are derived from Pax6⁺ apical radial glia (aRG) located within the ventricular zone (VZ) (Noctor et al., 2001; Anthony et al., 2004). aRG have apical processes that attach to the ventricular surface and basal processes that contact the pial basement membrane. aRG self-amplify through symmetrical divisions to produce two identical multipotent daughter cells. At the onset of neurogenesis, aRG divide asymmetrically to produce a self-renewed aRG and a post-mitotic neuron (Taverna et al., 2014). In addition, some Pax6⁺ aRG lose their apical processes and delaminate from the VZ to become basal (outer) radial glia (bRG) in a secondary germinal area called the subventricular zone (SVZ). These basal neuroprogenitor cells show similar gene expression compared with aRG (i.e. Pax6), and can also self-renew through symmetrical divisions while maintaining contact with the pia (Hansen et al., 2010; Reillo et al., 2011). As neurogenesis proceeds, RG divide asymmetrically to produce Tbr2⁺ basal intermediate progenitors (bIPs). The division of these transit amplifying cells, which are located in the SVZ along with bRG, gives rise to postmitotic projection neurons (Noctor et al., 2004).

Studies in ferrets, non-human primates and transgenic mice suggest that increasing the local abundances of bRG and bIPs in an expanded SVZ leads to regional differences in the levels of neurogenesis, increased cortical surface area and folding (Chizhikov et al., 2019; Heide et al., 2020; Ju et al., 2016; Matsumoto et al., 2017; Rash et al., 2013; Roy et al., 2019; Stahl et al., 2013; Florio et al., 2015; Wang et al., 2016; Nonaka-Kinoshita et al., 2013). bRG help induce folding because their basal processes fan out to contact the pia, promoting the dispersion of migrating neurons along the tangential axis (i.e. lateral dispersion) (Reillo et al., 2011). Studies in ferrets also suggest the relative abundance of basal progenitors across the tangential axis causes regional differences in neurogenesis and shapes the topology of gyri and sulci (Toda et al., 2016). Importantly, in gyrated versus lissencephalic brains, basal progenitors have greater capacity to self-renew, which increases the size of the progenitor pool and gives rise to more neurons, which also promotes folding (Nonaka-Kinoshita et al., 2013). By contrast, bRG are rare and most basal progenitors are bIPs in lissencephalic (smooth brained) animals, which typically divide only once in a self-consuming manner to produce two post-mitotic neurons (Kelava et al., 2012).

Gyrencephaly is often assumed to have evolved from lissencephaly. However, it is an evolutionarily labile trait, as evidence suggests lissencephaly in mammals such as mice and marmosets arose from loss of folding present in gyrencephalic ancestors (secondary lissencephaly) (Kelava et al., 2013). Factors that promote secondary lissencephaly in mice include non-cell autonomous BMP4 signaling from embryonic cranial mesenchyme and cell-autonomous expression of adhesion proteins FLRT1 and FLRT3 in migrating neurons (Chizhikov et al., 2019; Del Toro et al., 2017). Loss of BMP4 signaling in Lmx1a^−/−/Lmx1b^−/− embryos causes the formation of gyri in the dorsolateral telencephalon via transient upregulation of Wnt/β-catenin signaling in the cortical hem (Chizhikov et al., 2019). This causes accumulation of aRG and, subsequently, their delamination to produce abundant bRG and greater numbers of bIPs. Conversely, double knockout (DKO) of Flrt1 and Flrt3 in mice causes sulci to form in the dorsolateral telencephalon through a process independent of progenitor cell amplification. The loss of these adhesion proteins alters radial migration of neurons and causes abnormal neuronal clustering along the tangential axis of the cortical plate, in a manner similar to tangential dispersion (Del Toro et al., 2017). Findings in Lmx1a^−/−/Lmx1b^−/− and Flrt1/Flrt3^DKO mutants, and those in Trnp1 shRNA knockdown mice, which develop gyri via expansion of bRG (Stahl et al., 2013), suggest that signaling processes that increase self-renewal and expansion of bRG and/or bIPs, or perturbing adhesive forces that alter the tangential distribution of neurons, can induce cortical folding.

We have previously shown that loss of the transcription factor Twist1 from osteoprogenitor cells and cranial mesenchyme using the Sm22a-Cre driver causes skull malformations, hypoplastic dura, and affects the growth and expansion of blood and lymphatic vessels in dura (Ang et al., 2022; Tischfield et al., 2017). Through the course of those studies, we serendipitously discovered that cortical folding is present in Twist1^FLX/FLX:Sm22a-Cre mice. We now provide evidence that loss of balanced retinoic acid (RA) signaling from the meninges induces cortical folding in these mice. Cell proliferation in the primitive meninx is reduced in Twist1^FLX/FLX:Sm22a-Cre embryos, causing loss of arachnoid fibroblasts that express Raldh2, an enzyme required for RA synthesis (Siegenthaler et al., 2009). Mechanistically, we show that reduced Raldh2 expression in the dorsolateral meninges during late gestation (E16.5) leads to expansion of Pax6⁺ aRG and tangential forebrain lengthening. At this time, the regional densities of Tbr2⁺ bIPs along the tangential axis begins to vary, affecting the levels of neurogenesis and tangential distribution of neurons. This pattern of events is maintained such that by E18.5, regions with relatively less neurogenesis and neuronal density along the tangential axis buckle and form sulci. Furthermore, maternal RA supplementation rescues cortical folding by normalizing the distribution of aRG and neurogenesis across the tangential axis. Thus, our findings implicate proper meningeal development, downstream of Twist1, and balanced RA signaling for maintaining lissencephaly in mice via processes that affect neurogenesis and the tangential distribution of neurons.

We examined the effects of meningeal Twist1 expression on forebrain development by conditionally inactivating Twist1 in cranial mesenchyme using Sm22a-Cre. Sm22a-Cre is active by E10.5 in populations of neural crest and cranial mesoderm that produce the three meningeal layers surrounding the forebrain, diencephalon and cerebellum (El-Bizri et al., 2008; Tischfield et al., 2017). Lineage labeling using R26:Ai14:Sm22a-Cre revealed Ai14:td-Tomato expression in preosteoblasts, dura and connexin 43(+) arachnoid fibroblasts surrounding the developing forebrain at E14.5, by which time the three meningeal layers are recognized (Fig. 1A,B). By contrast, Ai14:td-Tomato expression appeared sparser in developing pia mater underlying the connexin 43⁺ arachnoid membrane (Fig. 1C).

Twist1⁺ cells were found in cranial dermis and the presumptive connexin 43⁺ dura and arachnoid mater at E12.5 (Fig. 1D). Of note, connexin 43 labels both the dura and arachnoid mater in rodent embryos before becoming restricted to the latter in adults (Ruangvoravat and Lo, 1992; Farmer et al., 2021). Twist1⁺ cells also colocalized with proliferating Ki67⁺ cells in the presumptive dura and arachnoid (Fig. 1D′). By contrast, Twist1⁺ cells were lower in the pia and absent in blood vessels and neurons (Fig. 1D). In Twist1^FLX/FLX:Sm22a-Cre embryos, Twist1⁺ cells were less abundant in cranial mesenchyme surrounding the dorsolateral forebrain, including differentiating connexin 43⁺ dura/arachnoid mater (Fig. 1E,F). Twist1⁺ cells were also reduced in cranial mesenchyme surrounding the basolateral forebrain, but to a lesser extent compared with dorsolateral regions (Fig. S1A-C). Twist1⁺ cells in the pia were also reduced, but not to the extent seen in dura/arachnoid (Fig. 1E,F). Taken together, Twist1^FLX/FLX:Sm22a-Cre produces a model in which Twist1⁺ cells are lost in the leptomeninges (arachnoid and pia), but more so in the arachnoid and throughout dorsolateral versus basolateral cranial mesenchyme.

We examined cell proliferation in cranial mesenchyme and developing leptomeninges at E12.5 and E14.5. In E12.5 Twist1^FLX/FLX:Sm22a-Cre embryos, we found fewer proliferating cells in the dorsolateral and basolateral cranial mesenchyme after a 1 h EdU pulse. EdU⁺ cells were reduced in the presumptive connexin 43⁺ dura/arachnoid mater, whereas EdU labeling was comparable in the pia (Fig. 1G,H, Fig. S1D,E). Twist1 protein was significantly downregulated by E14.5 and only a few Twist1⁺ cells were found in the meninges (Fig. S1F). Cell proliferation, however, remained decreased throughout dorsolateral and basolateral regions after a 1 h EdU pulse. The number of EdU⁺ cells co-labeled with connexin 43 was reduced, whereas EdU labeling in the pia, as visualized by p75NTR staining (DeSisto et al., 2020), remained unaffected (Fig. 1I-K, Fig. S1G-I). Dura and arachnoid mater were hypoplastic in both basolateral and dorsolateral regions by E14.5, whereas the pial layer, which showed normal amounts of EdU⁺ cells, was unaffected (Fig. 1L, Fig. S1J). Thus, consistent with Sm22a-Cre activity and Twist1 expression, cell proliferation was reduced in cranial mesenchyme giving rise to dura and arachnoid mater, whereas proliferation in the pia was comparable between mutants and Twist1^FLX/FLX controls. The leptomeninges differentiated by E14.5 and E16.5, as assessed by p75NTR and connexin 43 staining (Fig. 1M,N, Fig. S1K,L). The dura/arachnoid mater remained hypoplastic in E16.5 mutants, which did not show signs of increased apoptosis (Fig. 1O,P, Fig. S1M). These results show that loss of Twist1 affects cell proliferation and attenuates meningeal expansion, although the meninges continue to differentiate and express layer-specific markers.

Loss of Twist1 in cranial mesenchyme produced macroscopic cortical abnormalities in postnatal mice. Compared with the smooth cortex characteristic of mice, many Twist1^FLX/FLX:Sm22a-Cre mutants showed tapering of the anterior cortex, and bumps and dips on the cortical surface (Fig. 2A-C). A smaller fraction appeared to have caudal shortening of the cortex along the rostro-caudal axis (Fig. 2B, Fig. S2A). Tapering of the anterior cortex corresponded to the shape of the skull, which was shorter with misshapen frontal bones (Fig. 2D). However, reconstructed computed tomography scans of Twist1^FLX/FLX:Sm22a-Cre skulls did not show indentation or extrusion matching the shape of the cortex (Fig. S2B-C″). We postulated that bumps on the cortical surface could have resulted from neuronal breaches of the pial basement membrane. Sections through the cortex of P10 mice, however, revealed that cortical layering was intact beneath the bumps, as assessed by Satb2 (layers 2-4), Ctip2 (layer 5) and FoxP2 (layer 6) staining (Fig. 2E-F″). Furthermore, the pial basement membrane was intact at E16.5 and P10, as assessed by laminin α1 staining (Fig. 2H-I′,J), and nestin⁺ radial glia endfeet were properly anchored to the pial basement membrane by E16.5 (Fig. 2H-I″). Rather, large macroscopic bumps on the cortical surface in some P10 Twist1^FLX/FLX:Sm22a-Cre mutants appeared to result from focal expansion of cortical deep layers (Fig. 2G).

Strikingly, ∼46% of Twist1^FLX/FLX:Sm22a-Cre mice developed cortical folding that persisted postnatally and in young adults (Fig. 3A-I). Folds were typically present in one cortical hemisphere (but could be bilateral in some mice) and located in the posterior dorsolateral cortex at or near the hippocampus (equivalent to the interparietal sulcus in gyrencephalic animals). Folding was less frequent in regions equivalent to the Sylvian sulcus (Fig. 3I). Folding more anteriorly, at the level of the striatum, was also less frequent (4/39 mice) (Fig. 3G). Sulci were present with and without cortical invagination (Fig. 3B′-D′). Cortical layering was grossly normal and the overlying pial basement was intact (Fig. 3E-F′,H). The telencephalic fissures appeared to be bona fide sulci as they maintain hallmarks of proper cortical folding: folding of the cortical layers and pial surface without disrupting the ventricular surface (Borrell, 2018). Collectively, loss of Twist1 in developing dura/arachnoid mater, especially in dorsolateral regions surrounding the posterior telencephalon, induces cortical folding in mice.

The first signs of cortical folding were apparent at E16.5 in the form of small invaginations that corresponded to the eventual locations of sulci (Fig. 4A,B). Although we could detect occasional Ctip2⁺ cells that appeared to have migrated further into the cortical plate at regions adjacent to presumptive sulci (Fig. 2H), deep layer neurons were not found in superficial layers, or vice versa, as commonly seen in neuronal heterotopias when radial glia endfeet detach from the pial basement membrane (Zarbalis et al., 2007; Myshrall et al., 2012; Inoue et al., 2008). Reelin expression in Cajal-Retzius cells was also normal in the marginal zone beneath the pial surface (Fig. S2D,E). At E16.5, average cortical thickness was reduced and cortical layers were thinner in presumptive sulci compared with adjacent regions and the equivalent region in controls (Fig. 4C,D). Cell counts adjacent to sulcal regions showed comparable numbers of Tbr1⁺, Ctip2⁺ and Satb2⁺ neurons at this time. In presumptive sulci, the numbers of Ctip2⁺ and Tbr1⁺ deep layer neurons were generally reduced, accompanied by a larger reduction of Satb2⁺ upper layer neurons (Fig. 4E).

Cranial mesenchyme and the meninges proper secrete morphogens that regulate cortical development, including retinoic acid (RA) and bone morphogenetic proteins (BMPs) (Siegenthaler et al., 2009; Choe et al., 2012). RA signaling regulates neurogenesis by promoting asymmetrical division of aRG, and later facilitates the transition from multipolar to bipolar morphology required for neurons to radially migrate (Siegenthaler et al., 2009; Haushalter et al., 2017; Choi et al., 2014). Raldh2, an enzyme required for RA synthesis, is produced in arachnoid fibroblasts and is deficient in Foxc1^hith/hith mice, which have hypoplastic meninges plus skull and cortical abnormalities, similar to Twist1^FLX/FLX:Sm22a-Cre mice (Siegenthaler et al., 2009). Foxc1^hith/hith mice, and especially more severely affected Foxc1^−/− embryos, however, show pseudo-gyrification with folding of the ventricular surface, breaches in the pial membrane and cortical layering defects (Siegenthaler et al., 2009). Nonetheless, we asked whether Raldh2 expression and meningeal-derived RA signaling was affected in Twist1^FLX/FLX:Sm22a-Cre embryos. At E14.5, we did not detect a significant decrease in the numbers of Foxc1⁺ cells in the leptomeninges (Fig. 5A,B,I, Fig. S3A-B′,E). We also did not detect loss of Raldh2 expression at this time (Fig. 5C,D). Interestingly, by E16.5, the earliest time point when we could detect presumptive sulci, the numbers of Foxc1⁺ cells in the dorsolateral meninges were significantly reduced, whereas the numbers of cells in the basolateral meninges were only slightly reduced (Fig. 5E,F,J, Fig. S3C,D,F). Furthermore, we detected regionalized loss of Raldh2 expression in the dorsolateral meninges overlying regions where presumptive sulci form (Fig. 5G,H). Although Raldh2 expression was reduced in dorsolateral regions, expression at and adjacent to the midline was unaffected (Fig. 5M-O). Expression was similarly reduced at E18.5 (Fig. 5K,L,P). Thus, loss of Twist1 in the primitive meninx attenuates cell proliferation in the arachnoid, leading to regionalized loss of Foxc1⁺ cells and Raldh2 expression in the dorsolateral meninges.

Inactivating Foxc1 in mice causes meningeal hypoplasia and loss of Raldh2 expression, similar to Twist1. Lack of RA signaling prolongs symmetrical division of Pax6⁺ aRG at the expense of generating Tbr2⁺ bIPs. This causes forebrain lengthening in hypomorphic Foxc1^hith/hith embryos, with greater effects seen in Foxc1 null embryos (Siegenthaler et al., 2009). Given loss of meningeal Raldh2 expression overlying the dorsolateral telencephalon, we asked whether neurogenesis was affected in Twist1^FLX/FLX:Sm22a-Cre embryos. At E14.5, when Raldh2 expression was comparable between mutants and controls, we did not detect changes to the numbers of Pax6⁺ aRG or Tbr2⁺ bIPs, including those co-labeled with EdU (Fig. 6A-E). However, by E16.5, concomitant with the reduction of meningeal Raldh2 expression, the length of the VZ and dorsal telencephalon had expanded, with a corresponding increase in the total numbers of Pax6⁺ aRG along the tangential axis (Fig. 6F-I). Binning the cortex radially showed that the density of Pax6⁺/Tbr2⁻ cells in the VZ/SVZ was increased closest to the cortical midline (which undergoes expansion), but otherwise cell density was normal (Fig. 6J). Pax6⁺/Tbr2⁻ cells outside of the VZ were located within the normal confines of the SVZ, which did not appear to be enlarged, as seen in other transgenic models that develop gyri (Fig. 6F′,G′) (Stahl et al., 2013). We also found that the total number of Tbr2⁺ bIPs was slightly decreased at this time, as cell density was reduced adjacent to the midline and center regions (Fig. 6F″,G″,K,L). This suggests regionalized loss of RA production favors symmetrical division of Pax6⁺ aRG at the expense of producing Tbr2⁺ bIPs. These findings are similar to those in Foxc1^hith/hith embryos (Siegenthaler et al., 2009), but occur later during neurogenesis (∼E16.5) after the levels of Raldh2 have significantly decreased.

We next assessed regional changes to neurogenesis at E18.5 when cortical folding was well under way. Regional cell density differences once again correlated with cortical folding patterns. At and adjacent to the cortical midline, where greater numbers of Pax6⁺ cells were found, the numbers of Tbr2⁺ bIPs and Ki67⁺ cells were now increased, suggestive of a neurogenic ‘catch up’ period (Fig. 7A-B′,E). In developing sulcal regions, however, the numbers of bIPs were decreased compared with controls (Fig. 7E). Thus, neurogenesis at E18.5 was significantly increased at and adjacent to the midline compared with neighboring sulcal regions. We also tracked the trajectories of radially migrating neurons by injecting pregnant dams with EdU at E16.5 and analyzing the distribution of cells at E18.5 or P0.5. We found that proportionately more EdU⁺ cells were migrating towards the midline and immediately adjacent regions in mutants (Fig. 7C-D′, Fig. S4A,B). By contrast, fewer EdU⁺ cells migrated into developing sulcal regions along the dorsolateral cortex (Fig. 7F). Average cortical thickness was also reduced at this time, especially in sulcal regions (Fig. 7G-H′,I). Notably, the cortical midline was expanded compared with controls, and the average thickness ratio of the midline versus the dorsolateral cortex was much greater in mutants (Fig. 7G″,H″,J). P0.5 brains from newborns with significant forebrain lengthening and expansion of the cortical midline were folding (Fig. S4C,E,F). By contrast, cortices from newborns with milder forebrain lengthening and less midline expansion did not fold, although folding near the sylvian sulcus was possible (Fig. S4D). Thus, cortical folding in Twist1^FLX/FLX:Sm22a-Cre mice coincides with expansion of Pax6⁺ aRG and forebrain lengthening, coupled with regionalized changes to neurogenesis that affect the tangential distribution of neurons across the cortical plate. This causes the midline to expand more rapidly versus the dorsolateral cortex, which is predicted to cause buckling and folding.

Folding was less common in more anterior cortical regions at the level of the striatum. Thus, we asked whether neurogenesis was affected to the same extent as seen more posteriorly. At E16.5 and E18.5, Raldh2 expression was decreased in the dorsolateral meninges but preserved at the midline (Fig. S5A-C). Surprisingly, however, the total numbers of Pax6⁺ cells and the size of the VZ appeared relatively normal in anterior sections. The total number of Tbr2⁺ cells, and the relative densities in binned regions along the SVZ, were also more consistent compared with more posterior sections (Fig. S6A-C). Thus, in contrast to the anterior cortex, folding more posteriorly is likely influenced by regional changes to neurogenesis along the tangential axis that affect neuronal distribution and cortical expansion. At E18.5, we detected minor differences in the distribution of Tbr2⁺ cells across the tangential axis (Fig. S6D′,E′,G). Interestingly, EdU labeling (E16.5 pulse) showed more EdU⁺ cells within the intermediate zone (Fig. S6D-E′,F). Considering the large macroscopic bumps that are sometimes present along the dorsolateral cortex at the same axial level, these findings suggest late-stage neurogenesis and/or radial migration from the intermediate zone is affected more anteriorly, but to a different extent from that seen more posteriorly.

Regionalized loss of Raldh2 expression prompted us to ask whether maternal supplementation with RA during pregnancy could restore lissencephaly. We supplemented pregnant dams with either 0.175 mg or 0.35 mg of RA/gram of food from E12.5 to E17.5 before collecting embryos at E18.5. Supplementing with 0.175 mg of RA/gram of food resulted in macroscopic cortical abnormalities (e.g. cortical bumps) in all mutant embryos (n=10), similar to untreated mutants (Fig. 8A-C). Supplementing with 0.35 mg of RA/gram of food, however, resulted in 50% of embryos (6/12) that, macroscopically, resembled controls (Fig. 8D). The cortex was smooth and shaped normally in these rescued embryos (but still slightly tapered) and also did not show signs of caudal shortening, as observed in some Twist1^FLX/FLX:Sm22a-Cre mice. Two of the remaining RA-treated embryos had elongated cortices whereas the other four embryos still displayed small bump(s) (Fig. 8E). Sections through the posterior telencephalon did not reveal the presence of folding in 8/12 embryos supplemented with 0.35 mg of RA/gram of food (Fig. 8F,G). The remaining four embryos with small bumps showed shallow dips in the cortical layers, although cortical folding to the extent seen in non-RA treated E18.5 mutants was absent (Fig. 8H). Average cortical layer thickness along the dorsolateral cortex was similar between RA-treated controls and mutants, although the midline was still slightly expanded (Fig. 8F′-H′,I). Notably, the ratio of the average cortical thickness at the midline versus the dorsolateral cortex was now similar to non-RA treated controls (Figs 7I and 8J).

Next, we asked whether maternal supplementation with 0.35 mg of RA/gram of food from E12.5 to E16.5 rescued neurogenesis. The total numbers of Pax6⁺ aRG and regionalized densities of Pax6⁺/Tbr2⁻ cells at E16.5 were now similar to controls treated with RA (Fig. 9A-D,G). The total numbers of Tbr2⁺ bIPs and regionalized densities within the SVZ were also comparable with control littermates receiving RA (Fig. 9E-G). Furthermore, EdU labeling (E16.5 pulse) showed the tangential distribution of radially migrating neurons in E18.5 brains was now comparable with RA-treated controls, as neurons no longer preferentially migrated towards the midline at the expense of lateral regions (Fig. 9H-I′,L). The numbers of Tbr2⁺ bIPs distributed across the tangential axis were comparable with RA-treated controls, although more cells were still observed near the midline (Fig. 9L). More anteriorly, the numbers of EdU⁺ cells in the intermediate zone were comparable with controls (Fig. S7A-C). Collectively, these results show that supplementing Twist1^FLX/FLX:Sm22a-Cre embryos with RA suppresses folding and helps balance neurogenesis and neuronal distribution across the tangential axis. Thus, regionalized loss of RA signaling during late stages of neurogenesis (i.e. ∼E16.5 onwards) affects neurogenesis and the relative distribution of neurons across the cortical plate, and is at least partly responsible for folding.

Cortical folding evolved in large part from regionalized changes to the levels of neurogenesis in the neocortex. Symmetrical division of aRG is prolonged, producing more aRG that delaminate to increase the local abundance of basal progenitors, which continue to proliferate in an expanded outer SVZ (Borrell, 2018; Chizhikov et al., 2019). In mice and other lissencephalic mammals, bRG are rare and bIPs have limited capacity to self-renew (Wang et al., 2011; Shitamukai et al., 2011). Interestingly, although the relative abundance of bRG in marmosets, which have near lissencephalic brains, is significantly higher by comparison with mice, they also have less capacity to self-renew versus bRG in gyrated brains. This suggests the ability of basal progenitors to delaminate and also self-renew are important contributing factors for cortical folding (Kelava et al., 2012). Furthermore, increasing the pool of aRG and/or the numbers of bIPs in the mouse neocortex does not necessarily lead to folding per se (Stahl et al., 2013; Tuoc et al., 2013), as factors that promote tangential dispersion of radially migrating neurons (such as the fanned out basal processes of bRG) and regional differences in neuron density along the tangential axis also appear to be important.

Our findings show similarities to, but also notable differences from, mouse transgenic models with induced cortical folding and effects on neurogenesis. For example, shRNA knockdown of Trnp1 causes gyrus formation and folding in the intraparietal sulcus by promoting delamination of aRG, local accumulation of bRG and bIPs in an expanded SVZ, and increased levels of neurogenesis – but without amplifying effects on aRG (Stahl et al., 2013). Lmx1a^−/−:Lmx1b^−/− mutants also develop gyri and folds but show increased radial thickness of the VZ (without tangential lengthening) and more Pax6⁺ aRG at E12.5. These apical progenitors subsequently delaminate from the VZ and continue to divide as bRG in an expanded SVZ, causing regionalized expansion of bIPs and increased neurogenesis (Chizhikov et al., 2019). In contrast to these models, we do not see obvious accumulation of Pax6⁺ bRG in a significantly expanded SVZ underlying developing gyri in Twist1^FLX/FLX:Sm22a-Cre embryos, although slightly more Pax6⁺ cells are present in the VZ/SVZ near the midline. Cortical folding in our model also appears more consistent with formation of sulci.

Increased self-renewal of aRG and tangential lengthening of the VZ is reported to cause cortical folding in some instances. For example, conditional loss of centrosomal protein 83 (CEP83) using Emx1-Cre causes folding and sulcus formation at the midline. Similar to Twist1^FLX/FLX:Sm22a-Cre embryos, these mice show increased self-renewal of aRG and tangential cortical lengthening, with increased numbers of Tbr2⁺ bIPs and heightened levels of neurogenesis at late embryonic stages (Shao et al., 2020). The cortex is also enlarged in the mediodorsal region, and without local accumulation of proliferating bRG in an expanded SVZ. By contrast, however, folding at the intraparietal or Sylvian sulcus is absent and neurogenesis is more uniformly increased across the tangential axis. As such, regionalized differences in radial expansion, as seen in Twist1^FLX/FLX:Sm22a-Cre embryos, are much less prominent in Cep83^FLX/FLX:Emx1-Cre embryos, possibly explaining why cortical folding is milder by comparison.

Our findings in Twist1^FLX/FLX:Sm22a-Cre embryos seem to converge more closely with those in FGF2-injected embryos (Rash et al., 2013). Intraventricular injection of FGF2 protein at E11.5 causes folding in the rostrolateral cortex near the Sylvian sulcus. FGF2-injected embryos also show tangential expansion of the VZ and forebrain lengthening via increased self-renewal of aRG, and at the expense of producing neurons. Interestingly, after increased self-renewal of aRG, there is a neurogenic ‘catch-up’ period as greater numbers of bIPs are found by E13.5 in regions that undergo folding, and without localized accumulation of bRG in an expanded SVZ (Rash et al., 2013). Similar to FGF2-injected embryos (but during later stages of neurogenesis), Twist1^FLX/FLX:Sm22a-Cre embryos show tangential lengthening of the VZ with more Pax6⁺ aRG, initially at the expense of producing Tbr2⁺ bIPs. Moreover, there also appears to be a neurogenic ‘catch-up’ period such that by E18.5, the numbers of bIPs, especially adjacent to the midline, are greater in Twist1^FLX/FLX:Sm22a-Cre embryos. We also do not see increased numbers of bRG in an expanded outer SVZ, similar to findings in CEP83 conditional knockout mice.

Flrt1/Flrt3^DKO mice also develop folding in both the Sylvian and intraparietal sulcus, and with a similar spatiotemporal pattern compared with Twist1^FLX/FLX:Sm22a-Cre mice (Del Toro et al., 2017). Cortical folding in Flrt1/Flrt3^DKO mice, however, is independent from effects on neurogenesis. Instead, these mice show changes to the clustering and migration speed of neurons along the tangential axis, leading to progressive cortical folding and the formation of sulci (Del Toro et al., 2017). These findings agree with differential tangential expansion theory: regions with high neuronal density expand less compared with regions with lower density, and mechanical instability resulting from differential expansion leads to ‘buckling’ of the cortex (Ronan et al., 2014; Garcia et al., 2018). Thus, in Flrt1/Flrt3^DKO mice, abnormal neuronal clustering promotes mechanical instability and cortical folding by forming areas of high density (clusters of fast migrating neurons) and low density (neurons with normal tangential distribution and migrating speed). Perhaps in a somewhat similar manner, regionalized effects on neurogenesis alter the distribution of neurons along the tangential axis in Twist1^FLX/FLX:Sm22a-Cre embryos, forming regions with higher and lower densities, and causing changes to cortical thickness and surface area expansion. Cortical expansion at and adjacent to the midline is also greater compared with that in more dorsolateral regions; this type of asymmetric cortical expansion could be expected to induce mechanical instability, buckling of the cortex and folding (Llinares-Benadero and Borrell, 2019). Notably, regional differences in neurogenesis along the tangential axis were less apparent at more anterior levels of the cortex where folding was less frequent.

Our findings show that Twist1 maintains secondary lissencephaly in mice, at least in part, by promoting expansion of the meninges and the proliferation of arachnoid fibroblasts that produce RA. In Twist1^FLX/FLX:Sm22a-Cre embryos, cell proliferation is reduced in the dorsolateral meninges, causing loss of arachnoid fibroblasts (i.e. Foxc1⁺ cells) that express Raldh2, a key enzyme required for RA synthesis. Interestingly, cortical folding is not reported in conditional Raldh2 null mutants, in which expression is ablated throughout the meninges (Haushalter et al., 2017). Our results therefore suggest that balanced RA signaling is needed to maintain lissencephaly in mice. Regionalized loss of RA affects the generation and relative densities of bIPs along the tangential axis, causing regional imbalances in the levels of neurogenesis. This appears to induce cortical folding by affecting the tangential distribution of neurons in the cortical plate, especially as radial expansion is reduced in regions that buckle and form sulci. By contrast, local accumulation of basal progenitors in proximity to the midline enhances radial expansion and cortical thickness with respect to adjacent dorsolateral regions (Fig. 10). In mutant embryos supplemented with RA, regional differences in the levels of neurogenesis are rescued and radial expansion is also more uniform across the tangential axis, comparable with non-RA and RA-treated controls. As a result, cortical folding is absent or attenuated compared with non-RA rescued mutants.

Blocking RA signaling also affects radial neuron migration, causing some neurons to stall in the intermediate zone because the transition from multipolar to bipolar morphology is affected (Choi et al., 2014; Haushalter et al., 2017). Although more conclusive evidence is needed, we speculate that regionalized loss of RA in Twist1^FLX/FLX:Sm22a-Cre embryos may also impact radial migration, potentially also affecting the distribution of neurons across the tangential axis (perhaps also similar to Flrt1/Flrt3^DKO mice). For example, neurons may preferentially migrate towards regions with higher exposure to RA (or migrate faster), increasing the relative density in that area versus regions with lower exposure to RA. Assuming this scenario, cortical folding would also be absent in conditional Raldh2-null models, in which RA signaling is uniformly downregulated across the cortex, unlike the regionalized loss seen in Twist1^FLX/FLX:Sm22a-Cre mice.

In more anterior sections where large macroscopic bumps are sometimes present, we saw an increase in the numbers of EdU-labeled cells in the intermediate zone at E18.5. This suggests loss of RA signaling has different effects on neurogenesis and/or radial migration along the anterior-posterior axis in Twist1^FLX/FLX:Sm22a-Cre mice. Furthermore, in regions where macroscopic bumps are present, the thickness of cortical deep layers was increased. In conditional Raldh2 null mutants, neurons that stall in the intermediate zone are respecified into deep layer neurons at the expense of more superficial layers (Haushalter et al., 2017). Thus, we speculate that, as the levels of RA decrease and the numbers of aRG diminish in Twist1^FLX/FLX:Sm22a-Cre mice, some neurons may fail to radially migrate and instead are specified to become deep layer neurons.

Our findings shed more light on the role of meningeal RA signaling for cortical neurogenesis. In Foxc1-null or hypomorphic embryos, loss of meningeal Raldh2 expression prolongs symmetric division of Pax6 aRGCs in the VZ, affecting the timely generation of Tbr2⁺ IPs (Siegenthaler et al., 2009). This causes cortical thinning, especially in superficial layers. Furthermore, maternal RA supplementation partially restores neurogenesis, as seen in Twist1^FLX/FLX:Sm22a-Cre mice. As discussed, others have reported that conditional loss of Raldh2 from the meninges does not affect neurogenesis. Instead, it impairs the onset of radial migration in the intermediate zone, affecting specification of neurons normally destined for superficial cortical layers (Haushalter et al., 2017). In agreement, another study used electroporation to express a dominant-negative RA receptor in cortices, which also affected radial migration and neuron specification, but without changes to neurogenesis (Choi et al., 2014). The notable differences between these studies versus ours and that of Siegenthaler et al. (2009) is that meningeal hypoplasia is a shared phenotype between Twist1^FLX/FLX:Sm22a-Cre and Foxc1 mutants, whereas the meninges are not affected in the other models. Thus, RA does appear to have significant effects on neurogenesis – but perhaps more so when the meninges are hypoplastic – suggesting perturbations to other diffusible factors and/or signaling pathways may synergize with loss of RA to affect neurogenesis. Alternatively, the effects of RA on neurogenesis may also be dose dependent.

In conclusion, our findings suggest that balanced RA production from the meninges is necessary to suppress cortical folding and maintain lissencephaly in mice. It is also possible that changes to other signaling pathways (e.g. BMPs, Wnts) contribute to cortical folding in Twist1^FLX/FLX:Sm22a-Cre mice. Studies have shown that differences in gene expression or the densities of basal progenitors roughly correlate with the locations of gyri and sulci in ferrets, and thus may contribute to folding patterns (Toda et al., 2016; de Juan Romero et al., 2015). Considering our findings, we speculate regionalized differences in meningeal RA expression in gyrencephalic animals may vary the levels of neurogenesis across the tangential axis. However, given that there are numerous ways to regulate neurogenesis and the tangential distribution of neurons, it is also possible that regionalized differences in RA expression do not play a major role in the evolution of cortical folding. Rather, these differences become important in pathogenic states or with disruptions to the developing meninges.

The following transgenic mouse lines were used: Twist1^FLX (RRID: MMRRC_016842-UNC), Rosa26:Ai14^tdTomato (RRID:IMSR_JAX:007914) and Sm22a-Cre (RRID:IMSR_JAX:017491). Male and female mice were included for all experiments. Mice were maintained on a mixed background (C57Bl/6:FVB). Embryos obtained from timed matings were considered 0.5 days old upon observance of a vaginal plug. The ages of animals in this study include embryonic day (E) 12.5, E14.5, E16.5 and E18.5, and postnatal days (P) 0.5, P10 and P30. Experiments were approved and carried out under IACUC protocol PROTO201702623 (M.A.T.).

For embryonic samples, pregnant dams were checked for plugs daily. The day of vaginal plug observance was designated E0.5. Pregnant dams were euthanized by CO₂ inhalation. At E12.5 timepoints, embryos were fixed overnight in 4% PFA at 4°C on a rocker. Fixed tissue was cryoprotected in 30% sucrose, and 20 µm sections were collected using a Leica CM1950 cryostat. For meningeal staining and cell counts at E14.5, E16.5 and E18.5, embryos were collected and decapitated. Embryonic heads (with intact skulls) were incubated in PBS with 0.025% heparin, with continuous shaking for 10 min at room temperature before overnight fixation in 4% PFA at 4°C. The tissue was washed and prepared for paraffin wax embedding by dehydration in ethanol. Matched sections were antigen retrieved in boiling citrate buffer (pH 6.0) before staining. For postnatal timepoints P10 and P30, mice were anesthetized with ketamine/xylazine and perfused with PBS followed by 4% paraformaldehyde (PFA). Tissue was embedded in 3% agarose and coronal sections were collected at 60 µm using a Leica VT1000S vibratome. For the P0.5 timepoint, newborns were decapitated and brains were dissected and post-fixed in 4% PFA overnight at 4°C. Fixed tissue was cryoprotected in 30% sucrose, and 20 µm sections were collected using a Leica CM1950 cryostat.

EdU labeling was performed according to manufacturer's instructions using the Click-iT Plus EdU Alexa Fluor 488 kit (ThermoFisher Scientific, C10637). Pregnant dams were injected with 50 mg/kg EdU at E12.5, E14.5 or E16.5. One-hour post-injection (for E12.5 and E14.5), the pregnant dams were euthanized and embryos were collected and decapitated. Embryo heads were incubated with PBS with 0.025% heparin, with continuous shaking for 10 min at room temperature before overnight fixation in 4% PFA at 4°C. For EdU labeling at E18.5 and P0.5, pregnant dams were injected with 50 mg/kg EdU at E16.5. Embryos were collected and/or the brains from P0.5 newborns were dissected, and fixed overnight in 4% PFA at 4°C.

Antibodies were diluted in PBST with 0.1% Triton and 5% normal goat serum and applied overnight at room temperature. The following antibodies were used: rabbit anti-Pax6 (1:100, Novus NBP2-19711), rabbit anti-Tbr1 (1:100, Abcam ab31940, RRID:AB_2200219), rat anti-Ctip2 (1:1000, Abcam ab11370, RRID:AB_297976), mouse anti-reelin (1:500, Millipore, MAB5364, RRID:AB_1293544), mouse anti-Satb2 (1:50, Abcam ab51502, RRID:AB_882455), rabbit anti-Foxc1 (1:50, Cell Signaling 8758, RRID:AB_2797657), mouse anti-Twist1 (1:50, Santa Cruz sc-81417, RRID:AB_1130910), mouse anti-Crapb2 (1:200, Millipore MAB5488, RRID:AB_2085470), rabbit anti-p75NTR (1:200, Cell Signaling 8238, RRID:AB_10839265), chicken anti-nestin (1:500, Novus, NB100-1604, RRID:AB_2282642), rabbit anti-Laminin1a (1:1000, Novus, NB300-144, RRID:AB_10001146), rat anti-Tbr2/EOMES (1:300, Invitrogen 14-4875-82, RRID:AB_11042577), rabbit anti-Raldh2 (1:500, Sigma Aldrich HPA010022, RRID:AB_1844723), mouse anti-connexin 43 (1:200, Santa Cruz sc-59949, RRID:AB_1121832), rabbit anti-connexin 43 (1:1000, Abcam ab11370, RRID:AB_297976), rat anti-Ki67 (1:100, Abcam 15580, RRID:AB_443209), rat anti-Endomucin (1:200, Santa Cruz sc-65495, RRID:AB_2100037), rabbit anti-cleaved caspase 3 (1:500, Cell Signaling 9661, RRID:AB_2341188) and rabbit anti-RFP (1:1000, Rockland 600-401-379S, RRID:AB_11182807). For secondaries, goat anti-rat Alexa Fluor 647, goat anti-rabbit Alexa Fluor 546 and goat anti-mouse Alexa Fluor 488 (all 1:1000, ThermoFisher) were used.

Image data were acquired using a 10×0.45NA objective for E18.5, P10 and P30 samples, and using 20×0.80NA and 40× (1.4NA) objectives for E12.5, E14.5, E16.5 and E18.5 samples on a Zeiss LSM800 confocal microscope, with a z-step size of 3-5 µm. Images were analyzed using Zen software and ImageJ/FIJI.

Graphing and statistical testing were carried out in Graphpad Prism 9. Analyses were performed blinded to genotype. For E12.5 samples, two consecutive sections of 20 µm were used for quantification. EdU or Twist1 cells were quantified within a 200 µm by 100 µm window in the dorsolateral or basolateral region. To differentiate cells in the leptomeninges, cells that overlapped with connexin 43 staining were considered dural/arachnoidal cells, and those directly below the connexin 43 signal were considered pial cells. Sidak multiple comparison tests were performed in GraphPad Prism. For E14.5 samples, two consecutive sections of 20 µm were used for quantification. EdU or Foxc1 cells were quantified within a 300 µm×100 µm area in the dorsolateral or basolateral region. EdU⁺ cells were considered pial or dural/arachnoidal cells if they colocalized with p75NTR or connexin 43, respectively. For Pax6 and Tbr2 cell counts at E14.5, two consecutive sections of 20 µm were used for quantification, with a counting area of 250 µm×100 µm in the dorsolateral telencephalon. For E16.5 samples, total counts were obtained by tracing the entire ventricular and subventricular zones. Sulcal counts depended on the location of the sulcus, with the counting area being set to 300 µm×100 µm. Only cells in the cortical plate were counted. For cortical layer measurements at E16.5, three measurements per ROI (counting window of 250 µm by 100 µm) and layer were taken and averaged per sample, and the measurements spanned from the bottom of the cortical plate to the top most cell. For the cortical length measurements at E16.5, the entire cortical length from the base to the dorsal midline was calculated from each cortical hemisphere and averaged to produce one value/embryo. For E16.5 Pax6 and Tbr2 cell counts in Fig. 6I-L, maximum intensity projection files imaged for both Pax6 and Tbr2 were imported into Fiji software and each channel was separately thresholded. Virtual composite stacks were then formed by merging the Pax6 and Tbr2 channels. ROI boxes were placed tangential to the ventricular surface that enclosed the ventricular zone, subventricular zone and intermediate zone. After standard watershed separation of cells, the Pax6⁺ population was counted using the analyze particles function, creating an overlay of counted cells. Counted cells co-labeled with Tbr2 were then manually removed from the total count to achieve a final Pax6⁺, Tbr2⁻ cell count. To calculate cell density in defined ROIs from posterior cortical sections, three boxes were created for Pax6 (225 µm×125 µm) and TBR2 (300 µm×125 µm), corresponding to regions shown in Fig. 6F. Density was calculated as a measure of number of cells/volume in μm³ (all sections were 10 μm). For EdU, Tbr2 and Ki67 cell counts at E16.5 and E18.5, four boxes (200 µm×108 µm) were placed in the regions annotated in Fig. 7C. For cortical length measurements at P0.5, the entire length of the dorsal cortex was calculated (rostral to the insular cortex) by tracing the Ctip2 signal in stained sections. For meningeal thickness measurements in the dorsolateral and basolateral regions, the shortest distance that is orthogonal to the pia (labeled by p75NTR) and the arachnoid (labeled by connexin 43) was measured three times, and the averages recorded. For E16.5 and E18.5 Raldh2 mean fluorescence intensity, mean gray values were obtained at three different ROIs (closest to midline, area near Raldh2 reduction and lateral) in FIJI. For cortical ratio measurements (midline/dorsolateral cortex), the mean total thickness of the midline was divided by the mean total thickness of a sulcal region along the dorsolateral cortex, or an equivalent region in controls.

Pregnant dams were removed from the mating cage 10 days after observance of the vaginal plug and then weighed daily. Upon separation, food pellets were removed and replaced with 5 g of cherry flavored Nutra-Gel diet packs (Bio-Serv) in an opaque plastic container. Food was replenished daily. At E12, the Nutra-Gel was supplemented with all-trans retinoic acid (atRA, Sigma R2625-100MG). For 0.175 mg atRA/g food, 10.5 mg of atRA was dissolved in 2 ml of corn oil and mixed with 13 g of Nutra-Gel. For 0.35 mg atRA/g food, 0.004 g of atRA was dissolved with 2 ml of corn oil and mixed with 12 g of Nutra-Gel. All preparations involving atRA were carried out in the dark, and the mice were fed daily at 6pm. Pregnant females consumed all 5 g of Nutra-Gel with atRA daily until the morning or noon on day 16 or 18, at which point the female dam was euthanized and the pups were collected. Whole brain tissue was dissected and incubated in 4% PFA overnight at 4°C before being washed with PBS and sunk in 30% sucrose for cryo-embedding.

Animals were sacrificed via transcardial perfusion with 4% paraformaldehyde (PFA). The heads were then decapitated and post-fixed overnight in 4% PFA. Hair and skin were removed before imaging. Images were obtained using a Bruker Skyscan 1272 X-ray microscope. The following scan conditions were used: Image pixel size=13.5 μm, camera=1632 columns×1092 rows, rotation step=0.4 degrees, Frame averaging=3, Filter=1 mm Al. The resulting images were reconstructed and converted to dicom format with Skyscan Ctan software. Dicom files were opened in Vivoquant for segmentation of teeth and bone from less dense soft tissues.

The authors thank Kush Desai for technical assistance, Marianne Polunas, who provided assistance from the Rutgers Research Pathology Services Core, and The Rutgers Molecular Imaging Center (D. Adler and P. Buckendahl) for assistance with skull 3D x-ray microscopy using the Skyscan 1272 micro-CT scanner (funding by NSF Major Research Instrumentation Award 1828332).

This article has an associated ‘The people behind the papers’ interview with some of the authors.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201381.reviewer-comments.pdf.