High-mobility group nucleosomal binding domain 2 protects against microcephaly by maintaining global chromatin accessibility during corticogenesis

The surface area of the human cerebral cortex undergoes dramatic expansion during late fetal development, leading to cortical folding, an evolutionary feature not present in rodents. Microcephaly is a neurodevelopmental disorder defined by an abnormally small brain, and many gene mutations have been found to be associated with primary microcephaly. However, mouse models generated by ablating primary microcephaly-associated genes often fail to recapitulate the severe loss of cortical surface area observed in individuals with this pathology. Here, we show that a mouse model with deficient expression of high-mobility group nucleosomal binding domain 2 (HMGN2) manifests microcephaly with reduced cortical surface area and almost normal radial corticogenesis, with a pattern of incomplete penetrance. We revealed that altered cleavage plane and mitotic delay of ventricular radial glia may explain the rising ratio of intermediate progenitor cells to radial glia and the displacement of neural progenitor cells in microcephalic mutant mice. These led to decreased self-renewal of the radial glia and reduction in lateral expansion. Furthermore, we found that HMGN2 protected corticogenesis by maintaining global chromatin accessibility mainly at promoter regions, thereby ensuring the correct regulation of the transcriptome. Our findings underscore the importance of the regulation of chromatin structure in cortical development and highlight a mouse model with critical insights into the etiology of microcephaly.

The surface area of the human cerebral cortex undergoes dramatic expansion during late fetal development, leading to cortical folding, an evolutionary feature not present in rodents. Microcephaly is a neurodevelopmental disorder defined by an abnormally small brain, and many gene mutations have been found to be associated with primary microcephaly. However, mouse models generated by ablating primary microcephaly-associated genes often fail to recapitulate the severe loss of cortical surface area observed in individuals with this pathology. Here, we show that a mouse model with deficient expression of highmobility group nucleosomal binding domain 2 (HMGN2) manifests microcephaly with reduced cortical surface area and almost normal radial corticogenesis, with a pattern of incomplete penetrance. We revealed that altered cleavage plane and mitotic delay of ventricular radial glia may explain the rising ratio of intermediate progenitor cells to radial glia and the displacement of neural progenitor cells in microcephalic mutant mice. These led to decreased self-renewal of the radial glia and reduction in lateral expansion. Furthermore, we found that HMGN2 protected corticogenesis by maintaining global chromatin accessibility mainly at promoter regions, thereby ensuring the correct regulation of the transcriptome. Our findings underscore the importance of the regulation of chromatin structure in cortical development and highlight a mouse model with critical insights into the etiology of microcephaly. 5 in the embryonic ventricular zone are multipotent stem cells and the major source of cortical excitatory neurons, contributing to construction of the highly evolved neocortex. RGCs in the ventricular zone divide symmetrically to self-renew, expanding the ventricular surface laterally, or divide asymmetrically to generate neurons directly or indirectly via intermediate progenitor cells (IPCs) (1). In humans, it has been suggested that detachment of RGCs from the VZ to the outer subventricular zone leads to the generation of the outer RGCs, resulting in the expansion and folding of the neocortex (2). The cell fate determination and maintenance of RGCs are closely associated with their apical-basal polarity and cell cycle length (1). Extracellular matrix and apical junction complex are among the key components that establish and maintain the apical-basal polarity of RGCs (3).

Radial glial cells (RGCs)
Dysregulation of the division of RGs could lead to one of the most severe neural developmental disorders, microcephaly. This is defined as a significantly reduced occipito-frontal head circumference of more than three S.D. values below the mean of the same age, sex, and ethnicity, because normal head sizes vary greatly among individuals. The cause of microcephaly can be congenital or environmental, autosomal-recessive (autosomalrecessive primary microcephaly, MCPH), or associated with other genetic disorders (4). One difficulty in studying the etiology of microcephaly lies in the tolerance and variance of the human genome. As a healthy individual carries more than 46 nonsense mutations on average (5), 3-24 disease-causing mutations in homozygous state (6), and multiple copy number variations (7), identifying microcephaly-associated genes is extremely challenging, not to mention the fact that the phenotype spectrum of patients with MCPH gene mutations is wider than previously suggested (4). Another difficulty lies in the evolutionary divergence between humans and mice. Although mutations in MCPH genes such as ASPM (abnormal spindlelike microcephaly-associated) have been identified in microcephaly patients, mice deficient in these genes sometimes only show mild abnormality (8 -10).
Emerging evidence has suggested the importance of chromatin remodelers in central nervous system development (11,12), whereas it remains to be investigated whether the change of global chromatin state is associated with neocortical development and whether this change could drive the differentiation of embryonic radial glia (13)(14)(15). High-mobility group nucleosomal binding domain (HMGN) is a family of nucleosomebinding proteins, of which the two major variants, HMGN1 and HMGN2, have been shown to affect histone modification and play a role in the establishment and maintenance of regulatory sites in chromatin (16). Knockout of either Hmgn1 or Hmgn2 induces several mild abnormalities in mice, due to the functional compensation for HMGN variants (17). Interestingly, Hmgn2 is predicted to be highly likely haploinsufficient in the human genome (18,19) and is known to be included in a heterozygous copy number variation duplication of seven genes in a patient with microcephaly (DECIPHER ID 256762) (20). Therefore, we designed experiments to determine whether HMGN2 plays a role in corticogenesis and whether this is accomplished by maintaining global chromatin accessibility in mice.

HMGN2 is highly conserved and abundantly expressed in embryonic brain of humans and mice
The high-mobility group (HMG) family includes the HMGN, HMGA, and HMGB genes. The HMGN protein family comprises HMGN1-5 in humans and Hmgn1, -2, -3, and -5 in mice (21). We performed protein alignments to compare the conservation of protein sequences in HMGNs among six vertebrates (Fig. 1A). HMGN2 was the most conserved protein in the HMGN family, with higher scores than the two essential protein references, Kruppel-like factor 7 (KLF7) and cyclin B1 (CCNB1) (22), suggesting an important function. We then examined the expression pattern of HMGN2 in the developing forebrain of humans and mice by immunostaining and found that it was expressed in the V-SVZ and the cortical plate in humans at

HMGN2 maintains chromatin accessibility against microcephaly
gestational week 14 (GW14) and throughout the embryonic stages in mice (Fig. 1, B and C). Specifically, RGCs expressing NES and LeX were positive for HMGN2 in human V-SVZ and the outer SVZ ( Fig. 1D) and in the mouse VZ (Fig. 1E). HMGN2 was also detected in IPCs, which were labeled by the specific markers EOMES and ASCL1, and in newborn neurons expressing the neuronal markers PSA-NCAM or DCX in both species (Fig. 1, D and  E). In cortical cells of embryonic day 0.5 14.5 (E14.5) mice and GW13 humans, the transcriptomic profiles of all canonical HMG genes show a very similar pattern (23) (Fig. S1, A-D).
In postnatal mice, HMGN2 was continuously expressed in the SVZ of the lateral ventricle and the subgranular zone of the hippocampal dentate gyrus, where adult neurogenesis takes place (Fig. S1E). In adults, HMGN2 was expressed in both GFAP ϩ cells and newborn neurons expressing PSA-NCAM in the SVZ and subgranular zone (Fig. S1F). In summary, HMGN2 is conserved not only in terms of the protein sequence but also in its expressional profile in the embryonic cortex in both humans and mice.

Ablating Hmgn2 is subviable for mice in embryonic and postnatal stage
Based on the conservation of the protein sequence and the expression pattern of HMGN2 in human and mouse brain, we hypothesized that investigation of the function of HMGN2 in mouse brain development may shed light on understanding its function in human brain development and diseases. To test this hypothesis, we obtained an Hmgn2-KO mouse line, Hmgn2 tm1a(KOMP)Wtsi , generated by the Knockout Mouse Project of the University of California (Davis, CA) (24). The mutant allele was initially generated as a nonexpressive form by inserting a poly(A) signal before the coding sequence of Hmgn2 ( Fig. 2A). The knockout efficiency was nearly 100% by E10.5, as shown by immunostaining (Fig. 2B), further verified by Western blotting on E14.5 and by RNA-Seq on postnatal day 14 (P14) (Fig. S2, A and B). We inbred the heterozygous Hmgn2 tm1a and systematically analyzed the genotypes of 250 -300 offspring to see if they followed Mendel's law (Fig. 2C). At E18.5, in contrast to the theoretical Mendelian rate of 25%, only 17.5% of the embryos recovered were KO mice, meaning that the viability of the KO embryos was only ϳ70%. By weaning, the survival rate of KO mice had decreased further to 54%. Hence, Hmgn2 is a subviable gene satisfying the definition in a previous report for the variable lethality after knockout (22,25). Our result indicates that Hmgn2 is essential for both embryonic and postnatal development with incomplete penetrance. The green signal in the KO cortex is autofluorescence from blood vessels and the pia. C, table of genotype distributions of the progeny of in-bred heterozygotic parents at E18.5 and at weaning, along with the expected Mendelian distribution. The percentage of each genotype is listed, with the corresponding number of mice in parentheses. The KO survival rate decreased dramatically at E18.5 and further decreased at weaning (comparisons assessed using the 2 test). D, photographs of the brain and body Hmgn2-KO and WT mice at P14. The cortical area of the WT hemisphere circumscribed by the dashed line was superimposed onto the KO brain to show the reduction in brain size. The threshold of microcephaly, which is defined by brain weight Ͻ (mean WT Ϫ 3S.D. WT ), is shown as the dashed line in the bar graph. Mice with brain weights below the threshold were categorized as mKO (labeled in red). Data indicate mean Ϯ S.D. (error bars) (WT, n ϭ 36; KO, n ϭ 21). E, measurements of the WT and mKO brains at P14 by length (anterior-to-posterior), width (medial-to-lateral), and whole-brain and forebrain "area" (i.e. the area occupied by the corresponding structure from photographs of the dorsal aspect of the brain of P14 mice). Units for each group of columns in order are as follows: 1 mm, 1 mm, 1 mm 2 , 1 mm 2 (WT, n ϭ 5; KO, n ϭ 6). F, coronal sections from WT and mKO P14 mice stained with DAPI (white). G, area, surface length, ventricular length, and thickness (from the ventricular surface to the pial surface) of the neocortex of P14 WT and mKO mice at an anterior-posterior position similar to that in F. mKO brains show more marked reduction laterally than radially. Units for each group of columns in order are as follows:

HMGN2 maintains chromatin accessibility against microcephaly
Hmgn2 deficiency leads to microcephaly in the mouse model The KO pups died from birth to weaning, with a peak during the first 2 days. We found that the total brain weight was significantly reduced at P14 in KO mice, which also exhibited growth retardation ( Fig. 2D and Fig. S2C), with a larger variation than the WT mice. The growth retardation was consistent with many other mouse models of depletion of MCPH genes (8 -10). The definition of human microcephaly is a head circumference that is Ͻ3 S.D. values from the mean of normal infants; here, we defined microcephalic KO (mKO) mice as those with brain weights Ͻ 3 S.D. values from the mean of WT brains. According to this criterion, we found that microcephaly was manifested in ϳ30% of the KO mice (Fig. 2D), which did not survive to P21, indicating incomplete penetrance and variable expressivity, which is often the case for human diseases (26). To better illustrate the working mechanism of how losing Hmgn2 caused microcephaly, we further analyzed and compared mKO brains with WT brains as they were abnormal vera. We found that mKO brains had decreased size, with similar reductions in the cortical anterior-to-posterior and medial-to-lateral axes (length and width in Fig. 2E). The degree of reduction in mediolateral length from anterior to posterior did not vary much (Fig. S2D), so we selected the position around bregma Ϫ0.46 to precisely measure the reduction in the length along medial-to-lateral axes (cortical length) and in cortical thickness in brain sections (Fig. 2F). We found that the neocortical thickness in mKO mice was reduced to 91% of the WT, whereas the reduction in cortical ventricular length (87% of WT) and cortical surface length (83% of WT) was greater. Because both anteroposterior and mediolateral length contribute to the lateral expansion of the cerebral cortex, the decrease in lateral expansion (74% of WT) was more responsible than radial expansion (cortical thickness, 91% of WT) for the microcephaly of mKO mice (Fig. 2H). This phenomenon was consistent with what was observed in human microcephaly patients and the ferret ASPM-knockout model (2), suggesting that the RGCs responsible for cortical expansion may be lost in mKO mice.

Microcephalic KO mice retain generally normal radial corticogenesis
During cortical histogenesis, NSCs produce neurons in an inside-out pattern sequentially from the deeper layers to the upper layers, resulting in radial expansion. We then investigated whether cell number and neuron production in a radial unit (an RGC and its progeny) is affected in P14 mKO mice. We found that neither the number of cell nuclei nor CUX1 ϩ neurons changed significantly per unit of ventricular length in mKO compared with that in WT mice (Fig. 3 (A and B) and Fig. S2H). Consistently, the production of upper-layer neurons labeled by bromodeoxyuridine (BrdU) injection at E15.5 was not changed per unit of ventricular length in Hmgn2-null mice when analyzed at P14 (Fig. 3, C and D). In addition, we found that BrdU ϩ neurons reached the correct positions in the upper layers in mKO mice (Fig. 3, C and E), and the distribution of the layer markers CUX1, CTIP2, and SATB2 in mKO mice was similar to that in WT mice (Fig.  3A), suggesting that cortical lamination was normal in mKO.
Notably, cell density in the entire cortex or different cortical regions in mKO mice was not significantly different from that in the WT; however, in layers V/VI, the cell density was higher. As the number of deeper layer cells was not changed (Fig. S2E) in mKO mice, this increase in density could be explained by the reduced thickness of layers in mKO mice (Fig. S2, F and G). Our results indicate that within an expanding radial unit, the generation and migration of neurons are not affected by Hmgn2 deficiency. However, as the ventricular length was reduced, the overall cell number was decreased, leading to a microcephalic phenotype.
To determine whether gliogenesis was affected by loss of Hmgn2, we labeled the entire lineage of E15.5 RGCs with the piggyBac-transposon system by in utero electroporation. This system, consisting of a helper plasmid expressing piggyBac transposase (PBase) and a piggyBac donor plasmid (PB-GFP), enables the stable integration of donor genes into all progeny of E15.5 RGs after plasmid transfection (27). Glia were distinguished from neurons according to morphology labeled by GFP. The ratio of glial cells in GFP ϩ cells was higher in P14 mKO mice than in the WT (Fig. 3, F and G). In addition, stunted littermates with KO already exhibited microcephaly ( Fig. S2I) before gliogenesis reached its peak (28,29), indicating that microcephaly in Hmgn2 mKO mice is not caused by glia loss. We found very little cleaved caspase-3 staining in P4 and P14 mKO cortex (Fig. S2J), suggesting that cell death does not account for the microcephaly in Hmgn2-null mice. Taken together, due to a reduction in overall lateral expansion, mKO mice exhibit microcephaly despite the generally normal production of cortical cells within a radial unit.

Loss of Hmgn2 results in enhanced differentiation and altered cell division of radial glial cells
As Hmgn2-deficient mice were underrepresented at E18.5, we determined whether microcephaly had already taken place by that time. The average brain weight at this stage was significantly lower in the KO group, and some mice also met the criteria for mKO (Fig. 4, A and B). The cortical area and surface length were significantly lower in mKO mice, whereas cortical thickness and ventricular length did not differ from the WT (Fig. 4C). Both the thickness of VZ and SVZ expanded, whereas the cortical plate was thinner in mKO mice (Fig. 4D). Consistently, the number of deeper-layer neurons (CTIP2 ϩ ) decreased (Fig. 4, E and F), suggesting that early neurogenesis (before E14.5) was attenuated. Whereas the RG (PAX6 ϩ ) pools did not change in mKO mice, the IPCs (EOMES ϩ ) and the transitional cell type (PAX6 ϩ EOMES ϩ ) expanded in E18.5 mKO mice, suggesting that the self-renewal of RGs was attenuated (Fig.  4, G and H). Notably, ectopic IPCs and PAX6 ϩ EOMES ϩ transitional cells that appeared outside the V-SVZ dramatically increased in mKO mice (Fig. 4H), suggesting that the progenitor cells were displaced from their original niche. The precocious withdrawal of VZ neural progenitors from the germinal zones is reminiscent of the ASPM Ϫ/Ϫ ferret model of microcephaly.
We next assessed the proliferative behavior of cortical radial glia in E18.5 mKO cortex. First, we found that the proportion of progenitor cells on the surface of the ventricle in M-phase (labeled by the phosphorylated histone H3, pH3) HMGN2 maintains chromatin accessibility against microcephaly was higher in mKO than that in WT mice (Fig. 4, I and J). We then measured the fraction of dividing cells along the ventricular surface in prophase and pro-metaphase versus those in metaphase, anaphase, and telophase, reflecting the progression of mitosis. We found that the fraction of prophase/ pro-metaphase cells increased in mKO mice (Fig. 4, IЉ and K), indicating that the duration of prophase/pro-metaphase was prolonged. Therefore, the increase in the number of pH3 ϩ RGCs was more likely caused by cell-cycle delay rather than increased proliferation. Furthermore, more ventricular RGCs with an oblique spindle orientation were detected in mKO than in WT mice (Fig. 4L), indicating that fewer ventricular RGCs underwent symmetrical cell divisions, which may lead to decreased self-renewal of RGs and further the decrease in cortical lateral expansion. RGCs with more oblique orientation of the cleavage plane could leave the ventricular zone (30), explaining the detachment of IPCs and PAX6 ϩ EOMES ϩ transitional cells.

Hmgn2 deficiency impairs gene expression by diminishing global chromatin accessibility
To determine how HMGN2 protects against microcephaly, we evaluated the global chromatin accessibility and its influ-ence on the transcriptome of WT and mKO cortical cells at P14 using an assay for transposase-accessible chromatin followed by sequencing (ATAC-Seq) (31) and RNA-Seq. To identify the features of HMGN2 targeting specificity, we analyzed the deepsequencing data of ChIP-HMGN2 in neural progenitor cells differentiated from mouse embryonic stem cells based on a previous study (17). We found that HMGN2 predominantly bound to gene promoter regions (54% of 25,594 total peaks of 12,684 genes Fig. S3A), and this binding was associated with greater chromatin accessibility and higher transcriptional activity in WT P14 cortical cells (Fig. S3, B and C).
Using ATAC-Seq, we captured 29,217 peaks as open chromatin regions (OCRs; Fig. 5A) in P14 WT cortical cells. Compared with the WT, mKO cortical cells had significantly fewer OCRs (22,511 peaks, 77% of the WT) and dramatically reduced signals of ATAC-Seq at transcription start sites (Fig. 5, A and  B). The reduction of OCRs mainly occurred at promoters, consistent with the binding preferences of HMGN2. Furthermore, the chromatin accessibility greatly reduced at HMGN2-binding sites (Fig. 5C), indicating that HMGN2 indeed maintains the OCRs it binds accessible for regulatory elements, such as transcriptional factors (TFs). The overall reduction in chromatin accessibility resulted in the up-regulation of 798 genes and

HMGN2 maintains chromatin accessibility against microcephaly
down-regulation of 592 genes revealed by RNA-Seq (Fig. 5D). The number of up-regulated genes that were directly targeted by HMGN2 was nearly the same as that of down-regulated genes (Fig. 5E), suggesting that during corticogenesis, HMGN2 serves as a modulator of transcription rather than an activator or repressor. In agreement, compared with all expressed genes, the chromatin accessibility of the transcription start sites of HMGN2-targeted genes (both the up-and downregulated) was more susceptible to Hmgn2 depletion (Fig.  5F). Chromatins of differential expressed genes that were targeted by HMGN2 were not as closed as all targeted genes, probably because this relatively open state was beneficial for, or even the consequence of, the approach of TFs and chromatin remodelers to modulate transcription. Taken together, the binding of HMGN2 guarded genome-wide chromatin accessibility predominantly around transcriptional start sites, ensuring both transcription activation and suppression of targeted genes during corticogenesis.

HMGN2 regulates functional gene sets to protect against microcephaly
We next looked for the causes of microcephaly after HMGN2 ablation using gene ontology analysis and gene set enrichment analysis (GSEA). The functional classification of up-regulated genes by gene ontology analysis resulted in an enrichment of terms relating to neuronal activity, suggesting dysregulated neuronal differentiation (Fig. 6A). The terms proteinaceous extracellular matrix, cell adhesion, and apical plasma membrane were highly enriched in the genes that were down-regulated after Hmgn2-KO, and genes for apical junction assembly were generally down-regulated in mKO cells (Fig. 6B). The above data explained why the spindle orientation of dividing RGs and the displacement of neural progenitor cells occurred in E18.5 mKO cortex, given that apical junctions and extracellular matrix of RGCs play an important role in maintaining RG cell fate (32). Surprisingly, neither the overall nor individual expression level of microcephaly-associated genes

HMGN2 maintains chromatin accessibility against microcephaly
was decreased in mKO compared with WT cells (Fig. S3D). Rather, the expression of genes responsible for stem cell proliferation was down-regulated in mKO cells, indicating that the appearance of increased numbers of neural progenitor cells in V-SVZ of E18.5 mKO cortex was unlikely to be the result of an increase in proliferation. We found that 15 essential genes identified in a mouse knockout study (22) were down-regulated in mKO cells (Fig. S3E), elucidating the cause for the lower survival rate of Hmgn2-null mice.
As 1000 genes greatly changed their expression in mKO cortical cells, and dozens of genes are responsible for cell proliferation, it is conceivable that HMGN2 regulates a cohort of downstream effectors during neocortical development. By searching genes from the GO terms "cilium movement" and "cell proliferation," We found three possible direct downstream effectors that meet the following three criteria: expressed in embryonic neocortical neural progenitor cells, the promoters of which HMGN2 binds to, and with decreased chromatin accessibility and transcription level in mKO cells (Fig. 6, C and  E). Klf4 and Jag1 are required for the maintenance of radial glia and cell proliferation (33,34). Cep85, on the other hand, protects centriole duplication, PLK4 activation, and the efficient centriolar targeting of STIL and SASS6, both of which are microcephaly-associated genes (35). On the other hand, proneural genes Neurod1, Neurod2, and Neurod6 were all up-regulated (only Neurod1 was directly targeted), but without clear changes in ATAC-Seq signal. As NeuroD genes are expressed mainly in postmitotic neurons and ectopic expression of NeuroD genes in progenitor cells causes cell cycle withdraw by up-regulating cell cycle inhibitors and down-regulating cell cycle progression factors (36 -38), they were also likely to be responsible for the mKO phenotype.
We further explored whether HMGN2 binds to the sites where specific TFs are located during neural stem cell development and whether these TFs bind the three predicted direct downstream effectors. We identified abundant and significant enrichment of KLF7, RBPJ, retinoid X receptor (RXR), and NFYA at the HMGN2-binding sites using de novo motif analysis (Fig. 6D). Klf7 is an essential gene for mouse embryonic development and important for neuronal morphogenesis (22,39,40). RBPJ is a TF that plays a central role in Notch signaling and is essential for the maintenance of neural stem cells during corticogenesis (41). RXRs affect proliferation of neural progenitors (42) and are among the receptors of retinoic acid signaling which is a key signaling controlling the symmetric to asymmetric division of neural stem cells during corticogenesis (43). NF-Y is important for mouse early neural development, cell proliferation, and chromatin accessibility (44 -46). Consistently, we found that these motifs were predicted to be located within the HMGN2binding sites of the three potential target genes (Fig. 6E).
In conclusion, HMGN2 binds mainly to promoter regions in the genome, maintaining the global chromatin accessibility of P14 cortical cells for proper transcription. Furthermore, we propose that HMGN2 preferentially binds to the sites occupied by TFs of key signaling pathways of RGC development, facilitating the approach of these TFs to the downstream executors, and consequently maintains their transcription as well as functions against microcephaly.

Discussion
In this study, we suggest that the expression of HMGN2 in neocortex is required for maintaining cortical volume, especially proper lateral expansion. HMGN2 preserves neurogenesis by maintaining the self-renewal of apical RGCs, possibly by maintaining their localization and mitotic progress. We demonstrate that HMGN2 maintains chromatin accessibility in

HMGN2 maintains chromatin accessibility against microcephaly
cortical cells, especially at the promoters of genes responsible for proliferation and cell polarity.
Humans and mice are conserved in genome and brain structure, which is often the basis for making mouse models to study the etiology of human diseases. For example, KO mouse models of primary microcephaly-associated genes were generated to study the cellular and molecular mechanisms underlying human microcephaly. However, the divergence between the two species results in differences in phenotypes in some of the mouse models. Consequently, investigators have often confused "microcephaly" with "significant reduction of brain size." A recent study reports that a ferret model with ASPM knockout manifested patient-like microcephaly better than mouse models, as it has normal radial expansion of the neocortex but reduced lateral cortical expansion (2). By carefully measuring the S.D. in brain size of dozens of WT mouse brains and comparing it with Hmgn2-KO brains according to the human definition of microcephaly (4), we showed that the Hmgn2-KO brain not only had a "significant reduction of size," but also exhibited "microcephaly." The microcephalic KO brain showed impaired lateral expansion and almost normal radial expansion similar to the ferret model. As ferrets are much bigger, more expensive, and have a longer lifespan, the Hmgn2-KO mice model offers a better option to study the etiology of microcephaly.
The investigators in the ASPM-KO ferret study observed the premature displacement of ventricular radial glia to the outer subventricular zone and suggest that this is the cellular mechanism that drives human primary microcephaly (2). However, it is not clear what is the cellular cause of the displacement, and it would be interesting to know how the increase of total neural progenitor cells caused microcephaly in ferrets. Here, we found a cellular phenotype of marked displacement of PAX6 ϩ EOMES ϩ transitional cells and IPCs occurring in E18.5 mKO cortex, which is similar to that found in the ASPM-KO ferret. Importantly, we found prolonged mitosis of RGCs and decreased expression of many stem cell proliferation genes in the mKO brain, explaining how the apparent increase in the number of progenitor cells causes microcephaly. Furthermore, we found altered spindle orientation in mKO RGs, which may trigger the detachment of progenitor cells and the decreased selfrenewal of RGCs. Consistently, altered spindle orientation, detached progenitor cells, and increased M-phase RGCs are also observed in many other mouse models of primary microcephaly (47)(48)(49). Thus, our Hmgn2-KO mouse model bridges the gap between the cellular behaviors of the human-like ferret model and the cellular mechanism of many mouse models for primary microcephaly.
Although chromatin remodelers are known to be required for the maintenance of radial glia, the notion that embryonic RG differentiation is driven by global restriction of the chromatin state is not well-established, because the tools measuring global chromatin state are quite limited (14,15). By using ATAC-Seq for the easy capture of open chromatin regions  (23)). E, representative ATAC-Seq and RNA-Seq tracks for the genes in P14 WT and mKO cortical cells shown in C, accompanied by HMGN2 ChIP-Seq (line 3) for neural progenitor cells, histone H3 Lys-4 methylation (line 1) and histone H3 Lys-27 acetylation (line 2) ChIP-Seq data, DNase-Seq (line 4) data for E14.5 whole-brain from ENCODE, and predicted binding sites for KLF7 (blue), RBPJ (red), RXR (black), and NFYA (green).

HMGN2 maintains chromatin accessibility against microcephaly
genome-wide in WT and mKO cells and integrating the ChIP-Seq and RNA-Seq data, we supported this idea by showing that the ablation of Hmgn2 induced a dramatic reduction of chromatin accessibility and increased RG-to-IPC transition. Cell type-specific ATAC-Seq will be required in future studies, although fresh RG and IPCs cannot be differentiated by flow cytometry at present due to the limitation of cell surface markers.
The most-studied HMG family members, HMGAs and HMGBs, are known to control transcription and epigenetics through regulation of chromatin. HMGAs bind to AT-rich regions of the minor groove of the DNA, compete, and displace linker histone H1 from the chromatin, thus unfolding the chromatin (50). HMGBs bind to DNA without sequence specificity and remodel nucleosomes by binding and remodeling various ATP-dependent chromatin-remodeling machineries as well as histone chaperones (51). HMGNs modulate transcription in complex ways. They bind only nucleosomal DNAs mainly at open chromatins without sequence specificity, but in a tissue-specific manner (52). HMGNs regulate transcription by competing with histone H1 for its targets, by affecting histone modifications, by inhibiting ATP-dependent chromatin remodeling, or by specific interactions with certain transcription factors; all of these could result in either transcription enhancement or inhibition (52). The binding specificity of HMGN2 in embryonic neural progenitors and its roles in regulating transcription globally and locally were quite elusive before. By careful examination of the multi-omics data, we identify that in embryonic neural progenitor cells, from a global view, HMGN2 is predominantly located at promoter regions, and the binding sites are abundant in motifs of several TFs functioning during RGC development. The occupancy of HMGN2 thereby maintains global chromatin accessibility, keeping genes transcriptionally activated or repressed in a similar amount. At the level of individual genes, we pinpoint three potential direct targets of HMGN2 in RGC and propose that HMGN2 preserves transcriptional activation of these genes by keeping chromatin accessibility at their promoters to allow control of governing TFs. These in-depth analyses enrich the current understanding of transcriptional control through regulation of chromatin architecture.
Our study encountered a situation not rare but much less reported: incomplete penetrance and variable expressivity. Not all of the Hmgn2-deficient mice meet the criteria of microcephaly, although mouse models of other microcephaly-associated genes have similar problems (9,10,53). This limitation prevented us from linking the genotype of Hmgn2 depletion to the microcephalic phenotype at earlier neurogenic stages, as the majority of KO brains were not far smaller than the mean of the WT by E18.5. This relatively mild embryonic phenotype could be explained by the tolerance and redundancy of the genome, as patients carrying the same mutations have different types and levels of brain malformation (54 -56). Another explanation is functional redundancy in the HMGN family. HMGN1 and HMGN2 normally target different genes but can compensate for the binding to chromatin for each other to maintain the chromatin landscape without being up-regulated in embryonic stem cells and mouse embryonic fibroblasts (17). A further analysis of double or triple knockout of Hmgns would help illu-minate the functional role of HMGNs in cortical neurogenesis. In conclusion, our work demonstrates that HMGN2 promotes lateral cortical expansion by maintaining global chromatin accessibility against microcephaly.

Biological samples
Human brain samples at GW14 were obtained from a tissue bank in Beijing Tian Tan Hospital of Capital Medical University, following the instructions of Tian Tan Hospital and approved by the Institutional Review Board of Tsinghua University (Project 20160013). The study abides by the Declaration of Helsinki principles.

Animals and plasmid construction
The Hmgn2 tm1a(KOMP)Wtsi line was generated by the Knockout Mouse Project (KOMP) Repository of the University of California (Davis, CA) (24). All of the information of this knockout-first construct has been published online. Briefly, in the target allele, the first intron was substituted by the FRT-En2SA (splicing acceptor)-IRES-LacZ-poly(A)-loxP-hbactPneo-poly(A)-FRT-loxP cassette, and a third loxP was inserted in the sixth intron, so that only the first 5 amino acids of HMGN2 could be fused with lacZ and expressed. The Hmgn2 heterozygote strain was created on the C57BL/6N background. Sperm of heterozygotic mice was purchased from KOMP, and in vitro fertilization was conducted in the animal care facility at the Center of Biomedical Analysis in Tsinghua University, with C57BL/6J females as egg donors. The mutant allele was then kept in the C57BL/6J strain by subsequent mating with C57BL/6J WT mice. Only progeny from inbred heterozygotic mice were tested for genotype distribution. As mKO mice were a subpopulation of the progeny of inbred heterozygotic mice, to increase the numbers of mKO mice, KO mice were also generated from pairs of apparently normal KO parents. WT mice were generated in a similar way.
Mice were bred and kept in the animal care facility at the Center of Biomedical Analysis in Tsinghua University. All animal protocols were approved by the IACUC of Tsinghua University and conducted in accordance with guidelines of the IACUC. The laboratory animal facility of Tsinghua University has been accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. For staging embryos, the day when a vaginal plug was detected before noon was defined as E0.5. All adult mice used were between 6 and 12 weeks old, and the postnatal ages of pups are indicated throughout. PBase (to provide transposase) and PB-IRES-EGFP plasmids were constructed as described previously in the experimental procedures (57).

Immunostaining
Embryonic heads (E10.5-E18.5) were fixed at 4°C in 4% paraformaldehyde (PFA) in 0.1 M PBS overnight. Postnatal pups and adults were perfused transcardially with PBS and then 4% PFA, followed by post-fixation of isolated brains in 4% PFA overnight at 4°C. After fixation, the brains were equilibrated in 30% sucrose and then embedded and frozen in Tissue Tek OCT HMGN2 maintains chromatin accessibility against microcephaly (Sakura Finetek, Torrance, CA) for cryosectioning. Coronal cryostat sections were cut at 20 m for embryonic tissue and at 50 m for postnatal brain.
For staining, citrate buffer antigen retrieval was applied to the sections using the solutions (catalog no. P0088 or P0083) following the manufacturer's protocols (Beyotime). Then the cryosections were blocked in 5% BSA/PBST (1ϫ PBS and 0.3% Triton X-100) and incubated with primary antibodies at 4°C for 2 days. After washing with PBST, the sections were incubated with fluorescence-conjugated secondary antibodies (Alexa Fluor) overnight and then mounted with Fluoromount GTM (SouthernBiotech). The antibodies are listed in Table S1.

Image analysis and cell counting
Images were acquired using a slide scanner for fluorescence (Zeiss Axio Scan, Z1), laser-scanning confocal microscopes (Zeiss LSM710 or LSM780 or Leica TCS SP5), or the cameras on a smartphone (iPhone 7 Plus). Images were analyzed using Zeiss software ZEN2 or ImageJ. Figures were processed using Microsoft PowerPoint and assembled with Adobe Illustrator CC 2017. Total marker-positive cells were counted in 1-4 sections (two imaging fields per section) in regions that were anatomically matched between experimental groups.

Cell dissociation
Brain cortices were dissected and digested into a single-cell suspension using 10 units/ml papain (Worthington, catalog no. LS003126) for 30 min at 37°C in Dulbecco's modified Eagle's medium as described previously (57).

BrdU labeling and detection
BrdU (50 mg/kg) was injected intraperitoneally into pregnant mice to label S-phase cells of E15.5 embryos. At postnatal day 14, the litters were sacrificed, perfused, and fixed. Brains were removed and cryosectioned. Frozen sections were processed for detection of BrdU (after 2 N HCl treatment) using immunofluorescence.

In utero electroporation
Timed-pregnant mice were anesthetized with isoflurane and kept warm on a heated pad. The uterine horns were exposed, and 0.5-1 l of plasmid solution (2-3 mg/ml) was injected into the lateral ventricle of the E15.5 brain with a fine glass micropipette. Embryonic heads were then clamped between tweezer-type disc electrodes 5 mm in diameter (LF650P5; BEX). Five 36-V electrical pulses lasting 50 ms at intervals of 999 ms were delivered using an electroporator (CUY21VIVO-SQ; BEX). The uterus was then returned to the abdominal cavity to allow normal embryonic development. The phenotypes were analyzed at P14.

ATAC-Seq and RNA-Seq
WT or mKO brains at P14 were sliced on a vibratome in hibernation medium (58). Only cortices dorsal to the hippocampus between Bregma Ϫ0.22 to Ϫ2.7 were isolated. These cortices were digested into a single-cell suspension for further library preparation. The sequencing depth was 7.4 -30 million reads for each replicate.
For RNA-Seq, cells were lysed with 1 ml of TRIzol reagent (Life Technologies, Inc., catalog no. 15596018) according to the manufacturer's instructions. Library construction and sequencing were conducted by ORI-GENE using the NEB Next Ultra Directional RNA Library Prep Kit for Illumina and the HiSeq X Ten System. The sequencing depth was between 46.0 and 134.9 million reads for each replicate.
For ATAC-Seq, cells were further washed twice. Five thousand cells were used for ATAC-Seq library preparation, as described previously, with minor modifications (59). In brief, samples were lysed in lysis buffer (10 mM Tris-HCl (pH 7.4), 10 mM NaCl, 3 mM MgCl 2 , 0.1% NP-40, and protease inhibitor mixture (Sigma, P8340) for 10 min on ice to prepare nuclei. The nuclei were then spun at 700 ϫ g for 3 min to remove the supernatant. They were then incubated with the Tn5 transposome solution (Vazyme Biotech) at 37°C for 45 min. Stop buffer was added directly into the reaction to end the tagmentation. PCR was performed to amplify the library for 12 cycles according to the manufacturer's instructions. Libraries were selected with 1.2ϫ AMPure (Beckman) beads after PCR and then sequenced on Illumina Hiseq X Ten System following the manufacturer's instructions.

Bioinformatics
FastQC and Cutadapt were used for filtering and trimming RNA-Seq reads for further analysis. Reads were aligned to the mouse genome (UCSC mm10) with hisat2 and SAMtools for RNA-Seq and with Bowtie2 for ATAC-Seq and ChIP-Seq. Stringtie and DESeq2 were used for assembling transcripts and analyzing the differential expression of genes. The differentially expressed genes (DEGs) were identified using a threshold of FDR Յ 0.05 with ͉log 2 (-fold change)͉ Ͼ 1.
Peak calling of ATAC-Seq and ChIP-Seq was performed using MACS. DeepTools was used to analyze and visualize ATAC-Seq signals around transcription start sites. The gene ontology (GO) enrichment analysis of DEGs was performed using DAVID (https://david.ncifcrf.gov/) (60,61). GSEA (http://software.broadinstitute.org/gsea/index.jsp) 6 (62) was performed using all genes expressed with default parameters (permutation number ϭ 1000, "log2_ratio_of_classes" to calculate ranking of the genes). The criteria of significant enrichment were determined as NES Ͼ1, p value Ͻ 0.05, and FDR Ͻ 0.05. To visualize the ATAC-Seq, ChIP-Seq, and RNA-Seq signals in the Integrative Genomics Viewer, we extended each read by 100 bp and calculated the coverage for each base.

Quantification and statistical analysis
Sequencing data were analyzed based on the negative binomial distribution, whereas data characterizing cellular phenotypes were assumed to follow normal distribution without formal tests. For Figs. 2C and 4L, the 2 test was used for comparisons. Other statistical analyses were performed using Prism (GraphPad Software) with unpaired Student's t test except for Fig. 2E, where paired Student's t test was used. Unless otherwise stated, data are presented as mean Ϯ S.E. p Ͻ 0.05 was considered significant. The replicate numbers for each graph are stated in the figure legends.