Expression of Genes in the 16p11.2 Locus during Development of the Human Fetal Cerebral Cortex

Abstract The 593 kbp 16p11.2 copy number variation (CNV) affects the gene dosage of 29 protein coding genes, with heterozygous 16p11.2 microduplication or microdeletion implicated in about 1% of autism spectrum disorder (ASD) cases. The 16p11.2 CNV is frequently associated with macrocephaly or microcephaly indicating early defects of neurogenesis may contribute to subsequent ASD symptoms, but it is unknown which 16p11.2 transcripts are expressed in progenitors and whose levels are likely, therefore, to influence neurogenesis. Analysis of human fetal gene expression data revealed that KIF22, ALDOA, HIRIP3, PAGR1, and MAZ transcripts are expressed in neural progenitors with ALDOA and KIF22 significantly enriched compared to post-mitotic cells. To investigate the possible roles of ALDOA and KIF22 proteins in human cerebral cortex development we used immunohistochemical staining to describe their expression in late first and early second trimester human cerebral cortex. KIF22 protein is restricted to proliferating cells with its levels increasing during the cell cycle and peaking at mitosis. ALDOA protein is expressed in all cell types and does not vary with cell-cycle phase. Our expression analysis suggests the hypothesis that altered neurogenesis in the cerebral cortex contributes to ASD in 16p11.2 CNV patients.


Introduction
Large, recurrent Copy Number Variations (CNVs) are implicated in many neuropsychiatric disorders including autism spectrum disorders (ASD), epilepsy, intellectual disability (ID), and schizophrenia (McCarthy et al. 2009;Girirajan and Eichler 2010;Levy et al. 2011;Sanders et al. 2011;Malhotra and Sebat 2012). The 16p11.2 CNV (OMIM 611913) encompasses a 593 kb DNA sequence in the p11.2 region of human chromosome 16 (BP4-BP5). This region harbors 29 protein coding genes and is strongly linked to neurodevelopmental disorders (NDDs) including ASD (Kumar et al. 2008;Bijlsma et al. 2009;Rosenfeld et al. 2010;Shinawi et al. 2010;Zufferey et al. 2012). This 16p11.2 region is flanked by two homologous 147kbp sequences that arose after the evolutionary divergence of humans from other primates, generating a hot-spot for mis-aligned recombination that explains the high frequency of the 16p11.2 CNV in the human population and also the high frequency of de novo 16p11.2 CNV (Nuttle et al. 2016). In humans, the 16p11.2 microdeletion is associated with transient infant brain overgrowth (macrocephaly) and focal thickening of the cerebral cortex, while the 16p11.2 microduplication is associated with reduced brain size (microcephaly) (Qureshi et al. 2014;Blackmon et al. 2018). The early manifestation of anatomical phenotype in newborns, along with the onset of ASD symptoms in infancy, suggests crucial roles for 16p11.2 genes during neural development. 16p11.2 is the most prevalent CNV associated with ASD, approximately 1% incidence, making this CNV particularly intriguing and providing motivation for investigating the role played by 16p11.2 genes in brain development and function (Weiss et al. 2008). Available lines of evidence from 16p11.2 rodent models, 16p11.2 patient derived lymphoblastoid cell lines, and human induced pluripotent stem cells genetically engineered to harbor the 16p11.2 CNV indicate that all 16p11.2 messenger RNAs' (mRNAs') levels reflect the altered gene dosage of 16p11.2 genes (50% in microdeletion and 150% in microduplication heterozygotes) (Horev et al. 2011;Blumenthal et al. 2014;Pucilowska et al. 2015;Tai et al. 2016). This indicates that multiple 16p11.2 transcript levels are affected by the 16p11.2 CNV and that the pathology of the 16p11.2 CNV could stem from altered dosage of one or more them.
The cellular mechanisms by which the 16p11.2 CNV causes the patient phenotype are poorly understood. One plausible hypothesis is that disrupted neurogenesis causes changes in neuronal output, which produce a brain with abnormal cell number or composition and that this contributes to the 16p11.2 pathology. Consistent with this hypothesis, the 16p11.2 deletion mouse model exhibits proliferation defects in cortical progenitors during prenatal brain development and subsequently develops ASD-like symptoms (Horev et al. 2011;Pucilowska et al. 2015). However, it is unknown which of the proteins produced by 16p11.2 CNV genes are expressed by progenitor cells in the developing human cerebral cortex and are therefore candidates for regulating neurogenesis.
Excitatory (glutamatergic) neurons in the human cerebral cortex are derived from progenitors that reside in the ventricular and subventricular zones of the dorsal telencephalon while inhibitory (GABAergic) interneurons are derived from progenitors that reside in the ventral telencephalon and migrate into the cerebral cortex (Clowry et al. 2010) (Bystron et al. 2008) (Ma et al. 2013) Here we focused on the potential for the 16p11.2 CNV to affect neurogenesis of excitatory neurons in the developing human cerebral cortex by identifying 16p11.2 genes that are highly expressed in cerebral cortex progenitors in the ventricular zones and down-regulated as cells become post-mitotic. We analyzed previously published human fetal cortex single cell RNA sequencing (scRNA-seq) data (Pollen et al. 2015) (Zhong et al. 2018) to identify candidate genes and characterize their expression in sections of developing human fetal cerebral cortex from the late first and early second trimester.

Human Tissue
Human embryos ranging in age from 12 to 16 post-conceptual weeks (PCW) were obtained from the MRC/Wellcome-Trust funded Human Developmental Biology Resource at Newcastle University (HDBR, http://www.hdbr.org/) with appropriate maternal written consent and approval from the Newcastle and North Tyneside NHS Health Authority Joint Ethics Committee. HDBR is regulated by the UK Human Tissue Authority (HTA; www.hta.gov.uk) and operates in accordance with the relevant HTA Codes of Practice.

scRNA-seq Analysis
The publicly available scRNA-seq data sets (Pollen et al. 2015;Zhong et al. 2018) were used to identify candidate genes. Prior to dataset publication the reads were aligned, we normalized the RPKMs as log (x + 1). Analysis was performed using R studio.
To determine genes with significant changes a Wilcox test by FindAllMarkers in Seurat package was used. Monocle2 R package was used to order cells in pseudotime. To identify cell-cycle phase specific transcripts we used function CellCycleScoring from Seurat R package.
For colourmetric stains, sections were incubated 1 h at room temperature with biotinylated secondary antibody (1/200) followed by incubation for 1 h with ABC (Vector Labs) and developed with diaminobenzidine solution (Vector Labs), washed, counterstained with nuclear fast red, dehydrated and then mounted using DPX.

Microscopy and Imaging
DAB and in situ hybridisation images were taken using a Leica DMNB microscope with an attached Leica DFC480 Camera. Fluorescence images were obtained with a Leica DM5500B epifluorescence microscope with a DFC360FX camera. Confocal images were obtained using Nikon A1R FILM microscope and analyzed in ImageJ.

Image Analysis and Quantification
For DAB stains and in situ hybridisation the images were stitched in ImageJ using the stitching plugin (Preibisch et al. 2009).
For KIF22 analysis of DAB stains rectangular counting boxes (34x88μm) were overlaid across the section. Using ImageJ cell counting plugin cells in each box were counted and denoted KIF22+ (brown) or KIF22-(red). The distinction between the regions (VZ, SVZ, and IZ/CP) was determined anatomically by cell density. The count for each box was averaged with other boxes in the region to provide the final value.
For analysis of KIF22/KI67 double staining counting boxes (×20145 μm) were overlaid over the VZ and SVZ (determined based on cell density). For determining intensity cells were randomly selected on the DAPI channel, the nucleus outlined and intensity of KIF22 and KI67 recorded. 20 cells were selected per box and the counts from individual boxes combined to give final values.
For subcellular ALDOA analysis counting boxes (20 × 145 μm) were overlaid over the VZ and SVZ. Cells were randomly selected on the DAPI channel, far enough apart to ensure their cytoplasm would not overlap, the Z plane through the center of the cell was used and the nucleus outlined. The KI67 and ALDOA intensity was measured constituting the nuclear value. To obtain ALDOA cytoplasmic intensity the nuclear outline was duplicated and extended 4 pixels allowing a reading of just the cytoplasmic area to be obtained (see Fig. 6e). This was performed for 10 cells in each box and the counts from individual boxes combined to give final values.

Data Analysis and Statistics
Where error bars are shown they are expressed as mean ± SEM. Boxplots show median and upper and lower quartiles. Statistical comparison between two groups was performed with a t test. Statistical comparison between more than two groups was performed with ANOVA followed by post hoc test. P < 0.05 was considered statistically significant. Analysis was performed using GraphPad Prism.

Analysis of scRNA-seq Data Identifies KIF22 and ALDOA as Progenitor-Enriched 16p11.2 Transcripts in the Developing Human Fetal Cerebral Cortex
The 16p11.2 CNV involves microduplication or microdeletion of a 593 kb locus on human chromosome 16 containing 29 protein coding genes (Fig. 1a). The aim of the current study is to identify 16p11.2 genes that are potential candidates for being involved in neurogenesis in the developing human cerebral cortex (Fig. 1b,c) and whose altered dosage in 16p11.2 microdeletion or microduplication patients may disrupt neurogenesis and contribute to the CNV phenotype. We reasoned that 16p11.2 genes important for neurogenesis would be highly expressed in proliferating progenitor cells and downregulated as cells became postmitotic.
We took advantage of a published single cell RNA-sequencing (scRNA-seq) data-set acquired from 393 cells of the ventricular zone (VZ) and subventricular zone (SVZ) of gestational week (GW) 16-18 human fetal cerebral cortex (equivalent to post conception week (PCW) 14-16) to perform an unbiased screen to identify 16p11.2 transcripts that matched this expression profile (Pollen et al. 2015). Dimensional reduction of the scRNA-seq data separated the cells into three clusters based on transcriptome similarity ( Fig. 1d-each dot on the tSNE plot represents an individual cell) that were subsequently identified as the three cardinal cell classes of progenitors (blue), postmitotic neurons/principal cells (red) and interneurons (green), by expression of cell-type specific transcripts. We next used the monocle2 R package to order the cells in pseudotime using the normalized expression levels of selected differentially expressed genes (DEGs) as input to order the cells (Trapnell et al. 2014;Qiu et al. 2017) (Fig. 1e) moving from the progenitor state (left) to post mitotic state (right) along the X-axis. We plotted the average expression of each 16p11.2 transcript at each pseudotime-point on the Y-axis. We found that two genes, KIF22 (blue line) and ALDOA (brown line), were notable for having high expression in progenitors that declined as cells became post-mitotic. A Wilcox test identified ALDOA and KIF22 as the only 16p11.2 transcripts that were significantly higher in progenitor than neuronal populations (P < < 0.05). Although not significantly enriched in progenitors HIRIP3 (orange line), MAZ (red line), and PAGR1 (green line) were expressed in progenitors at higher levels than the remaining 16p11.2 transcripts (shown as gray lines), many of which were barely expressed at all.
Violin plots of the numbers of cells expressing different levels of KIF22 mRNA in the different cardinal cell classes show that KIF22 is expressed predominantly in progenitors ( Fig. 1f ) and mapping the expression level of KIF22 onto the tSNE plot ( Fig. 1g) revealed that KIF22 expression is highest in a subset of the progenitor cluster (arrow in Fig. 1g) with a substantial proportion of progenitor cells expressing relatively low levels of KIF22. Very few post-mitotic neurons, both interneurons and principal cells, express appreciable levels of KIF22 (Fig. 1f,g). The expression of KIF22 in a subset of progenitors prompted us to ask whether its expression was related to the cell-cycle phase. We used the expression of cell-cycle phase specific transcripts using function CellCycleScoring from Seurat R package to divide the cells into three classes (Macosko et al. 2015;Tirosh et al. 2016), G1/S, G2/M, and post-mitotic neurons, and compared KIF22 transcript levels between these three groups using a violin plot (Fig. 1h). We found that the majority of cells in G2/M phase expressed higher levels of KIF22 (red plot), cells in G1/S expressed lower levels (blue plot) while the vast majority of post-mitotic cells expressed low levels of KIF22 (green plot). Similar analysis for ALDOA show that while a greater proportion of cells expressing the highest levels of ALDOA are progenitors (blue plot) there are also a substantial number of principal cells (red plot) expressing similarly high levels of ALDOA transcripts although very few interneurons (green plot). Mapping ALDOA expression level onto the tSNE plot ( Fig. 1j) shows cells expressing high levels of ALDOA are evenly distributed throughout the progenitor cluster with appreciable numbers of principal cells expressing high levels of ALDOA and a much lower proportion of interneurons. In contrast to KIF22, there is no clear difference in the partitioning of ALDOA expression level between different phases of the cell-cycle (Fig. 1k). We performed the same analysis on another developing human cortex scRNAseq data set that spanned a wider developmental interval (PCW10-28) and also included cells from all layers of the cerebral cortex (Zhong et al. 2018). This analysis produced similar results. KIF22 transcripts showing them to be enriched in progenitors (Supplementary Figure 1b, e) and in G2/M phase of the cell cycle ( Supplementary Fig. 1g). Although ALDOA transcript levels were highest in progenitors they did not decline as much as KIF22 ( Supplementary Fig. 1b,f ) in postmitotic cells and showed no clear difference between different phases of the cell-cycle ( Supplementary Fig. 1h).
We next used in situ hybridisation to visualize the expression of KIF22, ALDOA, HIRIP3, PAGR1, and MAZ transcripts in the different layers of the 12 PCW brain. We identified the VZ and SVZ based on cytoarchitecture, however, to validate our delineation we stained a section with PAX6, expressed by progenitor cells (Fig. 2a-a' * ). This confirmed the absence of progenitor cells in the IZ and CP and allowed us a guidance for estimating cellular location in other sections.
Consistent with the scRNAseq data KIF22 and ALDOA transcripts are the most clearly differentially expressed between zones containing progenitors (VZ and SVZ) and the more superficial layers that are mainly composed of postmitotic cells (IZ and SP/CP). KIF22 mRNA expression is most prominent in the VZ and SVZ with a few expressing cells in the IZ and SP/CP (Fig. 2b,b  ). Although ALDOA mRNA expression is most prominent in the proliferative VZ and SVZ there are substantial numbers of ALDOA expressing cells in the SP/CP (Fig. 2c,c ). HIRIP3 (Fig. 2d,d ), PAGR1 (Fig. 2e,e ), and MAZ (Fig. 2f,f ) mRNAs are expressed in the VZ/SVZ and also in cells of the SP/CP. In addition to being expressed in the proliferative zones of the cerebral cortex KIF22, ALDOA, HIRIP3, PAGR1, and MAZ are also expressed in the ventricular zone of the ganglionic eminences where interneuron progenitors reside suggesting the hypothesis that interneuron development may be affected by alterations in their dosage in the 16p11.2 CNV.
To conclude, of all the 29 16p11.2 transcripts, five, KIF22, ALDOA, HIRIP3, PAGR1, and MAZ, are expressed in the ventricular and subventricular zones at higher levels than in post-mitotic cells. Of these only two, KIF22 and ALDOA, are significantly enriched in progenitors compared to post-mitotic cells making them candidates for having specific roles in neurogenesis in the developing human fetal cerebral cortex. Although both are enriched in progenitors, KIF22 and ALDOA transcript expression shows notable differences: KIF22 transcripts are more restricted to progenitors and their levels vary as the cell-cycle progresses. We next describe the expression of KIF22 and ALDOA protein over a range of developmental stages.

KIF22 Protein is Expressed in Germinal Zones of 12, 14, and 16 PCW Cortex
Here we characterize KIF22 protein expression during human corticogenesis. Coronal cortex sections spaced along the rostralcaudal axis were immunostained for KIF22 protein and counterstained with Nuclear Fast Red (NFR) to show cytoarchitecture. KIF22 + (brown) and KIF22 − (red) cells were counted for each region in the telencephalic wall (VE, VZ, SVZ, IZ, and CP) (see methods for details of sampling) and lamination was identified by cell density Altman 2002, 2005). These data are shown for three developmental stages, 12 PCW ( Fig. 3ad), 14 PCW (Fig. 3a -d ), and 16 PCW (Fig. 3a  *  -d  *  ). At all stages and rostro-caudal positions KIF22 expressing cells appear most abundant in the VE followed by the VZ and SVZ with the IZ and CP presenting a very low to complete absence of KIF22 (Fig. 3c,c ,c * with higher magnification of boxed regions from each zone shown in d, d , and d * respectively, green arrows indicate examples of individual KIF22 + cells).
We next pooled KIF22 + cell count data in two ways to compare between all ages (Fig. 3e) and anatomical regions (Fig. 3f ) and found that the percentage of KIF22 + cells in the VE (40-50%) was consistently higher than other regions, followed by the VZ (20-30%) and SVZ (10%), with even fewer cells (<10%) in the IZ, and CP (Fig. 3e,f ). This result describes KIF22 protein as predominantly restricted to a subset of cells in the germinal zones of the developing cortex at all stages studied.

KIF22 Protein Expression is Restricted to Proliferating Cells
KIF22 protein expression is almost exclusively restricted to a subset of cells in the proliferative regions. From the scRNA-seq data, we expect these to be progenitor cells (Fig. 1). To identify these KIF22 positive cells we performed double immunofluorescence for KIF22 and KI67 (a protein expressed in all proliferating cells (Scholzen and Gerdes 2000;Miller et al. 2018)). KIF22 + cells were predominantly located in the VE, VZ, and SVZ (Fig. 3), therefore these regions were examined for analysis. Low magnification of KI67/KIF22 staining is shown in 12 PCW (Fig. 4a) and 14 PCW (Fig. 4b) with higher magnification showing individual cells in Fig. 4c. Cell counts for KIF22 + /KI67 + -labeled cells show that the majority (80-90%) of KI67 + cells also express KIF22 both at 12 (Fig. 4d) and 14 PCW (Fig. 4e) or across the rostral-caudal axis. Combining the data for anatomical locations and ages revealed significantly more KIF22 + /KI67 + cells than KIF22 + /KI67 − and KIF22 − /KI67 + cells (Fig. 4f ).

KIF22 Levels Vary with Cell-Cycle Phase
From the scRNA-seq analysis, and the variable KIF22 protein levels in KI67 + cells, we hypothesized that KIF22 protein levels change throughout the cell cycle. To test this, we quantified nuclear immunofluorescence intensity of KIF22 and KI67 in two 12 PCW brains (see methods for details of sampling procedure). KI67 protein levels vary during the cell cycle: lowest in G1 phase, increasing through S and G2 to peak in mitosis (Fig. 5o) (Scholzen and Gerdes 2000;Miller et al. 2018). We found a strong correlation between KIF22 and KI67 intensity (Brain 1 rostral R 2 = 0.8095, middle R 2 = 0.8139, and caudal R 2 = 0.6691. Brain 2 rostral R 2 = 0.7489, middle R 2 = 0.7447, caudal R 2 = 0.7763) (Fig. 5d-f ). To ensure the correlation observed was not a result of nucleus size changing with cell cycle, we confirmed that KIF22 protein levels did not correlate with nuclear size by DAPI staining (Brain 1 rostral KIF22 R 2 = 0.104, middle KIF22 R 2 = 0.0874, caudal KIF22 R 2 = 0.2969. Brain 2 rostral KIF22 R 2 = 0.1183, middle KIF22 R 2 = 0.0512, and caudal KIF22 R 2 = 0.2287). A strong positive correlation was also observed at 14 PCW (Brain 1 rostral R 2 = 0.7465, middle R 2 = 0.6668, caudal R 2 = 0.634, Fig. 5j-l). Again, we confirmed that KIF22 protein levels did not correlate with nuclear size (rostral KIF22 R 2 = 0.1239, middle KIF22 R 2 = 0.0599, caudal KIF22 R 2 = 0.0229). This demonstrates that the  correlation between KI67 and KIF22 is consistent between ages and rostral-caudal location. Combining all values of KI67/KIF22 nuclear intensity values showed that KIF22 was expressed at significantly higher levels in KI67 + cells (Fig. 5m) with a strong correlation (R 2 = 0.7236) between KIF22 and KI67 levels (Fig. 5n).
Although KIF22 expressing cells were scattered throughout the VE, VZ, and SVZ there was a general trend for the cells expressing the highest levels of KIF22 to be closest to the apical surface (yellow colored dots on scatterplots Fig. 5d-f,j-l,n) with lower expressing cells tending to be further from the apical surface (blue colored dots on scatterplots Fig. 5d-g, j-l,n). During interkinetic nuclear movement radial glial cell nuclei move to the apical surface to perform mitosis so this spatial distribution suggests KIF22 is expressed at high levels by radial glial cells undergoing mitosis at the apical surface of the VZ. Cerebral cortex progenitor that do not undergo interkinetic movement (intermediate progenitors and outer sub-ventricular zone progenitors) cannot be assigned to cell cycle phase using their position as we were able to do for radial glial cells which do undergo interkinetic movement. As KIF22 transcript levels were at highest in all progenitors in M-phase (Fig. 1h) we suspect that cells expressing high levels of KIF22 and KI67 further away from the apical surface are cortical progenitors in M-phase not undergoing interkinetic movement but further experiments would be needed to show this.
From these data we show that, for radial glial cells, KIF22 protein levels change throughout the cell cycle in positive correlation with KI67: KIF22 is present in G1 and increases through S and G2 phase to peak in mitosis (Fig. 5p).

ALDOA Protein is Highest in the Germinal Zones of the Cortex
Bioinformatics analysis and in situ hybridisation show ALDOA mRNA levels decrease as progenitor cells move towards a neuronal fate (Fig. 1). Here we used immunofluorescence to characterize ALDOA protein expression across the telencephalic wall at 3 developmental time points; at 12, 14, and 16 PCW, ALDOA immunofluorescence is most intense in the VZ and SVZ before decreasing in the cortical plate (Fig. 6a-c). Double immunofluorescence for KI67 and ALDOA viewed at high magnification shows that ALDOA protein is primarily localized outside DAPI + nuclei in the cytoplasm and that ALDOA is expressed by KI67 + progenitor cells and also by cells that do not express KI67 (Fig. 6d). The schematic (Fig. 6e,e ) illustrates the areas used for quantification of nuclear and whole cell ALDOA fluorescence presented in Fig. 6.

ALDOA Protein Levels Do Not Correlate with Proliferation
Although examination of ALDOA mRNA expression indicated it was enriched in progenitors we were unable to find a significant difference in ALDOA protein levels between KI67 + and KI67 − cells at 12 (Fig. 7a), 14 (Fig. 7b), and 16 PCW (Fig. 7c) in the human cortex. To look for any fluctuation in ALDOA levels with the cell cycle we quantified immunofluorescence for KI67 and cell body ALDOA (nucleus and adjacent cell body) using the same analysis as that described above for KIF22, and found no correlation or discernible pattern at 12 (Fig. 7e) (R 2 = 0.018), 14 ( Fig. 7f ) (R 2 = 2e-4) or 16 PCW (Fig. 7g) (R 2 = 0.00992). These data show that in human cortex development, cellular ALDOA protein levels do not correlate with proliferation or fluctuate with cell cycle.
Previous work in different models demonstrated nuclear ALDOA level is greater in proliferating cells (Mamczur et al. 2010;Mamczur et al. 2013). To see if this was the case in human cortex development, we quantified nuclear ALDOA and KI67 (Fig. 6e) but found no significant difference in nuclear ALDOA fluorescence between KI67 + and KI67 − cells at 12 (Fig. 7i), 14 (Fig. 7j), or 16 PCW (Fig. 7k). We next tested if nuclear ALDOA levels in proliferating cells varied with cell cycle. Analysis of ALDOA and KI67 nuclear intensity established no correlation or pattern at 12 (Fig. 7m) (R 2 = 5e-05), 14 (Fig. 7n) (R 2 = 0.0365), or 16 PCW (Fig. 7o) (R 2 = 0.0723). This shows nuclear ALDOA levels do not increase with proliferation, nor fluctuate with cell cycle. We combined results across the 12, 14, and 16 PCW. There was no significant difference between KI67 + and KI67 − cells when examining ALDOA protein intensity in the whole cell (Fig. 7d) or the nucleus (Fig. 7i). Using the pooled data, there was no correlation or discernible pattern when nuclear ALDOA intensity was graphed against nuclear KI67 level for the whole cell (Fig. 7h) (R 2 = 0.0049) or the nucleus (Fig. 7p) (R 2 = 6e-04).

16p11.2 Transcript Expression during Human Neurogenesis
The 16p11.2 CNV is a polygenic mutation that causes NDDs and the current study identified a number of the 29 16p11.2 transcripts expressed in progenitor cells of the cerebral cortex. In addition to ALDOA and KIF22 that are significantly enriched in progenitors, several other transcripts (e.g., HIRIP3, PAGR1, and MAZ) are also expressed in progenitors albeit at lower levels and are not significantly down-regulated as cells become postmitotic. The simultaneous expression of multiple 16p11.2 genes in cells undergoing neurogenesis suggests that these cells may be particularly vulnerable to simultaneous alteration in their dosage as a consequence of the 16p11.2 microdeletion or microduplication. This lends support to the hypothesis that neurogenesis is disrupted in 16p11.2 CNV patients and that this contributes to subsequent development of NDDs.
In this study we focused on the expression of 16p11.2 genes in progenitors of the developing cerebral cortex that will give rise to excitatory neurons. However, KIF22, ALDOA, HIRIP3, PAGR1, and MAZ transcripts are also expressed in the ventricular zone of the ganglionic eminences where the progenitors of the inhibitory interneurons that subsequently migrate into the cerebral cortex reside. Simultaneous increase or decrease of these transcripts in interneuron progenitors resulting from the 16p11.2 CNV may therefore have an impact on the neurogenesis or differentiation of interneurons.

KIF22
KIF22 is a multifunctional protein that can regulate cell proliferation through at least two distinct mechanisms. First, KIF22 is a kinesin-like microtubule-based motor that binds microtubules and chromosomes during mitosis and regulates mitotic spindle microtubule stability and symmetric/asymmetric cell division (Tokai et al. 1996;Tokai-Nishizumi et al. 2005;Sun and Hevner 2014). Second, KIF22 regulates the expression of the cell-cycle regulator CDC25C. During cell division, CDC25C dephosphorylates CDK1, thus activating the CDK1-cyclinB complex while the CDK1-cyclin B complex phosphorylates CDC25C, causing an amplification loop to drive cells to mitosis (Nilsson and Hoffmann 2000). KIF22 directly transcriptionally represses CDC25C and inhibits mitosis; this transcriptional repression of CDC25C is dependent on KIF22 being phosphorylated on Thr463 (Ohsugi et al. 2003;Yu et al. 2014). KIF22 depletion in a tumor cell line accelerates the G2/M transition and slows M/G1 transition (Yu et al. 2014).
Overall, it therefore appears that KIF22 can act at several different points in the cell cycle making it difficult to predict how increased or decreased dosage of KIF22 in the 16p11.2 microduplication or microdeletion respectively would impact cell cycle in the specific context of cerebral cortex neural progenitors especially in light of the concomitant altered dosage of other 16p11.2 genes co-expressed with KIF22 in progenitors. Our observation that KIF22 mRNA and KIF22 protein levels both increase during the cell cycle to achieve the highest levels in G2/M phase that drop as cells enter G1 phase implies that KIF22 protein does not persist for long after it is translated and is degraded at the end of M-phase suggesting that both transcriptional and posttranscriptional mechanisms regulate its levels. A clear outcome of our study is that KIF22 levels positively correlate with KI67 in neural progenitors and steadily rise as the cell progresses Figure 5. Quantification of KIF22 protein levels. a, b, c) 12 PCW quantification of KIF22 f luorescence intensity in KI67+/KI67-cells (raw data transformation = +1(log), unpaired t-test with Welch's correction,P = < 0.001). d, e, f ) intensity correlations of KIF22 and KI67 nuclear f luorescence intensity at 12 PCW. g, h, i) 14 PCW quantification of KIF22 f luorescence intensity in KI67+/KI67-cells (raw data transformation = +1(log), unpaired t-test with Welch's correction, P = < 0.001). j, k, l) intensity correlations of KIF22 and KI67 nuclear f luorescence intensity at 14 PCW. m) quantification of KIF22 f luorescence intensity in KI67+/KI67-cells 12 and 14 weeks combined (raw data transformation = (log), paired t-test, P = 0.0122). n) intensity correlations of KIF22 and KI67 nuclear f luorescence intensity for rostral-caudal points at 12 and 14 PCW with distance from apical surface indicated by dot color. o) diagram of KI67 protein levels throughout the cell cycle. p) model based on our results of KIF22 protein levels throughout the cell cycle. Figure 6. ALDOA protein expression in the cortex. ALDOA protein expression across the telencephalic wall at a) 12, b) 14 and c) 16 pcw. Scale bars = 100 μm. White arrow indicates non-specific binding to blood vessels. d) high magnification immunof luorescence of ALDOA and KI67 proteins, scale bar = 10 μm. e) low power image showing how cells were randomly selected for analysis using the DAPI channel. e') high power image showing how the nucleus and cytoplasm were delineated for analysis through G1 > S > G2 > M phases culminating in the maximum level during M-phase. One possibility is that KIF22 is required to reach a threshold for mitosis to occur, after which its levels must decrease sufficiently to allow mitotic exit. Whether cells undertake proliferative or neurogenic divisions is a process heavily controlled by cell cycle length (Borrell and Calegari 2014). Perturbing KIF22 gene dosage as a consequence of the 16p11.2 CNV might affect the timing of KIF22 protein reaching this threshold in neural progenitors and therefore affect cell-cycle kinetics and perturb neurogenesis and neuronal output. Our Figure 7. ALDOA protein quantification. a-d) Cell body ALDOA protein f luorescent intensity in KI67+ and KI67− cells at a) 12 PCW (raw data transformation = +1(log), bimodal distribution, Mann-Whitney test, p = 0.3702), b) 14 PCW (raw data transformation = +1(log), normal distribution, unpaired t-test with Welch's correction, P = 0.2032), c) 16 PCW (raw data transformation = +1(log), normal distribution, unpaired t-test with Welch's correction, P = 0.3523). d) ALDOA cell body protein f luorescent intensity in KI67+ and KI67-cells, 12, 14, and 16 PCW individual datasets averaged, (raw data transformation = +1(log)), paired t-test, P = 0.0836. e-h) ALDOA cellular protein intensity levels ls correlated to nuclear KI67 protein intensity at e)12, f )14 and g)16 PCW with distance from ventricular edge indicated. h) ALDOA whole cell protein intensity levels correlated to nuclear KI67 protein intensity pooled 12, 14, 16 PCW. i-l) Nuclear ALDOA protein f luorescent intensity in KI67+ and KI67− cells at i) 12 PCW (raw data transformation = +1(log), normal distribution, unpaired t-test with Welch's correction, P = 0.7543), j) 14 PCW (raw data transformation = +1(log), normal distribution, unpaired t-test with Welch's correction, P = 0.0694), k) 16 PCW (raw data transformation = +1(log), normal distribution, unpaired t-test with Welch's correction, P = 0.0772). l) ALDOA nuclear protein f luorescent intensity in KI67+ and KI67− cells, 12, 14, and 16 PCW individual datasets averaged, (raw data transformation = +1(log)), paired t-test, P = 0.1330. m-p) ALDOA nuclear protein intensity levels is correlated to nuclear KI67 protein intensity at m)12, n)14, and o)16 PCW with distance from ventricular edge indicated. p) ALDOA nuclear protein intensity levels correlated to nuclear KI67 protein intensity pooled 12, 14, and 16 PCW. q) schematic demonstrating ALDOA protein is predominantly in the cytoplasm and lower in the nucleus in both KI67+ proliferating cells and KI67-post mitotic cells. r) model showing ALDOA levels do not change with the cell cycle. results suggest the hypothesis that KIF22 regulates neurogenesis in the human developing cortex through cell-cycle regulation.

ALDOA
The process of brain development requires a vast and consistent supply of energy. Glucose is the predominant energy substrate for the fetal brain (Gustafsson 2009), therefore efficient and controlled glycolysis is essential for normal brain development. ALDOA is required for the fourth step of glycolysis, conversion of fructose 1,6-biphospate to dihyroxyacetone phosphate and gluteraldehyde 3-phosphate. The metabolic role of cytoplasmic ALDOA is well established, and ALDOA also has other nonglycolytic 'moonlighting' roles such as regulating mitochondrial function and cytoskeleton stability (Orosz et al. 1988;Pagliaro and Taylor 1992;Kao et al. 1999;Jewett and Sibley 2003;Buscaglia et al. 2006). In addition to its cytoplasmic roles, ALDOA has been observed in the nucleus (Mamczur and Dzugaj 2008;Mamczur et al. 2010;Mamczur et al. 2013) where it has been suggested to impact cell cycle by positively regulating cyclin D1 expression to mediate G1/S progression (Ritterson Lew and Tolan 2012;Fu et al. 2018). Cell-culture studies show ALDOA sub-cellular localization depends on the availability of energetic substrates, with addition of glucose driving ALDOA protein to the cytoplasm (Mamczur et al. 2013). Therefore, it is likely the primary role for ALDOA is metabolic when cells require, and have available to them, large amounts of energy. The majority of ALDOA studies have used highly abnormal cancer tissue, or artificial cell culture systems in which glycolytic enzymes have been shown to be increased (Ritterson Lew and Tolan 2012;Mamczur et al. 2013;Fu et al. 2018;Pollen et al. 2019). How these observations of ALDOA in a variety of systems relate to its role in human cerebral cortex development is unclear.
Altering ALDOA dosage in the developing brain will likely impact energy metabolism by altering the flow of metabolites through the glycolytic pathway and impacting subsequent pathways which feed on outputs of glycolysis. Disruption to energy metabolism during development has previously been linked to ASD and ADHD (Rash et al. 2018). The offspring of hyperglycemic mice presented microcephaly, a phenocopy of the microcephaly observed in 16p11.2 microduplication patients (Rash et al. 2018) and disruptions to energy metabolism may contribute to the microcephaly seen in the offspring of Zika infected mothers (Gilbert-Jaramillo et al. 2019). No homozygous null ALDOA patients have been identified suggesting it is essential for life, but patients with changes to ALDOA levels have been identified; one patient with reduced ALDOA activity presented microcephaly (Kreuder et al. 1996) and another presented intellectual disability (Beutler et al. 1973). Of particular interest is the finding of schizophrenia patients with upregulated cortical ALDOA levels (Beasley et al. 2006) and 16p11.2 microduplication is strongly associated with risk of schizophrenia. This information, coupled with our results that ALDOA is expressed in all cell types, make it clear that any changes to ALDOA dose will perturb energy metabolism at many stages in the brain, impacting its development.
ALDOA is much more abundant in the cytoplasm and we also found no clear relationship between cell body ALDOA levels and cell proliferation status. Nuclear ALDOA has been linked to cell proliferation (Mamczur et al. 2010;Mamczur et al. 2013;Fu et al. 2018) but we found no clear relationship between nuclear ALDOA protein levels and cell proliferation status. Although ALDOA mRNA levels are higher in proliferating cells compared to non-proliferating cells, quantitative analysis of ALDOA protein revealed that ALDOA protein persists once cells exit mitosis. Therefore, while ALDOA protein is abundant in progenitor cells of the developing human cerebral cortex and so may play a role in neurogenesis phenotypes, the persistent expression of ALDOA protein as cells become post-mitotic argues against a specific role in neurogenesis and raises the additional possibility that ALDOA also plays roles in differentiated neurons.

Conclusion
Our study of 16p11.2 gene expression in developing human fetal cerebral cortex indicates that altered dosage of KIF22, ALDOA, HIRIP3, PAGR1, and MAZ caused by the 16p11.2 microduplication or microdeletion may impact on neurogenesis in the developing human cortex and we identified KIF22 being a strong candidate for having a specific role in neurogenesis. Further studies are required to unpick the mechanisms involved, but given the nature of the tissue, the scope for studying this in vivo is currently limited. However, growth of new model systems in which 16p11.2 gene expression can be manipulated such as human cerebral organoids will provide the opportunity address these questions.

Supplementary Material
Supplementary material can be found at Cerebral Cortex online.

Notes
We thank Katherine Howe for her assistance with designing in situ probes and undergraduate student Emma Fowler (EF funded by a WR Henderson Scholarship) for her assistance with quantification.