Cortex development is controlled by temporal patterning of neural progenitor (NP) competence with sequential generation of deep and superficial layer neurons, but underlying mechanisms remain elusive. Here, we report a role for heterogeneous nuclear ribonucleoprotein A3 (HNRNPA3) in regulating the division of early cortical NPs that mainly give rise to deep-layer neurons via direct neurogenesis. HNRNPA3 is expressed at high levels in NPs of mouse and human cortex at early stages, with a unique peri-chromosome pattern. Intriguingly, downregulation of HNRNPA3 caused chromosome disarrangement, which hindered normal separation of chromosomes during NP division, leading to mitotic delay. Furthermore, HNRNPA3 is associated with the cohesin-core subunit SMC1A and controls its association with chromosomes, implicating a mechanism for the role of HNRNPA3 in regulating chromosome segregation in dividing NPs. Hnrnpa3-deficient mice exhibited reduced cortical thickness, especially of deep layers. Moreover, downregulation of HNRNPA3 in cultured human cerebral organoids led to marked reduction in NPs and deep-layer neurons. Thus, this study has identified a crucial role for HNRNPA3 in NP division and highlighted the relationship between mitosis progression and early neurogenesis.
During neocortical development, distinct subtypes of excitatory neurons are born in well-ordered temporal patterns via direct or indirect neurogenesis pathways and migrate to form multiple layers with ‘inside out’ manner to form deep and up layers sequentially (Angevine and Sidman, 1961; Marin-Padilla, 1978; Noctor et al., 2004; Rakic, 1974). Radial glia cells (RGs) localized in the apical ventricular zone (VZ) can undergo either symmetric proliferative division, generating two daughter RGs and expanding the progenitor pool, or asymmetric neurogenic division to produce a RG and a postmitotic neuron or an intermediate progenitor (IP) residing in the subvetricular zone (SVZ) (Haubensak et al., 2004; Malatesta et al., 2000; Miyata et al., 2004; Noctor et al., 2001, 2004). Intriguingly, several heterogeneous types of neural progenitors (NPs) with high proliferation potency, including IPs in the inner SVZ (iSVZ) and so called ‘outer’ or ‘basal’ radial glia cells (oRGs or bRGs) in the outer SVZ (oSVZ) are prominent in primates (Betizeau et al., 2013; Fietz et al., 2010; Hansen et al., 2010), and these basal progenitors (BPs) in the expanded SVZ are assumed to be associated with cortex expansion (Lui et al., 2011; Otani et al., 2016; Smart et al., 2002; Sun and Hevner, 2014). Similar to IPs, oRGs also arise from ventricular RGs (vRGs) (Ju et al., 2016; Pollen et al., 2015; Wang et al., 2011). Despite the variations in the lineage of NPs, vRGs or apical progenitors (APs) are the primary types of NPs, the division of which provides the driving force for the following cortex development process. Nevertheless, molecule mechanisms that tightly control successive NP division and temporal progression of cortical neurogenesis are not fully understood.
Cell cycle progression is one of the key determinants in NP maintenance and cortical neurogenesis (Florio and Huttner, 2014; Sun and Hevner, 2014; Taverna et al., 2014). Notably, most of the genes associated with autosomal recessive primary microcephaly (MCPH), including MCPH1, CDK5RAP2 (CDK5 regulatory subunit-associated protein 2), ASPM (abnormal spindle-like microcephy-associated) and WDR62 (WD-repeat-containing protein 62), encode centrosome-associated proteins that govern normal cell divisions (Jayaraman et al., 2016; Kaindl et al., 2010; Nicholas et al., 2010; Sun and Hevner, 2014). Loss of these genes in mice either singly or combinatory caused centriole duplication or chromosomal organization defects and impaired mitosis, subsequently leading to reduction in the number of NPs and in cortical size (Barrera et al., 2010; Gruber et al., 2011; Jayaraman et al., 2016; Lizarraga et al., 2010; Pulvers et al., 2010). By contrast, enhanced cell cycle re-entry in NPs has been shown to promote the expansion of the NP pool and enlarge the cortical size (Chenn and Walsh, 2002; Lange et al., 2009; Mairet-Coello et al., 2012; Nonaka-Kinoshita et al., 2013). Normal progression of NP division requires high fidelity of sister chromosome segregation. However, the molecules that are responsible for maintaining chromosome integrity during cortical NP division remain unclear.
The HNRNP family was originally identified as nuclear RNA-binding proteins and many of them are suggested to be involved in multiple cellular functions, such as mRNA processing and transport, transcription, DNA repair, and telomere maintenance, through DNA- or RNA-protein or protein-protein interactions (Dreyfuss et al., 1993; Han et al., 2010; Krecic and Swanson, 1999). Among them, HNRNPA3 is less well studied. Phenotypic data from on-line DECIPHER database (decipher.sanger.ac.uk/) suggest that HNRNPA3 is extremely intolerant of loss-of-function variation and potentially haplo-insufficiency in populations. Notably, HNRNPA3 is highly expressed in oSVZ of developing human cortex (Miller et al., 2014). Prompted by these observations, we decided to investigate the role of HNRNPA3 in cortex development.
We found that HNRNPA3 is highly expressed at early stages of cortical development, in particular in RGs of mouse and human. Genetic ablation of Hnrnpa3a in mice prolonged mitosis of vRGs and caused a marked decrease in the production of deep layer neurons. Downregulation of HNRNPA3 in cultured human organoids also caused mitotic delay of NPs and a marked decrease in the number of RGs and their neuronal progeny. Finally, we found that HNRNPA3 is distributed in peri-chromosome regions of mitotic progenitors, and the loss or downregulation of HNRNPA3 impedes the chromosome resolution of sister chromatids of dividing NPs, probably via modulating the cohesin-association with chromosomes. These results suggest a conserved role of HNRNPA3 in the division of cortical NPs.
Expression of HNRNPA3 in cortical neural progenitors of mouse and human
Transcriptome analysis of prenatal human brain for principle features of different proliferative layers has led to identification of a few genes that are enriched in oSVZ compared to iSVZ, and HNRNPA3 is one of these genes (Miller et al., 2014). In line with this result, we found that, at gestational week (GW) 16.5, HNRNPA3 is highly expressed in cells at oSVZ and VZ regions compared with that in iSVZ (Fig. 1A,B). Notably, HNRNPA3 was more frequently observed in early (PAX6+TBR2−) and late (PAX6+TBR2+) RGs than in IPs (PAX6−TBR2+) in either VZ or SVZ regions (Fig. 1B,C), and many RGs labeled by HOPX also expressed HNRNPA3 in VZ and SVZ regions of human fetal cortex (Fig. 1D), and cultured cerebral organoids (Fig. S1A). To analyze the expression of HNRNPA3 in various types of NPs and differentiated neurons, we analyzed accessible online datasets of single cell RNA-sequencing (scRNA-seq) information with reads covering full-length RNA from human fetal brain and cultured cerebral organoids (Camp et al., 2015), and extracted data using methods described previously (Kageyama et al., 2018). We found that vRGs, APs and BPs exhibit higher expression of HNRNPA3 compared with differentiated neurons in the fetal brain (Fig. S1B) and in cultured organoids (Fig. S1C). Further analysis of dynamic expression patterns of individual genes, as reported recently (Camp et al., 2015; Kageyama et al., 2018; Telley et al., 2016), indicates that HNRNPA3 shares parallel expression waves with PCM1, which is involved in centromere assembly, and this pattern is conserved in mouse and human fetal brain, and in cultured human cerebral organoids (Fig. S1D). The gradual decrease in the level of PAX6 and increase in DCX, a marker for differentiated newborn neurons, indicates reliability of the analysis. Unlike HNRNPA3, the close subfamily member HNRNPA1 displayed ubiquitous expression throughout development, in particular in human fetal brain (Fig. S1D).
In rodents, RGs are the major cortical NPs that can undergo multiple rounds of proliferative divisions and IPs can undergo only one round of neurogenic division (Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004). Interestingly, temporal expression patterns of pulse-labeled apical progenitors and their daughter cells reported in a recent study (Telley et al., 2019) revealed that Hnrnpa3 was expressed at higher levels in APs at early neurogenesis stage than at later stages (Fig. S1E). Indeed, scRNA-seq data showed that Hnrnpa3 exhibited the highest expression in RGs and then declined in IPs and post-mitotic neurons (Fig. S1F) (Telley et al., 2016). Consistent with gene expression patterns, the level of Hnrnpa3 protein also gradually decreased from E12.5 to E15.5 in VZ and SVZ regions (Fig. 1E). In line with this notion, we detected high levels of Hnrnpa3 signal in Pax6+Tbr2− RGs in the VZ region and relatively low levels in Pax6−Tbr2+ IPs in both the VZ or SVZ of mice at E13.5 (Fig. 1F). Taken together, the expression level of HNRNPA3 is positively associated with cortical NPs with higher proliferation potency.
Hnrnpa3a deletion impaired cortical development in mice
It has been reported that there are two protein isoforms derived from Hnrnpa3 – a longer Hnrnpa3a isoform and a shorter Hnrnpa3b isoform – and the a-isoform predominates in human tissues (Papadopoulou et al., 2012). As shown in Fig. S1G, Hnrnpa3a was widely expressed in various tissues, including cerebrum, cerebellum, lung, pancreas and spinal cord, whereas Hnrnpa3b was expressed in limited tissues such as cerebrum, kidney and skeletal muscle. To investigate whether Hnrnpa3 is involved in cortical development, we took advantage of CRISPR/Cas9-mediated genome editing to generate Hnrnpa3a knockout (KO) mice (Fig. 2A). For this purpose, sgRNA was designed to target Hnrnpa3a at exon 1, which is absent in Hnrnpa3b (Fig. 2A). Compared with wild-type (WT) littermates, homozygous Hnrnpa3a KO mice showed marked reduction in the level of Hnrnpa3a, but not Hnrnpa3b, in the brain at E14.5 (Fig. 2B). Notably, the Hnrnpa3a KO mice exhibited barely detectable levels of Hnrnpa3 protein in VZ/SVZ regions (Fig. 2C), suggesting that Hnrnpa3a is the major isoform expressed in cortical NPs. Although there was no apparent abnormality in the morphology of whole body and the brain (Fig. S2A,B), the Hnrnpa3a KO mice died shortly after birth, probably owing to neonatal hypoxia caused by lung developmental defects (Fig. S2A,D,E). The neuromuscular synapses in the diaphragm were roughly normal (Fig. S2C).
Next, we determined cortical lamination by staining with layer V marker Ctip2 and layer II-IV marker Cux1 in mice at P0. We found that the Hnrnpa3a KO mice exhibited a mild but significant reduction in the thickness of the dorsal lateral cortex compared with that of wild-type littermates (Fig. 2D,E). Notably, the layer thickness and numbers of Ctip2+ cells, but not that of Cux1+ cells, were reduced in Hnrnpa3a KO mice (Fig. 2F-I). Consistent with this result, Hnrnpa3a KO mice at E18.5 also displayed a significant reduction of early formed layer VI labeled by Tbr1 (Fig. 2J-L). The number of Tbr2-labeled IP cells, which give rise to superficial layer neurons, was similar in wild-type and KO mice at E14.5 (Fig. 2M,N). Thus, Hnrnpa3a plays an important role in the neurogenesis of cortical neurons, in particular of Ctip2+ deep layer neurons. The specific effects on early-born deep layer neurons might be attributable to the temporal high level expression of Hnrnpa3a in RGs at early stages of cortex development.
Hnrnpa3a ablation leads to mitotic retardation in apical cortical progenitors
In light of aforementioned findings, we next asked whether the loss of Hnrnpa3a affected the proliferation or differentiation process of NPs. First, we used an antibody against phosphorylated histone 3 (PH3) to label mitotic cells in mice at various embryonic stages. We found that the number of PH3+ cells was similar in wild-type and Hnrnpa3a KO mice at E12.5, but markedly increased in Hnrnpa3a KO mice at later stages at E14.5 or E17.5 (Fig. 3A-D). Interestingly, unlike wild-type PH3+ cells that were aligned and attached to apical surface of the VZ, many PH3+ cells were stacked to form clusters at the VZ regions in Hnrnpa3a KO mice at E14.5 (Fig. 3E). By judging chromosome arrangement according to distribution of DAPI signals in dividing cells marked by phosphorylated-vimentin (p-Vim) (Insolera et al., 2014), we measured the proportions of VZ cells in different phases of mitosis, and found that Hnrnpa3a KO mice at E12.5 exhibited marked increase in the percentage of cells in prophase and pro-metaphase (pro/pro-metaphase) (Fig. 3F,G). These results suggest that the increase and abnormal distribution of mitotic cells might be due to cell cycle arrest. To further consolidate this conclusion, we pulse-labeled cycling NPs by BrdU injection at E12.5 and analyzed the percentage of PH3+BrdU+ cells among the BrdU+ cells, which represents cells arrested in mitosis (Ye et al., 2017). The increased mitotic index in Hnrnpa3a KO mice also suggests delayed mitotic progression in NPs (Fig. 3H,I).
Next, we measured the proportion of mitotic cells, which displayed condensed Ki67 full labeling associated with the chromosome, among the Ki67+ cycling progenitors, according to the methods described previously (Betizeau et al., 2013; Pilaz et al., 2016) (Fig. 3J,K). Although total numbers of proliferative RGs were similar between wild-type and Hnrnpa3a KO mice at E14.5 (Fig. 3L), the percentage of RGs at mitotic phase was increased in Hnrnpa3a KO mice (Fig. 3M), indicating again a mitotic delay.
To directly observe the cell division process, we performed live imaging of mitotic RGs in cultured brain slices from mice at E13.5 (Fig. 4A). We found that the percentage of RGs with prolonged M-phase duration was higher in Hnrnpa3a KO mice than in wild-type mice (Fig. 4B), providing direct evidence that loss of Hnrnpa3a impeded RG division by causing mitotic delay.
Early NPs in the mouse VZ can undergo proliferative symmetrical divisions if they occur in a vertical cleavage plane relative to the VZ surface or self-consumptive asymmetrical divisions if they occur in a horizontal plane (Mora-Bermúdez and Huttner, 2015; Sun and Hevner, 2014). Based on distribution patterns of chromosomes in mitotic RGs, we analyzed division angles of RGs at VZ regions of wild-type and KO mice (Fig. 4C). We found that NPs in Hnrnpa3a KO mice displayed a notable increase in oblique division (30-60°) at the cost of near-vertical division (60-90°) (Fig. 4C,D). The changes in cleavage plane relative to VZ surface might be due to mis-arrangement of chromosomes and subsequent disruption of spindle apparatus.
It has been reported that NPs with prolonged mitosis are usually committed to apoptosis or accelerated neurogenic competence (Pilaz et al., 2016). However, the Hnrnpa3a KO mice did not exhibit apparent apoptosis at least at the time of our observation (Fig. S3A-C), although DNA damage response reflected by the signals of γH2AX, which marks DNA damage-driven histone modification, was evoked in KO mice (Fig. 4E,F). The activation of DNA repair pathway, which was measured by phosphorylated-Atm (pS1981) (Fig. 4G,H), may have protected cells from death. Thus, the loss of Hnrnpa3a hindered neurogenic cell division of early NPs and impaired neuronal competency.
Downregulation of Hnrnpa3 causes abnormal chromosome arrangement
The protracted mitosis of RGs in Hnrnpa3a KO mice prompted us to determine subcellular distribution of Hnrnpa3 in mitotic cells. Interestingly, we found that Hnrnpa3 was enriched in peri-chromosome regions, especially in regions between two sister chromosomes of dividing NPs in the VZ (Fig. 5A, top lane). The signal was specific for Hnrnpa3a, because it was not observed in Hnrnpa3a KO mice (Fig. 5A, bottom row). Interestingly, much more NPs of Hnrnpa3a KO mice exhibited chromatin bridges at anaphase, as shown in fixed samples (Fig. 5B,C) and in live imaging of cultured brain slices (Fig. 5D). Notably, the NPs with chromatin bridges still finished the mitotic process, and finally gave rise to two daughter cells, although this process was abnormally prolonged (Fig. 5D). In addition to chromatin bridges, many mitotic NPs in KO mice also exhibited chromosome fragments (Fig. 5E), which might be derived from chromosome mis-alignment or mis-segregation, as shown previously (Crasta et al., 2012).
Next, we determined the HNRNPA3 localization and functions in human neural stem cells. In the human neural stem cell ReN at M-phase, HNRNPA3 was also distributed in peri-chromosome regions (Fig. S4A). However, upon treatment with nocodazole, which was used to induce cell division arrest by causing microtubule depolymerization and spindle disassembly, HNRNPA3 was segregated from the peri-chromosome regions (Fig. S4A). In mitotic ReN cells transfected with construct encoding small interference RNA against HNRNPA3 (shA3), which was able to downregulate the expression of HNRNPA3 (Fig. S4B), the percentage of cells exhibiting chromosome misalignment and spindle disorganization was dramatically increased compared with control cells transfected with construct encoding scrambled sequence (shScr) (Fig. S4C,D). These phenomena were unlikely to be caused by impairment of spindle assembly checkpoint activity, because the crucial mitotic checkpoint protein BUB-related 1 (BubR1) remained associated with chromosomes in HNRNPA3 knockdown ReN cells at prometaphase (Fig. 6A). Chromosome segregation errors can trigger the formation of DNA breaks that may evoke DNA damage responses (Crasta et al., 2012). Indeed, ReN cells with HNRNPA3 knockdown exhibited marked increase in the levels of γH2AX revealed by either immunostaining or immunoblotting (Fig. S4E,F).
As a RNA-binding protein, how does HNRNPA3 regulate chromosome organization during mitosis? In dividing cells, sister chromatids are held together by cohesin complexes, which are then removed from DNA during prometaphase, and the cohesion release is necessary for the separation of sister chromatids (Haarhuis et al., 2014). Notably, HNRNPA3 was found in a protein complex containing cohesion proteins, including the structural maintenance of chromosomes 1 (SMC1) – a key component of cohesin complex, which has been shown to be needed for mitotic progression (Yeh et al., 2015; Ma et al., 2013). As shown in Fig. 6B, immunoprecipitation (IP) using SMC1A antibody caused co-IP of HNRNPA3 in the lysates of ReN cells, confirming their association. In ReN cells at prophase, SMC1A was associated with chromosomes, where it was largely colocalized with Hnrnpa3a tagged with EGFP (Fig. 6C). During the transition from prophase to metaphase, SMC1A is gradually released from chromosomes, then associated with bipolar spindle apparatus and finally recruited to bipolar centromeric spindle pole (Guan et al., 2008; Wong and Blobel, 2008). The phosphorylation of cohesin complex proteins, including SMC1A, has been shown to stimulate their release from chromosomes and then binding to spindle apparatus (Nishiyama et al., 2013; Sumara et al., 2002; Wong and Blobel, 2008). Intriguingly, HNRNPA3 knockdown ReN cells exhibited a reduced level of phosphorylated SMC1A (p-SMC1A), which was localized in centrosomes labeled by γ-tubulin at prometaphase (Fig. 6D,E). Unlike control cells that showed SMC1A association with spindle, extensive SMC1A signals still remained attached to chromosomes in HNRNPA3 knockdown cells at metaphase (Fig. 6F). These results suggest that HNRNPA3 modulates dynamic association of cohesin with chromosomes and spindle apparatus to precisely control chromosome segregation during cell division.
Role of Hnrnpa3a in early direct neurogenesis in mice
To further determine the outcomes of mitotic delay caused by the loss of Hnrnpa3a in mice, we pulse labeled cycling NPs by BrdU injection at E12.5 or E14.5 and analyzed cell cycle exit and re-entry after 24 h by immunostaining with cell cycling marker Ki67. Notably, with BrdU injection at E12.5, the percentage of BrdU+Ki67− cells among total BrdU+ cells was significantly decreased in E13.5 Hnrnpa3a KO mice compared with wild-type littermates (Fig. 7A,B), suggesting a reduction in NP progeny that had left the cell cycle. However, this phenomenon was not observed in E15.5 Hnrnpa3a KO mice after BrdU pulse labeling at E14.5 (Fig. S5A,B). Furthermore, the percentage of BrdU+Ki67+ cells among Ki67+ cells was decreased in Hnrnpa3a KO mice at E13.5 (BrdU injection at E12.5) (Fig. 7C), but not E15.5 (BrdU injection at E14.5) (Fig. S5C). Thus, the effects on cell cycle progression caused by the loss of Hnrnpa3a may be limited to NPs at early periods. To directly trace the cell cycle re-entry of progenitors after division, we labeled the mitotic cells with carboxy-fluorescein esters (CFSE) injected at E12.5 followed by an EdU pulse 1 day later (Fig. 7D). We found that Hnrnpa3a KO mice displayed a marked reduction in the percentage of mitotic progenitors that entered a new round of the cell cycle (Fig. 7E). The decrease in cell cycle re-entry was most likely caused by mitotic delay, as mentioned above.
We then asked whether the mitotic delay had any effect on neuronal fates. To this end, we performed immunostaining for new-born neuronal marker Dcx at E13.5 after CFSE injection at E12.5. We found the percentage of Dcx+CSFE+ cells, which represents neuronal progeny via direct neurogenic division, was decreased in Hnrnpa3a KO mice compared with wild-type littermates (Fig. 7F,G). This result suggests that the loss of Hnrnpa3a in NPs impairs neuronal progeny. To further assess the effects of Hnrnpa3a ablation on neuronal specification, we analyzed neuronal fates in mice at P0 after BrdU pulse at E12.5, by calculating the percentage of cells labeled by pan-neuronal marker NeuN, or layer marker Ctip2 or Cux1, among cells with fully retained BrdU (BrdUfull), which were considered as daughter neurons directly derived from NPs after the last round of division (Cárdenas et al., 2018) (Fig. 7H-K and Fig. S5D,E). We found that the percentage of Ctip2+BrdUfull among BrdUfull cells was significantly decreased in Hnrnpa3a KO mice compared with wild-type littermates (Fig. 7H,J), whereas the percentage of Cux1+BrdUfull or NeuN+BrdUfull among BrdUfull cells had no difference (Fig. 7H,K; Fig. S5D,E). These results support the conclusion that mitotic delay caused by the loss of Hnrnpa3a impairs early neurogenesis rather than neuronal fate specification.
To determine whether Hnrnpa3 acts cell autonomously in NPs, E12.5 mice were subjected to in utero electroporation (IUE) with vectors encoding shRNA against Hnrnpa3 (shA3) or scrambled sequence (shScr), together with GFP plasmid to label transfected cells, and observed at E16.5 for neuronal fates (Fig. 8A). We found that the percentage of Ctip2+ cells in transfected GFP+ cells was markedly reduced in shA3 mice compared with shScr control mice (Fig. 8A,B), whereas the percentage of superficial layer neurons labeled by Satb2 in GFP+ cells had no difference (Fig. 8A,C). Notably, the majority of shA3-expressing neurons were localized in regions beneath the Ctip2+ cell band (Fig. 8A). This migration delay might be due to neurogenesis retardation. To further consolidate this conclusion at the single clone level, we analyzed the clone trajectory of individual RGs, which were sparsely labeled at E12.5 with low-titer retrovirus expressing shA3 or control sequence together with GFP (Fig. 8D). We found a marked reduction in Ctip2+ neurons and an arrest of NPs at the VZ of mice at E16.5 (Fig. 8D,E). These results suggest a cell-autonomous effect of Hnrnpa3 on the neurogenesis of early born neurons in mice. In line with a positive association between the level of Hnrnpa3 and NP stemness, overexpression (OE) of Hnrnpa3a in E14.5 mice via IUE caused an increase in Pax6-labeled RG cells in E16.5 mice (Fig. 8F,G). Furthermore, HNRNPA3 OE in ReN cells promoted cell proliferation as measured by EdU incorporation (Fig. S6A,B). Thus, the level of HNRNPA3 expressed in NPs is positively associated with mitotic progression.
Role of HNRNPA3 in the development of human cerebral organoids
Next, we investigated roles of HNRNPA3 in human cortex development by using the cerebral organoid culture system, which has been developed to mimic cortex development and related diseases (Lancaster and Knoblich, 2014). Embryonic bodies (EBs) were derived from H9 human embryonic stem cells infected by lentivirus encoding doxycycline-inducible expression of RFP reporter and shA3 or a control sequence (Ctrl) (Fig. S7A,B), following the protocol introduced previously (Li et al., 2017). To avoid potential effects on early pluripotent stages, the expression of shA3 or shCtrl was turned on at the stage of neural induction (around day 12) by adding doxycycline (1 μg/ml) to the culture medium (Fig. S7A,C), and development of cerebral organoids was examined at different days after neural induction (Fig. S7D,E). Compared with control organoids, shA3-expressing organoids displayed apparent reduction in PAX6-labeled RGs, as well as TBR2-labeled IPs (Fig. 9A,F and Fig. S7D). The defects seemed to be more severe at later stages (day 52 and day 78) than early stages (day 33) (Fig. 9A,F and Fig. S7D). Besides, the thickness of neuroepithelium composed of PAX6+ cells akin to VZ-like apical surface was also decreased in HNRNPA3 knockdown organoids compared with vehicle control organoids at D33 (Fig. 9B). With the extension of organoids culture, the overall size of the HNRNPA3 knockdown organoids was also smaller than controls (Fig. S7E). Remarkably, deep layer neurons were largely abolished in HNRNPA3 knockdown organoids (Fig. 9F), although residue differentiated NeuN+ neurons were present (Fig. 9G). This result is consistent with the notion that the Hnrnpa3 defect specifically affected early born deep layer neurons in mice.
We also examined the proportion of cycling NPs that remained at M phase by calculating the percentage of mitotic PH3+ cells among KI67+ cycling cells, and found that the HNRNPA3 knockdown organoids displayed a marked increase in mitotic NPs (Fig. 9D,E), although total number of KI67+ cells had little difference (Fig. 9C). Given that HNRNPA3 was also highly expressed in BPs of human fetal brains (Fig. S1B,C), we examined the number of cells labeled by HOPX, which has been shown to be enriched in oRGs (Pollen et al., 2015). As shown in Fig. 9G, HOPX+ cells were largely reduced in HNRNPA3 knockdown organoids. This might be due to delayed switch from vRG to oRG or to direct effects on oRG proliferation. The effects of HRNRPA3 downregulation on the pool of NPs were also reflected from reduced expression of a set of genes, which encode proteins either as markers or regulators of proliferative NPs involved in human cortex expansion (Fig. S7F). The activity of DNA damage or repair pathway displayed by γH2AX or phosphorylated ATM (pS1981) was also higher in HNRNPA3 knockdown organoids (Fig. S7G-I). In summary, these results suggest that HNRNPA3 plays a crucial role in human cortex development.
In this study we present direct in vivo evidence supporting the conclusion that HNRNPA3 plays an important role in early neurogenesis in cortex development. We found that HNRNPA3 is enriched in cortical NPs with higher proliferation potency, including RGs in both mouse and human, as well as BPs in human, at early cortical developmental stages. The loss or knockdown of Hnrnpa3a resulted in mitotic delay in NPs and impaired cortical neurogenesis (see Fig. 9H for a proposed model). Interestingly, the defects in mice were more pronounced in early-born deep layer neurons than in late-born superficial layer neurons. The temporal dynamic expression pattern of Hnrnpa3 may underlie its specific role, or alternatively, highly proliferative NPs might be more vulnerable to the loss of Hnrnpa3.
It is known that excitatory cortical neurons form six layers, with deep layer neurons (layers VI and V) born early and superficial layer neurons (layers II-IV) born later, and the successive process extends days in the mouse (E10.5-E18.5) to months in humans (Jabaudon, 2017; Marin-Padilla, 1978; Silbereis et al., 2016). During early corticogenesis periods (E11.5-E13.5 in mice), RGs in the VZ can either self-amplify or differentiate into early born deep-layer neurons and this direct neurogenesis gradually decreases as cortical development proceeds (Jabaudon, 2017). At later stages (about E14.5-E16.5 in mice), indirect neurogenesis via IPs in the SVZ dominate to generate superficial layer neurons. Interestingly, the loss of Hnrnpa3a mainly exhibited defects in deep layers, rather than superficial layers. Given that loss or downregulation of Hnrnpa3a caused mitotic delay in RGs at the mouse VZ, we predict that neurogenesis of early born neurons might be more vulnerable to the loss of Hnrnpa3a in NPs. Indeed, it has been hypothesized that fates of various types of cortical neurons might be predetermined by fate-restricted progenitors (Franco et al., 2012). This opinion was revised in a recent study that the fates of superficial layer neurons are affected by environmental signals rather than intrinsic transcriptional programming (Telley et al., 2019). In line with this notion, we did not observe changes in superficial layer neurons. This result is different from a recent study, which showed ectopic and increased neuronal fate specification under a condition with mitotic delay of NPs (Pilaz et al., 2016). This discrepancy might be due to either difference in temporal expression patterns or distinct functions in cell cycle phases. Indeed, it has been suggested that the neuronal fate might be pre-determined in NPs at G1 phase (Arai et al., 2011). Our results suggest that mitotic delay caused by Hnrnpa3a deficiency influences the replenishment of NPs and impairs neuronal competency.
HNRNPA3 was initially identified as a RNA-binding protein with a role in cytoplasmic trafficking of RNA (Ma et al., 2002). It has been reported that HNRNPA3 binds to hexanucleotide repeat (G4C2) RNA of the gene C9orf72, which is believed to be the most common cause of familial amyotrophic lateral sclerosis (ALS) or frontotemporal lobar degeneration (FTLD), and downregulation of HNRNPA3 increased the repeat RNA level and reproduced some features of C9orf72 pathology in neurons (Mori et al., 2013, 2016). Other studies suggest that it binds to telomeric DNA sequence and protects it from nuclease attack in vitro (Huang et al., 2010; Tanaka et al., 2007). Here, we found that HNRNPA3 is distributed in peri-chromosome regions of mitotic NPs (Fig. 5A and Fig. S4A). This localization appears to be dependent on the integrity of microtubules, because microtubule depolymerization disrupted its association with chromosome (Fig. S4A). Notably, loss or downregulation of HNRNPA3 affected normal separation of sister chromatids with the appearance of chromatin bridges in NPs at anaphase, leading to mitotic delay (Fig. 5B,D and Fig. 4A). These effects were most likely due to HNRNPA3 regulation of cohesion complex (Fig. 6). Furthermore, HNRNPA3 has been shown to be associated with telomeres directly (Tanaka et al., 2007). This association might protect integrity of chromosomes during NP division. It has been suggested that early neurogenesis from vRG is more vulnerable to chromosome defects (Lee et al., 2012a,b; Vargas-Hurtado et al., 2019). Considering the role of HNRNP family proteins in regulating alternative RNA splicing (Cáceres and Kornblihtt, 2002; Han et al., 2010), it remains possible that products of impaired alternative splicing may underlie the phenotypes caused by the loss of HNRNPA3. Indeed, evolutionarily or cell type specific alternative splicing has been shown to regulate cortex development (Bae et al., 2014; Zhang et al., 2016).
In humans and other non-human primates, the remarkable expansion of cortical surface is believed to attribute to increased proliferative capacity of NPs (Dehay et al., 2015). Unlike rodents, both apical and basal NPs in primates display high proliferation potency and exhibit similar gene expression patterns (Florio et al., 2015). In cultured human cerebral organoids, downregulation of HNRNPA3 markedly reduced both apical and basal progenitors (Fig. 9A), in line with the observation that HNRNPA3 is highly expressed in human VZ and oSVZ (Fig. 1B), two major regions containing heterogeneous types of proliferative NPs. It appears that the effects on human cortical development are more severe than on mice. One probable reason is that human corticogenesis requires more rounds of NP divisions to meet the need for lateral and vertical expansion, and the mitotic delay caused by downregulation of HNRNPA3 impaired the NP pool size.
Collectively, these results indicate that the role of HNRNPA3 is conserved in mammals and suggest that the NPs with higher proliferative potency are more vulnerable to retardation in cell cycle progression. Considering that defects in NP divisions are major causes of microcephaly, it would be of interest to link HNRNPA3 mutations, if any, with MCPH or other mental retardation diseases. It would also be interesting to check whether HNRNPA3 deletion contributes to mental retardation in individuals with deletion in 2q3.2 (OMIM #612345), where HNRNPA3 gene is located (Mencarelli et al., 2007).
MATERIALS AND METHODS
Fetal brain samples
Human fetal brain samples were obtained after medical pregnancy termination with informed consent, following standard procedure and institutional ethic guidelines in hospital, and with approval of the Shanghai Institutes for Biological Sciences Ethical Committee, Chinese Academy of Sciences (Approval Identifier Number: ER-SIBS-221506). Gestational age was checked by ultrasonic measurements. Fetal brain samples were temporarily preserved in ice-cold Leibowitz-15 medium (Gibco, 21083027) before further laboratory procedures.
The Hnrnpa3 knockout (KO) mice in the C57BL/6 background (MGI: 5812384) were generated with CRISPR/Cas9 technology as submission record (MGI Ref. ID: J:237332). The zygotes at the two-cell stage were transferred into oviducts of pseudo-pregnant females. Newborns were further bred in accordance with standard procedures and animal welfare. The genotype of offspring from heterogeneous crossings was determined by PCR with LA Taq (Takara, RR02AG). The sequences of primers were: 5′-GGATAGCACATCACGTGGTGACTGG-3′ (forward, F) and 5′-GCTAAGTCAGGCCGCGCAGC-3′ (reverse, R). The C57BL/6 pregnant mice were used for in utero electroporation experiments. All animal experiments abided by the guidelines of the Institutional Animal Care and Use Committee of the Institute of Neuroscience, Chinese Academy of Sciences and ShanghaiTech University.
Plasmids and virus preparation
To mark Hnrnpa3 localization, the Hnrnpa3-coding sequence (NM_198090) fused with EYFP sequence at the C terminus was cloned into the pCAG vector using ASCI and AgeI restriction enzymes (New England Biolabs). For overexpression experiments, the Hnrnpa3-coding sequence was cloned into the pCAGGS-EGFP plasmid. As the translation product of human HNRNPA3 is almost identical to that of Hnrnpa3a in mice, the mouse Hnrnpa3 construct was introduced into human cells for the analysis of subcellular localization. The shRNA targeting against conserved sequence between mouse and human HNRNPA3 (5′-AGAGAGCTGTTTCTAGAGA-3′) (Huang et al., 2010) or scrambled sequence (5′-GGTGAGCTATGTACGTAAA-3′) was cloned into pSUPER vector for in utero electroporation experiments, and cloned into the lentivirus vector pTRIPZ, which harbors puromycin resistance, a doxycycline-inducible tetracycline turn-on element and RFP reporter gene, for the infection of H9 human embryonic stem cells.
Cell culture and transfection
Human neural progenitor ReNeuron (ReN) cells were maintained on 15-40 μg/ml laminin (Sigma-Aldrich, L2020-1MG)-coated dishes in the medium containing 2% B27 without VA (Gibco, 12587010), 10 units/ml heparin (Sigma-Aldrich, H3149-100KU), 20 ng/ml EGF (Millipore, GF155) and 10 ng/ml bFGF (STEMCELL, 78003) in DMEM/F12 (Gibco, 11330057). After forced expression of HNRNPA3 in ReN cells, culture medium was replaced with differentiation medium [0.5×N2 supplement, 0.5×B27 supplement (Gibco, 17504044), in 50% DMEM/F12, 50% Neurobasal (Gibco, 12348017)]. ReN cells were transfected with Lipofectamine 3000 Transfection Reagent (Invitrogen, L3000015). ReN cells were treated with nocodazole (Sigma, M1404-2MG) (100 ng/ml) for 6 h to induce microtubule de-polymerization.
RNA extraction and qPCR analysis
Total RNA extracted from two or three organoids were isolated with RNeasy Plus Micro Kit (Qiagen, Germany, 74034). For reverse transcription, 2 μl total mRNA was processed with GoScript Reverse Transcription Kit (Promega, A5001). Quantitative PCR using SYBR Green PCR mix (Biotool, B21702) was performed on Roche LightCycler with primers listed in Table S1. For each transcript, raw values were normalized with that of GAPDH.
Human ES cell and brain organoid culture
H9 human embryonic stem (hES) cells were infected by lentivirus with pTRIPZ vector encoding siRNA against HNRNPA3 or control sequence and cultured for 3 days on Matrigel (Corning, 354277)-coated dishes in mTeSR1 medium (STEMCELL, 5850), followed by addition of puromycin (1 μg/ml) for 4 days to select infected clones. After at least two rounds of selection, active proliferative clones were digested with enzyme-free passage reagent ReLeSR (STEMCELL, 05872), passaged in a 1:10-1:40 ratio and cultured in puromycin-containing mTeSR1. Then, hES clones were digested with Accutase (Sigma-Aldrich, A6964-100ML) to generate a single cell suspension, followed by incubation with mTeSR1 containing 10 μM Y27632 to allow embryonic body (EB) formation. Around 7000 cells were seeded into each well of V-bottom 96-well plate (NUNC, 277143) coated with lipidure (Nof, CM5206). The time point of EB induction was designated as day 0. After 2-3 days, the culture medium was replaced with the medium containing 20% Knockout Serum Replacer (Gibco, 10828028), 1×MEM-NEAA (Gibco, 11140050), 3.5 μl/l β-mercaptoethanol (Sigma-Aldrich, M3148), 1×Glutamax (Gibco, 35050061), 2.5 μM dorsomorphine (Tocris, 3093/10) and 2 μM A83-01 (Tocris, 2939/10), in DMEM/F12 for ectodermal induction. After another 4 days, EBs were placed in neural induction medium containing 1×N2 Supplement (Gibco, 17502048), 1×Glutamax (Gibco), 1×MEM-NEAA (Gibco) and 1 μg/ml heparin in DMEM/F12 with half the medium renewed every other day. To turn on the siRNA expression, 1 μg/ml doxycycline was added into the medium from the second day of neural induction. After 6 days of neural induction, the neuroepithelial organoids were embedded into growth factor-reduced Matrigel droplet, as described previously (Lancaster and Knoblich, 2014), and fed with differentiation medium [0.5×N2 supplement, 0.5×B27 supplement without vitamin A, 3.5 μl/l β-mercaptoethanol, 250 μl/l Insulin (Sigma-Aldrich, I9278), 1×Glutamax and 0.5×MEM-NEAA, 1×Antibiotic-Antimycotic (Gibco, 15240096), in 50% DMEM/F12 and 50% Neurobasal medium)], with medium renewed every other day. After 4 days in differentiation medium, the organoids were transferred into maturation medium [0.5×N2 supplement, 0.5×B27 supplement (Gibco, 17504044), 3.5 μl/l β-mercaptoethanol, 250 μl/l insulin, 1×Glutamax and 0.5×MEM-NEAA, and 1×Antibiotic-Antimycotic, in 50% DMEM/F12, 50% Neurobasal medium] and then placed on a shaker in a 5% CO2 incubator and cultured at 37°C for different days with medium renewal every 2-3 days.
Control or shA3-expressing retrovirus (1 μl, ∼1×107 TU/ml) was injected into the telencephalic ventricles of E12.5 mice. Four days later, embryonic brains were fixed, dehydrated and cryostat-sectioned at 35 µm, followed by immunostaining. GFP+ cells aligned along a single vertical trajectory from apical to basal axis were considered as the progeny of a neural progenitor.
In utero electroporation
Pregnant mice at different gestation days were anesthetized with pentobarbital sodium, then plasmids (2 μg/μl) mixed with 0.25% Fast Green (Sigma-Aldrich) in sterile PBS were gently injected into the telencephalic ventricles of each embryos using a glass micropipette (VWR International). Immediately after injection, embryonic brains were subjected to electroporation at given pulses (28 V, 5×50 ms pulses with 1 s interval) with two 3 mm (in radius) disc electrodes (BEX, Japan, LF650P3) using an electroporator (BTX, ECM830).
Cell-cycle kinetics analysis
To label NPs at S-phase, the pregnant female mouse was injected intraperitoneally with BrdU (Sigma-Aldrich) at 50 mg/kg. To trace mitotic progenitors, 1 μl CellTrace reagent carboxy-fluorescein esters (shorted as CFSE) (Invitrogen, C34554) was injected into the telencephalic vehicle of E12.5 mice; 22 h later, pregnant mice were subjected to intraperitoneal EdU (100 μg) injection. At E13.5, brain slices were stained with antibodies against EdU or Dcx to measure cell cycle re-entry or neuronal differentiation. To measure mitotic index, embryos were fixed 6 h after EdU pulse at E12.5. To label proliferative ReN cells, EdU (10 μM) was added into differentiation medium and cultured for an additional 2 h. The immunofluorescence signals of BrdU and EdU were measured following the manufacturer's protocol (Abcam).
Histological analysis, immunostaining, immunoprecipitation and immunoblotting
Tissues from mice and organoids were fixed with 4% PFA in 1×PBS at 4°C overnight, followed by dehydration with 20-30% sucrose in PBS. Tissues were then embedded in OCT and cryosectioned at 20-40 µm thickness. Lung sections were stained with Hematoxylin and Eosin (HE) for morphological analysis. To optimize the immunofluorescence signals, sections were incubated in antigen retrieval solution (Beyotime Biotechnology, P0090) for 20-60 min at room temperature and washed three times with PBS, followed by incubation in 0.3% Triton X-100 for around 30 min. To avoid background staining, sections were blocked with 5% BSA in PBS at room temperature for 1 h or at 4°C for more than 2 h. Slices were incubated with primary antibodies in blocking solution overnight at 4°C; after extensive washes, they were incubated with secondary antibodies for about 24 h at 4°C. For immunofluorescence observation of cultured cells, the fixation was shortened to 15 min at room temperature and permeabilization was adjusted to 0.25% Triton X-100 for about 10 min. For organoid sections, the incubation of primary antibodies was prolonged to 2 days. HNRNPA3 antibody was conjugated with Cy3 using Fast Conjugation Kit (Abcam, ab188287). For immunoprecipitation, 30 μl protein A/G magnetic beads (Bimake, B23201) were washed with RIPA buffer and incubated with antibody against SMC1A or IgG at 4°C overnight followed by three washes using RIPA buffer to clear unbound antibody, and then incubated with cell lysates with rotation overnight at 4°C. After three RIPA washes, bound proteins were eluted from beads by boiling with 1× loading buffer for 10 min.
For immunoblotting, adherent cells were harvested using 0.05% Trypsin and washed several times with ice-cold PBS. Cell pellets collected by centrifugation were lysed in RIPA buffer [50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate] containing Protease Inhibitor Cocktail Set III (Millipore, 539134-1SETCN) for 20 min on ice. Further protein solutions were prepared by centrifugation at 16,000 g for 15 min at 4°C and boiled with 5×SDS sample buffer for 5 min at 95°C. For tissue samples, we used mechanical cutting and RIPA treatment for a longer time to enhance the lysis. For immunoblotting, lysates were loaded in 12% SDS-PAGE and then transferred onto 0.45 µm PVDF membranes at 100 V. Subsequently, PVDF membranes were blocked in 5% BSA in TBS-T for 1 h at room temperature, then incubated with primary antibodies diluted in 3% BSA/TBS-T at 4°C for more than 16 h, followed by three washes in TBS-T and incubation with HRP-conjugated secondary antibodies diluted in 3% BSA/TBST at room temperature for 2 h or 4°C overnight. Finally, blots were washed several times in TBS-T and imaged using a Tanon 4600 Chemiluminescent Imaging System. The primary antibodies used are listed in Table S2.
Live imaging of cortical slice was performed as described previously (Pilaz and Silver, 2014) with minor modifications. Briefly, embryonic brains were sectioned at 250 μm using a Leica VT1200S vibratome. After Hoechst (ThermoFisher, H3570) staining for 8 min, slices were mounted on glass-bottomed dishes (WPI) with Matrigel. Slices were cultured in slice culture medium (Pilaz and Silver, 2014) containing ProLong Live Antifade Reagent (ThermoFisher, P36975). Live imaging was performed at 37°C in 5% CO2 on FV10 microscope (Olympus).
Quantification and statistical analysis
For phenotypic analysis, Hnrnpa3 KO mice and their wild-type littermates from at least three mothers were analyzed. For in utero electroporation and virus injection experiments, each experiment was performed in three or four pregnant mice in each group. The results were obtained from slices at the same lateral-medial level and the rostral-caudal axis and all images were processed with Fiji software. Raw digital data were analyzed by Graphpad Prism 6 and results are presented as mean±s.e.m. Quantitative differences were determined using the Kolmogorov–Smirnov test, an unpaired t-test, an unpaired t-test with Welch's correction or a Mann–Whitney test. The statistical significance was indicated as: *P<0.05; **P<0.01; ***P<0.001. All tests were two-tailed.
Conceptualization: Z.-G.L.; Methodology: M.-Y.O., X.-C.J., Y.-J.C., X.-Y.S.; Validation: M.-Y.O., J.-F.W.; Formal analysis: M.-Y.O., X.-C.J.; Investigation: M.-Y.O., Y.-J.C., X.-Y.S., J.-F.W., X.-Q.F., Q.S.; Resources: Q.S., Z.-G.L.; Data curation: M.-Y.O., X.-C.J., J.-F.W., X.-Q.F.; Writing - original draft: M.-Y.O., Z.-G.L.; Writing - review & editing: Z.-G.L.; Supervision: Z.-G.L.; Project administration: Z.-G.L.; Funding acquisition: Z.-G.L.
This study was partially supported by grants from the National Natural Science Foundation of China (31490591 to Z.-G.L. and 31871034 to X.-C.J.), the National Key R&D Program of China (2017YFA0700500), the Frontier Key Project of the Chinese Academy of Sciences (QYZDJ-SSW-SMC025), the Shanghai Municipal Science and Technology Major Project (2018SHZDZX05) and the start-up support of ShanghaiTech University.
Peer review history
The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.185132.reviewer-comments.pdf
The authors declare no competing or financial interests.