mRNA stability fine-tunes gene expression in the developing cortex to control neurogenesis

To characterize the mRNA stability landscape of the developing cortex, we performed thiol(SH)-linked alkylation for metabolic sequencing (SLAM-seq) [36] at 3 different developmental stages. We chose E11.5, E14.5, and E16.5 as representative stages of early, middle, and late neurogenesis, respectively. E11.5 cortices are predominantly composed of RGCs, while E14.5 and E16.5 cortices contain increasing numbers of IPs and neurons, along with non-RGC derived cells. Cortices from E11.5, E14.5, and E16.5 embryos were dissociated into single cell suspensions using 3 independent biological replicates per stage and cultured in vitro using conditions permissive for proliferation [37], and 4-thiouridine (4sU) was added to culture media for 20 h to label nascent RNAs. Cells were then collected for RNA extraction 0, 2, 4, 8, and 12 h following a chase with excess unmodified uridine to monitor degradation of labeled RNAs over time (Fig 1B). Following sequencing, we obtained reproducible half-life measurements for 10,842 transcripts at E11.5, 10,671 transcripts at E14.5, and 11,832 transcripts at E16.5 (Fig 1C and S1 Table). Principal component analysis (PCA) showed that the biological replicates within each stage largely clustered together but were distinct between the stages (S1A Fig). Comparison of z-score normalized half-lives revealed high correlations between stages (r > 0.87), indicating that the relative stability of individual transcripts is highly consistent across cortical development (S1B and S1C Fig).

To further compare the RNA stability landscape between stages, we assessed the top 10% most and least stable transcripts at E11.5, E14.5, and E16.5. We observed a high degree of similarity between the 3 stages. Among the most stable transcripts, 456 were shared among all 3 stages (Fig 1D). Similarly, 491 of the least stable transcripts were shared across all 3 stages (Fig 1E). Notably, more transcripts were shared between all 3 stages than were unique to each stage. Gene ontology (GO) analysis of the most and least stable transcripts at each stage revealed consistently enriched terms shared across developmental stages (S1D and S1E Fig). Stable transcripts were enriched for functions such as ATP synthesis and protein synthesis, while unstable genes encoded regulators of transcription and cell fate determination. Importantly, these functional classes for stable and unstable transcripts are similar to those observed in other cell types in diverse organisms [36,38–41].

We next sought to identify cis-acting factors that shape RNA turnover dynamics in cortical cells. We first focused on 5′ and 3′ UTRs as they are known to contain regulatory elements important for RNA stability [42,43]. We examined 3′ UTR sequences of the same top 10% most stable and least stable transcripts at each stage [44]. The most stable RNAs had shorter, more GC-rich, 3′ UTRs (S1F and S1G Fig). Analysis of 5′ UTR sequences revealed a similar relationship between length and stability, with stable transcripts having shorter 5′ UTRs (S1H Fig). However, no relationship was observed between 5′ UTR GC content and stability (S1I Fig). The preferential impact of 3′ UTR GC content on RNA stability may reflect the enrichment of regulatory elements within these regions [45].

In addition to regulatory elements within UTRs, codon content within coding regions has also been implicated in mRNA stability regulation in multiple biological contexts [46–51]. To determine whether mRNA stability in the developing mouse cortex is associated with codon content, we calculated the codon stabilization coefficient (CSC), a measure of the correlation between codon usage and mRNA half-life [50]. We found robust correlations between codon usage and mRNA stability across all 3 stages, with CSC values ranging from 0.17 (stabilizing/optimal) to −0.2 (destabilizing/non-optimal) (Fig 1F). CSC values for each codon were virtually identical between E11.5, E14.5, and E16.5 (S1J and S1K Fig). To further assess the impact of codon content on mRNA stability, we binned mRNAs into quartiles based on their optimal codon content and then compared the half-lives of each quartile. We found a clear and consistent trend across all 3 developmental stages showing that mRNAs with a higher proportion of optimal codons are more stable (Fig 1G).

Finally, we examined the impact of m⁶A, an abundant epitranscriptomic mark associated with multiple aspects of mRNA metabolism, including mRNA stability [52]. m⁶A was previously shown to be present in >2,000 transcripts in E14.5 mouse primary cortical cells, where it was implicated in RNA degradation [19]. To assess the destabilizing impact of m⁶A in our data set and to determine whether its role changes across development, we compared half-lives of mRNAs predicted to contain m⁶A to those without m⁶A at each stage [19] (Fig 1H). This analysis showed significantly lower half-lives for m⁶A-containing transcripts, with similar extents of destabilization at all stages. This strongly implicates m⁶A in RNA stability throughout neurogenesis, reinforcing previous findings [19]. Collectively, our data highlight multiple cis-acting features that shape RNA stability in cortical cells.

Although we did not observe widespread shifts in RNA half-lives across development, some variation among individual transcripts was observed (S1B and S1C Fig). We therefore aimed to identify transcripts with developmentally regulated RNA half-lives. For this, we employed a regression-based method [53] to assess RNA degradation kinetics across time in our SLAM-seq data. We first filtered the data to include only the 9,490 transcripts expressed at all 3 stages, then converted half-lives to z-scores to normalize the data between stages and allow for direct comparisons. Regression analysis was then used to identify transcripts with significant changes in stability across development. Consistent with our previous observations, half-lives for most transcripts were unchanged. However, a small subset of 294 transcripts (3.1%) had significant differences in stability across all 3 stages assayed. Hierarchical clustering of this subset based on their half-life dynamics identified 7 distinct clusters (Fig 2A and S2 Table). The temporal half-life profile for each cluster was unique (Fig 2B). Stability of cluster 2 and 3 RNAs gradually increased, while the stability of cluster 5 and 7 RNAs gradually decreased. Cluster 1 and cluster 4 stability was greatest at E16.5 and E14.5, respectively, while cluster 6 transcripts were least stable at E14.5. Altogether, these data demonstrate that turnover rates for individual transcripts are on average constant across cortical development but are dynamically regulated for a subset of transcripts.

Fig 2. Dynamic patterns of mRNA half-lives across cortical development.

(A) Heatmap showing z-score normalized half-lives for 294 significantly changing transcripts at E11.5, E14.5, and E16.5 (n = 3 biological replicates per stage). Numbers to the right indicate cluster labels. (B) Z-score normalized half-lives for transcripts found in the indicated clusters. *p p S2 Data.

https://doi.org/10.1371/journal.pbio.3003031.g002

Neural progenitors dynamically remodel their transcriptomes over developmental time and upon differentiation [7]. Yet, the extent to which RNA stability influences these transcriptional dynamics is largely unknown. To investigate this question, we assessed the stability of RNAs whose expression significantly changes across development. Because SLAM-seq is conducted using standard RNA-seq, our data set measures RNA expression levels at each stage, in addition to their half-lives. Using this data, we evaluated differential expression across the 3 stages. Using cutoffs of fold change >2 and p_adj Fig 3A and S3 Table). Transcripts that were up-regulated from E11.5 to E14.5 included neuronal markers (e.g., Neurod2, Satb2, Dcx), while down-regulated transcripts included progenitor markers (e.g., Sox2, Nes, Hes5). These gene expression changes may reflect the increased abundance of neurons at later stages of development. We further identified 962 up-regulated and 730 down-regulated genes at E16.5 compared to E14.5 (Fig 3B and S3 Table). Similar to changes from E11.5 to E14.5, we observed increased expression at E16.5 of neuronal markers (e.g., Satb2, Cux2, Pou3f3/Brn1) and decreased expression of progenitor and cell cycle markers (Ccnd2, Sox2, Btg2). Notably, these data are consistent with transcriptional dynamics that are expected to occur in vivo during cortical development.

Fig 3. Distinct stability profiles for developmentally regulated and cell type enriched mRNAs.

(A, B) Volcano plots showing transcript expression changes at E14.5 compared to E11.5 (A), or E16.5 compared to E14.5 (B). n = 3 biological replicates per stage. (C, D) Half-lives of differentially expressed transcripts at the indicated developmental transitions. (E) Half-lives for a subset of transcripts enriched in either progenitors (n = 56) or neurons (n = 57). **p p S3 Data.

https://doi.org/10.1371/journal.pbio.3003031.g003

To determine whether there is a relationship between RNA stability and gene expression dynamics across development, we compared half-lives of differentially expressed transcripts. We found that RNAs whose expression is significantly up-regulated at E14.5 compared to E11.5 were marginally, but significantly, more stable compared to those that were down-regulated (Fig 3C). The same trend was observed for RNAs that were differentially expressed at E16.5 compared to E14.5 (Fig 3D). These data suggest that expression dynamics and RNA stability are coordinated. Developmentally down-regulated transcripts undergo rapid turnover, while up-regulated mRNAs are degraded more slowly to facilitate their accumulation.

Some of these developmentally regulated transcripts are known to have cell-specific expression. Given this, we investigated whether this held true across more of the transcriptome. For this, we mined a previously published scRNA-seq data set of the developing cortex [6] to assess expression levels of developmentally regulated transcripts in specific cell types. At both transitions (E11.5 to E14.5 and E14.5 to E16.5), transcripts with decreasing expression were enriched in progenitors (RGCs and IPs), while those with increasing developmental expression were enriched in neurons (S2A and S2B Fig). Together, this suggests that gene expression differences in our data sets are linked to changes in cell composition across development.

Our data demonstrate that some developmentally regulated transcripts exhibit biased cell-type expression profiles (S2A and S2B Fig) and distinct stability profiles (Fig 3C and 3D). Therefore, we next sought to determine whether transcripts enriched in progenitors and neurons have distinct RNA half-lives. To address this, we generated a stringent list of transcripts that are highly enriched in either progenitors (n = 56) or neurons (n = 57) across 2 independent scRNA-seq data sets [6,7] (S4 Table). In our SLAM-seq data set, these transcripts were expressed at all 3 stages, with progenitor-enriched transcripts most highly expressed at E11.5, and neuron-enriched transcripts most highly expressed at E16.5 (S2C and S2D Fig). These expression patterns are consistent with known cell composition at each stage. Notably, progenitor-enriched transcripts were significantly less stable than neuron-enriched transcripts (Fig 3E). This was true across all 3 developmental stages. Collectively, these data demonstrate cell-specific layers of RNA stability during cortical development and strongly suggest that RNA stability influences developmental expression changes for a subset of transcripts. These data indicate developmental roles of RNA stability for shaping the transcriptome.

Our data thus far indicate that RNA stability regulation is a key feature of cortical development, implicated in gene expression changes across stages. Our data also indicate that codon optimality is highly correlated with mRNA stability in the embryonic cortex (Fig 1G). Mechanistically, codon content and its effects on translation are sensed by the CCR4-NOT complex to drive degradation of non-optimal mRNAs [54]. This led us to investigate the in vivo functional consequences of mis-regulating the CCR4-NOT complex during cortical development. To address this, we focused on CNOT3, a central component of the CCR4-NOT deadenylase complex (Fig 4A). CNOT3 is essential for deadenylation [21–23] and is linked to neurodevelopmental pathologies [24–28], making it a strong candidate for investigating the role of the complex in cortical development. Importantly, inspection of scRNA-seq data sets of the developing mouse cortex revealed broad Cnot3 expression in both progenitors and neurons [7]; however, its function in these cell types is unknown.

Fig 4. Cnot3 controls cerebral cortex size and neuronal number.

(A) Schematic of CCR4-NOT complex with CNOT3 indicated in red. (B) cKO strategy using Emx1-Cre to target RGCs and their progeny, including IPs and neurons. (C) Western blot using whole cortex protein lysates from E12.5 control (ctrl), conditional heterozygous (cHet), or cKO embryos. Quantifications shown to the right (n = 4 embryos per genotype). (D) Whole mount images and coronal cross sections of E18.5 brains. Quantifications of cortex area and thickness shown to the right (n = 4 embryos per genotype). (E–H) Immunofluorescence for indicated neuronal markers in E18.5 cortical sections, with quantifications shown to the right. n = 3–5 embryos per genotype. *p p p S4 Data. cKO, conditional knockout; IP, intermediate progenitor; RGC, radial glial cell.

https://doi.org/10.1371/journal.pbio.3003031.g004

We generated conditional knockouts (cKOs) of Cnot3 in RGCs and their progeny by crossing Cnot3^lox/lox mice with Emx1-Cre [55] (Fig 4B). At E12.5 (approximately 3 days after the onset of Cre-mediated recombination), we observed an 84% and 82% reduction in CNOT3 protein and RNA expression, respectively, in cKO cortices (Figs 4C and S3A). Interestingly, expression of CNOT1 was also reduced. Reduction of CNOT1 protein likely occurs via translational and/or posttranslational mechanisms, as we observed no significant change in Cnot1 RNA measured by RT-qPCR (S3A Fig). No statistically significant changes were observed for CNOT2 expression at either the protein or RNA levels. As CNOT1 is essential for scaffolding the complex, this suggests that the integrity of the entire CCR4-NOT complex may be impaired in this mouse model. These data demonstrate the Cnot3 cKO can be faithfully used to assess roles of this complex in cortical development.

To determine the requirement of Cnot3 for cortical development, we first examined brains at E18.5, when neurogenesis is largely complete. cKO mice exhibited profound microcephaly, with a 38% reduction in cortical area, and a similar 38% reduction in cortical thickness (Fig 4D). In contrast, conditional heterozygous (cHet) brain sizes were comparable to control. This indicates that CNOT3 is required for overall cortical size. The cortex forms a six-layered laminar architecture, with early born neurons present in deep layers and later born neurons primarily present in superficial layers [3]. To examine the extent to which Cnot3 ablation affects glutamatergic projection neurons, we performed immunostaining for TBR1 (Layer VI), CTIP2 (Layer V), RORβ (Layer IV), and LHX2 (Layer II/III). In cKO brains compared to control, we observed a ~28% reduction in TBR1+ neuron density, a near complete loss of CTIP2+ and RORβ+ neurons, and a 65% reduction in LHX2+ neurons (Fig 4E–4H). cHet brains were similar to control. In addition to reduced density, neuronal distribution for some markers was also disrupted in cKO brains (S3B–S3E Fig). These data indicate that CNOT3 is essential for number and laminar position of glutamatergic neurons produced throughout cortical development.

To understand how Cnot3 controls brain size and neuron number, we next examined its requirements for neurogenesis. At E14.5, mid-neurogenesis, we observed an 18% reduction in cortical thickness in Cnot3 cKO mice, indicating a role for Cnot3 in development at earlier stages (Fig 5A and 5B). We performed immunofluorescence against SOX2 and PAX6 to assess RGCs. Both markers were qualitatively reduced in the proliferative zones (S4A Fig). Indeed, we observed a 26% reduction in apical SOX2+ RGCs (Fig 5C). We also assessed IPs using TBR2 expression. This showed a 43% reduction in IPs in cKO cortices compared to control (Fig 5D). This demonstrates that Cnot3 is critical for proper number of both RGCs and IPs.

Fig 5. Cnot3 is required for progenitor cell number and cell survival.

(A) Immunofluorescence for SOX2 and TBR2 in E14.5 cortical sections, with quantifications shown in (B–D). n = 3–6 embryos per genotype. (E) Immunofluorescence for CC3 in E11.5 (top) and E12.5 (bottom) cortical sections. (F) Quantification of CC3+ cells from genotypes in (E) (n = 3–5 embryos per condition). (G) Immunofluorescence for CC3 and TUJ1 in an E12.5 cKO cortical section. (H) Quantification of percentage of CC3+ cells that were either TUJ1+ or TUJ1- in E12.5 cKO cortices (n = 3 embryos). *p p p S5 Data. cKO, conditional knockout; CC3, cleaved caspase 3.

https://doi.org/10.1371/journal.pbio.3003031.g005

The reduction in cortical thickness and decreased progenitor and neuronal cell density could be associated with alterations in proliferation or excessive cell death. We first assessed the latter possibility by measuring apoptosis. Staining for cleaved caspase 3 (CC3) revealed minimal cell death in control, cHet, and cKO cortices at E11.5 (Fig 5E and 5F). However, beginning at E12.5, we observed widespread cell death that was predominant in cKO cortices (Fig 5E and 5F). Apoptotic cells were present throughout the cortex and were both TUJ1+ and TUJ1- (Fig 5G and 5H). This indicates that Cnot3 loss induces apoptosis of both newborn neurons and presumed progenitors.

We next sought to determine whether neuronal cell death was strictly a result of Cnot3 loss in progenitors, or if Cnot3 was also required in neurons for viability. For this we crossed Cnot3^lox/lox mice with Nex-Cre (S4B Fig), which is active in post-mitotic glutamatergic neurons beginning at E12.5 [56]. Staining for CC3 at E18.5 indicated apoptosis in Nex-Cre cKO cortices (S4C Fig). This indicates that neuronal cell death is due, at least in part, to an intrinsic requirement of Cnot3 for newborn neuron viability. Staining for TBR1, CTIP2, RORβ, and LHX2 at E18.5 revealed reductions of 33%, 20%, 18%, and 10%, respectively, with only the deep layer neurons being significant (S4D–S4H Fig). The attenuated impact on superficial layers is most likely due to the later birthdate of these neurons and the time it takes for CNOT3 protein to be depleted after Cre-mediated recombination. Collectively, these data underscore the intrinsic requirement of CNOT3 for viability of multiple cell types of the developing cortex, including both neural progenitors and post mitotic neurons.

Given that Cnot3 is essential for cell viability, we next assessed the extent to which apoptosis contributes to the loss of both progenitors and neurons in Emx1 cKO brains. Cell death in the cortex is often associated with up-regulation of p53 signaling [10,57–59]. Consistent with this, we observed accumulation of p53 protein in E12.5 cKO cortices (Fig 6A), indicating active p53 signaling at this stage. Given this, we generated Emx1-Cre;Trp53^lox/lox;Cnot3^lox/lox double cKO (dcKO) mice to abrogate p53-dependent apoptosis. Rescue of p53-dependent apoptosis was confirmed by the quantification of p53 protein accumulation and CC3+ cells in E12.5 dcKO cortices (S5A and S5B Fig). Examination of dcKO brains at E18.5 revealed a striking rescue of cortical size (Fig 6B and 6C). Notably, cortical thickness in dcKO brains was virtually unchanged compared to cKO (Fig 6D). The density of early born, deep layer neurons marked by TBR1 and CTIP2 was similar between dcKO and cKO mice (Fig 6E and 6F). In contrast, the density of later born neurons marked by RORβ and LHX2 was restored in dcKO mice (Fig 6G and 6H). This suggests that p53-dependent apoptosis explains the loss of late-born, but not early-born neurons in Cnot3 cKO mutants. As with density, the compound mutants also failed to rescue lamination patterns for deep layer markers, but did rescue lamination of superficial layers markers (S5C–S5F Fig). These data indicate both p53-dependent and p53-independent mechanisms for CNOT3 regulation of neuronal populations.

Fig 6. p53-dependent apoptosis controls upper but not deep layer neurons.

(A) Immunofluorescence for p53 in E12.5 cortical sections. (B) Whole mount images and coronal cross sections of E18.5 brains. (C, D) Quantifications of cortex area and thickness (n = 3–6 embryos per genotype). (E–H) Immunofluorescence for indicated neuronal markers in E18.5 cortical sections, with quantifications shown to the right (n = 3–6 embryos per genotype). (I) Immunofluorescence for SOX2 and TBR2 in E14.5 cortical sections. (J–L) Quantification of cortical thickness, SOX2, and TBR2 at E14.5 (n = 2–7 embryos per genotype). *p p p S6 Data.

https://doi.org/10.1371/journal.pbio.3003031.g006

The reduction in superficial layer neurons may be due to apoptosis of progenitors at stages when these neurons are being produced. To probe this possibility, we quantified neurogenesis of compound mutant brains at E14.5. Similar to E18.5, cortical thickness at E14.5 was unchanged in dcKO brains compared to cKO (Fig 6I and 6J). However, the densities of apical SOX2+ RGCs and TBR2+ IPs were rescued in dcKO brains (Fig 6K and 6L). Thus, the decrease in progenitor density in E14.5 Emx1-cKO brains is due to p53-dependent apoptosis. This observation is consistent with the rescue of upper layer neurons, whose peak production occurs around E14.5-E15.5. Collectively, our analysis of dcKO mice suggests that Cnot3-dependent microcephaly and neuronal loss are due in part to massive apoptosis.

Given the findings above, we next asked if Cnot3 controls the rate at which early-stage progenitors produce neurons. For this, we employed a previously established in vivo semi-cumulative labeling strategy to measure total cell cycle (Tc) and S-phase (Ts) durations at E12.5 [60]. BrdU was intraperitoneally injected into pregnant dams to label DNA in S-phase progenitors, followed by a second injection with EdU 1.5 h later (Fig 7A). Embryonic brains were dissected at t = 2 h and stained for BrdU, EdU, and Ki67 (proliferative cells). Total cell cycle and S-phase in control cortices were determined to be ~10 h and ~5 h, respectively (Fig 7B). Importantly, both measurements are in close agreement with previous measurements [61,62]. In contrast, in cKO cortices, total cell cycle and S-phase durations were both significantly increased to ~14 h and ~6.5 h, respectively (Fig 7B). The ratio of Ts/Tc was unchanged across the 3 genotypes, indicating that cell cycle lengthening was not biased towards S-phase (Fig 7C). These data demonstrate that Cnot3 loss from early-stage progenitors is associated with overall slower cell cycle.

Fig 7. Cnot3 controls neural cell fate by modulating progenitor cell cycle duration and neurogenic divisions.

(A) Top, schematic of experimental approach for semi-cumulative labeling at E12.5. Bottom, immunofluorescence for the indicated markers in E12.5 cortical sections. (B) Quantification of total cell cycle duration (T_c), S-phase (T_s) (n = 4 embryos per genotype). (C) Quantification of the ratio between S-phase and total cell cycle (T_s/T_c). (D) Top, schematic of experimental approach for live imaging of E12.5 primary progenitor divisions. Bottom, DIC images at the beginning (t = 0 h) and end (t = 20 h) of the live imaging session and immunofluorescence after fixation for SOX2, TBR2, and TUJ1 showing representative proliferative and neurogenic divisions (n: Ctrl = 4 brains (446 divisions), cHet = 3 brains (331 divisions), cKO = 3 brains (350 divisions)). (E) Quantification of cell divisions: Proliferative (2 SOX2+TUJ1-, 2TBR2+TUJ1-, or one of each); Asymmetric neurogenic (1 SOX2+TUJ1- or TBR2+TUJ1- and 1 TUJ1+); Symmetric neurogenic (2 TUJ1+). *p p p p-values (E). Error bars represent standard deviation. Scale bars: 50 μm (A), 10 μm (D). Underlying data for this figure can be found in S7 Data. cKO, conditional knockout.

https://doi.org/10.1371/journal.pbio.3003031.g007

We next investigated the extent to which Cnot3 controls early stage progenitors’ ability to produce neurons. We first measured if the reduced number of early stage progenitors is due to cell death using p53 mutants. In dcKO brains, we observed rescue of proliferative cell number, suggesting p53 contributes to E12.5 RGC number (S5G and S5H Fig). Notably, in contrast to this phenotype, deep layer neurons are not rescued by p53 (Fig 6E and 6F). Given this, we tested the possibility that CNOT3 controls fate of newborn neurons at early stages when these neurons are being generated. For this, we performed a live imaging assay to monitor the fate of dividing E12.5 progenitors and their progeny (Fig 7D and S1 and S2 Movies). Primary cell cultures were prepared from E12.5 control, cHet, and cKO cortices, and progenitors were imaged every 10 min for a period of 20 h, as previously [63,64]. After 20 h, cells were fixed and stained for SOX2 (RGCs), TBR2 (IPs), and TUJ1 (neurons) to assign the fate of progeny. In control progenitors, approximately 60% of divisions were proliferative (2 SOX2+ or TBR2+ progeny), approximately 10% were asymmetric neurogenic (one SOX2+ or TBR2+, and one TUJ1+ progeny), and approximately 30% were symmetric neurogenic (two TUJ1+ progeny). In contrast, divisions of cKO progenitors were skewed towards producing more progenitors, with approximately 80%, 5%, and 15% of divisions being proliferative, asymmetric neurogenic, and symmetric neurogenic, respectively (Fig 7E). cHet progenitors had an intermediate phenotype that was not statistically significant. The observation that progenitors undergo fewer neurogenic divisions in cKO cells suggests that Cnot3 loss from progenitors directly impairs neuronal production. This, together with prolonged cell cycle, suggests potential mechanisms to help explain the p53-independent loss of deep layer neurons.

Next, we aimed to understand the molecular consequences of Cnot3 loss in the developing cortex. For this, we isolated RNA from E12.5 control, Cnot3 cKO, and Cnot3;Trp53 dcKO cortices and performed bulk RNA-seq. We chose E12.5, the earliest onset of phenotypes in cKO embryos, with the goal of understanding primary molecular deficits. Using cutoffs of p_adj 2, we identified 1,216 transcripts that were differentially expressed in cKO cortices compared to control, and 1,164 differentially expressed transcripts in dcKO cortices (Fig 8A and S5 Table). Targets of p53 transcriptional activation, including Sesn2, Eda2r, Ccng1, and Cdkn1a, were among the most up-regulated genes in cKO cortices (Fig 8A and 8B). In dcKO cortices, these transcripts were each expressed at levels similar to control, confirming attenuation of p53-dependent apoptosis in the compound mutant mice. These expression changes were independently validated by RT-qPCR (Fig 8B). Importantly, this data provides additional molecular support for p53-dependent apoptosis as a major mechanism of Cnot3 cKO phenotypes. Given the altered cell cycle duration of progenitors in cKO cortices (Fig 7A–7C), we also investigated expression levels of transcripts involved in cell cycle regulation. Inspection of RNA-seq data revealed that Ccnd1, Ccnd2, and Cdk4, were down-regulated in cKO and dcKO cortices (S6A Fig). These data provide molecular support for cell cycle phenotypes reported above and suggest potential underlying mechanisms.

Fig 8. Transcriptomic and stability analysis reveals molecular targets of CNOT3.

(A) Volcano plots showing differentially expressed transcripts in cKO (left) and dcKO (right) cortices versus control at E12.5 (n = 3 embryos per genotype). (B) Expression of p53 target genes as measured by RNA-seq (left) or qPCR (right). (C) Venn diagram showing overlap between differentially expressed genes in cKO and dcKO cortices. (D) Log₁₀(expression) of cKO/dcKO shared up-regulated genes (Up; n = 431), shared down-regulated genes (Down, n = 289), and all other genes (NS, n = 13,089) in Ctrl cortices. (E) Percentage of optimal codons (CSC > 0) for shared differentially expressed genes. (F) Log₂(fold change) in dcKO versus control cortices for all genes, binned into quartiles based on percentage of optimal codons. (G) Left, schematic of experimental approach for transcriptional shutoff of E12.5 primary cultures with actinomycin D (Act D). Right, half-lives of the indicated transcripts determined by RT-qPCR (n = 4 embryos per genotype). *p p p p-values from DESeq2 (B (RNA-seq)). One-way ANOVA with Tukey’s HSD post hoc test (B (qPCR), D–H), Welch’s two-sample t test (G). Error bars represent standard deviation. Underlying data for this Figure can be found in S8 Data. cKO, conditional knockout; CSC, codon stabilization coefficient; dcKO, double conditional knockout.

https://doi.org/10.1371/journal.pbio.3003031.g008

Among the 686 transcripts that were up-regulated in cKO cortices, 431 (63%) were also up-regulated in dcKO cortices (Fig 8C). A similar overlap was observed for down-regulated transcripts, with 289 in common out of the 530 down-regulated transcripts in cKO cortices (55%). We focused our subsequent analysis on these shared differentially expressed transcripts as they are most likely to be regulated by Cnot3 in a manner independent of apoptosis. Comparing our data with previously published scRNA-seq data [6] showed that differentially expressed transcripts were not biased for expression in either progenitors or neurons (S6B and S6C Fig).

To further define the molecular features of these shared Cnot3-dependent transcripts, we assessed their expression levels in control cortices. Up-regulated transcripts were expressed at significantly lower levels in control, compared to Cnot3-independent transcripts (Fig 8D). In comparison, down-regulated transcripts had significantly higher levels of expression in control brains (Fig 8D). Taken together, these data suggest that CNOT3 plays a role in helping to suppress a subset of transcripts that are typically expressed at low levels across multiple cell types.

We next sought to understand the mechanisms by which Cnot3-dependent transcripts are targeted. Previous work has shown that the CCR4-NOT complex is recruited to mRNAs with low codon optimality in a manner that is dependent on CNOT3 interaction with the ribosome E-site, and deletion of the yeast homolog of CNOT3 (Not5) preferentially stabilizes non-optimal mRNAs [54,65]. We thus evaluated the codon content of Cnot3-sensitive transcripts using CSC values calculated from our SLAM-seq data in WT cells (Fig 1H). Comparing optimal codon content among differentially expressed transcripts showed that up-regulated transcripts were significantly less optimal than unchanged transcripts, while down-regulated transcripts were more optimal (Fig 8E). This data is consistent with preferential targeting of non-optimal RNAs for degradation by the CCR4-NOT complex. We next asked whether this phenomenon could be observed transcriptome-wide or if it was specific to transcripts that were significantly altered in mutant cortices. Binning all mRNAs into quartiles based on percentage of optimal codons revealed that the most non-optimal mRNAs tended to be the most up-regulated in Cnot3 mutants (Fig 8F). Collectively, these data implicate Cnot3 in regulating RNA expression in a manner dependent on codon content, with preferential up-regulation of non-optimal mRNAs in mutant cortices. This further supports RNA stability control as a mechanism by which Cnot3 regulates cortical development.

Disruption of the CCR4-NOT complex in Cnot3 cKO cortices is expected to impair deadenylation and stabilize target transcripts. To evaluate whether increased expression of non-optimal mRNAs is due to changes in mRNA stability, we measured mRNA half-lives in primary cortical cells using transcriptional shutoff with Actinomycin D (ActD). Primary cultures were generated from E12.5 control and cKO cortices and cells were harvested 0, 2, and 8 h after addition of ActD to assess RNA degradation by RT-qPCR (Fig 8G). We examined the decay kinetics of 5 mRNAs whose expression increased in cKO cortices (S6D Fig) and that fell into the lowest quartile of codon optimality. All 5 transcripts were stabilized in cKO cells, consistent with defective mRNA turnover in the absence of Cnot3 (Figs 8G and S6E). Of note, three of these genes (Klf8, Kif18a, and Clk1) have roles in cell cycle regulation [66–68]. A fourth gene (Vldlr) is a reelin receptor important for cortical neuron positioning [69]. In sum, these data reveal up-regulation of non-optimal, poorly expressed RNAs in cKO cortices. Transcriptional shut-off analysis of select transcripts confirms impaired RNA degradation in the absence of Cnot3. These observations directly implicate the CCR4-NOT complex in controlling gene expression in the developing cortex.

Our data define the landscape of RNA degradation across developmental time. We provide evidence for 2 distinct modes of RNA stability regulation in the cortex. First, a dynamic half-life model in which the stability of a given RNA varies across development. Second, a static half-life model in which RNA degradation rates are constant across development. While these models are not mutually exclusive, our data indicate that a static half-life model is predominant in the developing cortex (Fig 9A), as RNA decay kinetics for most transcripts are unchanged across development. While dynamic half-life regulation occurs for a subset of transcripts (Fig 2), this appears to be the exception rather than the rule.

Fig 9. Models for relationship between RNA turnover and gene expression, and the role of Cnot3 in cortical development.

(A) Transcripts that are down-regulated across development are relatively unstable, while those that are up-regulated across development are relatively stable. For the majority of transcripts, RNA half-lives are static across development. Cortical development in WT (B) and Cnot3 cKO embryos (C). Loss of Cnot3 impacts development by p53-dependent and p53-independent mechanisms leading to apoptosis of progenitors and neurons, reduced neurogenic divisions, and longer cell cycle duration. Our data suggest that apoptosis is especially relevant for reduced number of upper layer neurons. cKO, conditional knockout.

https://doi.org/10.1371/journal.pbio.3003031.g009

We discovered an unexpected relationship between RNA half-life and expression changes across time. Specifically, we find that transcripts that increase in expression at later stages of cortical development tend to have longer half-lives, while transcripts that decrease in expression tend to have shorter half-lives (Fig 9A). We propose that transcripts required during early, but not late, developmental stages have short half-lives to help promote their rapid clearance when they are transcriptionally silenced at later stages. In contrast, the relative higher stability of transcripts needed later in development may promote their accumulation over time. Our data suggests that RNA stability acts as an additional regulatory layer to help ensure proper timing of gene expression in the developing cortex.

Our data also give some insight into the extent to which RNA half-lives vary across cortical cell types. We measured RNA half-lives as the cellular composition of the cortex transitions from progenitor-enriched at E11.5 to neuron-enriched at E16.5. If the half-life of a given transcript greatly differed between progenitors and neurons, we expect to have observed a shift in RNA stability kinetics from E11.5 to E16.5. However, we did not observe widespread changes in half-lives across these developmental transitions. Importantly, this suggests that turnover rates for most transcripts may be similar, regardless of the cell type in which they are expressed. It is important to note we do not rule out the possibility of cell type specific stability regulation for a small number of transcripts. Further, while we used scRNA-seq data sets and knowledge of cell composition changes from E11.5 to E16.5 to inform these conclusions, our study did not directly compare RNA decay in progenitors and neurons. Single cell analysis of RNA stability has successfully been employed in other complex model systems, including intestinal organoids and zebrafish embryos [70,71]. Adapting these techniques for in vivo analysis of RNA stability across diverse cell types of the developing mammalian cortex is an exciting direction for future investigation.

Our study used an orthogonal genetic approach to demonstrate essential roles of RNA stability in cortical development by focusing on the CCR4-NOT complex. To that end, we used progenitor and neuronal Cre drivers to demonstrate requirements of Cnot3 in both neural progenitors and newborn neurons of the developing cortex. While we specifically targeted CNOT3 in progenitors, expression of the central scaffold CNOT1 was also affected, suggesting the entire complex may be affected (Fig 4C). Cnot3 cKO brains are microcephalic, with fewer progenitors and neurons. We show these phenotypes are due in part to widespread p53-dependent apoptosis of both populations. Interestingly, Cnot3 cKO does not induce apoptosis in mouse ESCs or male germ cells [33,35], but does in B cells by directly regulating the stability of p53 mRNA [72]. These observations collectively suggest that the requirement of Cnot3 for mammalian cell viability may be cell type specific.

Using compound mutants, we mechanistically define how Cnot3 controls cortical neurogenesis, in both a p53-dependent and p53-independent fashion (Fig 9B and 9C). By analyzing cKO and dcKO brains at different stages, we defined the timing of when these mechanisms are predominant. We observed a preferential rescue of superficial layer neuron density at E18.5, coupled with the rescue of progenitor density at E14.5, when superficial neurons are being born. This suggests that p53-dependent mechanisms are especially prevalent at later stages of cortical neurogenesis. In contrast, loss of early born deep layer neurons was largely p53-independent.

This raises the question of what mechanisms control deep layer neuron number. We observe fewer proliferating cells in E12.5 cKO cortices, and this phenotype was rescued in dcKO brains. However, in contrast, loss of deep layer neurons at E18.5 is not rescued by p53. This suggests there must be apoptosis-independent mechanisms at play. Importantly, our semi-cumulative labeling and live imaging data in cKO cells (Fig 7) demonstrate that E12.5 progenitors take longer to divide and tend to produce more progenitors at the expense of neurons. We expect that these progenitor cell fate changes ultimately lead to production of fewer deep layer neurons. How Cnot3 controls progenitor cell cycle and neurogenic divisions is a fascinating question for future studies. It is notable that similar phenotypes have been observed for mutations in other RNA-binding proteins [73], reinforcing the role of posttranscriptional control in cell fate specification.

To understand the molecular underpinnings of these cortical development phenotypes, we measured transcriptomic changes using RNA-seq. While the biological implications for the majority of dysregulated transcripts remain elusive, this analysis provided several key insights into potential roles of CNOT3-mediated gene regulation in the cortex. First, we observed increased expression of p53 transcriptional targets in cKO cortices, including Eda2r, Sesn2, and Ccng1. This data underscores p53-dependent mechanisms as a major driver of cortical defects in cKO embryos, including the reduction in superficial layer neurons. We also observed decreased expression of several cell cycle-related transcripts, including Ccnd1, Ccnd2, and Cdk4. These observations align with our in vivo measurements of prolonged cell cycle, as well as previous reports indicating a role for CNOT3 in cell cycle regulation [32,74,75]. Of note, Cdk4 is also down-regulated upon knockdown of CNOT3 in human cardiomyocytes, suggesting a conserved mechanism of cell cycle regulation [32].

Although the CCR4-NOT complex has roles in multiple aspects of RNA metabolism [20], it is primarily described as a regulator of mRNA stability. While we cannot rule out the possibility that cKO phenotypes are caused in part by disruption of other molecular processes, several lines of evidence suggest that RNA turnover is impaired in cKO cortices. First, gene expression changes in cKO and dcKO cortices correlated with codon content, with non-optimal mRNAs showing the highest up-regulation of transcript levels. In light of previously established roles of the CCR4-NOT complex in degrading non-optimal RNAs [54,76], this is consistent with widespread RNA stabilization in mutant cortices. Second, using transcriptional shutoff experiments we directly show stabilization of non-optimal mRNAs in cKO cells. Overall, this data highlights the importance of maintaining control of RNA degradation in the developing cortex.

The CCR4-NOT complex, including CNOT3, is increasingly implicated in human disease. De novo mutations in human CNOT3 cause severe neurodevelopmental deficiencies characterized by a spectrum of intellectual disability, speech delay, seizures, and behavioral problems [24–30]. Further CNOT3 is a high confidence syndromic Autism gene with a SFARI score of 1S, while CNOT1 is considered 2S. Importantly, recent work has demonstrated that many autism risk and intellectual disability genes converge on a common pathway characterized by impaired neurogenesis including asynchronous development of deep layer neurons [77,78]. Along these lines, we found aberrant neurogenesis including of deep layer neurons in Cnot3-deficient cortices. Together, these data suggest that alterations in cortical development may be an important contributor to neurodevelopmental pathologies associated with CNOT3 mutation. Further study of this cKO mouse model may thus provide important insights into the etiology of CNOT3 pathologies. Additionally, an important next step for future studies will be to investigate how germline nonsense and missense mutations of Cnot3 impact the developing cortex.

Images were captured using a Zeiss Axio Observer Z1 equipped with an apotome for optimal sectioning at 5× and/or 20×. For each experiment, 2 to 3 sections were imaged per brain. Images were captured with identical exposures. For image processing, maximum intensity projections were generated from Z-stacks and brightness was adjusted equivalently for all images using FIJI software. For quantification, images were cropped to 200 μm wide radial columns and cells were counted manually using FIJI cell counter or automatically using QuPath [80]. For QuPath quantification, parameters were adjusted as follows: requested pixel size = 0.1 μm, background radius = 5 μm, minimum area = 10 μm², cell expansion = 2 μm, include cell nucleus and smoother boundaries boxes were unchecked. The threshold was set independently for each channel and maintained for each image in an experiment. For binning analysis, spatial coordinates for each detected cell were exported from either FIJI or QuPath and cells were assigned to one of 5 equally sized bins spanning from the ventricle to the pia to calculate the distribution.

E12.5 cortices were dissected and dissociated as described above for SLAM-seq sample preparation with the following modifications: only 1 embryo was used per biological replicate, trypsinization was performed for 5 min, after resuspension in neural progenitor media, 175,000 cells were plated onto a poly-D-lysine treated 24-well glass-bottomed plate (MatTek). Cells were allowed to adhere to the plate for 2 h, then images were captured every 10 min for 20 h using a Zeiss Axio Observer Z1 microscope fitted with a Pecon incubation chamber as previously described [85]. After 20 h, cells were fixed in PFA for 20 min, permeabilized with 0.1% Triton-X100 for 10 min, blocked in 5% NGS for 30 min, incubated with primary antibody for 1 h, and secondary antibody for 30 min. All immunostaining steps were performed at RT. The following primary antibodies were used: SOX2 (Thermo Fisher 14-9811-82, 1:1,000), TBR2 (Abcam ab183991, 1:500), and TUJ1 (Biolegend 801202, 1:2,000).

E12.5 embryos were dissected in cold PBS and cortical tissue from single embryos were microdissected and flash frozen on dry ice. Three biological replicates were used per genotype. Tissue was homogenized in RLT buffer (Qiagen) and RNA was extracted using the RNeasy Plus Micro Kit (Qiagen). Libraries were prepared using the TruSeq Stranded mRNA kit (Illumina) and sequenced on the NovaSeq platform with 150 bp paired-end reads. Read quality was assessed with FastQC and adapter trimming was performed using Cutadapt. Reads were aligned to the mm39 genome using STAR [86]. Mapped reads were counted and assigned to genes using featureCounts [87], and differential expression in cKO and dcKO versus control samples was determined using DESeq2 [83].

E12.5 cortices were dissected and dissociated as described above for SLAM-seq sample preparation with the following modifications: only 1 embryo was used per biological replicate, trypsinization was performed for 5 min, after resuspension in neural progenitor media each replicate was split into 3 wells of a poly-D-lysine coated 24-well plate. At t = 0, actinomycin D (Sigma A9415) was added to the media to a final concentration of 5 μg/ml. Cells were lysed at t = 0, 2, 8 h by addition of Trizol (Thermo Fisher 15596026) and RNA was extracted according to manufacturer’s instructions. cDNA was generated from 500 ng total RNA using iScript cDNA Synthesis Kit (Bio-Rad). qPCR was performed using iTaq SYBR Green Supermix (Bio-Rad). qPCR primers are listed in S6 Table. Decay data was fitted to the single exponential decay equation (y = y_o * e^-kt) using the nls function in R to determine the degradation rate constant, k. Half-life was calculated as t_1/2 = (ln(2))/k.

Sample collection, data acquisition, and quantifications were performed blind with respect to genotype. Statistical tests, p-values, and sample size for each analysis are reported in S7 Table.