Nrf2 dictates the neuronal survival and differentiation of embryonic zebrafish harboring compromised alanyl-tRNA synthetase

The sophisticated synthesis and precise modification of nascent proteins are tightly regulated by a set of intracellular structures, particularly the endoplasmic reticulum (ER) (Pobre et al., 2019). Errors in protein synthesis and protein folding usually lead to ER stress and unfolded protein response (UPR) (Walter and Ron, 2011). When the UPR is mild and can be handled properly, adaptive degradation of the misfolded protein is promoted by active spliced X-box-binding protein 1 (XBP1), which is mainly produced by the transautophosphorylated iron-responsive element 1α (IRE) (Grandjean et al., 2020). In parallel, the ATF6 axis is involved in the attenuation of the UPR to promote cell survival (Yu et al., 2017). When intensive and/or chronic stress is overloaded, cell apoptosis occurs via induction of PERK (EIF2AK3) through its oligomerization and the phosphorylation of multiple targets, in particular EIF2A, to inhibit a general panel of proteins (Wang and Kaufman, 2016). ATF4 selectively activates the cell apoptosis program through cooperation with its target, CHOP (DDIT3) (Hu et al., 2018). The three branches of the UPR pathway coordinately determine the cell conditions (Binet and Sapieha, 2015).

The transcription factor NRF2 is encoded by the nuclear factor erythroid-derived 2-like 2 gene (NFE2L2). It has been proposed to be a master regulator of external stresses to promote cell survival (Baumel-Alterzon et al., 2021). However, multiple functions of NRF2 have recently been reported. Phosphorylated NRF2 has been linked to PERK in the immunosuppression of tumors (Mohamed et al., 2020). The activation of nfe2l2a, a zebrafish paralog of Nfe2l2, attenuates ER stress in pmm2-deficient larval livers (Mukaigasa et al., 2018). Several studies have reported that activated NRF2 serves as a substrate for caspase-3 (-like) proteases and functions together with P38 or P53 (TP53) to induce apoptotic processes (Bonay et al., 2015; Ohtsubo et al., 1999). The various outcomes of NRF2 indicate its complicated and context-dependent effects (Menegon et al., 2016).

The AARS1 gene encodes alanyl-tRNA synthetase (AlaRS), which is composed of three domains: an N-terminal aminoacylation domain, an editing domain and a C-terminal domain. It conducts a conserved aminoacylation reaction to charge alanine amino acids on the corresponding cognate tRNAs. This process is essential for the translation of the genetic code in protein or peptide synthesis. In the Aars1^sti/sti mutant mice that harbored a C-to-A mutation at amino acid 734 in the editing domain of aminoacyl-tRNA synthetase (Aars1), notable loss of Purkinje cells and the development of neurodegenerative-like syndromes were observed (Lee et al., 2006). This phenotype was caused by the production of serine-mischarged-tRNA^Ala, which led to UPR in the cerebellar Purkinje cells. Ankrd16, a vertebrate-specific protein containing ankyrin repeats, partially alleviated the degeneration of cerebellar Purkinje cells in Aars1^sti/sti mutant mice by binding to the catalytic domain of AlaRS to capture the misactivated serine and prevent mischarging (Vo et al., 2018). In humans, 11 missense mutations in three functional domains of AARS1 have been identified in Charcot-Marie-Tooth type 2N, an autosomal-dominant inherited form of axonal neuropathy (McLaughlin et al., 2012; Motley et al., 2015; Simons et al., 2015). These phenomena indicate the indispensable role of AARS1 in the control of proteostasis and UPR intensity in neuronal development and neurodegenerative diseases.

Neurogenesis in zebrafish is initiated by the specification of neural progenitor cells (NPCs) after the establishment of the neuroectoderm (Schmidt et al., 2013). Deficiency in the production and maintenance of NPCs leads to compromised neurogenesis (Li et al., 2020). In the present study, we performed a large-scale mutant screening and identified an aars1^cq71/cq71 mutant zebrafish allele that manifested compromised neurogenesis caused by intensive cell apoptosis of NPCs. Together with the threonyl-tRNA synthetase mutant, tars1^cq16/cq16 (Cao et al., 2016), we revealed that compromised aminoacyl-tRNA synthetase in the NPCs led to UPR reactions and resultant cell apoptosis by the activation of Perk signaling. nfe2l2b, but not nfe2l2a, acted as a downstream executor by targeting p53. Our study reveals Nrf2 as a regulative molecule of p53 in Perk-mediated cell apoptosis and provides clues on the progress and treatment of neurodegenerative disorders.

To explore the molecules involved in neurogenesis, we performed an unbiased large-scale screening of a candidate pool of ethylnitrosourea (ENU)-induced mutagenesis (Cao et al., 2016). We examined the signals of neurogenic differentiation 1 (neurod1), which is mainly expressed in postmitotic neuronal cells, by whole-mount in situ RNA hybridization (WISH). The recessive ENU mutant allele that presented an obvious reduction in neurod1⁺ signals was fortuitously recovered (Fig. 1A). We named it as ‘Chongqing number 71’ (cq71) mutant (referred to as Mut^cq71/cq71 herein). Several genes that encode components of neuronal and glial cells were further analyzed in Mut^cq71/cq71 by characterizing either corresponsive reporter lines or transcript levels. The signals of neural-specific beta-tubulin (NBT) in pan-neuronal cells, sox2 and neurog1 in NPCs, Elav-like neuron-specific RNA-binding protein 3 (HuC) and NGF/microtube-associated protein2 (map2) in young neurons, alpha-tubulin ortholog (tuba1b) in mature neurons, olig2 in eurydendroid neurons in the cerebellum, and glial fibrillary acidic protein (gfap) in astrocytes and radial glial cells showed marked reduction in the brains, but not spinal cords, of Mut^cq71/cq71 compared with their wild-type (WT) and heterozygous siblings at 2.5-3 dpf (Fig. 1A, Fig. S1A,B). The size of 3 dpf midbrain was 75,973±2162 μm² in Mut^cq71/cq71 larvae compared with 91,176±2063 μm² in WT counterparts (Fig. S1C). Mut^cq71/cq71 larvae did not survive beyond 7 dpf (Fig. S1D). The behavior recordings indicated that 3 dpf WT larvae spontaneously moved at 0.19±0.05 cm/s to cover a distance of 81.74±20.34 cm during a 5 min recording. However, Mut^cq71/cq71 larvae exhibited compromised spontaneous mobility (Fig. 1B, Fig. S1E,F, Movie 1). The WT larvae rapidly responded to the external stimuli of either needle-touch or light reduction and altered the speed to 0.22±0.02 cm/s or 0.15±0.02 cm/s, respectively. However, the Mut^cq71/cq71 larvae remained mostly frozen (Fig. S1G-L, Movies 2 and 3). Although Mut^cq71/cq71 larvae occasionally acted upon the stimuli, their movement speeds were markedly reduced compared with those of their WT siblings (Fig. S1I,L). Accordingly, neuronal activities indicated by Ca²⁺ transients of various types were frequently detected in WT transgenic Tg(HuC:GCaMP6s) midbrains; however, they were barely seen in the Mut^cq71/cq71 Tg(HuC:GCaMP6s) (Fig. 1C, Movie 4). These phenotypes indicated compromised neurogenesis in Mut^cq71/cq71 embryos, suggesting increased cell death and/or reduced proliferation of NPCs in this scenario. The staining results indicated that phospho-Histone 3 (PH3⁺) signals surprisingly displayed an increment in Mut^cq71/cq71 brains at 2.5 dpf (Fig. S1M,N). However, the terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL)⁺ signals that were rarely detected in Sox2⁺ NPCs in WT larvae were easily visible in Mut^cq71/cq71 counterparts (Fig. 1D). Simultaneously, Acridine Orange (AO) and Caspase-3 signals increased notably in the Mut^cq71/cq71 brains, but not in the spinal cord, when compared with those in their WT and heterozygous siblings (Fig. 1E,F, Fig. S1O-S). The expanded AO⁺ signals were collected but not effectively cleared by Coro1a-Dsred⁺ microglia (Li et al., 2012) and exhibited enlarged foci in the Mut^cq71/cq71 brains (Fig. S1S). These results indicated increased apoptosis of NPCs and compromised neurogenesis in the Mut^cq71/cq71 brains (Fig. 1G).

To unveil the gene responsible for Mut^cq71/cq71 mutants, we utilized a total of 1890 meioses to perform genetic linkage analysis and restricted the location to chromosome 18, with ∼1.35 and 0.92 centimorgan intervals to ZK225B14 and ZK261F4, respectively (Fig. 2A). One recombination was discovered by sequencing the selected single-nucleotide polymorphism (SNP) 1 and 2 in this region, further narrowing the mutation to a 0.13 Mb spacing (Fig. 2A). Four genes were annotated in this fragment according to the information from Ensembl (http://asia.ensembl.org/index.html). A point mutation (T-to-A) in exon 5 of aars1 was identified in the Mut^cq71/cq71 embryos (Fig. 2B, Fig. S2A). There are four isoforms of aars1 in zebrafish. However, only 201 (ENSDART00000076695.6), a short isoform, and 202 (ENSDART00000100401.6), encoding the whole functional domain of aars1, were successfully amplified from the cDNA library of 3 dpf embryos. The T-to-A mutation in aars1^cq71/cq71 caused an alteration from UGU to UGA at position 171, producing a truncated protein with a premature stop codon (Fig. 2B, Fig. S2B). Western blot analysis indicated that the mutant forms of HA-aars1 were unstable, and the protein levels of Aars1 were markedly reduced by 87%, in the aars1^cq71/cq71 brains compared with the brains of their WT siblings (P=0.0002) (Fig. 2C, Fig. S2C). Furthermore, WISH showed a dramatic reduction in aars1 transcripts in the aars1^cq71/cq71 mutants compared with the WT siblings (Fig. S2N). The transcription levels of the nonsense-mediated mRNA decay (NMD)-related genes, including upf1 and upf3b, but not upf3a, were significantly increased in aars1^cq71/cq71 mutants (Fig. S2D), implying an NMD (Cheruiyot et al., 2021) promoting rapid degradation of aars1 mRNA in aars1^cq71/cq71 mutants. These data indicated that the aars1^cq71/cq71 mutant was a null allele.

The zebrafish full-length Aars1 (Aars1²⁰²) protein shared 81.88% and 82.61% similarity with its human and mouse counterparts, respectively (Fig. S2E). Supplementation of aars1²⁰² (referred to as aars1 in subsequent sections) and even human AARS1 in the aars1^cq71/cq71 mutant embryos rescued the phenotypes of Sox2⁺ NPCs and reduced the TUNEL⁺ and AO⁺ signals (Fig. 2D, Fig. S2F,G). An additional aars1^Δ10/Δ10 mutant allele with a 10 bp deletion was generated. The aars1^Δ10/Δ10 mutation caused premature termination of the Aars1 protein in the amino acid activation domain (Fig. 2B, Fig. S2A), which led to a dramatic reduction in its protein levels to 15% of that in siblings (Fig. 2C, Fig. S2C). Similar to the aars1^cq71/cq71 mutant embryos, the aars1^Δ10/Δ10 mutants manifested intensive atrophy of the midbrain, an increase in AO⁺ signals, and a reduction in Sox2⁺, HuC⁺ and neurod1⁺ pools compared with their WT siblings (Fig. 2E,F, Fig. S2H). The result of a complementary assay by mating aars1^cq71/cq71 with aars1^Δ10/Δ10 allele indicated the same declined aars1 transcripts and neural-defective signatures as those seen in either aars1^cq71/cq71 or aars1^Δ10/Δ10 mutant alleles (Fig. S2I-K). Interference of aars1 using ribonucleoproteins (RNPs) (Cas9 protein/synthetic crRNA:tracrRNA duplex technology) (Kroll et al., 2021) was further investigated. Among the six predicted target sites by Integrated DNA Technologies (IDT), three targets of crRNAs at loci A, B and D showed the highest mutated efficiency of 94%, 90% and 65%, respectively, according to the ratio of head loop-PCR/standard-PCR (Kroll et al., 2021) (Fig. S2L). The levels of Aars1 protein were reduced by 77% in the aars1 F0 group after the combinational application of these crRNAs compared with those in the controls (P=0.0016) (Fig. S2M). Annexin V staining was applied as an indicator, because it has a high affinity for phosphatidylserine of dying cells (Balaji et al., 2013). A marked increase in Annexin V⁺ signals was detected in aars1 F0 embryos treated with aars1 RNPs compared with controls (Fig. 2G), corroborating the increased neuronal apoptosis and compromised neurogenesis upon Aars1 deficiency.

The aars1 transcript appeared even at the one-cell stage of the fertilized embryos and was almost evenly present in the cells at the 50% epiboly stage (Fig. S2N). Following this, the transcript was highly expressed in the midbrain and digestive organs (Fig. S2O). In the midbrain, double staining and quantitative PCR (qPCR) results indicated that the aars1 transcript was highly detected in the sox2⁺, NBT⁺ and HuC⁺ cells, detected to a lower extent in the neurod1⁺ and olig2⁺ cells, and was limited in the coro1a⁺ microglial cells (Fig. 2H, Fig. S2P), implying its enrichment in NPCs and intrinsically regulated neurogenesis. Therefore, we provided aars1 in the pan-neuronal cells by fusing its cDNA sequence to the NBT promoter (Thomas-Jinu et al., 2017). This assay could supply aars1 in Sox2⁺ NPCs, based on the clear detection of Sox2⁺ signals in the NBT-DenNTR⁺ cells (Fig. S2Q). Continual supplementation of NBT-aars1 in Tg(NBT:aars1) slightly lowered the AO⁺ signals but markedly mitigated the intensity of AO⁺ phenotypes in aars1^cq71/cq71 mutants to a level comparable with that in WT siblings (Fig. 2I,J). Furthermore, a transplantation assay was performed following previous instruction (Zou and Wei, 2010). An equal number of cells (30-50) was isolated from the blastomeres of WT or aars1 F0 animals harboring NBT-DenNTR background. These cells were transplanted into the recipients (Fig. 2K). Examination of Annexin V indicated that the signals did not present notable alterations in the aars1^cq71/cq71 mutant recipients after accepting the NBT⁺ blastomere cells from the WT siblings, but were elevated dramatically after receiving the donors of the aars1 F0 origin compared with their control groups (Fig. 2L,M). Taken together, these results indicate that aars1 is cell-autonomously required and functionally conserved for neuronal survival and development.

The molecular mechanisms underpinning aars1 in neurogenesis require further investigation. To this end, single-cell RNA sequencing (scRNA-seq) was performed (Fig. S3A). In total, 17,519 and 14,465 cells were obtained from 2 dpf siblings and aars1^cq71/cq71 mutant midbrains, respectively. Comparable sequencing depths (18,129 and 22,921 reads/cell), median unique molecular identifier (UMI) counts (1964 and 2844 UMI/cell), and median gene numbers (949 and 1211 genes/cell) were obtained from WT siblings and aars1^cq71/cq71 mutants. We classified the cell types of all samples and mapped the resulting reads to the zebrafish genome for unbiased clustering using Seurat software. Twelve major cell clusters were identified (Fig. S3B). Among them, nine neuronal clusters [GABAergic neurons, glutamatergic neurons, thalamic neurons, progenitor neurons (differentiating), progenitors, radial glial cells (RGCs), optic tectum neurons, vagal neurons and oligodendrocyte progenitor cells (OPCs)], based on the top ten genes in each cluster, were predominant (Fig. 3A-C). Intriguingly, cluster 4 progenitor cells were notably enlarged to 65% in the aars1^cq71/cq71 mutants, in contrast to the detectable shrinkage of other clusters (Fig. 3D). Evaluation of 20 aminoacyl-tRNA synthetases (ARSs) revealed that aars1 transcripts were singularly enriched in cluster 4 (Fig. 3E, Fig. S3C). Cluster 4 was further divided into three subtypes, and aars1 was more precisely positioned in subpopulations 1 and 2, the oldest stages of the progenitors that occupied more than half of this pool (Fig. 3E-G). These data were consistent with the intrinsic roles of aars1 in the survival of NPCs and suggested the stalling of progenitor differentiation before apoptosis in the aars1^cq71/cq71 mutants. Information regarding the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway from cluster 4 revealed pronounced elevations in aminoacyl-tRNA biosynthesis signaling, necroptosis and protein processing in the ER (Fig. 3H). Gene Ontology (GO) analysis indicated a marked alteration of factors in the UPR and apoptotic pathways (Fig. 3I, Fig. S3D), pointing to protein overload and ER stress in the NPCs of the aars1^cq71/cq71 mutants. To this end, we injected FLAG-tagged ubiquitin constructs into WT siblings and aars1^cq71/cq71 mutant embryos, which were then detected using anti-FLAG antibodies. The levels of ubiquitin-modified protein fragments were increased by 30% in the administered aars1^cq71/cq71 mutant embryos compared with those in the WT siblings (P=0.0021) (Fig. 3J). However, when the protein loads were attenuated by cycloheximide (actidione, naramycin A; CHX), an antifungal antibiotic that inhibits protein and RNA synthesis (Klinge et al., 2011), the AO⁺ and Sox2⁺ cells in the aars1^cq71/cq71 mutants recovered to the levels in the WT siblings (Fig. 3K,L). These results indicated that the intensive UPR by misfolded proteins induced NPC apoptosis in aars1^cq71/cq71 mutants (Fig. 3M).

Protein overload causes ER stress. A prolonged UPR eventually leads to apoptosis (Guha et al., 2017). Indeed, the protein levels of BiP (Hspa5), a protein-folding chaperone (Wang et al., 2017), increased by 117% in the aars1^cq71/cq71 mutant brains compared with those in the brains of their WT siblings (P=0.0004) (Fig. 4A, Fig. S4A). The ER structures were severely distended and dilated in aars1^cq71/cq71 mutant brains compared with those in WT sibling brains (Fig. 4B, Fig. S4B). To determine the role of aggravated ER stress in excessive apoptosis of aars1^cq71/cq71 mutant NPCs, two chemicals, thapsigargin (Tg), a potent ER Ca²⁺ pump inhibitor (Sehgal et al., 2017), and tunicamycin (Tn), a protein glycosylation inhibitor (Guha et al., 2017), were used to treat WT embryos. The application of either molecule promoted the appearance of substantial AO⁺ signals in the brains of cultured embryos (Fig. 4C). Conversely, significant mitigation of AO⁺ phenotypes was observed when aars1^cq71/cq71 mutant embryos were incubated with azoramide, an ER stress inhibitor (Fu et al., 2015) (Fig. 4D,E).

In mammalian cells, ER stressors activate proximal UPR pathways through three major trans-ER membrane proteins, IRE1, ATF6 and PERK (Metcalf et al., 2020). We investigated the kinetics of these pathways by measuring the protein volume of phosphorylated (p)-eIF2α (as a direct readout of Perk activation) (Humeau et al., 2020), and the transcript amounts of atf6 and ‘unconventional’ splicing of xbp1 (reflecting Ire1 activation) (Guo et al., 2014) in the pooled WT sibling or aars1^cq71/cq71 mutant embryos. The protein levels of eIF2α, especially its activated form, p-eIF2α, were markedly elevated by 48% in the aars1^cq71/cq71 mutant brains compared with those in the brains of their WT siblings (P=0.0058) (Fig. 4F), indicating the activation of Perk signaling upon aberrant aars1 function. Although the transcript levels of atf6 and ire1α were notably increased in aars1^cq71/cq71 mutants (Fig. S4C), neither pathway was effectively functional, based on the results of the defective xbp1 RNA splicing and the inability of the ATF6 inhibitor Ceapin A7 (Gallagher et al., 2016) to attenuate cell apoptosis in the aars1^cq71/cq71 mutants (Fig. S4D,E). To further assess the effects of Perk signaling, we interfered with Perk function by using GSK2656157, a specific Perk signaling inhibitor (Atkins et al., 2013), which decreased the p-eIF2α protein levels by 64% compared with the controls (P=0.0030) (Fig. S4F). Application of this chemical remarkably reduced the AO⁺ signals in the aars1^cq71/cq71 mutant midbrains, which were not present in the samples that were treated with controls (Fig. 4G). Collectively, Perk activation during ER stress plays a crucial role in NPC apoptosis in aars1-deficient embryos. Atf4 and the pro-apoptotic transcription factor chop are responsible for Perk and eIF2α activity (Walter and Ron, 2011). Their expression levels increased markedly in aars1^cq71/cq71 mutant brains compared with those in the brains of their sibling counterparts (Fig. S4A,G). Additionally, overexpressing chop mRNA in WT embryos markedly induced the appearance of AO⁺ signals (Fig. 4H). Conversely, injection of the designed chop RNPs, which efficiently mutated the F0 embryos by more than 70% in each target locus, significantly reduced the AO⁺ signals in aars1^cq71/cq71 mutants compared with those in WT embryos (Fig. 4I, Fig. S4H,I). An enlarged pool of NBT-DenNTR⁺ cells and increased sizes of the midbrain in the RNP-treated embryos were observed simultaneously (Fig. 4J, Fig. S4J). Autophagy has been reported to be regulated by Perk in zebrafish (Jia et al., 2015), and it is involved in tRNA synthetase-deficient cardioproteinopathy (Liu et al., 2014). However, examination of the protein levels of microtubule-associated protein light chain 3 (Lc3; Map1lc3b) and p62 by western blotting, together with the imaging of mCherry-Lc3-labeled autophagosomes and p62-characterized autophagy substrates in the neurons, indicated that their signals were limitedly altered in the aars1^cq71/cq71 mutants compared with those in the WT siblings (Fig. S4K-M). Collectively, these results indicated that activated Perk signaling in the aars1^cq71/cq71 mutants triggered NPC apoptosis via chop (Fig. 4K) but limitedly affected autophagy.

The mechanism by which the PERK–CHOP axis regulates cell survival remains unclear. To address this, we established an in vitro cell culture assay to explore the key factors involved. The addition of Tn and Tg to HEK293T cells successfully induced an increase in annexin V⁺ signals (Fig. 5A,B). In the groups treated with Tg and Tn, the levels of p-PERK elevated to 398% (P<0.0001) and 353% (P<0.0001), respectively, compared with the controls (Fig. 5A-C). We then focused on NRF2, a classical antioxidant transcription factor that plays a vital role in a wide spectrum of stress responses (Ma, 2013). Western blotting results indicated that the protein levels of NRF2 significantly elevated by 210% (P<0.0001) and 67% (P=0.0018) in the Tg and Tn groups, respectively, compared with the controls (Fig. 5D). Immunostaining data revealed a similarly notable increase, especially its phosphorylated form in the nucleus, upon application of these chemicals (Fig. 5D,E). Concordantly, treating HEK293T cell lines and WT zebrafish embryos with oltipraz, an Nrf2-specific activator (Magesh et al., 2012), exacerbated Annexin V⁺ and AO⁺ signals but reduced p-eIF2α protein levels by 32% relative to those in the control (P=0.0017) (Fig. 5F,G, Fig. S5A). There are two Nfe2l2 paralogs in zebrafish, nfe2l2a and nfe2l2b (Fig. S5B). Pattern characterization using WISH indicated that both factors were maternally expressed (Fig. S5C). They were similarly detected in the brain and eyes and colocalized with NBT-DenNTR⁺ cells (Fig. 5H, Fig. S5C). Interestingly, qPCR and fluorescence in situ hybridization (FISH) results indicated that the transcript levels of nfe2l2b were increased by 2.7-fold in aars1^cq71/cq71 NBT⁺ cells (Fig. 5H,I, Fig. S5D), in contrast to the reduced levels of nfe2l2a at the same time (Fig. S5D). Activation of nfe2l2a has been reported to attenuate ER stress in pmm2-deficient larval livers (Mukaigasa et al., 2018); however, the functions of nfe2l2b, a paralog of Nfe2l2, have rarely been investigated. Delivery of exogenous chop mRNA and oltipraz significantly elevated the transcript levels of nfe2l2b, but showed reduced influence on nfe2l2a transcript levels (Fig. 5J, Fig. S5E). The application of chop RNPs led to opposite outcomes (Fig. 5K). Supplying nfe2l2a to the aars1^cq71/cq71 mutants showed limited effects on the mitigation of AO⁺ and Sox2⁺ signals (Fig. S5F,G). These data suggest that nfe2l2b functions downstream of the Perk–Chop axis in the regulation of neuronal apoptosis. Therefore, we injected one-cell embryos with nfe2l2b mRNA and performed AO staining. Surprisingly, overexpression of full-length nfe2l2b mRNA significantly increased AO⁺ signals in WT midbrains but reduced the p-eIF2α protein levels by 26% (P=0.2332) (Fig. 5L, Fig. S5H). Conversely, we generated a nfe2l2b mutant allele with a 1 bp deletion, which led to a predictive premature stop codon in translation (Fig. S5I). The protein levels of HA-tagged-Nfe2l2b^Δ1/Δ1 were reduced by 96% compared with those in Nfe2l2b^wt (P<0.0001) (Fig. S5J). Homozygous nfe2l2b^Δ1/Δ1 mutants survived to adulthood, with no general difference in appearance compared with their WT siblings (Fig. 5M). However, compensatory disruption of nfe2l2b through the generation of aars1^cq71/cq71;nfe2l2b^Δ1/Δ1 mutant larvae attenuated the phenotypes of increased TUNEL⁺ signals and brain atrophy in the aars1^cq71/cq71 mutants (Fig. 5N,O). Intriguingly, the phenotypes of the neural populations, reflected by Sox2⁺, neurod1⁺ and even oligo2⁺ cells, were significantly recovered in the aars1^cq71/cq71;nfe2l2b^Δ1/Δ1 mutants to the levels in either the WT siblings or nfe2l2b^Δ1/Δ1 embryos (Fig. 5P-R). Taken together, these results indicate that Nfe2l2b serves as a sufficient downstream player of Perk signaling via transcript control by Chop and activity manipulation by Perk to regulate NPC survival and neurogenesis in embryonic zebrafish (Fig. 5S).

NRF2 is a transcription factor (Fernández-Millán et al., 2016), the targets of which we dissected. p53 was selected as a candidate based on the predicted Nrf2-binding site (GTAACAAAG) in its promoter sequence element (Fig. 6A). Chromatin immunoprecipitation (ChIP) and luciferase assays were then performed. The ChIP results indicated that Nfe2l2b is bound to the element (GTAACAAAG) located at the −639 site of the p53 promoter in vivo (Fig. 6B). Luciferase activity increased ∼3.5-fold after addition of the vectors harboring this element compared with the control (Fig. 6C). However, mutating this coding sequence (CDS) caused an obvious reduction in luciferase activity by ∼2-fold (Fig. 6C). Meanwhile, the WISH results showed that the p53⁺ signals were markedly decreased in the nfe2l2b^Δ1/Δ1 mutant brains, but significantly increased upon nfe2l2b mRNA injection (Fig. 6D,E). Overexpression of nfe2l2b mRNA in Tg(NBT:DenNTR) embryos followed by triple fluorescent labeling revealed that p53⁺ signals were ectopically detected in NBT-DenNTR⁺ cells, which colocalized with nfe2l2b and displayed TUNEL⁺ signals (Fig. 6F,G). Additionally, P53 protein levels increased markedly in cultured HEK293T cells treated with Tg in a dose-dependent manner (Fig. S6A). Similarly, the p53 transcript levels were significantly increased in NBT-DenNTR⁺ cells in the aars1^cq71/cq71 mutant brains but decreased markedly upon knocking out either nfe2l2b or chop (Fig. 6H-J, Fig. S6B). These data indicate that p53 is a direct target of Nrf2. We then generated a Tg(hsp70:p53) line in which full-length p53 is driven by the hsp70 promoter (Wu and Wang, 2020) (Fig. S6C). To facilitate screening, the lens protein Cryaa-Cerulean-BGHpA element was inserted in reverse sequence. This allowed us to temporally induce p53 expression at the desired time points via heating (Wu and Wang, 2020). We validated the successful overexpression of p53 transcripts based on the appearance of Cerulean⁺ signals in the retina and p53⁺ signals in the body (Fig. S6D,E). Appearance of AO⁺ signals in the brains and significantly decreased midbrain size were observed in heat-shocked 3 dpf Tg(hsp70:p53) embryos (Fig. S6F,G). In contrast, transient knockdown of p53 in aars1^cq71/cq71 embryos by p53 morpholinos (MOs) markedly attenuated the phenotypes of AO⁺ signals, enlarging the pool of NBT-DenNTR⁺ cells and midbrain volume (Fig. 6K, Fig. S6H,I). These p53 MO-treated aars1^cq71/cq71 embryos survived much longer than their untreated controls (Fig. 6L); however, no notable nfe2l2b transcript alterations were observed simultaneously (Fig. 6M), further supporting the epistatic effect of Nfe2l2b on p53. Similarly, we observed that p53 knockdown significantly decreased cell death in oltipraz-treated embryos (Fig. 6N). Taken together, these results showed that p53 acts as an executor of Nfe2l2b in the regulation of neuronal survival and neurogenesis in embryonic zebrafish (Fig. 6O).

We then investigated whether other aminoacyl-tRNA synthetases regulated neuronal survival and neurogenesis via the same Perk–Nrf2 axis. scRNA-seq data revealed the enrichment of several ARS members in various neuronal subpopulations, in which tars1 was highly expressed in clusters 4 and 5 (Fig. 7A). We characterized the phenotype of tars1^cq16/cq16 mutants (Cao et al., 2016). tars1^cq16/cq16 mutant embryos presented a notably smaller head than that of their WT siblings and a curled-up tail (Fig. 7B). The AO⁺ and TUNEL⁺ signals increased dramatically, but the numbers of NeuN⁺ (Rbfox3a⁺) cells and transcript levels of map2 and tuba1b were substantially reduced, in tars1^cq16/cq16 mutants relative to their WT siblings (Fig. 7C-E). These data indicate increased neuronal apoptosis and failed neurogenesis in tars1^cq16/cq16 mutants. The UPR of ER stress signaling was initiated in tars1^cq16/cq16 mutants, as indicated by elevated transcript levels of bip and chop and an obvious attenuation of cell apoptosis upon CHX application (Fig. 7F,G). Similar to in the aars1^cq71/cq71 mutant, the Perk pathway was selectively activated. The protein levels of p-eIF2α were increased by 136% in tars1^cq16/cq16 mutants compared with WT siblings (P=0.0009) (Fig. 7H). However, the splice of xbp1 transcript failed in the tars1^cq16/cq16 embryos (Fig. 7I). Concordantly, interfering with Perk functions in tars1^cq16/cq16 embryos by GSK2656157 treatment remarkably reduced the aggregated AO⁺ signals (Fig. 7J). Likewise, the expression levels of nfe2l2b and p53 were increased in tars1^cq16/cq16 mutant brains compared with those in the brains of their WT siblings (Fig. 7K). Transient interference with the functions of nfe2l2b or p53 by MO administration significantly repressed the AO⁺ and TUNEL⁺ signals in the tars1^cq16/cq16 mutant midbrain (Fig. 7L,M), implying that tars1 deficiency caused neuronal apoptosis through a similar Perk–Nrf2 pathway. Taken together, our study results demonstrated that abrogated protein metabolism caused by tRNA synthetase deficiency led to accelerated apoptosis of NPCs and compromised neurogenesis, as determined by Nrf2 and its target p53. Nrf2 function was doubly controlled. nfe2l2b transcript was regulated by Chop, with nuclear activation being promoted by Perk phosphorylation (Fig. 7N).

In the present study, we identified a recessive aars1^cq71/cq71 mutant zebrafish allele. aars1^cq71/cq71 mutant harbored a point mutation that led to a premature stop codon in the amino acid activation domain and generated a rapidly decayed truncated Aars1 protein. The abrogation of Aars1 functions in the present aars1^cq71/cq71 null mutant zebrafish was initially anticipated to cause a failure in protein synthesis; however, aggregated loads of ubiquitylated proteins and intensive ER stress were observed in the aars1^cq71/cq71 mutant brains. This result implied that other tRNA synthetases had a compensatory function, resulting in the production of large amounts of misfolded proteins. Indeed, the transcripts of nars1, tars1, eprs1, mars1, yars1, iars1, hars1, kars1, sars1, fars1 and gars1 were markedly increased in the aars1^cq71/cq71 mutant brains, based on the results of scRNA-seq and WISH (Fig. S6J,K).

scRNA-seq analysis revealed intensive enrichment of aars1 in NPCs, particularly in the primary stages. This different pattern of aars1 transcription from that of the other members of the ARS family points to a special and/or unique role of aars1 in harnessing the signatures of NPCs during embryonic neurogenesis. Concordantly, significant apoptosis of NPCs was observed in both the ENU screening and CRISPR/Cas9-created aars1 mutant zebrafish alleles, and this phenotype could be rescued by human AARS1, highlighting the solid and conserved roles of aars1. Interestingly, scRNA-seq analysis revealed a remarkably enlarged pool of neuronal progenitors in 2 dpf aars1^cq71/cq71 mutant brains before pronounced cell apoptosis. These data implied the blockage of differentiation and maintenance of quiescence when progenitors lacked proper protein supplementation. The maintenance of progenitor characteristics is controlled by a set of factors, and Notch activity is a key clue in this situation (Alunni et al., 2013). Consistently, our sequencing data indicated Notch signaling activation (Fig. S6L), but the reasons behind its activation upon compromised aars1 functions were still elusive. Whether and how Notch signaling is involved in ER stress and neuronal apoptosis remains an open question.

In the aars1^cq71/cq71 and tars1^cq16/cq16 mutants, intensive UPR reactions specifically activated the Perk signaling branch, causing a dramatic increase in apoptotic cells. PERK activation occurs when the UPR is beyond tolerance or adaptation, eventually leading to apoptosis through mitochondrial signaling (Ansari et al., 2018). QRICH1 was identified as an interesting factor downstream of the PERK–eIF2 axis as a distinct arm of ATF4 to determine cell fate through proteostasis modulation (You et al., 2021). However, the key executive targets of the atf4–chop axis in cell death are unclear. Taking advantage of this genetic manipulation, we recognized that nfe2l2b, a paralog of Nfe2l2, is an essential mediator of Perk signaling that regulates cell apoptosis upon UPR.

In previous studies, NRF2 was found to be a transcriptional factor that plays a critical role in defense against electrophiles and oxidative stress (Kawai et al., 2011). However, some reports have shown that ER stress activates NRF2 activity, whereas others have opposed these opinions (Chang et al., 2018; Kang et al., 2019). Interestingly, two paralogs of Nfe2l2 have been found in zebrafish. The nfe2l2a paralog was closer to Nfe2l2 than nfe2l2b. A recent zebrafish study demonstrated that nfe2l2a transcript was detected in the brain and liver, and its activation ameliorated mild ER stress in the pmm2 mutant liver (Mukaigasa et al., 2018). However, we did not observe a notable increase in the nfe2l2a paralog in the aars1^cq71/cq71 mutants, and its forced supplementation had limited effects. In contrast, nfe2l2b appeared in the midbrain, presenting marked elevation in its transcript level in the aars1^cq71/cq71 mutants. Unlike for the nfe2l2a paralog, oversupply of nfe2l2b induced cell apoptosis. Conversely, interfering with nfe2l2b in aars1^cq71/cq71;nfe2l2^Δ1/Δ1 mutants rescued the extensive neuronal apoptosis induced by ER stress in aars1^cq71/cq71 mutants. The distinct effects of nfe2l2a and nfe2l2b on zebrafish neuronal survival suggest a diversity of Nrf2 functions in a context-dependent manner. This alluded to a possible interpretation of the debate regarding the NRF2 response to ER stress. It is a common result that fish harbor duplicated forms of a plethora of factors. Thus, another possibility is that nfe2l2b was a unique duplicated factor of Nrf2 acquired by zebrafish during evolution. However, the pro-apoptotic role of nfe2l2b extends the understanding of crucial molecules in ER stress.

The target genes of NRF2 in ER stress are limited. A recent study has documented the interaction of NRF2 with P53 in promoting cancer cell death (Kang et al., 2019). Our study indicated that p53 was a direct target of Nrf2 and its transcript was significantly reduced in nfe2l2b^Δ1/Δ1 mutants but increased upon overexpression of nfe2l2b, in which p-NRF2 entered the nucleus to directly increase the levels of p53 to induce cell apoptosis. These data are in contrast to a previous report that an increase in p53 level was observed using nfe2l2b MO (Timme-Laragy et al., 2012). We believe that the difference was probably caused by the different methods and tissues used in the two studies. The present study examined the p53 transcripts in nfe2l2b⁺ cells using both WISH and FISH, in contrast to the previous bulk qPCR analysis of whole embryos (Timme-Laragy et al., 2012). Furthermore, this hypothesis supports organ sensitivity to NRF2. The homozygous nfe2l2b^Δ1/Δ1 mutants were able to survive to adulthood and were comparable to the WT siblings, suggesting the non-essential role of nfe2l2b in organ development. However, an obvious reduction in the p53 transcripts was observed in the nfe2l2b^Δ1/Δ1 mutant midbrain. Considering the wide spectrum of the effects of decreased p53, further studies should focus on the neuronal phenotypes in the nfe2l2b^Δ1/Δ1 mutants, especially in adult animals. Overall, our study extends the understanding of the regulatory mechanisms of UPR caused by compromised tRNA synthesis during neurogenesis. Misfolded proteins have been found to activate Perk signaling, leading to increased apoptosis of NPCs and termination of neurogenesis via the Nrf2–p53 axis. Nrf2 is a critical factor in the determination of cell fate and may be considered a target to attenuate neurodegenerative disorders with defective tRNA synthesis.

Complete information on the zebrafish (Danio rerio) strains is provided in Table S2. All experimental protocols were approved by the School of Life Sciences, Southwest University (Chongqing, China). The methods were approved by the guidelines. Fish were raised at 28.5°C under a 14 h light/10 h dark cycle. Embryos were obtained after natural spawning and were put in E3 buffer with 0.003% 1-phenyl-2-thiourea (PTU; Sigma) to prevent pigment formation. The zebrafish facility and study were approved by the Ethics Committee of the College of Life Science, Southwest University. The HEK293T cell line (see Table S2) was cultured following a standard protocol. The medium contained high-glucose Dulbecco's modified Eagle medium (DMEM) with glutamine and sodium pyruvate (Biological Industries), 10% fetal bovine serum (Biological Industries), 1% L-glutamine (Beyotime), 1 µg/ml Mycoplasma Removal Agent Plus (Beyotime) and 1% Penicillin-Streptomycin-Amphotericin B Solution (Beyotime). The cells were grown in an incubator (Shanghai Yiheng Technology) following a standard protocol.

ENU mutagenesis and positional cloning were performed as previously described (Trevarrow, 2011). The genetic mapping strain ENU/WIK was created by crossing heterozygous ENU mutant with WT WIK zebrafish. WT and mutant embryos were phenotypically collected from pairwise mating of heterozygous ENU/WIK zebrafish. The Mut^cq71/cq71 embryos presented an obvious reduction of neuronal markers at 2.5-3 dpf. We first located the mutation to linkage group (LG)18 by bulked segregation analysis, which indicated a close linkage of the mutant pools with the sequence-length polymorphism (SSLP) markers in LG18. SSLP markers and SNPs were used to perform the fine mapping to narrow the genetic interval. The cDNA of candidate genes was cloned and sequenced from pooled mutants, Finally, a point mutation in exon 5 of aars1 gene (Ensembl: ENSDARG00000069142) was identified in Mut^cq71/cq71 embryos based on the sequencing data.

aars1^Δ10/Δ10 and nfe2l2b^Δ1/Δ1 mutant alleles were generated by the CRISPR/Cas9 system (Chang et al., 2013). The gRNAs targeting the 5′-GGGACGGGTCAATGGTGTTCAGG-3′ in aars1 exon 4, and 5′-TGGATGTCTCCGGCAGAGGG-3′ in nfe2l2b exon 2, were in vitro synthesized by a mMESSAGE mMACHINE T7 transcription kit (Thermo Fisher Scientific). Cas9 protein was mixed with gRNA according to the manufacturer's instruction (Biolabs). The mixture was injected into the fertilized embryos at one-cell stage. The aars1^Δ10/Δ10 and nfe2l2b^Δ1/Δ1 mutant alleles were obtained by DNA sequencing. To construct the pT2AL2-NBT:aars1;cryaa:cerulean plasmid, the full length of the aars1²⁰² cDNA fragment was amplified and subcloned into the pT2AL2-NBT;cryaa:cerulean backbone (Tilton and Tanguay, 2008; Wu et al., 2018). Similarly, to construct Tg(hsp70:p53) plasmids, specific primers (Table S1) were used to amplify its cDNA sequence and clone into the pTol2-hsp70 backbone (Wu and Wang, 2020). To facilitate screening, a Cryaa-Cerulean-BGHpA element (Hesselson et al., 2009) was inserted reversely. The pT2AL2-NBT:aars1;cryaa:cerulean constructs (30 ng/μl) containing Tol2 transposase mRNA (250 ng/μl) were co-injected into the WT embryos at one-cell stage to build the F0 founders. The stable transgenic lines were recovered in the F1 generation based on the appearance of cerulean⁺ signals in the eyes and the same WISH patterns of aars1 as that of NBT. The Tg(hsp70:p53;cryaa:cerulean) and Tg(NBT:DenNTR) heritable zebrafish transgenic lines were generated by the same methods.

Embryos with Tg(hsp70:p53) transgenic background were heated to 38.5°C for 40 min and then incubated at 28.5°C. The efficiency of p53 expression after heat shock was verified by fluorescence signals in the eye and presentation of p53 by WISH.

RNA probes were synthesized by a DIG RNA Labeling Kit T3/T7/SP6 (Roche). WISH and FISH experiments were performed as described previously (He et al., 2020; Huang et al., 2019). TUNEL staining was performed using an In Situ Cell Death Detection Kit and TMR Red Kit (Roche) following the manufacturer's instructions or as reported previously (Lei et al., 2017). AO staining was performed as described (Thompson et al., 2015). Briefly, zebrafish embryos were incubated with 10 μg/ml AO solution in the dark at 28.5°C for 40 min. They were then observed by a Carl Zeiss Discovery V20 microscope. Immunofluorescence staining of zebrafish embryos was performed following standard protocols (Zhao et al., 2018). Briefly, embryos of the desired stages were fixed by 4% paraformaldehyde (4°C, overnight). The embryos were then co-incubated with the primary antibody (Table S2) at 4°C overnight. Afterwards, the samples were rinsed with PBDT (phosphate buffer containing 1% DMSO and 0.02% Tween 20) and then stained with Alexa Fluor 488/555/647-conjugated secondary antibodies (Table S2). The WISH images were captured by a Carl Zeiss Discovery V20 microscope. The immunofluorescence staining images were photographed by LSM700 or LSM880 confocal microscopes.

The behavioral study was performed according to a previous description (Palmér et al., 2017). Individual zebrafish larvae were placed in a plate containing 1 ml egg water. To record the movement of spontaneous swimming behavior, larvae were adaptive for 5 min, and their behaviors were then captured for 5 min by a scan camera (Basler, model ACA1300-60 GC). Their behaviors after light–dark and needle stimulation were monitored following reported studies (Marquez-Legorreta et al., 2020; Sztal et al., 2016). Briefly, to perform the light–dark stimulation experiment, the larvae were first adapted in a recording box with light for 30 min. Then, the light was turned off for 5 min and then turned on for another 5 min. The track of larvae in the 10 min light–dark cycle was monitored by an area scan camera. To conduct the needle stimulation, a sterilized syringe needle was used to touch the larval head, and the area scan camera was used to capture the swimming track for 3 min. All the imaging data were analyzed by the Ethovision XT15 animal tracking system (Noldus). The distance and movement speed were calculated accordingly.

Ca²⁺ activities of neurons were monitored in Tg(HuC:GCaMP6s) embryos. Embryos were immobilized in 1% low-melting point agarose in a 35 mm glass-bottom dish. In vivo uninterrupted time-series Ca²⁺ imaging experiments were performed with scanning (1.2 Hz at an excitation of 488 nm) under an LSM880 confocal microscope (Carl Zeiss), and the resolution was 512×512 pixels. The data were extracted by ZEN2012 software, and the movies were created by Corel Video Studio.

Annexin V staining was performed following previous instructions (She et al., 2018). Briefly, Alexa Fluor 488-conjugated Annexin V (Molecular Probes) was mixed with 1× binding buffer at a ratio of 1:1 (BioVision). The mixtures were loaded into the midbrain gaps of 3 dpf zebrafish, which were then incubated in the dark for 1.5 h. Similarly, after HEK293T cells were treated with 500 μl trypsin solution for 1 min, the cells were collected and washed twice with pre-cooled PBS. A mixture of 100 μl 1× binding buffer and 5 μl annexin V-FITC were added to the preparations of 1×10⁶ cells. These cells were then incubated (37°C) in the dark for 5 min. The signals were visualized by a LSM700 confocal microscope.

Cell transplantation was performed as previously described (Zou and Wei, 2010). Approximately 30-50 blastomere cells were drawn from 1000-cell stage WT and aars1 RNP F0 donors with Tg(NBT:DenNTR) background and loaded into needles. The blastomere cell-filled needles were injected into the animal pole of 1000-cell stage host embryos (aars1^cq71/cq71 or siblings). The cells were slowly released by a micromanipulator (Eppendorf CellTram 4r Oil).

TEM was conducted by Wuhan Servicebio Technology. We prepared the samples according to their instructions. Briefly, the anaesthetized 3 and 5 dpf zebrafish larvae were placed in pre-cooled 1× PBS, and the midbrains were microdissected using 0.33 mm micro-fine sterile needles. The samples were fixed in the electron microscope fixation solution for 2 h. They were then moved to 1% OsO₄ in 0.1 M phosphate buffer (PB) (pH 7.4) in the dark at room temperature for 2 h. After removing OsO₄, the tissues were rinsed with 0.1 M PB (pH 7.4) for 15 min, repeated three times. Fixed embryos were gradually dehydrated with ethanol and embedded individually into resin blocks. The resin blocks were cut to 60-80 nm thickness by an ultramicrotome (Leica UC7). The tissues were placed onto 150 mesh cuprum grids with formvar film, and then stained with 2% uranium acetate saturated alcohol solution and 2.6% lead citrate. Images were acquired using a Hitachi HT7800/HT7700 TEM.

Western blotting was performed as described previously (Zhao et al., 2018). Proteins were purified using cell lysis buffer (Beyotime). Subsequently, the membranes were blocked in Tris-buffered saline with 0.5% Tween 20 (TBST) containing 5% fat-free milk at 37°C for 1 h. The samples were then incubated with primary antibodies (Table S2), which were diluted by QuickBlock™ buffer (Beyotime) at 4°C overnight. After incubating with corresponding secondary antibodies (Table S2), the specific bands were detected by an Odyssey FC two-color infrared laser imaging system (LI-COR).

The fragments of WT and mutated CDS of aars1 were amplified by primers (Table S1). They were subcloned into a pCS2+ vector to synthesize mRNA using a mMESSAGEmMACHINE™ SP6 kit (Roche). chop and nfe2l2b mRNA were prepared by the same method. To perform the ChIP and luciferase experiments, we used a standard protocol to perform site-directed mutagenesis (Huang et al., 2019). Briefly, the primers were designed on both sides of the mutation sites, and 50 μl PCR reactions contained 100 ng template, 2 μl primer (F/R), 1 μl dNTPs, 1 μl fast-Pfu DNA polymerase and 5 μl reaction buffer. Each amplification cycle consisted of 95°C for 20 s, 50°C for 30 s and 68°C for 10 min. After that, the PCR products were treated with 1 μl DpnI (15 μl products) at 80°C for 10 min. They were then analyzed by agarose gel electrophoresis, and finally transformed into Escherichia coli DH5α by heat shock. The plasmids were isolated by a Miniprep Kit (Axygen) and then verified by sequencing. To perform chemical treatment, 1.5 dpf zebrafish embryos were incubated with GSK2656157 (65 μM), Tg (5 μM), Tn (1 μM), CHX (50 μM), oltipraz (10 μM) and Ceapin-A7 (50 μM). The HEK293T cells at designated time points were cultured with Tg (0.5 μM or 1 μM), oltipraz (40 μM) and Tn (6 μM) for 24 h.

ChIP assay was performed using 24 hpf embryos injected with HA-nfe2l2b mRNA. Because there was no Nfe2l2b-specific antibody for ChIP assay in zebrafish, we pulled down the target DNA fragments using antibody against HA. The eluted DNA (precipitated by anti-HA antibody) was assayed by PCR. The primers were designed according to the conserved Nfe2l2b-binding sites (Table S1). Non-specific primers and rabbit purified IgG were used as controls. The luciferase assays were performed using a Dual-Luciferase Reporter Assay System (Beyotime). The luciferase activity was measured by a GloMax 20/20 luminometer (Promega) according to the manufacturer's instructions. The −2.6 kb p53 promoter was amplified by the primers (Table S1) and cloned into a pGL3-basic vector (Promega). The p53-binding sites were mutated by site-directed mutagenesis using the designed site-specific oligonucleotide primers (Table S1). HEK293T cells were cultured in DMEM with 10% fetal bovine serum in 24-well plates. According to a standard protocol of Lipo8000™ Transfection Reagent (Beyotime), the plasmids (200 ng pGL3-p53 promoter, 200 ng pGL3-p53 promoter-MBS, 250 ng pCS2-nfe2l2b CDS, 250 ng pCS2 and 40 ng pRL-CMV) were transfected into cells and cultured for 36 h. Finally, the cells were harvested to perform luciferase assays using a Dual Luciferase Reporter Gene Assay Kit (Beyotime).

The genomic sequences of nrf2 and aars1 were acquired from the Ensembl database. The predicted zebrafish Nrf2 amino acid sequence was aligned with human (Homo sapiens) by the neighbor-joining method. The standard Kimura two-parameter model was used for the construction of phylogenetic tree using Molecular Evolution Genetics Analysis software (MEGA, version 5.1) (Tamura et al., 2011). The divergence was determined by the distance matrix method. The pairwise distance of Aars1 among zebrafish, chicken, human and house mouse was estimated (https://www.ebi.ac.uk/Tools/msa/clustalo/). The model diagram was created in biorender.com.

Quantitative RT-PCR (qRT-PCR) was carried out using total RNA isolated from the brains of aars1^cq71/cq71 or WT siblings according to a previous method (Wu et al., 2018). NBT-DenNTR⁺, Sox2-GFP⁺, Neurod1-EGFP⁺, Olig2-DsRed2⁺ and Coro1a-DsRed⁺ cells were sorted by fluorescence-activated cell sorting (Moflo XDP, Beckman). Then, cDNA libraries were generated by a REPLI-g WTA Single Cell Kit (Qiagen). All data were obtained by a Roche light cycler 96 machine and the expression levels of β-actin or ef1α were used as the reference. Table S1 lists information on the qRT-PCR primers.

scRNA-seq was conducted by AccuraMed. The embryonic midbrains of 2 dpf aars1^cq71/cq71 and WT siblings were extracted under a stereomicroscope by syringes and needles. They were then dissected in the ice-cold Dulbecco's phosphate-buffered Saline (DPBS) and dissociated using a Papain Dissociation Kit (Worthington) according to the manufacturer's instruction (Tambalo et al., 2020). Subsequently, the samples were centrifuged at 200 g for 4 min and washed twice with 1 ml DPBS. Then, the cells were filtered through a 40 μM cell strainer and finally resuspended in DPBS. The single-cell suspension was promptly loaded on a 10× Chromium System and sent to AccuraMed. The 10× libraries were prepared as per the manufacturer's instructions. The raw sequencing data were processed by the Cell Ranger software with the default options. The reads were aligned to zebrafish transcriptome reference (ENSEMBL Zv10, release 91). The resulting matrices were used as input for the downstream analysis by Seurat.

The analyses on the UMI count matrix in the two datasets were performed using the R package Seurat (version 4.0.1) (Butler et al., 2018) following the standard workflow in the Satija laboratory webpage (https://satijalab.org). The aars1^cq71/cq71 and sibling datasets were merged using the Seurat function Merge Seurat (https://www.rdocumentation.org/packages/Seurat/versions/3.1.4/topics/merge.Assay) (Stuart et al., 2019). Initial quality control was performed by filtering out genes expressed in less than one cell. Gene selection for dimensional reduction was accomplished using the Seurat function Find Variable Genes (https://www.rdocumentation.org/packages/Seurat/versions/2.3.4/topics/FindVariableGenes). Following gene selection, all log-normalized expression values were scaled and centered using Scale Data (https://github.com/satijalab/seurat). Principal components were chosen according to the principal component analysis elbow plot, which ordered principal components from the highest to the lowest based on the percentage of variance explained by each principal component. Next, we performed clustering on each set of principal components and a second round of dimensional reduction using t-distributed stochastic neighbor embedding (t-SNE). Markers of each cell type were identified based on their significant differential gene expression (adjusted P<0.05) using the Seurat function Find All Markers (https://www.rdocumentation.org/packages/Seurat/versions/4.1.1/topics/FindAllMarkers). Clusters were annotated based on the expression datasets of these cluster-specific marker genes on ZFIN (https://zfin.org). Both datasets contained non-specific cell types that contained markers for glial (pharyngeal cells), retinal (cone bipolar cells) and erythrocyte cells were removed after the initial clustering. Following these stepwise processes, we finally obtained the purified data of each sample. To identify cell trajectories on pseudotime, the progenitor cells of Seurat object was converted to Monocle (Durruthy-Durruthy and Heller, 2015). We used the genes of canonical correlation analysis to order Monocle and defined the maximum component as 2.

GO analysis was performed in Database for Annotation, Visualization and Integrated Discovery (DAVID) (Xue et al., 2019). KEGG pathway analysis of highly differentially expressed genes was performed on progenitor cell clusters of aars1^cq71/cq71 versus WT siblings by the ‘cluster Profiler’ R package (Yu et al., 2012).

The AO⁺ signals were quantified as described previously (Sidi et al., 2008). In brief, the images were analyzed by ImageJ software (National Institutes of Health, Bethesda, MD, USA). The targeted regions in the midbrains were outlined by the polygon selector. We reduced the noise using the ‘Image>Adjust> Threshold’ command accordingly (Sidi et al., 2008). After that, the ‘Analyze>Analyze Particles’ commands were executed, and the Size (Micro ^2) was set as 1-infinity. The data were obtained by running ‘Display Results, Clear Results’ and ‘Summarize’ programs at the same time. Finally, the ratio of the marked AO⁺ area in the whole outlined midbrain region was calculated using the ‘Area Fraction’ program. Midbrain area was measured as described previously (Chowdhury et al., 2018). In brief, zebrafish embryos at the desired stages were anaesthetized by tricaine. The brains were imaged by a digital camera (Axiocam 506 color, Carl Zeiss) in a Discovery.V20 microscope (Carl Zeiss). The images were analyzed by ImageJ software. The ‘Straight’ tool was used to describe the scale. The ‘Analyze>Set Scale’ option was used to calculate the actual size of the whole image. Then, the ‘Polygon Selection’ tool was applied to determine the targeted region and the volume was given out by running the ‘Analyze>Measure’ command. To evaluate the zebrafish survival ratio, embryos in each group were assigned to different tanks and cultured at 28.5°C. The numbers of surviving and dead larval fish were counted daily from 3 to 10 dpf. The fresh water for larval culture was supplied regularly to avoid environmental turbidity and hypoxia. Finally, GraphPad Prism 6.01 was used to draw the survival curve, and survival distribution was estimated by the Kaplan–Meier method.

The bands of western blotting and DNA gel electrophoresis were quantified by ImageJ software. Briefly, the ‘Process>Subtract Background>80’ command was executed to reduce the background. The rectangle tool was used to outline the measured bands. The gray values of the bands were given out by the ‘Analyze>Gels>Select First Lane> Select Last Lane>Plot Plane’ commands. To quantify the FISH fluorescence intensity, the ‘Image>Adjust>Auto Threshold’ instructions were followed. Then, the ‘Analyze>Set Measurements>Area>Mean Gray Value>Limit To Threshold>Mode Yen’ were selected. The intensity results were calculated by the ‘Analyze>Measure’ commands.

scRNA-seq statistical analyses were performed in RStudio (version 4.0.3/4.0.4). The quantified data were double confirmed and analyzed by GraphPad Prism 6.01. The detailed statistical methods are included in the corresponding figure legends. Briefly, unpaired two-tailed Student's t-test was performed to compare two data groups without any treatment; two-way ANOVA was employed to weight the two data groups in different conditions; and one-way ANOVA was performed to analyze the data of multiple groups. Survival ratio was calculated by Kaplan–Meier calculations. Error bars indicate s.e.m. All statistical tests, P-values and n numbers are stated in figure legends

We thank H. Li and M. Ma for biochemistry analysis and discussions, Z. Cao for the tars1^cq16 mutant, and H. Huang and H. Ruan for technical assistance.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.200342.reviewer-comments.pdf