To identify lincRNAs that may encode for micropeptides, we first analyzed ribosome profiles for previously published lincRNAs Ulitsky et al., 2011 in zebrafish embryos during early development (0–48 hr post-fertilization) (Bazzini et al., 2014), performed in situ hybridization on 21 of these candidates, and identified brain-enriched micropeptide candidates (Figure 1—figure supplement 1; Supplementary file 1). To identify the physiological role of ten of these putative micropeptides, we adapted an F0 CRISPR/Cas9 behavioral screening pipeline (Figure 1A; Kroll et al., 2021). CRISPR/Cas9 targeting efficiently induced a range of mutations in the targeted gene sequences, with inferred indel or large deletion rates with multiple guides estimated between ~40 and 100% per targeted locus, including frame-shift mutations (Figure 1—figure supplement 2; Supplementary file 1).

At 4–7 days post-fertilization (dpf), zebrafish display a repertoire of conserved, stereotyped baseline locomotor behaviors across day:night cycles (Prober et al., 2006; Rihel et al., 2010; Kroll et al., 2021). To quantitatively track locomotor activity of wild type and F0 mutant fish, single larvae from each condition were placed into individual clear wells of a clear 96-square well flat plate, then placed on a tracking platform that detects the change in pixels per frame for each well, between 4 dpf and 7 dpf (Figure 1A). We measured daytime and nighttime behavioral parameters, calculated the deviation (Z-score) of each F0 mutant larva from its wild type siblings, generated ‘behavioral fingerprints’ (Figure 1—figure supplement 3A), and measured the Euclidean distance between each larva and the mean fingerprint of its wild type siblings (Figure 1—figure supplement 3B).

This screen identified two candidate genes, linc-mipep and linc-wrb, that had a specific daytime hyperactivity phenotype and correlated behavioral fingerprints (r=0.67) when mutated in the ORF identified by ribosome footprints (Figure 1B; Figure 1—figure supplement 3C; Figure 1—figure supplement 4). Sequence analysis revealed that linc-mipep (current nomenclature si:ch73-1a9.3, ENSDART00000158245, also called lnc-rps25) and linc-wrb (current nomenclature si:ch73-281n10.2, ENSDART00000155252) (Bazzini et al., 2014; Ulitsky et al., 2011; Figure 1—figure supplement 5A, B) both had homology in their sORFs’ exon structure (Figure 1—figure supplement 5A, B) and mRNA sequences (BLAST identity score = 72%) (Supplementary file 1), as well as a highly conserved element (92% identical sequence) in their non-coding sequences (Figure 1—figure supplement 5C). While both linc-mipep and linc-wrb were originally identified as long non-coding RNAs, both genes have ribosome-protected fragments, suggesting they are likely encoding proteins 87aa and 93aa in size, respectively (Figure 1B; Figure 1—figure supplement 5D, E). In situ hybridization and RNA-sequencing revealed that transcripts for both genes are expressed throughout embryogenesis, through 5 dpf (Figure 1—figure supplement 5F, G). These results indicate that linc-mipep and linc-wrb might encode redundant or paralogous genes functioning as either lincRNAs or micropeptide-encoding genes involved in behavior.

Although linc-mipep and linc-wrb are transcribed and likely translated (Figure 1B; Figure 1—figure supplement 5D–F), ribosome profiling data is insufficient to distinguish between pervasive background translation and translation of functional proteins. For example, these could represent sORFs within enhancer RNAs or in noncoding RNAs that have acquired an ORF but yield a nonfunctional protein. Thus, to distinguish whether linc-mipep and linc-wrb function as regulatory DNA, noncoding RNA, or protein coding genes, we used CRISPR-Cas9 gene editing to generate stable deletion mutants that either target the full sequence, the translation start site, the putative coding region, or the conserved untranslated/non-coding region (Figure 1C–E ; Figure 1—figure supplement 6). Examining the behavioral profile of these mutants identified a consistent and specific increase in locomotor activity during the daytime in all mutants affecting the ORF for both linc-mipep and linc-wrb (Figure 1—figure supplement 6A–D). In contrast, deleting the highly conserved element in the untranslated region in linc-mipep, which could encode a conserved lincRNA sequence, did not result in any detectable morphological or behavioral phenotypes (Figure 1—figure supplement 6E).

First, we asked whether the coding part of these genes is necessary. Start codon mutations in linc-mipep (zygotic or maternal-zygotic linc-mipep^ATG-del6) resulted in a similar daytime hyperactivity phenotype as frameshift mutations (linc-mipep^del8) or deletion of most of the linc-mipep region (linc-mipep^del1.8kb) (Figure 1—figure supplement 6A–C, Figure 1—figure supplement 7A). These results indicate that the observed phenotypes are the result of protein coding function of linc-mipep rather than a non-coding transcript or a regulatory DNA sequence function. Double linc-mipep^del1.8kb; linc-wrb^del11 homozygous mutants display even higher daytime locomotor hyperactivity levels compared to linc-mipep; linc-wrb heterozygous or wildtype larvae (Figure 1—figure supplement 7C), with no significant changes in nighttime activity (Figure 1F), a phenotype that is maintained if we remove the maternal contribution in maternal-zygotic (MZ) linc-mipep^del1.8kb; linc-wrb^del11 animals (Figure 1G and H; Figure 1—figure supplement 7B). Each additional loss of a copy of either gene generally results in higher hyperactivity levels (Figure 1—figure supplement 7C), suggesting that these genes may work together in a dose-dependent manner.

Next, we asked whether the coding part of these genes is sufficient to drive behavior. To determine that the behavioral phenotypes observed in mutants result from the loss of coding function, we generated transgenic zebrafish that ubiquitously express the coding sequence (CDS) of linc-mipep and tracked their behavior (Figure 2A and B). The sORF encoded in linc-mipep was able to rescue the hyperactivity phenotypes in linc-mipep mutants (Figure 2C and D) without significant changes to wild type activity levels (Figure 2D) or in nighttime activity (magnified, Figure 2C). Moreover, linc-mipep expression was able to rescue the hyperactivity of linc-wrb heterozygous mutants to almost wild type levels (Figure 2—figure supplement 1A, B), suggesting that these proteins share properties that can rescue loss of the other.

Finally, to confirm and visualize the protein encoded by linc-mipep and linc-wrb, we developed custom antibodies (Figure 2—figure supplement 2A, B). The protein product of both transcripts are detected in developing wild type embryos and larvae (Figure 2—figure supplement 2C–K). We find that these proteins are expressed throughout early development, with stronger staining and broader expression pattern for the protein encoded by linc-mipep compared to that of linc-wrb (Figure 2—figure supplement 2G, H). We note nonspecific staining of the Linc-wrb antibody in embryos and in likely endothelial cells throughout early development, as staining is still detected in these cells in linc-wrb mutants (Figure 2—figure supplement 2I, K). We further observed that the protein products of both transcripts are enriched in non-dividing wild-type nuclei (Figure 2—figure supplement 2J, K) and absent in linc-mipep;linc-wrb loss-of-function mutant embryos (Figure 2—figure supplement 2J, K) and larval brains (Figure 2—figure supplement 2L, M). Together, these results indicate that linc-mipep and linc-wrb encode for nuclear-localized micropeptides that have a dosage effect to regulate locomotor activity and behavior in zebrafish.

Protein BLAST of both Linc-mipep (87aa) and Linc-wrb (93aa) ORFs identified conserved sequences across teleosts and other vertebrates, including humans, with homology to non-histone chromosomal protein HMG-14, or High Mobility Group N1 (HMGN1), and the related HMG-17/HMGN2 protein (Supplementary file 2; Bustin, 2001). Whereas the cDNA sequence showed some mild conservation (Figure 2—figure supplement 3A), the highly conserved proximal 3’UTR elements instead allowed us to identify homologous predicted lincRNAs, unannotated genes, pseudogenes, and HMGN1 genes across vertebrate species spanning over 450 million years (Figure 2—figure supplement 3B–C; Figure 2—figure supplement 4; Supplementary file 2; Kumar and Hedges, 1998).

We first identify that linc-wrb is syntenic to human Hmgn1 (Figure 2—figure supplement 5D–F). To identify the evolutionary origin of this gene, we traced back the synteny for sequences or expressed sequence tags (ESTs) that were identified between flanking genes that are syntenically conserved with humans, Get1/Wrb and Brwd1 (Figure 2—figure supplement 5E). Through these analyses, we were first able to identify an unannotated ORF in the basal agnathan (jawless vertebrate) lamprey, syntenic to human HMGN1, that encodes for an ancestral protein more similar to human HMGN2 (Figure 2—figure supplement 5B; Figure 2—figure supplement 3E). Though we did not identify any linc-mipep or linc-wrb protein-coding homolog in invertebrates (Supplementary file 2), in line with previous results (Johns, 1982), we did identify an ORF syntenic to linc-wrb in the invertebrate basal chordate lancelet (or amphioxus) genome (Figure 2—figure supplement 5A). When we analyzed whether there were any similarities between the sequence of this APEX1-like gene in the lancelet genome (Figure 2—figure supplement 5A), we found by BLAST that its N-terminal sequence (30aa) aligns to HMGN family members in various vertebrate species (Figure 2—figure supplement 3D). These results suggest that the N-terminal sequence of the gene in the ancestral location that would give rise to linc-wrb and human HMGN1 may have been co-opted to give rise to the HMGN gene and pseudogene families in vertebrates.

We next searched for the evolutionary origins of linc-mipep. The highly conserved 3’UTR suggested that linc-mipep and linc-wrb derived from the same ancestral gene, either before or after the teleost-specific genome duplication. To address this question, we analyzed the regions syntenic to human HMGN1 in spotted gar, a slowly evolving species whose lineage diverged from teleosts before the teleost genome duplication, and in coelacanth, a lobe-finned fish with the slowest evolving bony vertebrate genome that split from ray-finned fish such as gar and zebrafish (Braasch et al., 2016). In coelacanth, we only identified one protein sequence that aligns to HMGN1, syntenic to human HMGN1, with no ESTs or other sequences identified elsewhere (Supplementary file 2). In spotted gar, we found the gene syntenic to linc-wrb and human Hmgn1 (Figure 2—figure supplement 5C). Although we did not identify a gene syntenic to linc-mipep, we did identify the appearance of both Mipep next to Brwd1, and of Igsf5 next to Sh3bgr (Figure 2—figure supplement 5C). When analyzed compared to the genomic location of linc-mipep in zebrafish (Figure 2—figure supplement 5D), we suggest that linc-mipep may have resulted from a gene duplication of linc-wrb into the neighboring IGSF5 intronic region, which then rearranged to land next to Mipep in the teleost genome duplication (compare Figure 2—figure supplement 5C, D). We found that linc-mipep has been maintained in other teleost fish species (Figure 2—figure supplement 5; Figure 2—figure supplement 6). Together, these findings suggest that linc-mipep arose from a gene duplication from linc-wrb, and that linc-wrb arose from what we identify here as the basal HMGN gene in agnathan lineages.

Finally, to understand whether the proteins encoded by linc-mipep and linc-wrb share common functions with human Hmgn1, we asked whether the human HMGN1 homologous protein can rescue the hyperactivity of linc-mipep and linc-wrb mutants. We generated transgenic zebrafish that ubiquitously express the coding sequence (CDS) of human HMGN1 in each mutant background. Human HMGN1 was able to rescue the hyperactivity phenotypes in linc-mipep mutants (Figure 2F and G), without significant changes in nighttime activity (magnified, Figure 2F). We were unable to rescue the linc-wrb mutant phenotype with human HMGN1 (Figure 2—figure supplement 1D). These data suggest that genes encoded within linc-mipep and linc-wrb have some functional homology with each other, and that at least the protein encoded by linc-mipep has functional homology with, and can be rescued by, human HMGN1. Based on these results, we propose renaming linc-wrb as hmgn1a, and linc-mipep as hmgn1b, as their official nomenclature.

To gain insight into pathways regulated by linc-mipep and linc-wrb, we analyzed the behavioral fingerprints of each mutant compared to zebrafish larvae treated with 550 psychoactive drugs that affect different pathways (Rihel et al., 2010). We used hierarchical clustering (Rihel et al., 2010) to identify drugs that elicit a similar behavior to the linc-mipep and linc-wrb mutants (i.e. drugs that phenocopy across multiple day-night behavioral measurements) (Figure 3—figure supplement 1A, B, overlapping hits in blue text). We found that linc-mipep mutant behaviors most resembled those of WT fish treated with an NMDA receptor antagonist (Figure 3A), suggesting that NMDA signaling may be reduced in linc-mipep mutants. The linc-mipep and linc-wrb mutant phenotypes also resembled that of WT fish treated with glucocorticoid receptor activators (Figure 3A, Supplementary file 3), suggesting that downstream glucocorticoid signaling may be upregulated in the mutants.

The identified drugs may alter either common or parallel pathways as loss of linc-mipep. To distinguish between these possibilities, we first assessed the effect of glucocorticoid receptor agonist flumethasone on linc-mipep mutant behavior. These treatments further exacerbated the daytime locomotor activity of linc-mipep^-/- larvae above the control-treated linc-mipep mutant levels (Figure 3—figure supplement 2A), with higher nighttime activity levels in linc-mipep mutants treated with flumethasone (Figure 3—figure supplement 2B). Since both the daytime and nighttime effects of glucocorticoids were much stronger in the mutants than in similarly treated wild type controls, linc-mipep mutants are sensitized to glucocorticoid signaling. We found similar glucocorticoid sensitivity in linc-wrb mutants (Figure 3—figure supplement 2C, D).

Next, to test the NMDA receptor pathway, we compared the response of WT and linc-mipep mutant to L-701–324, an NMDA receptor antagonist at the glycine binding site. L-701–324 elicited a daytime locomotor hyperactivity in WT larvae to a level that was similar to that of linc-mipep mutant larvae and linc-mipep larvae treated with L-701–324 (Figure 3B). Yet, treatment with higher doses of L-701–324 did not affect or exacerbate the activity levels in linc-mipep mutants (Figure 3C). We found similar results with linc-wrb mutants treated with L-701–324 (Figure 3—figure supplement 2E, F). These non-additive results indicate that NMDA receptor antagonism and mutations in linc-mipep and linc-wrb share a common mechanism for inducing hyperactivity.

Given that linc-mipep and linc-wrb have protein domains with homology to nucleosome binding and chromatin unwinding domains of HMGN1 (Cuddapah et al., 2011; Deng et al., 2013), and given that both NMDA antagonism and glucocorticoid signaling alter immediate early gene expression, we hypothesized that the daytime hyperactivity might be due to altered chromatin accessibility in the mutants. To test the effect of full loss-of-function of both related proteins encoded by linc-mipep and linc-wrb on chromatin accessibility, we performed omni-ATAC-seq (Corces et al., 2017) at 5 dpf comparing WT and double mutant brains (Figure 3—figure supplement 3A, B).

We first observed a broad dysregulation of chromatin accessibility, with 2167 regions losing accessibility and 1220 regions gaining accessibility in linc-mipep;linc-wrb mutant brains (Figure 3D; Supplementary file 4), with most regions remaining unchanged (Figure 3—figure supplement 3C). CTCF/L transcription factor (TF) motifs were enriched in regions that lost accessibility, suggesting a possible dysregulation of 3D chromatin structure (Figure 3E). Enriched TF motifs at regions that lost accessibility were members of the ATF (activating transcription factor)/CREB (cAMP responsive element binding proteins) family, and AP-1 transcription factor components (Figure 3E, left panel). TFs binding at these motifs regulate the expression of immediate early response genes (IEG) such as c-fos, c-jun, and c-myc (Sheng and Greenberg, 1990). We confirmed reduced c-fos transcription in linc-mipep;linc-wrb brains at this timepoint by in situ hybridization and by qPCR (Figure 3—figure supplement 3D). We also found that the motifs for the glucocorticoid modulatory element binding protein 2 (GMEB2), and for interferon-stimulated transcription factor 3, gamma (ISGF3G, also called IRF-9), were enriched in regions that lost accessibility in linc-mipep; linc-wrb mutants. On the other hand, TFs most enriched in regions that gained accessibility were KLF/SP family members, which promote stem cell pluripotency and are downregulated during differentiation, and EGR family members (Figure 3E, right panel; Yamane et al., 2018). Altogether, these results indicate that linc-mipep; linc-wrb have altered accessibility for TF binding sites, which modify the expression of genes involved in neural activation.

Our molecular analyses of wild-type and mutant brains point to gene regulatory networks involved in global transcription rather than neural cell type-specific TFs. We hypothesize that the observed hyperactivity may instead be a result of defects in cells most susceptible to loss of linc-mipep and linc-wrb. To test this hypothesis, we used single-cell multiomics (transcriptomic and chromatin accessibility) and determined how single cell states are affected in mutant brains compared to sibling-matched WT brains at 6 dpf (Figure 4a). To circumvent batch effects from unmatched (non-sibling) samples that may skew single-cell analyses, and because our results so far indicated generally overlapping functions for linc-mipep and linc-wrb, we chose to analyze linc-mipep mutant brain cells and then to validate findings in vivo in linc-mipep; linc-wrb double mutants.

First, we used Weighted Nearest Neighbors (WNN) (Hao et al., 2021) on transcriptomic and chromatin accessibility data from both linc-mipep mutant and WT nuclei all pooled together. This analysis identified 43 clusters (Figure 4A, Figure 4—figure supplement 1A–C; Supplementary file 2). linc-mipep transcripts were detected in all WT clusters except microglia, with a slight enrichment in Purkinje cells, the inhibitory projection neurons of the cerebellum (Figure 4—figure supplement 1D, E), raising the possibility that these cells may be more affected in linc-mipep mutant brains. linc-wrb was detected in all WT clusters except cranial ganglia and ventral habenula cells, and was broadly expressed at lower levels than linc-mipep transcripts (Figure 4—figure supplement 1D, E). Each cluster was comprised of both WT and linc-mipep mutant cells, indicating that there was no complete absence of any cell type in mutants. In linc-mipep mutant cells, we note that an almost complete loss of expression of linc-mipep was observed in all clusters, without major changes in linc-wrb levels (Figure 3—figure supplement 2A).

Next, to identify the brain cells most significantly affected by linc-mipep mutations, we used Multiscale PHATE/Integrated Diffusion (Figure 4B and C, Figure 4—figure supplement 3A, B; Kuchroo et al., 2022; Kuchroo et al., 2021). This approach measures the effect of linc-mipep loss on cellular states by calculating the relative likelihood that any sampled cell state would be observed in either WT or mutant cells. When we analyzed the ‘Mutant Likelihood Score’ (from Figure 4B) for each cell by its respective cluster, we found that differentiating neuronal progenitor, glial progenitor, and cerebellar granule (excitatory) cell states were more likely to be represented in WT brains, while oligodendrocyte progenitor cell (OPC) states were more likely to be represented in linc-mipep^-/- mutant brains (Figure 4C, asterisks; Figure 4—figure supplement 3A, B; Supplementary file 6). Indeed, we find a subcluster of oligodendrocyte progenitor cell states much more likely to be found in linc-mipep^-/- samples (Figure 4C, dashed box). These data indicate that linc-mipep preferentially regulates oligodendrocyte and cerebellar cell states during development. Interestingly, these cell states correspond to evolutionarily newer vertebrate brain cell types (Lamanna et al., 2022).

In wild-type brains, Linc-mipep and Linc-wrb proteins are expressed throughout the brain. Linc-wrb antibody staining reveals an even expression pattern across the brain at 5 dpf, including in the cerebellar region and in olig2:GFP+ OPCs (Figure 4—figure supplement 4C, H–K). We found that Linc-mipep is more weakly expressed in the torus longitudinalis and tegmentum (as in Figure 4—figure supplement 4D–G) compared to Linc-wrb staining. These data suggest that both Linc-mipep and Linc-wrb are expressed in olig2 + cells and throughout the cerebellum.

To determine why the mutation affected these particular cell types, we next asked whether loss of linc-mipep caused any significant changes in chromatin accessibility and gene expression in single cell types (Figure 4D; Supplementary file 5). While all cell types are present in both mutant and wild-type brains, we found a strong dysregulation of chromatin accessibility and gene expression within multiple cell types (Figure 4—figure supplement 3C–E; Supplementary file 5; Supplementary file 8). When we examined TF motifs in regions of differential chromatin accessibility in each cluster, we found that linc-mipep regulates accessibility for key neurodevelopmental transcription factor families, including Sox, Stat, and Zic family members, in radial glial cells (clusters 3 and 7) and other cells; Esrra/b in midbrain glutamatergic neurons (cluster 13); and Egr and NeuroD members across various cell types (Figure 4—figure supplement 5A).

We then examined each cell type of interest more closely to better define cell type-specific changes, starting with cerebellar cell types. In cerebellar granule (excitatory) cells, we found 989 regions where chromatin accessibility was strongly dependent on linc-mipep function (645 regions with decreased accessibility, and 344 regions with increased accessibility) (Figure 4E). We specifically assessed linc-mipep granule cells and found that they lost accessibility at motifs known to bind the transcription factors Bhlhe22, Hic1, which is expressed in mature cerebellar granule cells and transcriptionally represses Atoh1 (Briggs et al., 2008), Neurod2, which required for survival of granule cells (Miyata et al., 1999), and Nfia, and gained accessibility at binding sites for Gfi1b and Nfatc1 (Figure 4—figure supplement 5B). Purkinje (inhibitory) neurons also showed significant differences in chromatin accessibility (Figure 4—figure supplement 3E), losing accessibility at motifs known to bind the transcription factor Gbx2, which is required for cerebellar development (Figure 4—figure supplement 5C; Wassarman et al., 1997). Furthermore, examining single-cell expression data, we found that, compared to wild-type Purkinje cells, linc-mipep mutant Purkinje cells exhibited a significant decrease in the expression of numerous genes, including roraa, rorb, foxp4, and prkcg, which are required for maturation or maintenance of Purkinje cells in zebrafish (Figure 4H; Supplementary file 5; Takeuchi et al., 2017). Consistent with these results showing cerebellar cell types are affected, Pol II ChIP-seq in 5 dpf brains showed that genes involved in cerebellar development, including zic2a, ascl1b, and atxn3, have reduced RNA Polymerase II binding in mutant brains (Supplementary file 7).

We next asked whether loss of linc-mipep caused any significant changes in chromatin accessibility and gene expression in OPCs. Like cerebellar granule and Purkinje cells, OPCs similarly showed a broad loss of chromatin accessibility in the absence of linc-mipep (Figure 4F). OPCs from linc-mipep mutants showed reduced accessibility at binding sites of E2f7, Elf1 (which is upregulated in differentiating oligodendrocytes), Fev, and Hinfp TFs, and increased accessibility at Sox10 binding sites (Figure 4—figure supplement 5D). These changes in accessibility were associated with shifts in gene expression levels consistent with defects in OPC development or maturation, as we found 136 genes that were down-regulated and 57 genes that were up-regulated in linc-mipep mutant OPCs relative to wild-type OPCs (Supplementary file 6).

To better understand how OPCs may be affected, we further analyzed omni-ATAC-seq analyses in wild type and linc-mipep;linc-wrb mutant brains. These analyses revealed differentially accessible regions downstream of olig2 (a transcription factor that activates the expression of myelin-associated genes), within a large intronic span of sgms2b (which synthesizes a component of myelin sheath), and upstream of fabp7a (which is important for OPC differentiation in vitro in mouse) (Foerster et al., 2020; Figure 4—figure supplement 6A–C). To validate that OPCs are affected in vivo, we found a significant 13% decrease (p=0.0053) in olig2 + oligodendrocyte progenitor cells’ signal in mutant brains compared to WT brains, with most of the loss coming from the optic tectum and the cerebellum of Tg(olig2:eGFP); linc-mipep^-/-; linc-wrb^-/- compared to control larvae (Figure 4G; Figure 4—figure supplement 4A).

Finally, we asked whether some of the genes that were differentially regulated in OPC, cerebellar granule cell, or Purkinje cell clusters could explain the dysregulation of NMDA receptor signaling and sensitization to glucocorticoids that we found in our earlier pharmacological profiling (Figure 3). Indeed, we found that some of the differentially regulated genes in single-cell analyses are known to be involved in NMDA receptor and glucocorticoid receptor signaling. For example, fkbp5, which is associated with glucocorticoid signaling, showed reduced expression in linc-mipep mutant OPCs, and scg5, which can mediate stress responses (Cao-Lei et al., 2014; Mbikay et al., 2001), showed reduced expression broadly (Figure 4H; Figure 4—figure supplement 7). These genes also showed changed chromatin accessibility in the linc-mipep;linc-wrb double mutant brains (Figure 4—figure supplement 6D and E). Similarly, numerous genes involved in NMDA receptor activity (aldocb, ttyh3b, slc1a2b, nrxn1a, grin1b, gpmbaa, atp1a1b) showed reduced expression in linc-mipep mutant OPCs relative to wild-type OPCs, consistent with a reduction in NMDA receptor signaling in mutants (adjusted p-value = 0.0294, GO Molecular Function from FishEnrichR analysis) (Supplementary file 6). For one of these genes, grin1b, we also observed associated changes in chromatin accessibility (Figure 4—figure supplement 6F). At the single-cell level, we find that the expression of grin1a and grin1b, which encode NMDAR subunits, are significantly downregulated in linc-mipep mutant OPCs relative to wild-type OPCs (Figure 4H; Figure 4—figure supplement 7 and Figure 4—figure supplement 8A, B). Some genes that are significantly misregulated specifically in linc-mipep mutant OPCs, such as erbb4, mag, qkia, and myt1b, are also specifically enriched in wild type granule cells, despite different developmental lineage origins and cellular progressions (Figure 4—figure supplement 9A–D). Together, these results suggest that loss of linc-mipep and linc-wrb preferentially affect the development of oligodendrocyte progenitor cells and cerebellar cells – evolutionarily newer vertebrate cell types - and these effects may mediate changes in NMDAR and glucocorticoid signaling through changes in chromatin accessibility and gene expression.

Fish lines were maintained in accordance with the AAALAC research guidelines, under a protocol approved by the Yale University Institutional Animal Care and Use Committee (IACUC Protocol Number 2021–11109). We have complied with all relevant ethical regulations under this protocol. Zebrafish husbandry and manipulation were performed as described, and all experiments were carried out at 28 °C. For all larval experiments, zebrafish embryos were raised at 28.5 °C in petri dishes at densities of 70 embryos/dish on a 14 hr:10 hr light:dark cycle in a DigiTherm 38 liter Heating/Cooling Incubator with circadian lighting (Tritech Research). Dishes of embryos were cleaned once per day with blue water (fish system water with 1 mg/L methylene blue, pH 7.0) until they were placed in behavior boxes (ZebraBox, Viewpoint), to ensure identical growing conditions. Normal development was assessed, and larvae exhibiting abnormal developmental features (no inflated swim bladder, curved) were not used.

Sequences for ribosome profiling were previously published (Bazzini et al., 2014 and Johnstone et al., 2016). Code for updated ribosome profiling plots available here. Updated mapping, including for new genome releases, is available here.

Synthetic guides were designed using CRISPRscan and ordered as sgRNAs through Synthego (Synthego Corportation, Redwood City, CA, USA). Target and scrambled (control) sequences are presented in Supplementary file 1. EnGen Spy Cas9 NLS protein (NEB, M0646) was used for F0 experiments. RNPs were formed by mixing 3 μM Cas9 protein, 300 mM KCl, and 10 mM of each synthetic sgRNA targeting one gene, incubating at 37 °C for 10 min, and cooling to room temperature for 5 min. One-cell stage zebrafish embryos were injected with 100pl of each respective mix early after fertilization into the yolk. Pools of 8 embryos at 24 hpf were collected and incubated in 50 μl of 100 mM NaOH at 95 °C for 20 min. Then, 25 μl of 1 M Tris-HCl (pH 7.5) was added to neutralize the mix. Two μl of these crude DNA extracts were used for genotyping with the corresponding forward and reverse primers (10 µM; Supplementary file 1) using a standard PCR protocol, and these products were then sent for Sanger sequencing to assess cutting efficiencies. Mutation efficiency was assessed using Inference of CRISPR Edits (Synthego Performance Analysis, ICE Analysis. 2019. v3.0. Synthego). We note that the linc-mettl3 target sites lie between highly repetitive regions, making it difficult to amplify the necessary length for ICE analysis. We provide PCR and Sanger sequencing results in this case, indicating efficient targeting and significant large genomic deletion.

CRISPR mutant generation was done following Moreno-Mateos et al., 2015. Briefly, CRISPRScan (crisprscan.org) was queried to identify appropriate target sequences (Moreno-Mateos et al., 2015). Primers were ordered and amplified with universal primer 5’- AAAAGCACCGACTCGGTGCCACTTTTTCAAGTTGATAACGGACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAAC-3’. sgRNAs were in vitro transcribed using the AmpliScribe T7 Flash kit, using the PCR product (with T7 promoter) as template. In vitro transcribed sgRNAs were treated with DNase I and precipitated with sodium acetate and ethanol. Cas9 mRNA was in vitro transcribed from DNA linearized by XbaI (pT3TS-nCas9n) using the mMESSAGE mMACHINE T3 kit (Ambion). In vitro transcribed Cas9 RNA was treated with DNase I and purified using the RNeasy Mini kit (Qiagen).

One-cell stage zebrafish embryos were injected with 50 pg of each respective sgRNA and 100 pg of cas9 mRNA. sgRNA and genotyping primers and target sequences are available in Supplementary file 1.

gBlocks (IDT) were ordered for the linc-mipep or human Hmgn1 coding sequence, plus a FLAG and HA tag at the C terminus, as follows:

Addgene plasmid #79885 (pMT-ubb-cytoBirA-2a-mCherry, a gift from Tatjana Sauka-Spengler Trinh et al., 2017) was digested with BamHI and EcoRV, and the resulting vector was used as the backbone for the construct. InFusion cloning (Takara Bio) was used to amplify the Linc-mipep-FLAG-HA coding sequence and ligate with the vector, using primers F: 5’- TTGTTTACAGGGATCgccaccATGCCTAAAAGGAGC-3’ and R: 5’- CTCTCCTGATCCGATagcgtaatctggaacatcgtatggg-3’. InFusion cloning (Takara Bio) was used to amplify the human Hmgn1-FLAG-HA coding sequence and ligate with the vector, using primers F: 5’ TTGTTTACAGGGATCCGCCACCATGCCCAAGAGG –3’ and R: 5’-CTCTCCTGATCCGATATCATCAGACTTGGCTTCTTTCTCTCC-3’. Sequence-verified plasmids were midi-prepped and injected into the cell of one-cell stage embryos at 20 ng/μl along with 200 ng/ul of Tol2 transposase capped mRNA.

The following stable fish mutant lines have been established in this study: linc-mipep^del1.8kb (ya126); linc-mipep^ATG-del6 (ya127); linc-mipep^del8 (ya128); linc-mipep^{3’UTR-del74} (ya129); linc-wrb^del11 (ya130). The following stable transgenic lines have been established in this study: Tg(ubb:linc-mipep-FLAG-HA-T2A-mCherry) (ya145); and Tg(ubb:human-Hmgn1-FLAG-HA-T2A-mCherry) (ya151). The following previously published transgenic line has been used in this study: Tg(olig2:egfp)^vu12.

At 4 dpf, single larvae from heterozygous linc-mipep mutant incrosses were placed into individual wells of a clear 96-square well flat plate (Whatman) filled with 650 μL of blue water (fish system water with 1 mg/L methylene blue, pH 7.0). Plates were placed in a Zebrabox (ViewPoint Life Sciences), and each well was tracked using ZebraLab (Viewpoint) in quantized mode, and analyzed with custom software as in Rihel et al., 2010 and at Rihel, 2023 and DOI: 10.5281/zenodo.7644073. Behavioral data were analyzed for statistical significance using one-way ANOVA followed by Tukey’s post hoc test (α=0.05), as previously described (Rihel et al., 2010). Each behavioral experiment presented was repeated 2–4 times. For analyses of maternal-zygotic linc-mipep;linc-wrb mutants, age- and size-matched wild type adult stocks (AB/TL) or linc-mipep;linc-wrb double-homozygous mutants were incrossed, collected simultaneously, and raised in identical conditions prior to quantitative locomotor activity tracking as described above.

As previously described (Kroll et al., 2021), the raw file generated by the ZebraLab software (ViewPoint Life Sciences) was exported into a series of xls files each containing 1 million rows of data. Each datapoint represented the number of pixels that changed grey value above a sensitivity threshold, set to 18, for one larva at one frame transition, a metric termed Δ pixels. These files, together with a metadata file labelling each well with a genotype, were input to the MATLAB script Vp_Extract.m (Ghosh and Rihel, 2020), which calculated the following behavioral parameters from the Δ pixels timeseries for both day and night: (1) active bout length (duration of each active bout in seconds); (2) active bout mean (mean of the Δ pixels composing each active bout); (3) active bout standard deviation (mean of the Δ pixels composing each active bout); (4) active bout total (sum of the Δ pixels composing each active bout); (5) active bout minimum (smallest Δ pixels of each bout); (6) active bout maximum (largest Δ pixels of each bout); (7) number of active bouts during the entire day or night; (8) total time active (% of the day or night); (9) inactive bout length (duration of each pause between active bouts in seconds). These measurements were then averaged across both days or both nights to obtain one measure per parameter per larva for the day and night. To build the behavioral fingerprints, we calculated the deviation (Z-score) of each mutant (F0) larva from the mean of their wild-type siblings across all parameters. Plotted in Extended Data Figure 1b and c for each parameter is the mean ± SEM of the Z-scores. We compared fingerprints between replicates (Extended Data Figure 1b) or between linc-mipep and linc-wrb (Extended Data Figure 1c) using Pearson correlation. The behavioral fingerprint of each larva can be conceptualized as a single datapoint in a multidimensional space where each dimension represents one behavioral parameter. To summarize the intensity of each phenotype across parameters, we measured the Euclidean distance between each larva and the mean fingerprint of its wild type siblings, set at the origin of this space by the Z-score normalization (Figure 1—figure supplement 3B). Generally, F0 mutants with more parameters affected, or with more extreme differences in the parameters affected, displayed larger Euclidian distances; those with few or with mildly-affected parameters displayed smaller Euclidean distances. Code for this analysis is available on GitHub (Kroll, 2022). Prism 9 (GraphPad) was used for statistics and plotting for Figure 1—figure supplement 4.

Correlation analysis was done in MATLAB (R2018a; The MathWorks) as previously described (Rihel et al., 2010). Behavioral phenotypes of wild-type fish exposed to a panel of 550 psychoactive agents from 4 to 7 dpf were ascertained as previously described (Rihel et al., 2010). To compare the behavioral fingerprints of WT larvae exposed to each drug and the linc-mipep mutant behavioral fingerprint, hierarchical clustering analysis was performed as in Rihel et al., 2010; Hoffman et al., 2016.

BLASTp, BLASTn, and the UCSC Genome Browser were used to find sequences (especially the highly conserved 3’UTR sequence) and proteins with sequence homology and/or synteny to human Hmgn1. Clustal Omega (through EMBL-EBI) was for multiple sequence alignments.

Three custom antibodies were designed (YenZym Antibodies, LLC) against: Si:ch73-1a9.3 (linc-mipep), C-Ahx-DDASATEDGDKKEDGE-COOH; Si:ch73-281n10.2 (linc-wrb), C-Ahx-EDAKPEAEEKTP-amide; and both Si:ch73-1a9.3 and Si:ch73-281n10.2: KRSKANNDAE-Ahx-amide. The last antibody designed to recognize both proteins was non-specific and not further used. Antibody specificity was confirmed by antibody staining in wild type and linc-mipep; linc-wrb mutants.

Embryos up to 24 hpf: Embryos were dechorionated and collected into room-temperature 4% PFA in PBS for 1 hr. Embryos were blocked rotating for 1 hr at room temperature in 10% normal goat serum (NGS) (Thermo Fisher Scientific, 50062Z), primary antibody stained for 1 hr at room temperature in 10% NGS, washed 3x5 min in 1xPBS with 0.25% Triton-X (PBST), incubated rotating and protected from light for 1 hr at room temperature, washed 3x5 min in PBST, and mounted in 0.7–1% low-melt agarose on glass-bottom dishes (MatTek) for imaging. Larvae: Larvae (up to 6 dpf) were maintained in a quiet environment. For assessment of olig2 + cells, the Tg(olig2:egfp)^vu12 line (Shin et al., 2003) was crossed to either wild type or double-homozygous linc-mipep;linc-wrb mutants. Subsequently, those olig2:egfp adults were outcrossed to either wild type or linc-mipep1;linc-wrb double homozygous mutants. To ensure rapid fixation at 6 dpf, larvae from each of these crosses were poured through a mesh sieve and immediately submerged into ice-cold 4% PFA (Electron Microscopy Sciences) /1 x PBS-0.25% Triton X-100 (PBST)/4% sucrose, in fix, as previously described (Randlett et al., 2015). Larvae were fixed overnight at 4 °C and washed three times for 15 min each in PBST. For whole larvae, pigment was bleached with a 1% H₂O₂/3% KOH solution (in PBS), washed 3x15 min in PBST, then permeabilized with acetone (pre-cooled to –20 °C) at –20 °C for 20 min, and washed three times for 15 min with PBST. For dissecting brains (critical for assessment of GFP + cells), following overnight fixation, larvae were washed 3x5 min in PBST, then brains were dissected by hand and transferred back into tubes with PBS. Brains were sequentially dehydrated 5 min each in 25% MeOH/75% PBS, 50% MeOH/50% PBS, 75% MeOH/50% PBS, and 100% MeOH, and stored at –20 °C for at least overnight. Brains were sequentially similarly rehydrated, then permeabilized with 1 x Proteinase K (10 mg/ml is 1000 x stock) in PBST for exactly 10 min. Brains were then rinsed 3 x with PBST, post-fixed in 4% PFA/PBST for 20 min at room temperature, and washed three times for 5 min in PBST. Samples were mounted in 0.7–1% low-melt agarose on glass-bottom dishes (MatTek) for imaging. Confocal imaging was performed using a Zeiss 980 AiryScan or a Leica SP8 confocal microscope. Images were processed and analyzed using FIJI software and plugins.

custom Linc-mipep (rabbit); custom Linc-wrb (rabbit); anti-GFP (mouse, A11120, Thermo Fisher Scientific, 1:500); acetylated α-tubulin (rabbit, 5335T, Cell Signaling Technology, 1:500). Alexa Fluor 488, 546 or 568 secondary antibodies against rabbit or mouse were used at 1:500 (Invitrogen). DAPI (for nuclear marking) was added at 1:10,000 during secondary antibody staining.

Template DNAs for antisense RNA probes were amplified from a pool of 6 hpf, 1 dpf, 2 dpf, and 5 dpf zebrafish cDNA using primers containing the T7-promoter sequence in the reverse primer. All sequences are listed in Supplementary file 1. Digoxigenin (DIG)-labeled RNA probes were synthesized using T7 RNA Polymerase (Roche) and purified using Monarch RNA Cleanup Kit (New England Biolabs). RNA in situ hybridization was performed as described (Giraldez et al., 2005; Thisse and Thisse, 2008). Briefly, embryos at the respective stages were dechorionated (if applicable) and fixed with 4% paraformaldehyde (PFA) overnight at 4 °C. Fixed embryos were washed 3 X with 1 x phosphate-buffered saline (PBS), then dehydrated with a methanol series (25%, 50%, 75%, and 100% methanol). Dehydrated embryos were stored in 100% methanol for at least 24 hr at –20 °C. Embryos were then rehydrated with a reverse methanol series and washed with 1 x PBS. Pre-hybridization and hybridization were performed at 65 °C for 3 hr and overnight, respectively. Embryos were washed extensively and blocked for 3 hr at room temperature, then incubated with anti-DIG antibody overnight at 4 °C. After antibody incubation, embryos were stained with BCIP/NBT, and staining was stopped with 4% PFA overnight at 4 °C. Embryos were then washed briefly, mounted with a glycerol series (50%, 70%, and 86%), and imaged in 86% glycerol with a Zeiss stereo Discovery.V12 microscope.

Data in Figure 1—figure supplement 5F was generated using publicly available RNA-sequencing data (White et al., 2017). For qPCR, larvae (n=10 per sample) were pooled and flash-frozen in liquid nitrogen and stored at –80 °C. Trizol (Invitrogen) was added to samples and homogenized with sterile pestles. Chloroform was then added, and samples were centrifuged at 4 °C for 15 min at 12,000 x g. The aqueous supernatant was placed into a new tube, and isopropanol was added along with 1 μl of GlycoBlue. Samples were left at –20 °C for 2 hours and centrifuged at 4 °C for 15 min at 12,000 x g. The pellet was washed two times with RNase-free 70% ice-cold ethanol, dried, and resuspended in RNase-free water. 1 μg of RNA was used to make cDNA with the SuperScript III First-Strand Synthesis system (Invitrogen). cDNA was diluted 1:3, and 1 μl was used for each qPCR sample using Power Sybr Green Master Mix (2 x) and respective primers, in technical triplicates. Primers for amplification: fosab (c-fos), 5′- GTGCAGCACGGCTTCACCGA-3′ and 5′- TTGAGCTGCGCCGTTGGAGG-3′; ef1a1l1, 5′-TGCTGTGCGTGACATGAGGCAG-3′ and 5′-CCGCAACCTTTGGAACGGTGT-3′ (Reichert et al., 2019). Expression of fosab (c-fos) was normalized to the expression of ef1a1l1 for each respective sample and timepoint, and relative expression levels were calculated using the ΔΔCt method.

Embryos from wild type or Tg(ubb:linc-mipep-FLAG-HA-T2A-mCherry) incrosses were dechorionated at 6 hpf, and 150 embryos were collected per sample per replicate. Water was removed, and embryos were deyolking in 500 μl Deyolking Buffer (55 mM NaCl, 1.8 mM KCl, 1.25 mM NaHCO3) by pipetting through a narrow tip to disrupt the yolk sac. Embryos were shaken at 1100 rpm for 5 min. Cells were then pelleted at 300 g for 30 s, and the supernatant was discarded. Two wash steps were performed using wash buffer (110 mM NaCl, 3.5 mM KCl, 2.7 mM CaCl2, 10 mM Tris/Cl pH 8.5), shaking two minutes at 1100 rpm and pelleting cells. The supernatant was then removed and samples were flash-frozen in liquid nitrogen. Cell pellets were then resuspended in 100 μl sample buffer (1 x NuPAGE LDS Sample Buffer supplemented with DTT). After heating for 10 min at 95 °C, protein samples (40 μl, ~60 deyolked embryos) were resolved on a 4–12% Bis-Tris gel with NuPAGE MOPS Running Buffer (Thermo Fisher Scientific) and transferred to a nitrocellulose membrane using the iBlot 2 Gel Transfer Device (Thermo Fisher Scientific). Membranes were blocked in 5% milk / PBS with 0.1% Tween-20 (PBST), incubated with primary antibody solution (each antibody at 1:2000) prepared in block solution, and then incubated with a peroxidase-conjugated secondary antibody solution prepared in block solution. Proteins were detected with SuperSignal West Pico PLUS Chemiluminescent Substrate (for Actin antibody) or SuperSignal West Femto Maximum Sensitivity Substrate (for FLAG antibody; Thermo Fisher Scientific).

At 4 dpf, single larvae from heterozygous linc-mipep mutant incrosses were placed into individual wells of a clear 96-square well flat plate (Whatman) filled with 650 μL of blue water. Respective pharmacological agents (from a stock of 5 or 50 mM depending on solubility) or corresponding vehicle controls (DMSO or water) were pipetted directly into the water to achieve the desired final concentrations at the start of the experiment (typically evening of 4 dpf). Since both linc-mipep and linc-wrb had similar hyperactivity profiles, we focused on linc-mipep to allow for drug analyses of mutant and wild type (WT) larvae with matched genetic backgrounds. Drug treatments, vehicles, and doses are described in Supplementary file 3.

After each behavioral tracking experiment, larvae were anesthetized with an overdose of MS-222 [0.2–0.3 mg/ml], transferred into 96-well PCR plates, and incubated in 50 μl of 100 mM NaOH at 95 °C for 20 min. Then, 25 μl of Tris-HCl 1 M pH 7.5 was added to neutralize the mix. Two μl of these crude DNA extracts were used for genotyping with the corresponding forward and reverse primers (10 µM; Supplementary file 1) using a standard PCR protocol.

Briefly, brains at peak daytime activity levels (Zeitgeber Time 4, i.e. 4 hr after lights on) were dissected from 5 dpf MZ-linc-mipep;linc-wrb or wild type zebrafish (for omni-ATAC-seq n=10 per sample, and ChIP-seq n=50 per sample) or 6 dpf zebrafish from one linc-mipep^-/- heterozygous incross (for single-cell Multiome, n=12 per sample) in ice-cold Neurobasal media supplemented with B-27 (Thermo Fisher Scientific), snap-frozen in a dry ice/methylbutane bath (to preserve nuclear structure), and stored at –80 °C until use. Trunks of linc-mipep fish from the heterozygous cross were genotyped, then wild type or linc-mipep^-/- brains as confirmed by genotyping were pooled together before proceeding with scMulitome.

For ChIP-seq experiments, brains were dissected and homogenized before treatment with 1% PFA (protocol adapted from Cotney and Noonan, 2015) and performed as previously described Miao et al., 2022 using 4 μg of RNA Polymerase II antibody (ab817, Abcam) per sample; 5% input samples were also collected and processed.

Omni-ATAC was performed on frozen brains from 5 dpf zebrafish based on published protocols (Buenrostro et al., 2013; Corces et al., 2017). Frozen brain tissue was homogenized in cold homogenization buffer (320 mM sucrose, 0.1 mM EDTA, 0.1% NP40, 5 mM CaCl2, 3 mM Mg(Ac)2, 10 mM Tris pH 7.8, 1×protease inhibitors (Roche, cOmplete), and 167 μM β-mercaptoethanol) on ice. The lysate was filtered with a tip strainer (Flowmi Cell Strainers, porosity 70 μm) into a new Lo-Bind tube. Nuclei were isolated using the gradient iodixanol solution as described (Corces et al., 2017). Nuclei solution was mixed with 1 ml of dilution buffer (10 mM Tris-HCl pH 7.4, 10 mM NaCl, 3 mM MgCl2, 0.1% Tween-20) and was then centrifuged at 500 x g for 10 min at 4°C. Transposition and library preparation were performed on the purified nuclei as described (Miao et al., 2022).

The supernatant was removed, and the purified nuclei were resuspended in the transposition reaction mixture (25 μl 2×TD Buffer, 2.5 μl Tn5 transposase, 22.5 μl Nuclease-Free water) and incubated for 30 min at 37 °C. DNA was then purified with the Qiagen MinElute Kit (Qiagen, 28004). Libraries were prepared using NEBNext High-Fidelity 2 X PCR Master Mix (NEB, M0541) with the following conditions: 72 °C, 5 min; 98 °C, 30 s; 15 cycles of 98 °C, 10 s; 63 °C, 30 s; and 72 °C, 1 min. Libraries were purified with Agencourt AMPureXP beads (Beckman Coulter Genomics, A63881) and sequenced with the Illumina NovaSeq 6000 System at the Yale Center for Genome Analysis.

LabxDB seq (Vejnar and Giraldez, 2020) was used to manage our high-throughput sequencing data and configure our analysis pipeline. Export to the Sequence Read Archive was achieved using the “export_sra.py” script from LabxDB Python. All sequencing datasets generated in this work have been deposited through NCBI, BioProject PRJNA945049. Detailed information about these datasets are also provided in Supplementary file 9.

Raw paired-end Omni-ATAC reads were mapped using LabxPipe (Vejnar, 2023b). Reads were adapter trimmed using ReadKnead (Vejnar, 2023d) and mapped to the zebrafish GRCz11 genome sequence Yates et al., 2020 using Bowtie2 (Langmead and Salzberg, 2012) with parameters ‘-X 2000, --no-unal, "--no-unal", "--no-mixed", "--no-discordant". The alignments were deduplicated using samtools markdup (Li et al., 2009). For genome-wide analysis, only uniquely mapped reads (with alignment quality ≥30) were used. Reads mapped to the + strand were offset by +4 bp and reads mapped to the – strand were offset by −5 bp (Buenrostro et al., 2013). Only fragments with insert size ≤ 100 bp (effective fragments) were used to determine accessible regions. Genome tracks were created using BEDTools (Quinlan and Hall, 2010) and utilities from the UCSC genome browser (Lee et al., 2022). For all the genome tracks in the paper, signal intensity was in RPM (reads per million). Fragment coverage on each nucleotide was normalized to the total number of effective fragments in each sample per million fragments.

Effective reads from three linc-mipep;linc-wrb double mutant replicates and three wild-type replicates were merged. Then narrow peaks were called on the merged data using MACS3 (Zhang et al., 2008) with the additional parameters ‘-f BEDPE --nomodel --keep-dup all’ with significance cutoff at p=10⁻²⁰. In total, 173,443 narrow peaks were called. Among them, 170,599 peaks were located within chromosomes 1–25; these regions were determined as accessible regions for further analysis. Differential analysis was performed using DESeq2 (Love et al., 2014), comparing fragment coverage of each accessible region in the three linc-mipep;linc-wrb double mutant replicates with that in the three wild-type replicates. A total of 3367 regions that were mapped to chromosomes 1–25 show a significant difference (false discovery rate (FDR)<0.01), with 2167 regions significantly up-regulated and 1200 regions significantly down-regulated. A total of 2928 unaffected regions (FDR >0.95; 1.005<linc-mipep;linc-wrb / WT <0.995) were used as control regions for plotting and motif enrichment analysis. Accessibility heatmaps and density plots were generated using deeptools (Ramírez et al., 2014).

This was performed on the up-regulated and the down-regulated regions, with unaffected regions as control, using AME in MEME suite (McLeay and Bailey, 2010) with default parameters (https://meme-suite.org/meme/tools/ame, motif database option: Vertebrates In vivo and in silico, Eukaryotic DNA). Motif heatmaps were generated using the R package gplots (Warnes et al., 2022). Tracks for omni-ATAC-seq of wild type or linc-mipep; linc-wrb mutant brains are publicly available at https://www.giraldezlab.org/data/tornini_et_al_2023_elife/.

Raw ChIP-seq reads were adapter trimmed, mapped, and deduplicated using the same method described in the previous section but using the default parameters for Bowtie2 for read mapping. GeneAbacus (all code available at Vejnar, 2023c) was used to create genomic profiles for creating tracks. Fragment coverage on each nucleotide was normalized to the total fragments in each sample per million fragments. For genome-wide analysis, only uniquely mapped reads (with alignment quality ≥30) were used.

Peaks were called using MACS2 Zhang et al., 2008 for ChIP-seq data. Narrow peaks were called using MACS2 with the additional parameters ‘-f BEDPE --nomodel --keep-dup all’ with the default significance cut-off (q=0.05, high threshold) and p=0.05 (low threshold). Peaks that are called at high threshold in one condition but not called at low threshold in the other condition are defined to be specific to the condition. Genes with promoter regions (+/-1 kb of transcription start site) that overlap with a peak are defined to be associated with that peak. Tracks for PolII ChIP-seq at 5dpf of wild type or linc-mipep; linc-wrb mutant brains are publicly available at https://www.giraldezlab.org/data/tornini_et_al_2023_elife/.

Flash-frozen pooled brains were prepared based on Protocol CG000366 – Rev D (Protocol 2) from 10 x Genomics (available here). It is critical to keep samples cold and/or on ice for all steps. Briefly, all samples were processed identically and simultaneously to minimize batch effects. Chilled 0.1 X Lysis Buffer (500 μl) was immediately added to frozen samples, and samples were homogenized using a glass dounce tissue grinder with glass pestle. Samples were incubated on ice for 5 min, gently pipetted 10 x, then incubated again for 5 min. Chilled Wash Buffer (500 μl) was gently added to samples. After pipetting the mix 5 x, the samples were passed through 70μm-porosity Flowmi tips into new ice-cold low-bind 1.5 ml tubes. Each suspension was subsequently passed through a 40μm-porosity Flowmi tip into a new ice-cold low-bind 1.5 ml tube. Samples were centrifuged at 500 rcf 5 min at 4 °C. The supernatant was gently removed without disturbing the nuclei pellet. Chilled Wash Buffer (1 ml) was added, and the nuclei were gently resuspended 5 x. This wash and resuspension step was repeated one more time. On the final step, nuclei were resuspended in Diluted Nuclei Buffer. Quality and number of nuclei (as assessed by >90% Trypan Blue staining and almost no cell clumps) for each sample was assessed using a hemocytometer and were immediately used for tagmentation step using the 10 x Genomics platform. Library preparation was performed following the standard 10 x Genomics protocol (available here).

Single nuclei from brains of wild type or linc-mipep mutant siblings were collected as described above (Brain collection for molecular analyses). The raw 10 x Genomics Multiome data of scRNA-seq and scATAC-seq were processed using the 10 x Genomics cellranger-arc pipeline (v1.0.1) with the genome, GRCz11. The total numbers of sequenced read pairs per sample for RNA and ATAC were between 197,900,000 and 268,400,000. The estimated numbers of cells for WT and mutant were 7,137 and 7,872, respectively. The mean numbers of raw read pairs per cell were (1) 27,742.56 for RNA and 37,593.97 for ATAC in WT and (2) 26,154.78 for RNA and 27,382.86 for ATAC in mutant. The median numbers of genes per cell for WT and mutant were 349 and 365, respectively. ATAC median high-quality fragments per cell for WT and mutant were 10,466 and 8,626, respectively.

For downstream analyses, we used the Weighted Nearest Neighbor (WNN) method in Seurat (Hao et al., 2021). The two experimental conditions of WT and mutant were first analyzed separately. Data filtering was based on visual inspection of data distributions. The number of RNA read counts per cell was filtered between 50 and 3000 for WT and between 50 and 5000 for mutant. The number of ATAC read counts per cell was filtered between 500 and 50,000 for WT and between 500 and 80,000 for mutant. The filtering threshold for mitochondrial fractions was 15% for both WT and mutant data. Other parameters were left to default values in Seurat (v4.0.2). The numbers of filtered cells in WT and mutant were 6942 and 7740, respectively. The numbers of filtered ATAC peaks in WT and mutant were 164,266 and 167,925, respectively. We then followed the standard Seurat pipelines, with default parameters, for RNA analysis (SCTransform and PCA) and ATAC analysis (TFIDF and SVD) to obtain a WNN graph as a weighted combination of RNA and ATAC data for each of WT and mutant data. Dimensionality reduction was done by UMAP, clustering by the shared nearest neighbor and smart local moving algorithms, and differential marker identification by Wilcoxon rank sum tests. For analyses of variation in chromatin accessibility and enriched motifs, we used chromVAR (Schep et al., 2017) and all motifs from the Fornes et al., 2020 database. We also performed a merged analysis of the two conditions in a similar way by merging the two datasets using the merge function in Seurat. We did not make any correction for batch effects because the two conditions did not show any distinct batch effects on UMAP plots of the merged data. Cell states, or types, were identified by cross-referencing with known markers on ZFIN and 5 dpf datasets from Raj et al., 2020.

For identification of condition-specific significant ATAC peaks in each cluster, intensity distributions of each peak in WT and mutant were statistically analyzed by the Wilcoxon rank sum and the Kolmogorov-Smirnov (KS) methods using one-tailed tests for each condition. Based on manual inspection of p-value distributions of all peaks, we chose raw p-value thresholds of 0.001 and 0.01 for the Wilcoxon and the KS tests, respectively, to deem peaks to be significant. No p-value correction was performed at this filtering step as a strategy of choice. Those significant peaks were further analyzed to identify enriched motifs as described above. In addition, for those clusters of interest, Clusters 8, 35, 38, 39, and 42, we performed a simulation for the number of significant peaks in each cluster by generating 1000 random peak intensity datasets by shuffling the intensity values between WT and mutant as many as the number of cells in the cluster in question. This simulation provided empirical null distributions of the number of significant peaks to obtain p-values. R code for data processing and analyses is available on GitHub (Lee, 2023).

The cells included after filtering from the Seurat analysis were used to perform integrated diffusion and MELD to keep the analyzed dataset consistent. These new techniques were implemented to analyze the data from a different approach. Integrated diffusion was used to combine multimodal datasets, specifically each cell’s RNA-seq and ATAC-seq data, to create a joint data diffusion operator. The 3D integrated PHATE was computed on this joint data diffusion operator as described previously (Kuchroo et al., 2022; Kuchroo et al., 2021). To color the plots by likelihood of a cell belonging to the wildtype or mutant sample, this integrated diffusion operator was used for MELD, outputting the likelihood score for each cell belonging to a wildtype or mutant sample. The notebook for this analysis is available on GitHub (Du, 2023).