To gain an unbiased view of the molecular changes resulting from dscaml1 deficiency, we compared the transcriptomic profiles between dscaml1 homozygous mutant (dscaml1−/−) and control (wild type) siblings using RNA sequencing (RNA-seq). cDNA from whole 3.5–4 days post-fertilization (dpf) dscaml1−/− and control larvae were sequenced using an Illumina next-generation sequencer. Using a threshold of at least two-fold change and an adjusted p-value of less than 0.01, we identified 25 upregulated and 79 downregulated genes (Figure 1A, Supplementary Datasheet S1).

Among the 25 upregulated genes in the dscaml1−/− animals, seven were secreted neuropeptides/hormones expressed in the hypothalamus or pituitary: corticotropin-releasing hormone b (crhb), parathyroid hormone 2 (pth2), somatolactin beta (smtlb), cocaine- and amphetamine-regulated transcript 3 (cart3), proopiomelanocortin a (pomca), arginine vasopressin (avp), and spexin hormone (spx). Among them, three (crhb, avp, pomca) are core regulators of the HPI axis (Alsop and Vijayan, 2009; Lohr and Hammerschmidt, 2011). crhb encodes the zebrafish CRH; avp encodes the neuropeptide AVP that controls osmolarity, blood pressure, and synergizes with CRH to promote cortisol release; pomca encodes the adrenocorticotropic hormone (ACTH), the primary pituitary hormone that triggers glucocorticoid release. Additionally, two genes involved in apoptosis are upregulated: BCL2 apoptosis regulator b (bcl2b) and phorbol-12-myristate-13-acetate-induced protein 1 (pmaip1/noxa). Protein class categorization analysis with PANTHER revealed “intercellular signal molecule” as the largest class (4 genes, 16%) (Figure 1B) (Mi et al., 2021).

The 79 genes downregulated in the dscaml1−/− animals are more diverse in function compared to the upregulated genes. Protein class categorization analysis with PANTHER identified the largest protein classes as metabolite interconversion enzymes (12 genes, 15.19%), protein-binding activity modulators (9 genes, 11.39%), transporters (8 genes, 10.13%), and defense/immunity protein (6 genes, 7.59%) (Figure 1C). We noted that 30 of the 79 (38%) downregulated genes are highly expressed in the liver, and 10 of these genes are involved in innate immunity and the complement cascade (Figure 1A, Supplementary Datasheet S2). These results suggest that liver function and innate immunity may be suppressed in the dscaml1 mutants.

To identify the signaling pathways affected by dscaml1 deficiency, we analyzed all differentially expressed genes (p < 0.01, 210 mapped genes) using the statistical enrichment test (PANTHER Classification System, version 17.0) (Mi et al., 2021). All significantly enriched (FDR < 0.05) Gene Ontogeny (GO) terms for molecular function relate to the parent GO term hormone activity (Figure 1D). The PANTHER pathway analysis also identified another significant stress-modulating neuropeptide, adenylate cyclase activating polypeptide 1b (adcyap1b, also known as PACAP) that is significantly upregulated (0.66 fold change, adjusted p < 0.0001) (Stroth et al., 2011).

Together, our transcriptomic analyses indicate that dscaml1 deficiency results in the upregulation of neuropeptide/hormonal signaling and the downregulation of liver and innate immune function. Suppression of liver and immune function is a hallmark of stress axis activation, which is consistent with the overexpression of the principal neuropeptides involved in the stress axis (crhb, avp, pomca, and adcyap1b).

To further investigate whether the stress axis is perturbed in dscaml1 mutants, we first examined the development of CRH neurons in the NPO (CRH^NPO neurons) (Figure 2A), which are localized in the dorsal preoptic area and characterized by their expression of crhb (Herget et al., 2014; Vom Berg-Maurer et al., 2016). We focused on three developmental stages (2, 3, and 5 dpf) that span the period between the first appearance of crhb + neurons in the NPO (2 dpf) and the onset of stress axis responsivity (4–5 dpf) (Chandrasekar et al., 2007; Alderman and Bernier, 2009; Clark et al., 2011). Sibling controls were used for all comparisons.

Using fluorescent in situ hybridization (FISH), we found that crhb expression pattern was initially similar between dscaml1−/− and wild-type (WT) animals at 2 dpf (Figure 2B, B'). At 3 dpf, we began to see higher crhb expression in the NPO of dscaml1−/− animals (Figure 2C, C'), and crhb expression in the NPO remains higher in dscaml1−/− animals at 5 dpf (Figure 2D, D”). Quantification of crhb FISH signal intensity among CRH^NPO neurons showed that crhb expression is higher in dscaml1−/− animals, compared to WT animals (Figure 2E). There was a significant difference in signal intensity per cell by developmental stage and genotype, and a significant interaction between stage and genotype (two-way ANOVA, Supplementary Table SI). Pair-wise comparisons with Holm-Sidak correction found a significant increase at 3 and 5 dpf but not at 2 dpf (Figure 2E).

In addition to changes in crhb expression levels, dscaml1 deficiency also increased the number of crhb-expressing CRH^NPO neurons. In WT animals, the number of CRH^NPO neurons increased over time, from 14.78 cells (2 dpf) to 18.85 cells (3 dpf) to 26.60 cells (5 dpf) per animal (Figures 2B–D, F). In dscaml1−/− animals, the number of CRH^NPO neurons increased at a higher rate, from 12.67 cells (2 dpf) to 19.46 cells (3 dpf) to 36.39 cells (5 dpf) per animal (Figure 2B', C', D', F). There was a significant difference in cell number by developmental stage and genotype, and there was a significant interaction between stage and genotype (Two-way ANOVA, Supplementary Table SI). Pair-wise comparisons with Holm-Sidak correction found a significant increase in cell number between dscaml1−/− and WT animals at 5 dpf but not at 2 and 3 dpf (Figure 2F, adjusted p values as shown).

Overall, we found significant increases in crhb expression and cell number in CRH^NPO neurons in dscaml1−/− mutants as compared to WT. These phenotypes were not due to the visual deficits in dscaml1−/− animals (Ma et al., 2020b), as similar phenotypes were observed in animals raised in the dark (Supplementary Figure S1). Together, these findings show that dscaml1 is essential for the normal developmental trajectory of CRH^NPO neurons.

To test whether other hypothalamic CRH neurons are affected similarly to CRH^NPO neurons, we examined two additional crhb-expressing hypothalamic regions. In zebrafish, the intermediate hypothalamus (IH) CRH neurons (CRH^IH neurons) regulate stress response and the perception of the emotional valence of light (Figure 3A) (Wagle et al., 2022). In the IH at 5 dpf, we did not observe any significant difference in crhb expression levels (FISH intensity) between dscaml1−/− and WT animals (Figures 3B–D). However, the number of CRH^IH neurons was significantly higher in dscaml1−/− animals (unpaired t-test, Figure 3E).

In addition to the IH, we have found consistent crhb expression in the rostral hypothalamus (RH) (Figure 3F). The roles of CRH neurons in the RH (CRH^RH neurons) are still unknown, but they may contribute to the rostral hypothalamus-optic tectum pathway (Heap et al., 2017). In the RH at 5 dpf, both the crhb expression levels and the CRH^RH cell number were higher in dscaml1−/− animals compared to WT (unpaired t-test, Figures 3G–J). Thus, dscaml1 is required broadly to regulate CRH neuron cell number in the developing hypothalamus, but dscaml1 deficiency only affects crhb expression in the RH and NPO.

DSCAM family proteins are known to reduce neuronal cell number by promoting PCD in the mammalian retina (Fuerst et al., 2012; Garrett et al., 2016). Additionally, our transcriptomic analysis showed that dscaml1 mutants express higher levels of apoptosis-regulating genes, bcl2b and pmaip1 (Figure 1A). Therefore, we hypothesized that dscaml1 deficiency might impair PCD in the developing hypothalamus.

To test this, we used an anti-activated caspase 3 (AC3) antibody to label apoptotic cells (Pyati et al., 2011). In wild-type animals, we saw widespread apoptosis in the nervous system, peaking at 3 dpf (Figure 4A–A”), consistent with previous findings using TUNEL (Cole and Ross, 2001). We found a decrease in AC3-positive cells in dscaml1−/− animals at 3 and 5 dpf, compared to wild-type sibling controls (Figure 4B–B”). To quantify the amount of cell death in specific hypothalamic regions, we used the Map-map calculation and the Z-Brain anatomical atlas (Randlett et al., 2015). In this method, individual confocal image stacks are registered to the Z-Brain reference brain; registered stacks are then grouped by genotype for voxel-wise comparisons using the Mann-Whitney U-statistic (WT n = 14; dscaml1−/− n = 21). The voxels with significant Z-scores, either higher in WT (WT over dscaml1) or higher in dscaml1−/− (dscaml1 over WT), were mapped to regions of interest (ROIs) and the mean voxel values in each ROI were calculated and displayed using a green-magenta color scale (green: higher in WT; magenta: higher in dscaml1−/−) (Figure 4C).

Overall, AC3 signal is higher in WT throughout the brain, compared to dscaml1−/− (i.e., many green regions and very few magenta ROIs). Out of 293 ROIs annotated in Z-Brain, 178 had voxels with significant WT over dscaml1 signal (mean signal = 1,384.90) (WT over dscaml1 sheet, Supplementary Datasheet S3). In contrast, only 65 ROIs had voxels with significant dscaml1 over WT signal (mean signal = 84.23) (dscaml1 over WT sheet, Supplementary Datasheet S3).

In the ROIs corresponding to the hypothalamic crhb expressing areas, the NPO (Z-Brain ROI: Diencephalon—Otpb Cluster 2) (WT over dscaml1 signal = 5,688.06; dscaml1 over WT signal = 0), IH (Diencephalon - Intermediate Hypothalamus) (WT over dscaml1 signal = 673.56; dscaml1 over WT signal = 36.40), and RH (Diencephalon - Rostral Hypothalamus) (WT over dscaml1 signal = 916.95; dscaml1 over WT signal = 0.70) all had higher AC3 signal in WT, compared to dscaml1−/− (Figures 4D–F). Together, these results suggest that reduced PCD in dscaml1−/− animals may contribute to the higher number of CRH neurons in the hypothalamus.

To further determine whether dscaml1 affects the survival of hypothalamic CRH neurons, we tracked the fate of individual CRH^NPO neurons using in vivo time-lapse imaging at 3 to 5 dpf, when cell number begins to diverge between dscaml1−/− and WT animals. To visualize CRH^NPO neurons in live zebrafish, we generated a crhb knock-in fluorescent reporter line using CRISPR-mediated genomic insertion (Figure 5A) (Kimura et al., 2014). A Cre-switchable hsp-LoxP-RFP-LoxP-GFP cassette was inserted 35 base pairs upstream of the first exon of crhb so that the expression of RFP (default) or GFP (with Cre-mediate recombination) would mark the endogenous crhb-expressing cells (Figure 5B). The resulting transgenic line, crhb:LoxP-RFP-LoxP-GFP (crhb:LRLG), has RFP and GFP expression pattern that matches the endogenous crhb transcript expression patterns reported by others (Figures 5C, D) (Shainer et al., 2022). To validate the fidelity of fluorescent reporter expression, we examined whether crhb:LRLG-labeled cells in the NPO express CRH protein. In animals not exposed to Cre (default RFP expression), we found that most RFP+ neurons are CRH immunopositive (84.71% ± 2.19%, Supplementary Figure S2A–B). Together, these results indicate that the crhb:LRLG line reliably labels CRH^NPO neurons.

In crhb:LRLG animals, dscaml1 deficiency increased the number of RFP-labeled CRH^NPO neurons (Figure 5E–F'). There were significant differences by developmental stage and genotype, with no significant interaction (two-way ANOVA, Supplementary Table SI). There was a significantly higher number of RFP-positive neurons in the dscaml1 mutants at 5 dpf but not 3 dpf (Figure 5G, multiple comparison test with Holm-Sidak correction). This result corroborates our crhb FISH results that show increased CRH^NPO neuron number in dscaml1 mutants (Figure 2F).

Using the crhb:LRLG line, we first performed time-lapse imaging in anesthetized animals. We induced partial Cre-mediated recombination by injecting CreER mRNA into crhb:LRLG animals at the 1-cell stage and adding 4-hydroxytamoxifen (4-OHT) to activate CreER at 6–24 hpf (Figure 5B). At 3–4 dpf, intermingled GFP+ and RFP+ cells can be seen in the NPO by confocal imaging (Figure 5H–H” and Supplementary Video S1). Over time, some labeled cells moved away from the CRH^NPO neuron cluster. These cells were likely dying cells being carried away by microglia, as seen in previous zebrafish studies (Mazaheri et al., 2014). To confirm this, we induced GFP expression (by injecting codon-optimized Cre mRNA) (Horstick et al., 2015) in all crhb:LRLG-labeled cells and labeled microglia with the mpeg1:mCherry transgene (Figure 5I–I” and Supplementary Video S2) (Espenschied et al., 2019). Indeed, mCherry+ microglia migrated toward the GFP+ cell cluster, engulfed GFP+ CRH^NPO neurons, and carried the engulfed cells away. The remnant of the engulfed cell can be seen inside a large vacuole within the microglia (Figure 5J, J'). This type of engulfment events was seen in three out of five animals between 3.5-4 dpf. These results suggest that CRH^NPO neurons undergo PCD and that dying cells are rapidly removed by microglia.

Next, to track the fate of individual CRH^NPO neurons, we performed two-photon imaging from 3 to 5 dpf on crhb:LRLG animals with partial Cre-mediated recombination (Figure 6A). To minimize the potential effects of stress on PCD (Irles et al., 2014), we anesthetized and immobilized the animals during imaging. In between imaging sessions, each animal is allowed to recover in individual wells of 12-well plates, under normal light-dark cycles. Cells that are present at the first time point (78 hpf) were tracked at four subsequent time points (84, 96, 108, and 120 hpf) and categorized as either persisting (present at the last time point) or lost (lost at any of the following time points) (Figure 6B–B”). Overall, we observe a trend of reduced cell loss in the dscaml1+/− and dscaml1−/− animals (Chi-square test for trend, p = 0.0398) (Figure 6C, total number of tracked cells as indicated). In WT animals, 16.41% of cells are lost, versus 10.91% and 7.53% for dscaml1+/− and dscaml1−/−, respectively.

Finally, considering the timing of cell loss, we plotted the survival curve of CRH^NPO neurons for each genotype. Again, there was a significant difference in the trends of cell loss (Logrank test for trend, p = 0.0098), with the dscaml1−/− animals consistently showing a higher survival rate (Figure 6D). Together, these results show that dscaml1 deficiency reduces PCD of CRH^NPO neurons.

Given the developmental perturbation of hypothalamic CRH neurons as well as the global changes in gene expression related to stress axis activation, we next determined whether the hormonal output of the stress axis—cortisol—is altered. Cortisol levels were measured using an enzyme-linked immunosorbent assay (ELISA) on homogenates made from pools of 5 dpf animals (30 animals per sample, 6 samples per condition). All animals were raised under standardized conditions, at the same density, and with a normal circadian cycle (14 h day/10 h night) (Yeh et al., 2013). Under this circadian cycle, dscaml1 mutant animals exhibit similar diurnal locomotor rhythms as wild-type animals (Ma et al., 2020b).

Baseline and stressed conditions were tested to evaluate potential alterations of cortisol in control (WT and dscaml1+/−) and dscaml1−/− siblings at 5 dpf. To measure baseline cortisol, we collected unperturbed animals within 30 min after light onset (zeitgeber time 0, ZT0) and in the afternoon (ZT6-8). To measure stress-induced cortisol, animals were exposed to either stirring stress (swirling water, 5 min) (Castillo-Ramirez et al., 2019) or hyperosmotic stress (250 mM NaCl, 20 min) at ZT6-8 (Yeh et al., 2013). These acute stressors are well-characterized and are comparable to the water current and salinity changes experienced by zebrafish larvae in their natural habitat (Clark et al., 2011).

At baseline, dscaml1−/− animals had significantly higher cortisol levels than control animals (Multiple Mann-Whitney test with Holm-Sidak correction. Adjusted p values shown in Figure 7A). At ZT0, the dscaml1 mutant baseline cortisol levels were 2.7-fold higher than that of controls (median 350.06 pg/mL versus 129.80 pg/mL). The difference in cortisol was less pronounced at ZT6-8, but dscaml1 mutants still had 2-fold higher cortisol levels than controls at baseline (median 238.51 pg/mL versus 118.52 pg/mL).

After acute exposure to stressors, we found that dscaml1 mutant animals exhibited attenuated cortisol induction. We assessed the extent of stress-induced cortisol production by normalizing cortisol levels to the baseline cortisol of the same genotype at the same circadian time (ZT6-8). In the control group, stirring stress and hyperosmotic stress produced 1.88 and 1.95-fold increases in cortisol over the control baseline, respectively (grey bars, Figure 7B) (non-normalized cortisol data are shown in Supplementary Figure S3). The response to hyperosmotic stress was more robust (p = 0.0113, Multiple Mann-Whitney tests with Holm-Sidak correction) than that generated by stirring stress (p = 0.0546). In the dscaml1−/− group, stirring stress and hyperosmotic stress only produced 1.29 and 1.4-fold increases in cortisol over the dscaml1−/− baseline, respectively (red bars, Figure 7B). These increases were not statistically significant (p > 0.9999 for stirring stress, p = 0.9768 for hyperosmotic stress).

Together, these results show that dscaml1 deficiency elevates baseline cortisol levels and impair responses to acute stressors. These findings indicate that dscaml1 is critical for establishing the normal function of the stress axis.

Some aspects of the stress axis-related phenotypes in dscaml1 mutants resemble the zebrafish glucocorticoid receptor (gr) mutants. In particular, both gr and dscaml1 mutants exhibit elevated baseline cortisol and crhb (Ziv et al., 2013). This resemblance raises the possibility that the dscaml1 mutant phenotypes may result from insufficient glucocorticoid receptor-mediated signaling. To test this, we examined whether dscaml1 mutants can respond transcriptionally to exogenously applied glucocorticoids. We examined the expression of crhb in the NPO, as it is under feedback control from glucocorticoid-glucocorticoid receptor signaling (Watts, 2005). Wild-type siblings were used for controls.

A synthetic glucocorticoid receptor agonist, dexamethasone (Dex), was added to the embryo media at a final concentration of 2 μM from 4 to 5 dpf (24 h). Vehicle (0.02% ethanol) treated siblings were used for comparison. For crhb transcript level in CRH^NPO neurons, we found significant differences caused by Dex treatment and genotype, with significant interaction between the two (two-way ANOVA, Supplementary Table SI). Multiple comparison tests found that Dex significantly reduced crhb transcript level in dscaml1−/− animals but not in wild-type animals (Holm-Sidak correction, Figure 7C with p values as indicated). These results show that dscaml1 deficiency does not result in a loss of glucocorticoid responsivity in CRH^NPO neurons. Instead, dscaml1 deficiency may render animals more sensitive to the inhibitory effects of glucocorticoids.

A major finding of this study is that DSCAML1 is necessary for the development of hypothalamic CRH neurons. To our knowledge, DSCAML1 is the first intercellular signaling molecule to be implicated in CRH neuron development. Our finding also provides the first link between DSCAML1 and hypothalamus development. We show that, similar to retinal neurons, the regulation of cell number by PCD is a DSCAML1-mediated process in hypothalamic CRH neurons (Garrett et al., 2016). Further investigations are required to determine whether other aspects of DSCAML1 function, such as the regulation of cellular spacing and synaptogenesis, are involved in CRH neuron development (Fuerst et al., 2009; Yamagata and Sanes, 2010; Sachse et al., 2019). Additionally, given that dscaml1 is expressed broadly in the nervous system (Ma et al., 2020b), including CRH neurons and non-CRH neurons (Supplementary Figure S4), it will be important to address whether DSCAML1 acts cell-autonomously in CRH^NPO neurons.

During development, PCD is critical for eliminating transient cell types, matching input and output cell populations, and maintaining cellular spacing (Yamaguchi and Miura, 2015; Wong and Marin, 2019). It has been hypothesized that PCD in the hypothalamus may specify neural circuit assembly and shape output activity (Simerly, 2002; Forger, 2009). Congruent with this idea, reduced hypothalamic PCD caused by early-life stress is associated with increased sensitivity to acute stressors in adulthood (Zhang et al., 2012; Irles et al., 2014).

As an intercellular signaling molecule, DSCAML1 may act to transduce extracellular cues for PCD. One potential cue is synaptic activity, which is critical for neuronal survival during development (Wong and Marin, 2019). In culture, DSCAML1 is localized to excitatory synapses (Yamagata and Sanes, 2010) and inhibits excitatory synaptogenesis when overexpressed (Sachse et al., 2019). It remains to be determined whether DSCAML1 deficiency increases excitatory synaptic transmission and activates activity-dependent cell survival pathways in CRH neurons (Wong and Marin, 2019).

It is worth noting that CRH neuron cell death may not be the sole mechanism by which dscaml1 affects CRH neuron number. Changes in the number or proliferative capacity of neuroprogenitor cells may contribute to the increase CRH neuron number we observed in dscaml1 mutants. Additionally, the higher crhb expression induced by dscaml1 deficiency may lead to more cells being identified as CRH neurons. Further work is needed to determine how these factors control CRH neuron number during development.

dscaml1−/− animals exhibit multiple signs of stress axis activation at baseline, including cortisol elevation and the suppression of immunity-associated genes. The elevated baseline cortisol levels likely resulted in attenuated responses to acute stressors, similar to animals under chronic cortisol administration (Barton et al., 1987; Johnson et al., 2006). A likely cause of these phenotypes is the overexpression of crhb. In mice, broad overexpression of CRH leads to elevated corticosterone (the stress glucocorticoid in rodents) and produces phenotypes similar to Cushing’s syndrome, a human disorder caused by the overproduction of cortisol (Stenzel-Poore et al., 1992; Arnett et al., 2016).

Beyond dysfunction of hypothalamic neurons and CRH signaling, a broader neurological imbalance can also activate the stress axis. For example, seizures have been shown to activate the HPA axis, which increases the likelihood of future seizures (O'Toole et al., 2014; Hooper et al., 2018). It has been reported that Dscaml1 mutant rats and human patients with DSCAML1 loss-of-function variants exhibit neuronal hyperactivation and seizures (Hayase et al., 2020). It is, therefore, possible that excitation-inhibition imbalance in extra-hypothalamic regions may result in increased hypothalamic CRH neuron firing in zebrafish dscaml1 mutants. Further studies on the cell-type-specific functions of dscaml1 are needed to understand the precise cause of stress axis hyperactivation in dscaml1−/− animals.

Facilitation and feedback are signature features of the stress axis (Dallman et al., 1992; Spencer and Deak, 2017). While CRH neurons control cortisol levels, cortisol signaling also affects the development of CRH neurons. Zebrafish gr mutants have phenotypes similar to that of dscaml1−/− animals, including slow visual-background adaptation, sluggish light onset response, elevated cortisol, and increased expression of crhb (Griffiths et al., 2012; Muto et al., 2013; Ziv et al., 2013). Surprisingly, rather than decreased glucocorticoid receptor signaling (as in gr mutants), dscaml1−/− animals have intact glucocorticoid receptor signaling, with dexamethasone exerting strong suppression of crhb expression in the NPO. Thus, despite the superficial phenotypic similarity, the underlying signaling mechanisms are distinct between dscaml1 and gr mutants.

Given the known roles of DSCAML1 in regulating synapse formation (Sachse et al., 2019), the increase in crhb expression may be caused by the hyper-activation of upstream stress-related neural inputs. In rodents, various paradigms of chronic stress usually result in an upregulation of CRH in the PVN (Herman and Tasker, 2016). The transcriptional activation caused by stress-related synaptic activity may overcome the negative feedback mediated by glucocorticoid receptor signaling. Additional work is needed to disambiguate the relationship between cortisol disturbances and developmental deficits. A possible approach would be to normalize cortisol levels by genetically ablating interrenal cells (i.e., genetic adrenalectomy) and supplementing with constant levels of exogenous cortisol (Gutierrez-Triana et al., 2015).

Zebrafish (all ages) were raised under a 14/10 light/dark cycle at 28.5°C. Embryos and larvae were raised in E3 buffer (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, 0.33 mM MgSO₄) (Nüsslein-Volhard et al., 2002). All zebrafish used in this study were in a mixed background of AB and TL wild-type strains (Zebrafish International Resource Center). Sex was not a relevant variable for the stages used in this study (0–6 dpf), as laboratory zebrafish remain sexually undifferentiated until 2 weeks of age (Maack and Segner, 2003; Wilson et al., 2014).

The dscaml1 ^vt1 loss-of-function allele contains a seven base pair deletion that results in frame shift and premature stop codon (Ma et al., 2020b). Animals used for live imaging were in homozygous nacre (mitfa) mutant background to prevent pigment formation (Lister et al., 1999). The microglia RFP line [Tg(mpeg1:Gal4;UAS:NTR-mCherry)] was obtained from Dr. John Rawls at Duke University (Espenschied et al., 2019). The crhb:LoxP-RFP-LoxP-GFP line was generated using CRISPR-mediated knock-in, as described by Kimura et al. (Kimura et al., 2014). The sgRNA sequence for the crhb knock-in locus is AGC​TCG​CGT​CTG​CGC​AGA​G. All group-wise comparisons (mutants versus controls) were done between siblings.

Progenies from heterozygous dscaml1 mutant parents were anesthetized and harvested at 3.5-4 dpf. The anterior half of the animal was used for RNA preparation using the RNA Miniprep Kit (Zymo). The posterior half was used for genotyping. Three biological replicates for each group were analyzed, each containing RNA from 6–11 animals. All samples had RIN ≥8.0 and were converted into a strand-specific library using Illumina’s TruSeq Stranded mRNA HT Sample Prep Kit (RS-122–2,103; Illumina) for subsequent cluster generation and sequencing on Illumina’s NextSeq 75 sequencer. Sequence data processing, alignment, read count, mapping, and quality control were performed as previously described (Ates et al., 2020). Differential expression was tested for significance using the false discovery rate (FDR) (Benjamini–Hochberg) corrected Likelihood Ratio Test (LRT) in the R-package DESeq2 (Love et al., 2014). 238 and 116 genes showed a significant difference in read counts at FDR < 0.01 and 0.001, respectively. Original sequence data have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al., 2002) and accessible through GEO Series accession number GSE213858.

Single and double whole-mount fluorescent in situ hybridization (FISH) was performed using protocols described previously (Pan et al., 2012). Probes were synthesized by in vitro transcription using the DIG and Fluorescein RNA Labeling Mix (Roche). DIG and fluorescein-labeled probes were detected with anti-DIG or anti-Fluorescein POD-conjugated Fab fragments (Roche) and Cy3 or Fluorescein TSA-plus Reagent (Akoya Biosciences). Plasmid template for crhb (Lohr et al., 2009) was provided by Dr. David Prober at Caltech. The dscaml1 probe was generated as described previously (Ma et al., 2020b).

Immunohistochemistry was performed as described previously (Ma et al., 2020a). ERK1/2 was stained with mouse anti-ERK1/2 (4696S; Cell Signaling Technology). Activated caspase three was stained with rabbit anti-cleaved caspase 3 (559565; BD Biosciences). Nuclei were stained with TOTO-3 Iodide (ThermoFisher). RFP was stained with chicken anti-RFP (600-901-379S, Rockland) or rabbit anti-RFP (PM005, MBL Life Science). GFP was stained with rabbit anti-GFP (598, MBL Life Science). CRH was stained with rabbit anti-CRH (PBL rC68) provided by P. Sawchenko and J. Vaughan from the Salk Institute. All FISH and immunohistochemistry samples were mounted in 1.5% low-melt agarose in glass-bottomed Petri dishes (P50G-1.5–14-F; MatTek) and imaged using a Nikon A1 upright confocal microscope.

A detailed protocol for cortisol extraction and ELISA is provided online (Supplementary Datasheet S4). At 4.5 dpf, dscaml1−/− and control animals were separated based on the darker pigmentation of the dscaml1−/− animal (Ma et al., 2020b). 30–35 animals were placed in each petri dish for cortisol extraction. The morning baseline (unstressed) sample collections were done at 15–30 min after light onset (08:15–08:45), and the afternoon sample collections were done between 14:30–16:00. Hyperosmotic stress and stirring stress experiments were done between 14:30–15:30. 6 biological duplicates—each containing a pool of 30 animals—were collected for each genotype (dscaml1−/− and control) and stress condition (morning baseline, afternoon baseline, stirring stress, osmotic stress). Mutants and control animals are tested side by side for each experiment.

Stirring stress was induced by creating a vortex water flow with a spinning magnetic stir bar (Castillo-Ramirez et al., 2019). A small magnetic stir bar was placed into a 100 mm petri dish containing 35 animals and 20 mL of E3 media. The stir bar was rotated at 300 rpm with a stirring microplate for 5 min. Hyperosmotic stress was induced by increasing salt concentration in the media (Yeh et al., 2013). 30 animals were placed in 8 mL of E3 media. Then, 2 mL of prewarmed 1.25 M NaCl was added to the media for a final concentration of 250 mM for 20 min.

Sample homogenization and cortisol extraction were performed as described by Yeh et al. (Yeh et al., 2013). Briefly, 5 dpf larvae were rapidly immobilized with ice-cold E3 media and then flash-frozen at −80°C. Once all samples were collected, cortisol from the frozen samples was extracted with ethyl acetate (33211-1L-R; Sigma-Aldrich). Cortisol concentration was measured using a commercial ELISA kit, following the manufacturer’s instructions (500360; Cayman Chemical). Sample plates were read with a microplate reader (FilterMax F3; MicroDevices) 90–120 min after initial development.

Cre-mediated recombination of the crhb:LRLG transgene was induced by injecting Cre mRNA into the embryo at the 1-cell stage. To achieve partial Cre-mediated recombination, ∼30 pg of CreER mRNA was injected at the 1-cell stage and 4-Hydroxytamoxifen (4-OHT) was added to the embryo media at 6 hpf (10 μM), followed by washout with E3 media at 24 hpf. To achieve complete Cre-mediated recombination, ∼50 pg of in vitro transcribed Cre. zf1 mRNA (#61391, Addgene) was injected into the embryos (Horstick et al., 2015).

To perform confocal live imaging, 3 dpf animals were anesthetized with 0.01% tricaine methanesulfonate (MS-222, Sigma) and embedded in 1% low-melting point agarose, with the dorsal side resting on the glass surface inside the glass-bottomed petri dish (P50G-1.5–14-F; MatTek) (Beier et al., 2016). The petri dish was then filled with E3 media containing 0.01% tricaine methanesulfonate (MS-222, Sigma). Confocal z-stacks were acquired every 15 or 30 min for 12 h on a Nikon A1 confocal microscope.

Two-photon live imaging was done on a custom Bruker two-channel two-photon microscope. A tuneable Ti:Saphire laser (Chameleon Vision II; Coherent) was tuned to 980 nm to excite RFP and GFP simultaneously. 78 hpf animals were anesthetized and embedded the same way as confocal live imaging, but with the dorsal side away from the cover glass. Each animal was imaged at 78, 84, 96, 108, and 120 hpf (Figure 6A). After each time point, imaged larvae were gently removed from the agarose and recovered in E3 media at 28.5°C under normal day/night cycles. After the last time point, genomic DNA was prepared for all imaged animals and genotyped.

Images were processed using Fiji—an open-source image processing software (Schindelin et al., 2012). For FISH images, images were convolved (kernel = 12) to enhance cell boundaries, and the center of each cell was manually tagged using the ROI manager tool. The cell number equals the number of ROIs in each animal. The signal intensity per cell was defined as the median signal intensity of all ROIs in a given animal. The number of RFP+ cells in crhb:LRLG animals were counted using the ROI manager tool without convolution. For confocal live imaging, images were pre-processed using the Denoise. AI function in the Nikon Elements software. For two-photon imaging, z-stacks from different time points were aligned using the “Correct 3D drift” function. Spectral overlap between the RFP and GFP channels was linearly unmixed. Individual cells were tracked with the MTrackJ plugin in Fiji (Meijering et al., 2012).

All statistical analyses were performed in GraphPad Prism (Version 9). For normally-distributed data, parametric tests (t-test or ANOVA) were used. For non-normally distributed data, non-parametric tests (Mann-Whitney) were used. The Holm-Sidak post-test was used to correct for multiple comparisons, and the adjusted p values are shown. All values are expressed as mean ± standard error, unless otherwise noted. Statistical tests were considered significant when p < 0.05.

Quantification of apoptotic (AC3+) cells was done by the MAP-map quantitation and Z-Brain annotation method, but Anti-AC3 was used in place of anti-phosphorylated ERK1/2 (Randlett et al., 2015; Brunal et al., 2020). All MAP-map processing procedures (e.g., CMTK, MATLAB scripts) were performed as described by Randlett et al. The NPO is not annotated as such in Z-Brain, but the “Diencephalon - Otpb Cluster 2” ROI fits the definition of the NPO, as it delineate the dorsal region of the preoptic area that expresses otpb (Herget et al., 2014).