Genetically defined nucleus incertus neurons differ in connectivity and function

The nucleus incertus (NI), a conserved hindbrain structure implicated in the stress response, arousal, and memory, is a major site for production of the neuropeptide relaxin-3. On the basis of goosecoid homeobox 2 (gsc2) expression, we identified a neuronal cluster that lies adjacent to relaxin 3a (rln3a) neurons in the zebrafish analogue of the NI. To delineate the characteristics of the gsc2 and rln3a NI neurons, we used CRISPR/Cas9 targeted integration to drive gene expression specifically in each neuronal group, and found that they differ in their efferent and afferent connectivity, spontaneous activity, and functional properties. gsc2 and rln3a NI neurons have widely divergent projection patterns and innervate distinct subregions of the midbrain interpeduncular nucleus (IPN). Whereas gsc2 neurons are activated more robustly by electric shock, rln3a neurons exhibit spontaneous fluctuations in calcium signaling and regulate locomotor activity. Our findings define heterogeneous neurons in the NI and provide new tools to probe its diverse functions.

Initial investigations in rodents indicate that the NI responds to stressful cues; NI neurons are enriched in receptors for the Corticotropin Releasing Factor (CRF) and upregulate c-Fos in response to CRF exposure (Bittencourt and Sawchenko, 2000;Potter et al., 1994).Placement in an elevated plus maze, exposure to an anxiogenic drug, foot shock, or water-restraint stress also induce expression of the neural activity marker c-Fos in the NI (Lawther et al., 2015;Passerin et al., 2000;Rajkumar et al., 2016).Other reports have implicated the NI in regulating baseline locomotor activity.For example, electrical microstimulation of the NI promotes locomotion in rats (Farooq et al., 2016), and optogenetic activation of a subset of neurons in the mouse NI that produce the neuropeptide neuromedin B increases locomotor speed (Lu et al., 2020).
In rodents, the NI contains the largest population of neurons in the brain that produce relaxin-3 (RLN3; Smith et al., 2011;Tanaka et al., 2005), a neuropeptide thought to mediate behavioral responses to aversive stimuli (Lawther et al., 2015;Ryan et al., 2013;Zhang et al., 2015).Although there are also NI neurons that do not produce RLN3 (Ma et al., 2013), their characteristics are not well distinguished from the RLN3 population.
Larval zebrafish are a powerful model to investigate neuronal diversity and connectivity because their transparency and genetic tractability are advantageous for monitoring and manipulating specific subpopulations.In zebrafish, the presumed analogue of the NI is the griseum centrale, situated on the ventral surface of the rhombencephalic ventricle (Agetsuma et al., 2010;Olson et al., 2017;Wullimann et al., 1996).Expression of relaxin 3a (rln3a) is restricted to two bilaterally paired clusters of neurons in the midbrain and two bilaterally paired nuclei bordering the hindbrain midline (Donizetti et al., 2008).It was proposed that the midbrain rln3a expression domains correspond to the periaqueductal gray (PAG), a region that produces RLN3 in rodents (Ma et al., 2017;Smith et al., 2010;Tanaka et al., 2005), and that the hindbrain rln3a neuronal clusters correspond to the NI (Donizetti et al., 2008).
In this study, we find that a small population of neurons defined by expression of the goosecoid homeobox 2 (gsc2) gene is closely apposed to rln3a neurons in the zebrafish hindbrain and distinct from neurons producing relaxin-3, cholecystokinin, and neuromedin B. Through CRISPR/Cas9mediated targeted integration, we generated transgenic driver lines to facilitate selective labeling and manipulation of the gsc2 and rln3a neuronal populations in the nucleus incertus, and found that they differ in efferent and afferent connectivity, calcium signaling, and control of locomotor behavior.

Identification of gsc2 neurons in the nucleus incertus
We initially identified the gsc2 gene through transcriptional profiling aimed at distinguishing genes with enriched expression in the midbrain interpeduncular nucleus (IPN).IPN tissue was microdissected from the brains of adult zebrafish harboring TgBAC(gng8:Eco.NfsB-2A-CAAX-GFP) c375 , a transgene that labels dorsal habenular (dHb) neurons and their axons with membrane-targeted GFP in the larval and adult brain (deCarvalho et al., 2013).Because GFP-labeled dHb axon terminals demarcate the IPN, they serve as a guide to locate and excise this midbrain structure.After comparing the transcriptional profile of pooled IPN samples with remaining brain tissue, gsc2 transcripts were identified as enriched approximately fivefold in the IPN region relative to the rest of the brain.The gsc2 gene encodes a protein that has homology to Goosecoid-related proteins in its homeobox domain-containing sequence.We note that the gsc2 sequence is not annotated in the latest genome assembly (GRCz11) and was initially identified by aligning reads to Zv9 (Ensembl release 77).
From whole-mount in situ hybridization (WISH), we found that gsc2 transcripts are restricted to bilateral clusters just posterior to the midbrain-hindbrain boundary and to a few sparsely distributed neurons anterior to the main cluster (Figure 1A and A').Double labeling with somatostatin 1.1 (sst1.1), a marker of IPN neurons (Doll et al., 2011), revealed that the bilateral clusters are not situated within the IPN but rather lie dorsal to it (Figure 1B and B').
Owing to the similar positions of gsc2 and rln3a (Donizetti et al., 2008) neurons in the larval hindbrain, we performed double-label WISH, and found that gsc2 neurons are a distinct population, located anterior to the rln3a neurons (Figure 1C,C' and D).
Other neuropeptides in addition to RLN3 have been detected in the rodent NI, including neuromedin B in mice (Lu et al., 2020), and cholecystokinin (Kubota et al., 1983;Olucha-Bordonau et al., 2003) and neurotensin (Jennes et al., 1982) in rats.To determine whether transcripts encoding each of these neuropeptides are present in the zebrafish NI, we performed WISH for the zebrafish cholecystokinin a (ccka), cholecystokinin b (cckb), neuromedin a (nmba), neuromedin b (nmbb), and neurotensin (nts) genes (Figure 1-figure supplement 1A-E').For cholecystokinin and neuromedin B, the combined expression of the two zebrafish paralogues closely resembles the overall expression pattern of each single rodent gene (Albus, 1988;Ohki-Hamazaki, 2000).Only cckb and nmbb transcripts were detected in the NI, and nmbb expression was also observed in the PAG (Figure 1figure supplement 1B, D).Using double-label fluorescent WISH, we found that the gsc2 neurons (48.33±2.33 neurons) fail to express any of these neuropeptides and comprise a unique population.We found that hindbrain nmbb neurons (8.33±1.45)are intermingled with rln3a neurons (10.67±1.33) in the NI, with a small subset expressing both neuropeptides (Figure 1E, Figure 1-figure supplement 2A'-C'').By contrast, rln3a and nmbb neurons exist as separate, adjacent populations in the PAG (Figure 1-figure supplement 2A).Hindbrain cckb neurons (4.5±1.1) are a distinct population located just posterior to rln3a and nmbb neurons (Figure 1F).From these results, we can construct a map of peptidergic neurons in the zebrafish NI (Figure 1G), with a discrete group of gsc2-expressing neurons, partially overlapping expression of rln3a and nmbb in cells posterior to the gsc2 neurons, and a distinct population of cckb neurons posterior to the rln3a and nmbb neurons.

gsc2 and rln3a transgenic lines drive expression in the NI
To verify that the gsc2 and rln3a neurons reside in the zebrafish analogue of the mammalian NI, we examined the properties of these closely apposed neuronal populations.Using CRISPR/Cas9mediated genome integration, we generated transgenic lines to selectively label and manipulate each group.The gsc2 and rln3a loci were independently targeted for integration of sequences encoding QF2 (Figure 2A and D), a modified transcription factor that binds to the upstream activating sequence (QUAS) in the bipartite Q transcriptional regulatory system of Neurospora crassa (Riabinina and Potter, 2016;Subedi et al., 2014).Tg(gsc2:QF2) c721 was generated by introducing the QF2 sequence into exon 2 of the gsc2 gene through non-homologous end joining (Kimura et al., 2014).Another method for homology-directed integration called GeneWeld (Wierson et al., 2020) was adapted to include a secondary reporter that, together with the QF2 sequence, was integrated into exon 1 of the rln3a gene to produce Tg(rln3a:QF2, he1.1:YFP) c836 .Identification of rln3a:QF2 transgenic carriers was facilitated by inclusion of a reporter consisting of a promoter from the hatching enzyme 1, tandem duplicate 1 (he1.1)gene (Xie et al., 2012) driving expression of yellow fluorescent protein (YFP) in hatching gland cells starting at 1 day post-fertilization (dpf).Because labeling is transient, the he1.1:YFP secondary reporter does not interfere with brain imaging of older larvae.
We confirmed that the Tg(gsc2:QF2) c721 and Tg(rln3a:QF2, he1.1:YFP) c836 driver lines recapitulate endogenous expression patterns of gsc2 and rln3a, respectively, at both larval (Figure 2B, C, E and  F) and adult (Figure 1-figure supplement 1A-G) stages.Consistent with their location in the NI, the rln3a and gsc2 neurons reside on the floor of the 4th ventricle (Figure 2-figure supplement 1G), with the gsc2 neurons anterior to the rln3a neurons and also distributed more ventrally up to the dorsal surface of the raphe nucleus (Figure 2-figure supplement 1C).Using TgBAC(gng8:Eco.NfsB-2A-CAAX-GFP) c375 to delineate dHb axon terminals at the IPN (deCarvalho et al., 2013), we confirmed that gsc2 neurons are located outside of the IPN in the adult brain, although a few scattered gsc2 neurons lie just posterior and lateral to it (Figure 2  The online version of this article includes the following figure supplement(s) for figure 1:     sparsely distributed anterior group of gsc2 neurons are anatomically distinct from the main cluster, and not within the nucleus incertus proper (Figure 2B and C), they were excluded from subsequent analyses.
To our knowledge, it has not been verified whether rln3a neurons in the periaqueductal gray are also GABAergic.We found that rln3a neurons in the PAG were not labeled by the glutamatergic reporter (Figure 3B), whereas an average of 81.67±3.81%showed labeling from the gad1b transgene (Figure 3E, E', E'' and G).This suggests that rln3a neurons possess similar neurotransmitter identity across neuroanatomical locations.

Distinct projection patterns of gsc2 and rln3a neurons
To compare the projection patterns of gsc2 and rln3a NI neurons, we expressed membrane-tagged fluorescent reporters in each group and acquired optical sections of their labeled processes using confocal microscopy.At 6 dpf, projections from gsc2 neurons were prominent in the cerebellum, IPN, raphe, diencephalon, and rostral and caudal hypothalamus (Figure 4-video 1, Figure 4A-E, Figure 2-figure supplement 2B, B').Sparse efferents from gsc2 neurons were also found in the medulla (Figure 4-video 1) and telencephalon (Figure 4D).Projections from rln3a neurons were found in the medulla, IPN, diencephalon, lateral hypothalamus, and optic tectum (Figure 4-video 2, Figure 4F-J), with some axons appearing to pass through the posterior commissure (Figure 4G).Sparse fibers were also observed in the raphe and telencephalon (Figure 4H and J).
Innervation of the IPN by rln3a neurons originates solely from the NI cluster, whereas the bulk of projections throughout the brain emanate from rln3a neurons in the PAG (Figure 4-video 2).This was confirmed by two-photon laser ablation of rln3a PAG neurons, which greatly reduced fibers in the medulla, diencephalon, hypothalamus and optic tectum, but spared innervation of the IPN (Figure 4K-M).Reduction of rln3a PAG neuronal projections enabled visualization of those from the NI, which exclusively target the IPN (Figure 4-video 3, Figure 4L).Accordingly, ablation of rln3a neurons solely in the NI eliminated innervation of the IPN without affecting the rest of the rln3a neuron projection pattern, including projections to the medulla, diencephalon, hypothalamus and optic tectum (Figure 4M).Efferents from gsc2 neurons were far more extensive than those of rln3a NI neurons, and were observed in regions not innervated by any rln3a neurons (e.g.cerebellum and caudal hypothalamus).Thus, the closely apposed gsc2 and rln3a NI neurons exhibit divergent and largely non-overlapping projection patterns.
To examine gsc2 and rln3a efferent innervation of the IPN more precisely, we used TgBAC(gng8:Eco.NfsB-2A-CAAX-GFP) c375 or TgBAC(gng8:GAL4FF) c426 ; Tg(UAS-E1B:NTR-mCherry) c264 to delineate the IPN by labeled dHb axon terminals.We confirmed the location of gsc2 and rln3a neuronal cell bodies dorsal to the IPN as visualized by nuclear-tagged reporters (Figure 5A, A', and C, C').Using membrane-tagged reporters, we identified projections from both populations to the IPN (Figure 5B,  D and E-H) and found that they innervate disparate regions.This is more readily observed in sections of the adult brain in which axons of gsc2 neurons terminate at the ventral IPN mainly along the midline neuropil (Figure 5B, E-F', I) and axons of rln3a neurons terminate at the dorsal IPN (Figure 5D, G-H',  J), as depicted schematically in Figure 5K.

Afferent input to the NI from the dHb-IPN pathway
Tracing studies in mice (Lu et al., 2020), rats (Goto et al., 2001;Olucha-Bordonau et al., 2003), and zebrafish (Agetsuma et al., 2010) suggest that NI neurons receive afferent input from the Hb-IPN pathway.However, it is unclear whether the Hb-IPN axis influences all NI neurons or specific populations.
We first confirmed that 561 nm light increases calcium signaling in ReaChR-expressing dHb neurons, but not in ReaChR-negative controls D,G,J).Next, we showed that ReaChR activation in the dHb increased calcium transients in gsc2 NI neurons, as they showed greater activation in response to 561 nm light in ReaChR-expressing larvae than in ReaChR-negative controls (Figure 6E-E'', Figure 6-figure supplement 1B, C, E, F).However, similar levels of calcium signaling were detected in the rln3a NI neurons of ReaChR-expressing larvae and negative controls in response to 561 nm light (Figure 6G-G'', Figure 6-figure supplement 1H, I, K, L,).Statistically significant differences in the activation of rln3a PAG neurons between ReaChRexpressing larvae and ReaChR-negative controls were also not detected Figure 6figure supplement 1M,N,O,P).These results show that activation of the dHb-IPN axis increases activity in gsc2 neurons but not in rln3a neurons, indicating that the latter do not directly mediate functions of the dHb-IPN pathway.

Spontaneous and evoked activity differs between gsc2 and rln3a neurons
In rodents, aversive stimuli, such as foot shock, air puff, water-restraint stress, exposure to an elevated plus maze, and the anxiogenic drug FG-7142 all increase neuronal activity in the NI ( Lawther et al., 2015;Lu et al., 2020;Passerin et al., 2000;Rajkumar et al., 2016;Szőnyi et al., 2019;Tanaka et al., 2005), yet whether NI neuronal subtypes show distinct responses to aversive stimuli is unclear.To determine whether gsc2 and rln3a neurons differ in their response to an aversive stimulus, we recorded calcium transients upon delivery of a mild electric shock (25 V, 200 ms duration) (Duboué et al., 2017) to immobilized larvae (Figure 7A, B and C).The gsc2 neurons showed little spontaneous activity and a robust increase in calcium signaling in response to shock (Figure 7D and D',Figure 7video 1).By contrast, rln3a neurons showed more spontaneous fluctuations in activity throughout the recording period (Figure 7E-H, Figure 7-video 1), producing a wider distribution of amplitudes (Figure 7F-G'), and their response to shock was shorter in duration than for gsc2 neurons (Figure 7I).The gsc2 and rln3a neurons therefore differ in their spontaneous activity and in the duration of their response to an aversive stimulus.Ablation of rln3a but not gsc2 neurons alters locomotor activity Previous reports have implicated the NI in regulating locomotor activity and proposed that an animal's increased movement after an aversive stimulus is, in part, mediated by increased activity in the NI (Farooq et al., 2016;Lu et al., 2020).We tested whether eliminating small populations of NI neurons (i.e.10.67±1.33 rln3a neurons or 48.33±2.33 gsc2 neurons) would be sufficient to influence baseline locomotor behavior or the response to electric shock, which normally elicits immediate hyperactivity in larval zebrafish (Duboué et al., 2017).
With GFP expression as a guide, we used a two-photon laser to selectively ablate gsc2 (Figure 8A  and A'), rln3a neurons in the NI (Figure 8B and B'), or rln3a neurons in the PAG (Figure 8C-C'''), at 6 dpf.We confirmed ablation by WISH (Figure 8-figure supplement 1A,, A', D, D'), and also verified selectivity by determining that rln3a NI neurons were spared in larvae with ablated gsc2 neurons (Figure 8-figure supplement 1B, B'), and, conversely, that gsc2 neurons were intact in larvae with ablated rln3a NI neurons (Figure 8-figure supplement 1C, C').One day later (7 dpf), we tracked locomotion of individual freely swimming ablated larvae and unablated siblings for 2 min.After recording baseline activity, we delivered a single electric shock (25 V, 200 ms duration) to each larva and measured the locomotor response (Duboué et al., 2017).
Both ablated and unablated larvae exhibited hyperactivity immediately following shock (Figure 8D-E), and statistically significant differences in the response to shock were not detected (Figure 8E).This suggests that neither the gsc2 neurons nor the neighboring rln3a neurons are required for the immediate behavioral response to shock.Unexpectedly, however, larvae that lacked rln3a NI neurons swam a greater overall distance during the pre-shock period and exhibited longer phases of activity than unablated controls, larvae with ablated gsc2 neurons, or larvae with ablated rln3a PAG neurons (Figure 8F-H, Figure 8-video 1).Phases of activity were defined by continuous movement of the larvae with no more than one second of prolonged immobility.The total number of phases was similar in all groups (Figure 8I).This indicates that ablation of rln3a NI neurons promotes prolonged phases of movement, rather than increasing the frequency of movement initiation.In addition to increased swimming, larvae that lacked rln3a NI neurons showed increased turning behavior.For larvae with ablated rln3a NI neurons, the change in angle of orientation measured per unit of distance traveled was greater than that of unablated controls, larvae with ablated gsc2 neurons, or larvae with ablated rln3a PAG neurons (Figure 8-figure supplement 2A).The unablated control larvae included both Tg(gsc2:QF2) c721 ; Tg(QUAS:GFP) c578 and Tg(rln3a:QF2, he1.1:YFP) c836 ; Tg(QUAS:G-FP) c578 siblings of ablated larvae, although repeating the analyses with the control group containing only one genotype or the other did not change the conclusions (Figure 8-figure supplement 2B, C and Figure 8-figure supplement 3).Overall, the results show that spontaneous locomotor activity and turning behavior are increased following ablation of the rln3a cluster of NI neurons, whereas swimming behavior is normal after ablation of the larger gsc2 population.

Discussion
The nucleus incertus ('uncertain nucleus'), first described in the human brain in 1903 (Streeter, 1903), remains an enigmatic structure that has been implicated in stress (Bittencourt and Sawchenko, 2000;Lawther et al., 2015;Passerin et al., 2000;Potter et al., 1994;Rajkumar et al., 2016;Tanaka et al., 2005), arousal (Lu et al., 2020), and memory (Ma et al., 2009;Szőnyi et al., 2019).As the NI is the primary source of relaxin-3 expressing neurons in the rodent brain, they have been a focus of interest even though not all NI neurons produce this neuropeptide (Ma et al., 2013;Nasirova et al., 2020).Here, we compare the properties of the cells expressing rln3a with a neighboring group of neurons in the zebrafish NI.Through transcriptional profiling, gsc2 transcripts were found to be enriched in samples dissected from the adult zebrafish brain that encompassed the IPN.However, closer examination of both larval and adult brains revealed that gsc2-expressing neurons are located outside of the IPN, just anterior to the rln3a neurons in the NI.The enriched transcripts were likely due to the presence of gsc2-positive neurons that lie just posterior and lateral to the IPN.Neurons expressing the Gsc2 murine homolog had previously been identified in the mouse brain, although there is conflicting information about their precise anatomical location (Funato et al., 2010;Gong et al., 2003;Gottlieb et al., 1998Saint-Jore et al., 1998).On the basis of our results, we suspect that Gsc2 neurons are not located within the rodent IPN as was previously concluded (Funato et al., 2010;Gong et al., 2003), but are likely situated adjacent to it.
We developed transgenic tools to characterize gsc2 and rln3a neurons in more detail.The Gal4-UAS system of yeast is widely used in zebrafish to express reporter genes in specific cell populations; however, its utility for small groups of neurons is limited because of mosaicism due to progressive methylation of CpG residues in multicopy upstream activation sequences (UAS), resulting in transcriptional silencing (Goll et al., 2009).The QF2/QUAS system of Neurospora (Riabinina and Potter, 2016;Subedi et al., 2014), coupled with CRISPR/Cas9 integration, enabled the generation of targeted driver lines and robust and selective expression of reporter genes in either gsc2 or rln3a neurons.By labeling with QUAS-driven fluorescent reporters, we determined that the anatomical location, neurotransmitter phenotype, and hodological properties of gsc2 and rln3a neurons are consistent with NI identity, supporting the assertion that the griseum centrale of fish is analogous to the mammalian NI.Both groups of neurons are GABAergic, reside on the floor of the fourth ventricle and project to the interpeduncular nucleus.However, these adjacent neuronal populations have distinct connections with the IPN and other brain regions, and also differ in their afferent input, calcium signaling, and influence on locomotor behavior (summarized in Table 1).Owing that the NI has been proposed to act in concert with the median raphe and IPN, in 'a midline behavior control network of the brainstem' (Goto et al., 2001), it is important to build the framework of neuronal subtypes that mediate such coordinated activity.
The online version of this article includes the following figure supplement(s) for figure 6:

Distinct patterns of calcium signaling by NI neurons
Distinct patterns of activity were observed in the neuronal populations of the NI, with gsc2 neurons having little spontaneous activity and rln3a neurons exhibiting continuous fluctuations in calcium signaling.A study in rats found that relaxin-3 neurons fire in synchrony with the ascending phase of the hippocampal theta oscillation (4-12 Hz), which has been implicated in spatial memory (Ma et al., 2013).Stimulation of NI neurons in rats and Nmb NI neurons in mice also increases hippocampal theta power (Lu et al., 2020;Nuñez et al., 2006).The oscillating calcium transients that we detected in rln3a neurons of larval zebrafish are on the order of seconds; however, consistent with infra-slow waves that occur at frequencies in the range of tens to hundreds of seconds, and within which fast oscillations are often nested (Palva and Palva, 2012).Infra-slow oscillations correlate with rhythmic fluctuations in performance observed in psychophysical experiments with humans, in which a subject performs a task of constant difficulty for several minutes.It has been proposed, therefore, that intraslow waves coordinate shifts between attentive and inattentive brain states (Palva and Palva, 2012).Given that ablation of rln3a NI neurons increases the length of phases of movement in zebrafish larvae, fluctuating activity in rln3a neurons may control transitions between phases of behavioral activity and inactivity.
Cell-type-specific roles for the NI Rodent studies have described the behavior of animals with null mutations in the gene encoding RLN3 (Smith et al., 2012), or its receptor, RXFP3 (Hosken et al., 2015), and found decreased voluntary wheel running, suggesting that the relaxin-3 system is involved in regulating locomotor activity.However, it is difficult to attribute mutant phenotypes to specific sub-groups of Rln3 neurons.Activation of the NI through microstimulation or chemogenetics increased movement in rats (Farooq et al., 2016;Ma et al., 2017), which implicates the NI region in regulating locomotor activity but does not identify the relevant neurons.
Strikingly, removal of rln3a NI neurons elicited hyperactivity in zebrafish larvae.Ablation of rln3a neurons in the PAG did not affect locomotion.This suggests that the role of the NI in regulating baseline locomotor activity is mediated by rln3a neurons.Because some nmbb neurons are interspersed with rln3a neurons in the NI, we cannot eliminate the possibility that loss of nmbb neurons also contributes to the hyperactivity phenotype.Previous studies in adult rodents indicate that enhanced NI activity promotes locomotion, but we find the opposite in larval zebrafish; NI neurons normally suppress spontaneous locomotor activity.Interestingly, a study of dopaminergic signaling in larval zebrafish also reported that dopamine suppressed spontaneous fictive swim episodes (Thirumalai and Cline, 2008), although dopamine is classically known for stimulating locomotor activity in adult rodents (Ryczko and Dubuc, 2017).Thus, differential roles for neuromodulators during development and adulthood could be a general feature of locomotor circuitry.
indicates standard deviation.(F, F', G, G') Histogram of %ΔF/F amplitudes for (F, F') gsc2 or (G, G') rln3a neurons during the (F, G) pre-shock or (F', G') post-shock period.(H, I) Average (H) number of peaks during the recording period (as depicted by arrows in examples D and E) and average (I) length of response for gsc2 neurons and rln3a neurons, defined as the time required for the %ΔF/F to return to a value equal to or less than the average %ΔF/F in the 100 frames (18.9 seconds) prior to shock.Black bars in (H) indicate mean peaks for gsc2 neurons (5.56±0.63,n=11 larvae) and rln3a neurons (9.91±1.18,n=10 larvae), **p=0.0035.Black bars in (I) indicate mean response times for gsc2 neurons (36.21±8.42,n=11 larvae) and rln3a neurons (10.63±3.27,n=10 larvae) *p=0.045.
The online version of this article includes the following video(s) for figure 7:  A number of studies have found that aversive stimuli promote expression of c-Fos in the NI (Lawther et al., 2015;Passerin et al., 2000;Rajkumar et al., 2016;Tanaka et al., 2005), leading researchers to evaluate the role of relaxin-3 in anxiety-like behaviors.In rats, intracerebroventricular infusion of a relaxin-3 receptor agonist increases entries to the open arms of an elevated plus maze and the amount of time animals spend in the light portion of a light-dark box (Ryan et al., 2013).Similar assays in mice showed that the relaxin-3 receptor agonist did not alter the basal behavioral state but rather reduced anxiety-like behavior induced by the anxiogenic drug FG-7142 (Zhang et al., 2015).However, a role for the NI in regulating the behavioral response to acute aversive stimuli has so far not been described.Lu et al., 2020 note that Nmb neurons in the mouse NI promote spontaneous locomotor activity and are activated in response to foot shock, a stimulus that elicits immediate locomotion, but whether they mediate the immediate locomotor response to this stimulus is unclear.Through selective ablation, we found that loss of either NI rln3a or gsc2 neurons was not sufficient to alter hyperactivity normally observed in zebrafish larvae post-shock.

Animals
Zebrafish were maintained at 27 °C under a 14:10 hr light/dark cycle in a recirculating system with dechlorinated, filtered and heated water (system water).All lines used are listed in the Appendix 1-key resources table.Larvae were screened for labeling by fluorescent proteins using an Olympus MVX10 Macro Zoom fluorescence microscope.For imaging, larvae were incubated in system water containing 0.003% phenylthiourea (P7629, Sigma-Aldrich) to inhibit melanin pigmentation.Most analyses were performed at the larval stage, before sex determination.Analyses performed at the adult stage included both males and females.All procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Dartmouth College (protocol #00002253).
To produce mRNA encoding Tol2 transposase, pCS-zT2TP (Suster et al., 2009) was digested with NotI and RNA was synthesized in vitro using the mMESSAGE mMACHINE Transcription Kit with SP6 polymerase (AM1340, Thermo Fisher Scientific).RNA was extracted with phenol/chloroform-isoamyl alcohol, re-extracted with chloroform, and precipitated with isopropanol.A solution containing QUAS plasmid DNA (25 ng/μl), Tol2 transposase mRNA (25 ng/μl) and phenol red (0.5%) was microinjected into one-cell stage zebrafish embryos that were raised to adulthood.Transgenic founders were identified by screening their F 1 progeny for fluorescently labeled hatching gland cells at 1 dpf and for labeling in the brain under QUAS control.
When applicable, injected embryos were screened for labeling by fluorescent proteins in the hatching gland.To verify successful integration, PCR was performed on genomic DNA from injected embryos using primers that flank the integration site, with the forward primer corresponding to genomic sequence and the reverse primer corresponding to donor plasmid sequence (gsc2_val_F, hsp70_R; rln3a_val_F, QF2_R, Appendix 1-key Resources Table ).Sanger sequencing confirmed the identity of PCR products.Transgenic founders were identified by breeding F 0 adults with a QUAS reporter line and screening progeny for fluorescent labeling of the hatching gland when applicable, and for labeling by QUAS-driven fluorescent reporters.PCR and sequencing were repeated in F 1 larvae to confirm integration at the correct target site.
For colorimetric double in situ hybridization reactions, larvae were hybridized with DIG and FITC probes simultaneously as previously described (Liang et al., 2000), and the DIG probe was first detected using NBT/BCIP as above.To inactivate alkaline phosphatase, larvae were post-fixed overnight at room temperature in 4% PFA, washed twice for 20 min each with MABT (100 mM maleic acid, 150 mM NaCl, 0.1% Tween-20, pH 7.5), incubated for 10 min at 70 °C in EDTA (10 mM in MABT), and dehydrated in methanol for 10 min.Samples were rehydrated stepwise in methanol/MABT, washed in MABT, and blocked for 1 hr in blocking buffer consisting of 20% sheep serum and 2% blocking reagent (11096176001, Roche) in MABT.Tissues were incubated overnight at 4 °C in alkaline phosphatasecoupled anti-FITC antiserum (11426338910, Roche) diluted 1:5000 in blocking buffer.Finally, samples were washed several times in MABT.FITC detection with BCIP and iodo-nitrotetrazolium violet was performed in alkaline phosphatase buffer with 10% polyvinyl alcohol.Samples were cleared in glycerol and mounted for imaging using a Zeiss Axioskop microscope fitted with a Leica DFC 500 digital color camera and Leica Applications Suite software.
For fluorescent double in situ hybridization, larvae were fixed in 4% PFA, dehydrated in methanol, and incubated in 2% hydrogen peroxide in methanol for 20 min.After rehydration and washing in PBT as above, larvae were digested for 30 min in 20 μg/ml proteinase K in PBT, post-fixed in 4% PFA, washed, prehybridized, and hybridized overnight at 70 °C in hybridization buffer with 5% dextran and 40 ng each of DIG and FITC probes.Stringency washes were performed as above, then larvae were washed in TNT [0.1 M Tris pH 7.5, 0.1 M NaCl, 0.1% Tween-20] and maintained for 2 hr in 2% blocking reagent (11096176001, Roche) in TNT.Larvae were incubated overnight at 4 °C in horseradish peroxidase-coupled anti-FITC antiserum (11426346910, Roche) diluted 1:500 in blocking solution, then washed several times in TNT.FITC detection was performed using TSA Plus fluorescein diluted 1:50 in amplification diluent (NEL741001KT, Akoya Biosciences).Samples were washed several times in TNT, incubated in 1% hydrogen peroxide in TNT for 20 min, washed again in TNT, blocked as above for 1 hr, and incubated overnight at 4 °C in horseradish peroxidase-coupled anti-DIG antiserum (11207733910, Roche) diluted 1:500 in blocking solution.Tissue was washed several more times in TNT and DIG detection was performed using TSA Plus Cyanine diluted 1:50 in amplification diluent (NEL744001KT, Akoya Biosciences).Fluorescently labeled samples were imaged using confocal microscopy.

Confocal imaging
Larvae were anesthetized in 0.02% tricaine and individually mounted in a droplet of 1.5% low melting point agarose (50100, Lonza) centered in a 60 mm x 15 mm Petri dish.After the agarose solidified, system water with 0.02% tricaine was added to each dish.Larvae were imaged using either a Leica SP5 with a 25 X (NA=0.95)water immersion objective, or a Zeiss LSM 980 with a 20 X (NA=0.5)water immersion objective.
Adult brains were fixed overnight in 4% PFA at 4 °C, rinsed in 1 X PBS, and mounted in 4% low melting point agarose (50100, Lonza) for sectioning (70 μm) by a Leica VT1000s vibratome.Sections were mounted in glycerol for imaging under either a Leica SP5 with a 20 X (NA=0.7)objective or a Zeiss LSM 980 with a 20 X (NA=0.8)objective. Z

Calcium signaling
Larvae were paralyzed by a 1 min immersion in α-bungarotoxin (20 µl of 1 mg/ml solution in system water; B1601, Thermo Fisher Scientific), followed by washing in fresh system water (Baraban, 2013;Duboué et al., 2017;Severi et al., 2014).Individual larvae were embedded in a droplet of 1.5% low melting point agarose (50100, Lonza) centered in a 60 mm x 15 mm Petri dish.After the agarose solidified, system water was added to the dish.For all calcium signaling experiments, larvae were imaged in xyt acquisition mode using a Zeiss LSM 980 with a 20 X (NA=0.5)water immersion objective and a 488 nm laser.
To record calcium transients in response to electric shock, a plastic ring holding electrodes that were connected to a Grass SD9 electrical stimulator (Grass Instruments), was placed in each dish.Images of gsc2 or rln3a NI neurons were acquired at 475 x 475 pixel resolution and a rate of 5.2 Hz.Calcium transients were recorded for 600 frames (115.4 s) for baseline measurements, then larvae were shocked once (25 V, 200 ms duration) and 1400-1800 more frames (269.2-346.2s) collected.
To record calcium transients in response to stimulation of the red-shifted opsin ReaChR (Lin et al., 2013) with 561 nm light, images were first acquired using a 488 nm laser at 310 x 310 pixel resolution and a rate of 2.6 Hz.The Z-depth was adjusted to the plane of the neuronal population being imaged (i.e.dHb, NI, or PAG brain regions).Spontaneous calcium transients were recorded for 200 frames (76.9 s), the 561 nm laser was activated at 5% power while 20 more frames (7.7 s) were acquired, and then calcium transients were recorded for another 150 frames (57.7 s).
For all calcium imaging experiments, individual frames were extracted in Fiji (Schindelin et al., 2012) using File ->Save As ->Image Sequence and imported to MATLAB, where mean fluorescence intensities for regions of interest (ROI) were calculated.Briefly, a high contrast image was generated for each larva by calculating a maximum intensity projection of its image series.ROIs were drawn manually using the high contrast image and the MATLAB function roipoly.For recordings of gsc2 or rln3a neurons, ROIs were individual neurons; for dHb recordings, each dHb nucleus was designated as an ROI.Mean fluorescence intensity of pixels within each ROI was calculated.ΔF/F was calculated according to the following formula: where F i indicates the mean fluorescence intensity in an ROI at each time point, and F max and F min are the maximum and minimum fluorescence values respectively for that ROI during the recording period.To calculate total activity before and after the stimulus, ΔF/F was averaged across ROIs for each larva and the total activity was obtained for the time period by calculating the area under the curve using the MATLAB function trapz.
The initial time point at which neuronal activity increased for a given ROI was calculated using the MATLAB findpeaks function (MinPeakProminence: 0.2, MinPeakWidth: 10).

Two-photon laser-mediated cell ablation
At 6 dpf, Tg(gsc2:QF2) c721 ; Tg(QUAS:GFP) c578 or Tg(rln3a:QF2, he1.1:YFP) c836 ; Tg(QUAS:GFP) c578 larvae were anesthetized in 0.02% tricaine and individually mounted within a droplet of 1.5% low melting point agarose (50100, Lonza) centered in a 30 mm x 10 mm Petri dish.After the agarose solidified, system water was added.GFP-expressing cells were located using a two-photon microscope (Bruker) with a 60 X (NA=1) objective.The laser was tuned to 885 nm and, using GFP labeling as a guide, was focused on the relevant neuronal population and activated for several seconds at maximum power until the GFP signal disappeared.Because the two-photon laser power is delivered to a restricted Z-plane, ablations were repeated at multiple depths to eliminate each cell population.For ablation of gsc2 neurons the laser was activated over an area of 600-2000 μm 2 on four Z-planes.The laser was activated over an area of 1000-1250 μm 2 on two Z-planes for ablation of rln3a NI neurons and over an area of 1200-1800 μm 2 on two Z-planes for removal of each rln3a PAG nucleus (left and right).

Locomotor assay
Behavioral experiments were performed blind to the ablation status of each larva being assayed.Unablated controls were a mixture of Tg(gsc2:QF2) c721 ; Tg(QUAS:GFP) c578 and Tg(rln3a:QF2, he1.1:Y-FP) c836 ; Tg(QUAS:GFP) c578 , and were siblings of ablated larvae.Behavioral tests were conducted in a temperature-controlled room (27 ° C) on individual 7 dpf larvae.The 6 cm 3 acrylic testing chamber had a 0.5 cm platform on which a 40 mm cell strainer (Falcon) was placed.The chamber was filled with fresh system water and set on top of an infrared illumination source (880 nm, ViewPoint Life Sciences).Locomotor activity was recorded by a high frame rate charged-coupled device (CCD) camera (Point Grey Research), which was connected to a computer (Dell).Tracking was performed in real time at 60 frames per second, using ZebraLab software (ViewPoint Life Sciences).Swimming behavior was recorded for 120 s, then each larva was shocked once (25 V, 200ms duration), and activity recorded for an additional 120 s.To analyze locomotor activity, the x and y coordinates of a larva's position in each frame were exported from ZebraLab.Activity was quantified using R statistical software (R Development Core Team, 2023) according to the following equation: where i indicates a single frame.Total distances were calculated by summing the distance for all frames over the relevant period of the recording.Total number of movement phases and average length of movement phases during the pre-shock period were calculated for each larva using R statistical software (R Development Core Team, 2023).Movement trajectories were plotted using MATLAB.
Phases of movement and their durations were extracted using R statistical software by first binning the distances for all frames within each second of the recording, then iterating through each subsequent second.A movement phase was determined to commence when there was movement within a given second and no movement in the preceding second, and the phase persisted for each subsequent consecutive second with movement.The phase of movement was determined to end when there was no movement within a given second following one with movement.Turning behavior was quantified using R statistical software by iterating through the x and y coordinates of a larva's position in each frame.Starting with the second frame, the size of the angle formed by the larva's change in orientation between the first and second frame, and the second and third frame, was calculated according to the law of cosines: where C is the angle formed by two lines of length a and b, and c is the length of the side opposite angle C. Total turning was calculated by summing the size, in degrees, of all calculated angles of all frames in which a larva was active.Total turning for each larva was divided by the total distance traveled to calculate a ratio of turning per unit of distance traveled.

Quantification and statistical analyses
All means are presented with standard error of the mean.Statistical details for all experiments are summarized in Table 2. Data structure was determined using Shapiro-Wilk tests.Statistical analyses were performed using either R statistical software (R Development Core Team, 2023) or MATLAB.Sample sizes were similar to those typically used in zebrafish behavior and calcium imaging studies (Agetsuma et al., 2010;Choi et al., 2021;Facchin et al., 2015;Muto et al., 2013;Wee et al., 2019).Data were plotted using the MATLAB library PlotPub (Habib Masum, 2022) or the R package ggplot2 (Wickham, 2016).Where applicable, larvae were randomized to the treatment or control group, and no larvae were excluded.

Figure supplement 1 .
Figure supplement 1. Subset of neuropeptides expressed in NI of larval zebrafish.

Figure supplement 2 .
Figure supplement 2. Partially overlapping expression of rln3a and nmbb in the zebrafish NI.

Figure supplement 1 .
Figure supplement 1. QF2 driver lines recapitulate gsc2 and rln3a expression patterns in the adult brain.

Figure supplement 2 .
Figure supplement 2. gsc2 neurons reside outside the IPN in the adult brain.

Figure supplement 1 .
Figure supplement 1. Calcium signaling in individual larvae and neurons.

Figure 6 continued
Figure 6 continued

Figure supplement 2 .
Figure supplement 2. Loss of rln3a NI neurons increases turning behavior.

Figure supplement 3 .
Figure supplement 3. Comparisons between ablated and unablated larvae with the same genotype.

Figure 8 -
Figure 8-video 1. Increased swimming behavior upon loss of rln3a NI neurons.https://elifesciences.org/articles/89516/figures#fig8video1 -stacks of the larval brain encompassing fluorescent signals included approximately 125 slices and 250 µm for dorsal views, or 75 slices and 150 µm for lateral views.Z-stacks focused only on the NI included approximately 35 slices and 70 µm from a dorsal or lateral view.Z-stacks of adult brain sections included 35 slices and 70 µm.

Table 1 .
Properties of gsc2 and rln3a NI neurons.