Population based forward genetic screen of mutagenized zebrafish identifies loci associated with nicotine preference and human smoking behavior

BACKGROUND Although there is clear evidence of genetic contributions to susceptibility to nicotine addiction, it has proved difficult to identify causal alleles and pathways from studies in humans. Mutagenesis in model species generates strong phenotypes not present in wildtype populations and can be used to identify biological mechanisms underlying quantifiable behaviours. We tested the hypothesis that a forward genetic screen for nicotine preference in zebrafish can predict loci and biological mechanisms influencing human smoking behaviour. METHODS A population-based forward genetic screen of ethylnitrosurea-mutagenized zebrafish was used to identify lines of fish showing altered nicotine preference. Immunohistochemical, behavioral and quantitative PCR analyses were used to characterize mutant larvae. Focussed SNP analysis of the homologous human locus in cohorts from the UK and a Finnish Twin study assessed the predictive validity of the zebrafish data for human smoking behavior. RESULTS We show nicotine preference is heritable in fish as in humans and identify loss-of-function mutations in the zebrafish Slit3 gene as associated with increased nicotine preference. Slit3 mutant larval fish showed altered sensitivity of habituation to acoustic startle to dopaminergic antagonists and increased Drd2 and Drd3 mRNA expression. Dopaminergic neuronal pathfinding was unaffected. Analysis of the SLIT3 locus in two independent human cohorts identified 2 genetic markers that predict level of cigarette consumption and likelihood of cessation. CONCLUSION These findings suggest a role for SLIT3 signaling in development of dopaminergic pathways affecting behaviours associated with nicotine dependence and confirm the translational relevance of the zebrafish model in exploring complex human behaviors.


INTRODUCTION
Tobacco smoking is the leading preventable cause of death worldwide and places a heavy social and financial burden on society (1)(2)(3). It is well established that aspects of nicotine addiction have a strong genetic component (4)(5)(6). However, identifying causal genetic factors and exploring the mechanisms by which they act is challenging in human studies: the field has been characterized by small effect sizes and lack of replication such that there are remarkably few genes and loci that can be confidently linked to nicotine addiction. The strongest evidence for causal effects in humans is for functional variants in CHRNA5 and CYP2A6, affecting amount smoked and nicotine metabolism respectively.
As approaches to identify genetic risk are difficult in humans, research has been facilitated by studies in animal models, with a focus on genomic analysis of inbred and selectively-bred naturally occurring genetic strains (7). This type of study produces quantitative trait loci (QTL) maps of multiple loci, each with a small impact on the phenotype. However, as with human studies, it is inherently difficult to identify relevant genes from QTL maps, as the overall phenotype cannot be predicted by individual genotypes. Mutagenesis studies in animal model systems may overcome these limitations: N-ethyl-N-nitrosourea (ENU) mutagenesis introduces thousands of point mutations into the genome, dramatically increasing variation with the potential to generate much stronger phenotypes than those occurring in a natural population.
Although the disruptive mutations generated by ENU mutagenesis are unlikely to occur in a natural population, they may identify pathways the attenuation of which, through 'weaker' variants in orthologous genes, can be reasonably expected to influence the same behavior in humans. Thus, ENU mutagenesis in model species could be used to distinguish novel, naturally occurring variants influencing human addictive behavior by identifying key genes and pathways affecting conserved behavioural phenotypes.
ENU Mutagenesis screens for addiction-related phenotypes have been undertaken in both invertebrates and vertebrates. They have identified fruit flies (Drosophila melanogaster) that vary in their alcohol consumption and response to alcohol (8).
Mutagenesis studies for acute effects of drugs of abuse in rodents have had limited success (9,10): they are limited by the small number of offspring in each generation that, given the inherent variability of behavior, makes identification of significant differences difficult. The prolific reproductive nature of zebrafish (up to 300 eggs in a single pairing) overcomes this difficulty but, to date, the predictive validity of behavioral screens for addiction-related phenotypes in fish for human behavior has not been established.
Here, we describe a population-based forward genetic approach to screen families of ENU mutagenized zebrafish for nicotine-induced conditioned place preference (CPP).
We identify a novel mutation influencing sensitivity to rewarding effects of nicotine and characterize the mutation using immunohistochemical analysis of neuronal pathways and behavioral responses to acoustic startle, a response known to be modulated by dopaminergic signaling and, in humans, associated with vulnerability to addiction (11)(12)(13). We then use focused SNP analysis of human cohorts to assess the predictive validity of findings in fish for human smoking behavior.

MATERIALS AND METHODS
Zebrafish: Fish were housed in a recirculating system (Techniplast, NL) on a 14h:10h light:dark cycle (0830-2230). All fish were age and weight matched for each part of the study and were 3-5 months old at the start of testing. The housing and testing rooms were maintained at ~25-28ºC. Fish were maintained in aquarium-treated water and fed three times daily with live artemia (twice daily) and flake food (once). All procedures were carried out under license in accordance with the Animals (Scientific Procedures) Act, 1986 and under guidance from the local animal welfare and ethical review board at Queen Mary University of London.

Human Cohorts:
Human subjects for studies in London were recruited from three clinical groups: patients with chronic obstructive pulmonary disease (COPD) (Cohort 1; n=272); patients with asthma (Cohort 2; n = 293); and residents and carers in sheltered accommodation, with neither condition (Cohort 3; n=298). The methods used for recruitment and definition of phenotypes are reported elsewhere (14)(15)(16). The studies were approved by East London and The City Research Ethics Committee 1 (09/H0703/67, 09/H0703/76 and 09/H0703/112) and written informed consent was obtained from all participants.
Details of the Finnish twin cohort are reported elsewhere (17)(18)(19). In brief, twin pairs concordant for moderate to heavy smoking were identified from the population-based Finnish Twin Cohort survey responders. The twin pairs and their siblings were invited to a computer-assisted telephone-based structured psychiatric interview (SSAGA) (17), to yield detailed information on smoking behavior and nicotine dependence defined as by Fagerström Test for Nicotine Dependence (FTND) and DSM-IV diagnoses. Human phenotypes to be investigated in relation to zebrafish nicotine seeking behavior were determined by consensus a priori. Detailed definitions of Finnish phenotypes can be found in the supplementary material.

Conditioned place preference (CPP):
Following habituation and determination of basal preference, fish were conditioned to a predicted threshold concentration of nicotine (Sigma, Catalogue number: N1019. Concentration of 5μM based on previous research (21)) over three consecutive days and assessed for a change in place preference the following day. The CPP procedure was carried out using a balanced assay as described previously (21-23) (see Supplementary Material for details). Of the 30 families included in the screen, four individuals were selected from each to make a total of 120 fish. from the same family were found to be within the top or bottom 5%. In these cases, only progeny from one member of the family were included in the second generation screen, for which a similar strategy was used.

Rescreening of outlier siblings and independent sa202
Slit3 allele: Once we had determined lines that clustered together in the extremes of the distribution, the siblings of the high (n=10) and low (n=14) responder lines were screened for CPP along with control fish. Larvae from the sa202+/-line were obtained from the Sanger Institute and reared to adulthood then assayed for CPP to 5µM nicotine (n=10) along with sibling controls sa202 +/+ (n=11). All fish were fin-clipped and genotyped using site-specific PCR after CPP.

Genomic DNA extraction and Polymerase Chain Reaction (PCR): Genomic DNA
was extracted from fin-clips using QIAGEN DNeasy® Blood and Tissue Kit according to manufacturer's instructions. Samples were eluted into distilled water and stored at -20°C until later use. Site specific PCR used for genotyping and quantitative PCR used for analysis of Drd2/Drd3 expression levels were conducted as described previously (24,25). Details can be found in Supplementary Material.  (26) and correspond to 50, 100 and 500 times its Ki value for the D2 receptor in mammals (27,28). To ensure that larvae were exposed to the drug for the same amount of time, amisulpride was added 15 minutes before undertaking the experiment. Care was taken regarding the distribution of concentrations and genotypes to ensure that experimental groups were randomly distributed in the plates. Plates were placed in a custom-made filming tower with a tapping device that applied 10 sound/vibration stimuli with two seconds interval between them. The setup for this device has been described elsewhere (29). Larval movement was recorded using Ethovision XT software (Noldus Information Technology, Wageningen, Netherlands) and data were outputted in one second time-bins.

Statistics:
Nicotine preference scores were calculated as proportion of time spent in the vicinity of each stimulus (i.e., for spots, Time(T)spots/ [Tspots + Tstripes]). Population means between generations were compared using independent two-sample t-tests, and effect-sizes ascertained using Cohen's d (30). For the rescreen of outlier siblings and sa202 line, mutant lines were compared with wildtype controls using an independent two-sample t-test.
Response to startle stimuli was defined as the distance travelled (mm) during one second after each tap. Linear mixed models were calculated to assess differences in baseline distance moved, distance moved one second after the first stimulus and distance moved during all the stimuli across groups. To assess habituation to the stimuli, we calculated the number of fish exhibiting a response ≤ 50% of the mean startle response to the first tap stimulus (set to 2.5mm because this was 50% of the mean distance travelled across all groups at tap one, and response to the first stimulus was not significantly different across experimental groups (n.s.)). Differences in the number of fish that moved ≤ 2.5mm across genotype and drug groups were compared using ANOVA. Results of all statistical analyses are reported with respect to a type-1 error rate of a=0.05. Post-hoc tests were conducted using Tukey's HSD.
Relative mRNA expression in qPCR was calculated against reference gene cyclethreshold (Ct) values, and then subjected to one-way ANOVA.

Human association analyses:
Association analysis for the London cohorts was performed using PLINK v1.07 (31). SLIT3 SNPs that had been previously associated with disease phenotype and induced an amino acid change were identified and the 20 with the highest linkage disequilibrium score selected for analysis. Of twenty SLIT3 SNPs, one departed from Hardy-Weinberg equilibrium (rs13183458) and was excluded. Linear regression was performed on average number of cigarettes smoked per day, controlling for age, sex and cohort. This analysis was repeated on heavy smokers (≥ 20 cigarettes per day) and light smokers (< 20 cigarettes per day) to investigate whether effects were related to intake level. Smoking cessation (current vs ever smokers) was analyzed using logistic regression controlling for age, sex and cohort. All the analyses were performed under the additive genetic model and multiple testing was taken into account using the Benjamini-Hochberg adjustment. Only individuals from European ancestry were included in the analyses.
Association analyses for the Finnish Twin Cohort were performed using GEMMA v0.94 (32) with linear mixed model against the allelic dosages controlling for age and sex. The sample relatedness and population stratification were taken into account by using genetic relatedness matrix as random effect of the model.

RESULTS
To assess heritability of nicotine preference in zebrafish we screened adult F2 and F3 offspring of ENU mutagenized fish (33) for nicotine CPP and bred from extremes of preference ( Figure 1A To confirm that loss of Slit3 function was related to nicotine seeking behavior we used a second allele, Slit3 sa202 , with a G>T transversion producing a premature stop codon at position 163. Heterozygous Slit3 sa202 fish showed enhanced nicotine conditioned place preference (P=0.03) compared to wildtype siblings ( Figure 2C & 2D). Both Slit3 sa1569 and Slit3 sa202 mutations affect splicing in the region before the second leucine rich repeat (LRR) domain in the encoded protein ( Figure 2F), which is essential for interaction with ROBO receptor proteins (34).
SLIT3 is a member of a family of proteins with established axon guidance properties and previously suggested to be involved in dopaminergic pathfinding (35). Therefore we performed immunostaining of the axonal projections in three-day-old zebrafish larvae. No differences between Slit3 sa1569 mutant fish and wildtype fish were observed in the axonal tracts labelled by anti-acetylated tubulin antibody (Supplementary Figure   1B) nor in the catecholaminergic tracts labelled by anti-tyrosine hydroxylase antibody (Supplementary Figure 1A).
As subtle effects on dopamine circuit formation may not have been detected by our antibody staining, we examined dopamine signaling in wildtype and mutant fish using habituation of locomotor response to acoustic startle, a response known to be sensitive to dopaminergic antagonists such as the D2/D3 antagonist amisulpride (36). were again associated with cigarette consumption together with rs11742567 (p=0.004715), (Table 1A & 1B). The SNP rs11742567 was associated with cigarette consumption in light smokers (<20 cigarettes per day, p=0.003909)) (Table 1C) and with quitting (Table 1D). No other SLIT3 polymorphisms were associated with smoking initiation, persistent smoking or cessation (Supplementary Tables 5 & 6).
We then investigated associations with more detailed smoking phenotypes in the Finnish twins (Table 2). Associations were observed between rs17734503 and DSM-IV nicotine dependence symptoms (p=0.0322) and age at onset of weekly smoking (p=0.00116) and between rs12654448 and age at onset of weekly smoking (p=0.00105). Associations were seen elsewhere between SLIT3 markers and FTND, cigarettes smoked each day, sensation felt after smoking first cigarette and time to first cigarette in the morning. In keeping with the London studies the minor allele was associated with a lower degree of dependence and decreased cigarette consumption.

DISCUSSION
The aim of this study was to use forward genetic screening in zebrafish to identify loci affecting human smoking behavior. We identified two loss of function mutations in the zebrafish Slit3 gene that were associated with increased nicotine place preference. We established the relevance in humans by identifying markers in SLIT3 where the presence of the minor allele was associated with fewer cigarettes smoked each day and with smoking cessation. Studies in a separate twin cohort showed that these same alleles were associated with DSM-IV nicotine dependence symptoms and age at onset of weekly smoking. Taken together these findings suggest that the alleles are linked in humans with a disruption of SLIT3 function which may affect the propensity to develop tobacco dependence.
We used a population-based pre-screen of ENU-mutagenized zebrafish lines followed by sibling re-screen to identify lines of fish showing altered nicotine preference. This proof of principle study indicates the relevance of zebrafish for human studies and emphasizes the advantage of using a population-based pre-screen to increase efficiency.
The classic three generation forward genetic screen examines phenotypes in groups of 20 or more individuals from each family (38). Logistical considerations make it difficult to apply such an approach to adult behavioral screens. Use of a populationbased pre-screening approach increases efficiency of the screen by initially screening a small number of individuals from a large number of F3 families and only selecting those families that occur at the extremes of the distribution for further analysis.
Although in the study described here we were able to confirm phenotypes using a relatively small population of siblings, re-screening of a larger number would increase the power of the analysis and allow more subtle phenotypes to be identified.
We identified the line showing increased nicotine preference as carrying a loss of function mutation in the zebrafish Slit3 gene and confirmed the phenotype in an independent line. SLIT molecules bind to ROBO receptors through a highly conserved leucine-rich repeat (LRR) domain (39). In the AJBQM1 line the gene breaking mutation causes a truncation at amino acid 176 and in the sa202 line at amino acid 163.
These are immediately adjacent to the LRR2 domain responsible for SLIT3's functional interaction with ROBO proteins (39) and would therefore be predicted to lead to formation of non-functional proteins. Initially identified as a family of axon guidance molecules, SLIT proteins are known to be expressed in a range of tissues and, by regulating cell polarity, to play major roles in many developmental process including cell migration, proliferation, adhesion, neuronal topographic map formation and dendritic spine remodeling (40). In vitro SLIT proteins bind promiscuously to ROBO receptors suggesting that the proteins may cooperate in vivo in areas in which they overlap. However, their restricted spatial distributions, particularly of SLIT3 in the CNS (41) suggest the individual proteins play subtly different roles in vivo.
Despite its neuronal expression, the most prominent phenotype seen in Slit3 deficient mice is postnatal diaphragmatic hernia (42,43) with no obvious neuronal or axon pathfinding defects having been reported. Similarly, we did not detect any major differences in axon pathfinding in Slit3 mutant zebrafish larvae. As suggested previously (44) it may be that overlap of expression with other SLIT molecules compensates for loss of SLIT3 in the brain preventing gross neuronal pathfinding defects. However, subtle differences in circuit formation and/or axon branching may have escaped our analysis.
In support of this suggestion, Slit3 mutants showed increased habituation to acoustic startle and altered sensitivity of the startle response to the D2/D3 antagonist amisulpride. As there were no differences in startle magnitude or mean distance travelled during the test period across groups, differences in habituation response cannot be due to differences in health or reactivity of the mutants and presumably reflect subtle differences in underlying neuronal circuits and/or their regulation. Acoustic startle has previously been shown to be sensitive to modulation of dopaminergic signaling in all species studied (36,45,46). Data from dopamine receptor loss of function mice indicate that the dopaminergic system regulates both sensitization and habituation of the startle response in an 'antithetical manner' (45); dopamine D1 receptors inhibit sensitization and augment habituation whereas D2-like receptors augment sensitization and inhibit habituation. Although results with dopamine agonists and antagonists in rodents are less clear cut (45), our finding that the selective dopamine D2/3 antagonist amisulpride increased habituation to acoustic startle in wildtype fish is consistent with this model, and is in agreement with the effect of amisulpride on habituation in humans (36) In contrast to results in wildtype fish, Slit3 mutants showed decreased habituation in the presence of low dose amisulpride. This finding, together with the observation that Slit3 mutants have increased Drd2a and Drd3 receptor expression indicate disruption of dopaminergic signalling pathways that may underlie both the differences in habituation response and the increased sensitivity to nicotine in our place preference assay. Although it is possible that an anxiolytic effect of nicotine contributed to the increased nicotine induced place preference, preliminary assessment of anxiety-like responses (tank diving) in Slit3 mutants (n.s) argue against this. The disrupted dopaminergic signalling and increased receptor expression supports a role of Slit3 in the regulation of nicotine reward processes. Further analysis is required in order to tease out the exact processes affected.
Whilst we confirmed the translational effects of Slit3 gene variants in a human study and the association was validated in another independent cohort, there are limitations to our findings: larger studies would be necessary to obtain greater precision on estimates of the effect size. Further studies are also required to determine the effects of genetic variation in SLIT3 on anatomical pathways in the human brain and their functioning with view to identifying people who are at high risk of developing dependence. This could be achieved by using imaging techniques to study brain activation in response to smoking related cues in smokers who have the SLIT3 polymorphisms linked to smoking (particularly rs12654448).
To our knowledge, this is the first report of a novel human functional polymorphism, identified using a forward genetic screen of adult zebrafish to uncover loci affecting a complex human behavioral trait. Taken together, these results provide evidence for a role for SLIT3 in regulating smoking behavior in humans and establish adult zebrafish as an animal model for exploration of addictive behaviors. Further work analyzing the cellular processes affected as a result of the Slit3 mutation may provide useful targets when designing tailored treatments to aid smoking cessation.     Table 1D.
Association analysis of SLIT3 SNPs with smoking cessation. Logistic regression of current smokers vs ever smokers controlling for age, sex and cohort. Odds ratio >1 indicates minor allele increases odds of persistent smoking relative to major allele. SE: standard error, L95: lower limit of 95% confidence interval, U95: upper limit. For all panels, associations in bold remained significant after adjustment for multiple comparisons using a Benjamin-Hochberg procedure to control false discovery rate at 10%.   (Table 2A), Fagerstrom scores (Table 2B), cigarettes smoked each day (CPD) (Table 2C), and sensation felt after smoking first cigarette and time to first cigarette in the morning (Table 2D). The three SNPs that were linked to smoking behavior in the London cohorts are shown in bold.