Copy number variation in small nucleolar RNAs regulates 1 personality behavior

20 Animals show behavioral traits that can collectively be called personality. We focus here on the role 21 of the Prader-Willi Syndrom gene region in regulating personality behavior. It includes two clusters of 22 tandem repeats coding for small nucleolar RNAs, SNORD115 and SNORD116. SNORD115 is 23 known to regulate splicing of the serotonin receptor Ht2cr and SNORD116 is predicted to interact 24 with the transcript of the chromatin regulator Ankrd11 . We show that both snoRNA clusters display 25 major copy number variation within and between populations, as well as in an inbred mouse strain 26 and that this affects the expression of their specific target genes. Using a set of behavioral scores 27 related to personality in populations of two species of wild mice, guinea pigs and humans, we find a 28 strong correlation between the snoRNA copy number and these scores. Our results suggest that the 29 SNORD clusters are major regulators of personality and correlated traits. 30


Introduction 37
The study of consistent individual differences in behaviour, termed "animal personality", has 38 flourished over the last decades, because it has been recognised as a major contributor to differences 39 in survival and fitness among individuals (Reale, 2007;Reale et al., 2010;Wolf and Weissing, 2012). 40 A major goal is to understand the origin of individual variation in behaviour and the mechanisms 41 generating this variation, especially since it is known to occur also in genetically isogenic strains 42 (Bierbach et al., 2017;Freund et al., 2013;Lewejohann et al., 2011;Wolf and Weissing, 2012). 43 Personality can be defined as behaviors which are consistent over time and across context (Reale,44 2007; Wilson, 1998 that influence personality are largely unknown and there is so far no molecular mechanism that could 48 explain the maintenance of such variation within populations. 49 A locus that has been implicated in behavioral traits in humans is the Prader-Willi syndrome (PWS) 50 region. PWS is a neurodevelopmental disorder which leads to several abnormalities in cognitive 51 behaviors such as social communication, speech, anxiety, intellectual ability and decision making, but 52 also to metabolic syndromes and craniofacial shape changes (Cassidy et al., 2012). The region is 53 subject to imprinting, i.e. parentally biased expression of genes (Nicholls et al., 1998). The protein 54 coding genes expressed in this region include Ube3a and Snprn, of which Ube3a is expressed from 55 the maternally provided chromosome and Snprn from the paternally provided one. This arrangement 56 has specifically evolved in mammals and is generally conserved among them, including humans 57 (Zhang et al., 2014). 58 In mice we have identified the PWS region as one of two imprinted regions that evolve particularly 59 fast between natural populations and that could be involved in behavioral differences between them 60 between SNORD115 expression and exon part Vb usage (Kishore and Stamm, 2006). Activation of 73 the receptor by serotonin inhibits dopamine and norepinephrine release in certain areas of the brain 74 (Alex et al., 2005). This regulates mood, anxiety, feeding, and motoneuron functions (see (Stamm et 75 al., 2017) for a recent review). SNORD116 shows a complementary sequence to Ankrd11 exon X, 76 allowing to suggest a direct regulation (Bazeley et al., 2008). ANKRD11 is a ankyrin repeat domain-77 containing protein that acts as a transcriptional co-factor with multiple possible target genes 78 (Gallagher et al., 2015). But this interaction has only computationally been predicted and has not yet 79 been experimentally confirmed. 80 Here we ask whether natural copy number variation in SNORD115 and SNORD116 could influence 81 behavioral traits in mice and other mammals. We used standardized tests that are mostly connected to 82 anxiety profiles, as well as comparable tests for wild guinea pigs and questionnaire-based tests for 83 humans. We find that there is indeed a strong correlation between copy numbers of the respective 84 SNORD genes and behavioral measures. Intriguingly, this is found not only for wildtype strains, but 85 also for the common laboratory inbred strain C57BL/6J. Using transcriptomic analyses, we show that 86 the predicted regulation of Ankrd11 can indeed be observed and that the network of genes affected by 87 the copy number variation of SNORD116 can explain the behavioral and osteogenic phenotypes. 88 These observations confirm a direct causative link. Since our data suggest further that new alleles 89 with different copy numbers are generated at an exceptionally high rate, we conclude that this 90 variation could be the basis for the long sought molecular mechanism for the high variance of 91 personality traits in families and populations. 92

Results 95
Given the possible molecular links to the regulation of behavioral responses, we first analyzed copy 96 number variation within the SNORD115 and SNORD116 clusters in house mice (M. m. domesticus). 97 Both snoRNAs are part of a larger transcript, from which they are processed (Cavaille, 2017). The 98 repeat unit length for SNORD115 is 1.97kb, for SNORD116 2.54kb. In the mouse reference genome 99 (Waterston et al., 2002), SNORD115 is annotated with 143 copies and SNORD116 with 70 copies, 100 but with an annotation gap. To type copy number differences in this range, we used droplet digital 101 PCR, which we had previously shown to allow accurate copy number determination across a broad 102 range of CNVs in mice (Pezer et al., 2015). We tested individuals derived from two populations of M. 103 m. domesticus that were originally caught in the wild in Germany (CB) and France (MC) and were 104 kept under outbreeding conditions (Harr et al., 2016). We find an average of 202 SNORD115 copies 105 in CB animals and 170 in MC. A similar difference exists for SNORD116 copies, with 229 in CB and 106 188 in MC ( Figure 1). Although there is a large variation within the populations, these differences 107 between the populations are significant (t-test; p=0.03 for SNORD115 and p=0.015 for SNORD116; 108 all data normally distributed, Shapiro-Wilk normality test p>0.21). Intriguingly, we observe also a 109 strong co-variation of copy number between the two SNORD clusters in both populations, with highly  ranges of copy numbers of the clusters in the different species analyzed in the present study. The co-119 variation regression coefficients (R 2 ) between the numbers in the two clusters are listed to the right. 120 The full data are provided in Table S1. 121 The measured copy number variation shows a highly significant regression with the expression of the 123 respective SNORDs in the brain (Figure 2A,B), suggesting a direct relationship between copy number 124 and expression. Note that given that this locus is subject to imprinting, only the paternal copies would 125 be expressed, implying that the correlation should be below 1, even when a perfect relationship exists 126 between copy number and expression. expression and the splice-site regulated target RNAs. All values were determined by ddPCR in brain 133 RNA preparations from eight outbred animals each of the CB and MC populations. These animals 134 constitute a subset of the 45 animals used for overall copy number variation and were chosen to 135 reflect the spread of the variation. The data were combined, since tests for each population on their 136 own yield similar correlations. The full data are provided in Table S2. 137 138 Further, we asked whether the target RNA expressions correlate with their respective SNORD 139 expression. The target of SNORD115 is the alternatively spliced exon V of Htr2cr (Kishore and 140 Stamm, 2006) and the predicted target of SNORD116 is exon X of Ankrd11 (Bazeley et al., 2008). 141 When assaying for the expression of these exons, we find indeed a high correlation with the respective 142 SNORD expression levels and exon expression levels ( Figure 2C,D). As a control, we assayed also 143 two transcript variants for each locus that should not be affected by alternative splicing. For these, we 144 find no significant regression with copy numbers of the respective SNORDs (Table S2). 145 146

Behavioral tests 147
To assess whether the SNORD gene copy number variation correlates also with the behavior of the 148 mice, we have employed an aggregate score consisting of three separate standard tests focusing on 149 anxiety. Anxiety is a consistent personality trait in mice (Freund et al., 2013;Lewejohann et al., 150 2011). Details on the tests and the derivation of the overall score are provided in the Methods section. 151 In short, the development of the score consisted of separate tests (Open Field Test, Dark/Light Box 152 and Elevated Plus Maze). Single test scores were tested for repeatability over the course of the 153 experiment using intra-class correlation coefficients. Those found to be repeatable were clustered and 154 these clusters were subjected to a principle component analysis. The first principle component was 155 then used as a combined single score for anxiety-related behavior. 156 We find a highly significant regression between these scores and SNORD copy numbers, with 157 coefficients between 0.42-0.52 ( Figure 3A-D). The results were analyzed separately for the different 158 populations, since overall copy numbers and overall behavioral scores differ between them. Still, both 159 populations show the same pattern. In each population, animals with higher copy numbers have a 160 higher relative anxiety score. 161 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made  Apodemus uralensis (N=32), F) and G) wild guinea pig Cavia aperea (N=19). Note that the 166 behavioral tests for cavies differ from the tests for the other species. Higher scores represent higher 167 anxiety, i.e. the X-axis is reversed for better comparability. Full data in Table S1. was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint

Inheritance pattern 173
Given the high variability of SNORD gene copy numbers in the mouse populations, one can ask 174 whether they are subject to frequent changes, such as unequal cross-over events during meiosis. Given 175 the total length of the repeat regions, this can not be directly tested, since the haplotypes can not be 176 distinguished by conventional techniques. However, we approached this question by comparing the 177 inheritance of copy numbers in nine families of the MC population. We find that the offspring shows 178 a large variation, exceeding the spread of numbers measured for the parents in 7 out of the 9 families 179 the MC population were tested and the copy numbers were determined for each individual (blue 187 diamonds). The parental copy numbers are marked with an arrow. Note that in 7 out of the 9 families 188 the spread of copy numbers exceeds that of the parents. The correlations with the behavioral tests for 189 these animals are provided in Table S3. 190 191 Hence, the data show that at least some new haplotypes are generated in every generation. The 192 experiment shows also that new large variation is generated in each generation, in line with the 193 general observation that personality traits tend to differ between parents and offspring, as well as 194 among the offspring. To confirm that this is also the case here, we subjected all the offspring to the 195 behavioral tests described above. We find indeed a large variation, as well as the expected significant 196 regression with copy numbers (SNORD115: R 2 =0.5, P << 0.001; SNORD116: R 2 = 0.41, P = 0.003) 197 (Table S3). 198 Further, we asked whether the variability could be so high that somatic variation would become 199 evident. We have therefore scored copy numbers of DNA isolated from different organs of a single 200 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint animal, but we did not find significant differences in this case (Table S3). While this does not rule out 201 the possibility of very high somatic variation such that every cell might be different, there is at least 202 no obvious pattern that distinguishes organs, including the brain. whether inbred mice would also show copy number variability for SNORD115 and SNORD116. We 208 used the C57BL/6J mouse strain for this purpose, which is known to harbor extremely little variation 209 in the form of SNP polymorphisms (Zurita et al., 2011). We genotyped 60 individuals of both genders 210 (30 female and 30 male). We find that there is indeed a large variation of copy numbers in these 211 inbred mice, comparable to the spread seen in the outbred cohorts (Table 1). 20 mice of each sex were 212 then subjected to the behavioral tests. We found no significant difference between males and females 213 in these tests, but we could confirm the strong correlation with copy numbers (Table S1). 214 215

Transcriptome analysis 216
To investigate transcriptome changes in response to copy number variation of the SNORD genes, we 217 choose ten C57BL/6J individuals representing the spread of copy numbers from low to high and 218 sequenced RNA from their brains. First, we sought to confirm that the target genes would show the 219 expected correlations and the corresponding non-target transcripts of these genes would not show this. 220 The first analysis focused on Ht2cr as target of SNORD115. We find a high positive correlation 221 between SNORD115 copy number and total read count from the target exon in Ht2cr-201. As 222 expected, there was no significant correlation between SNORD115 copy number and Ht2cr-204 and 223 Ht2cr-206 as non-target transcripts. Similar results are observed for the SNORD116 target Ankrd11. 224 The target exon of Ankrd11 shows significant correlation to SNORD116 copy number, while there 225 was no significant correlation between SNORD116 copy number and Ankrd11-204 and Ankrd11-207 226 as non-target transcripts (Table S4). Hence, the RNAseq results confirm the previous observations 227 from the ddPCR experiments (see Table S2). 228  (Table S4). 236 Gabrg1 is known to play a crucial role in anxiety regulation in both humans and mice (Nuss,  responses. SNORD116 CNV, on the other hand, has as direct target the chromatin regulator Ankrd11 247 and this affects a whole range of target genes. Among them is the GABA receptor Gabrg1, which is 248 also known to be involved in anxiety responses. Three other groups of genes regulated by Ankrd11 249 have effects on behavioral phenotypes as well, one group has general effects on cell differentiation 250 and development and one group on osteogenesis. The figure is drawn based on the results provided in 251  Table S4. intellectual disability, autism spectrum disorder, and craniofacial abnormalities (Ka and Kim, 2018). 256 Hence, this provides further support that the splice-site regulation of Ankrd11 through SNORD116 257 has a direct functional relevance for behavioral traits. 258 259 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint

SNORD116 copy number and craniofacial features 260
Given that one of the categories of genes affected by Ankrd11 expression level relates to osteogenesis 261 (see above) and a previous study of a particular mutation in the Ankrd11 gene has shown craniofacial 262 abnormalities (Barbaric et al., 2008), we tested whether SNORD116 copy number variation would 263 also have an effect on craniofacial features. We used a 3D landmarking approach to quantify 264 differences in skull shapes based on CT scans, following the procedures described in (Pallares et al.,265 2014; Pallares et al., 2016). A principle component analysis was then used to assess the results. We 266 find that PC1, which explains 35% of the variance, correlates indeed highly significantly with 267 SNORD116 copy number ( Figure 6). This suggests that not only behavioral, but also craniofacial  274  Table S5. 275 276

Other rodents 277
Given that the arrangement of the PWS region with the two clusters of snoRNAs is conserved 278 throughout mammals (Sato, 2017), we were interested to assess in how far one can trace the 279 correlation between copy number variation and personality traits also in other species. 280 We have first analyzed a second mouse species, the wood mouse Apodemus uralensis, which is 281 separated from Mus m. domesticus since about 10 million years and which has different ecological 282 adaptations. Still, given its general similarity to house mice, we applied the same phenotyping 283 scheme. From the currently available genomic data, we could only retrieve sufficient information for 284 the SNORD115 cluster, but this has even higher copy numbers than found in Mus musculus ( Figure  285 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint 1). We find indeed a significant regression with SNORD115 copy numbers and behavioral scores in 286 the wood mice as well ( Figure 3E). 287 As a second, even more distant rodent, we used a cavy species, Cavia aparea, the wild congener of 288 domesticated guinea pigs. For this species we used two behavioral tests that can be considered to 289 reflect anxiety behavior (Guenther and Trillmich, 2015). The current genome sequence of a close 290 relative, Cavia aperea f. porcellus, provides only the annotation for SNORD116 repeats, hence we 291 focused our analysis on this cluster. We find a much smaller range of copy numbers than in mouse 292 ( Figure 1), but still a significant regression with the anxiety scores from the two tests ( Figure 3F, G). To test a possible correlation between SNORD copy numbers and personality traits in humans, we 296 tested a subset of a cohort of 541 healthy individuals that had taken part in a study based on the 297 Tridimensional Personality Questionnaire (TPQ). From these individuals, we chose the top 48 each 298 for low and high anxiety scores and typed them for SNORD copy numbers. SNORD115 in humans is 299 represented by a single class only, while the SNORD116 family is split in three subclasses, for which 300 we designed different ddPCR assays. We find significant differences of copy numbers between the 301 two groups (Table 1) (Table S1). Particularly significant are SNORD115 and SNORD116_2. The 302 latter is the variant that is predicted to bind to Ankrd11 exon X, while the possible target genes for the 303 other two SNORD116 variants are not yet clear. These latter ones show generally only little copy 304 number variation (Table 1). However, we note that the direction of the correlation is different between 305 rodents and humans. In humans, the relatively higher anxiety group has the smaller number of copies, 306 while it is the other way around in the three tested rodent species (see above). 307 308

Discussion 314
Our data establish a link between behavioral scores and copy number variation at the SNORD115 and 315 SNORD116 loci within the PWS region across four species of mammals. This includes also an 316 isogenic mouse strain, which is not expected to carry many other genetic variants apart of the copy 317 number variants (Zurita et al., 2011). Further, our data show that new copy number alleles are 318 generated every generation, which provides an explanation for the high variances observed between 319 parents and offspring, as well as the low general heritability. Importantly, this observation implies 320 also that an artificial manipulation of copy numbers by engineering mouse strains would not be 321 possible, since strains with defined number could not be established. On the other hand, it is known 322 that whole deletions of parts of the PWS region including the snoRNAs lead to complex phenotypes, 323 including behavior, both in mice and humans. Further, the interaction of the snoRNAs with their 324 target genes via sequence complementarity provides a mechanistic explanation for the observed 325 regulatory effects. This, together with the consistency of the correlation across four species establishes 326 a causative link between snoRNA copy number and behavioral traits. Hence, our results establish a 327 new paradigm in the regulation of behavioral variance, namely the involvement of highly mutable 328 regulatory snoRNA clusters. We note that current association mapping protocols would not be able to 329 establish this, since they rely on a stable association with neighboring SNPs, which is not given in the family studies suggest that this may even occur every generation, given that we find more than four 343 length variants among the offspring. Imprinted regions are known to include hotspots of meiotic 344 recombination (Paigen and Petkov, 2010) and this has also been shown for the PWS region in humans 345 (Robinson and Lalande, 1995). However, such an extreme rate of change has not been described so 346 far. 347 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint Another very unusual finding is the covariation of repeat numbers between the SNORD115 and 348 SNORD116 clusters. The repeat units of these clusters do not show any sequence similarity and they 349 are separated by a non-repetitive region. We are not aware of any mechanism that could explain this 350 covariation, but it is consistently found in mice and humans, for which we could obtain the respective 351 data. Given the overlapping regulation of the same behavioral pathways by the two clusters, it makes 352 sense that they should show this covariation, but how this is achieved will require further studies. 353 354

Target pathways 355
Our transcript analysis data in response to SNORD115 copy number variation confirm the known 356 interaction and regulation of the alternatively spliced exon Vb of the serotonin receptor Ht2cr through 357 this snoRNA (Kishore and Stamm, 2006). We suggest that the corresponding predicted interaction of 358 SNORD116 with Ankrd11 may also be a direct interaction, based on our data. This can be inferred 359 from the specificity of the effect on the predicted interacting exon X, in comparison to the transcripts 360 of the locus that do not contain this exon and which do not show a dependence on copy number 361

variation. 362
There are a number of mouse knockout lines that delete more or less large parts of the PWS region. Interestingly, we find that not only behavior, but also craniofacial shape is under the control of the 370 SNORD116 gene cluster copy number variation. For humans it has been suggested that there is 371 indeed a link between personality and facial characteristics (Kramer and Ward, 2010; Squier and 372 Mew, 1981), but a possible causality of these observations was left open. Our data suggest such 373 causality for mice. But given that mice are essentially nocturnal animals that communicate mostly via 374 scents and ultrasonic vocalization, it is unclear whether they would even recognize different 375 craniofacial shapes among their conspecifics. However, there could be a general osteogenic effect on 376 the whole bone system that could be of relevance for being combined with the behavioral tendency. 377 For example, bold animals might profit from stronger bone structures, in case they get more involved 378 in fights. However, this connection will need further study, especially since the SNORD116 379 expression is confined to the brain. Qi et al. (Qi et al., 2016) have suggested that a general osteogenic 380 effect may be mediated via the metabolic pathways that are regulated by Ankrd11. 381 382 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint

Implications for the genetics of personality traits 383
Our findings can resolve a long standing controversy about the genetics and plasticity of personality 384 traits. Our data suggest that these traits are indeed genetically specified, but through a hypervariable 385 locus that would escape conventional genetic mapping approaches. Hence, it is not necessary to 386 invoke developmental or secondary epigenetic effects in explaining variance in personality traits, 387 given that there is a mechanistic pathway of modulation of target RNA expression through different 388 concentrations of snoRNAs. 389 The fact that we find significant regressions between personality scores and SNORD copy numbers 390 also in other rodents and even in humans, implies that the system is conserved throughout mammals. 391 In humans, we observed this even within a psychiatrically healthy control group, where we find that 392 extremes of personality traits for anxiety and activity are associated with significant differences in 393 copy numbers. The effects are somewhat weaker than for the rodents, but humans are evidently also 394 more subjected to environmental influences than the animals that were bred under controlled 395 conditions. Most intriguingly, however, the effect in humans is in the opposite direction, i.e. more 396 anxious individuals harbor lower copy numbers. This would suggest that the copy number variation 397 acts only as a general regulator, but the actual behavioral consequences are modified by downstream 398 pathways. The existence of such modifiers is also suggested in the comparison between the two 399 mouse populations in our study. MC mice have on average lower copy numbers than CB mice, but 400 they are overall more anxious when compared at a population level (Linnenbrink et al. unpublished 401 observations). Also the fact that a complete deletion of the SNORD116 cluster in inbred mice leads to 402 very anxious animals (Ding et al., 2008) suggests that modifiers must play a role as well. We suggest 403 that it should be possible to identify such modifiers in association studies when one controls for the 404 SNORD115 and SNORD116 copy numbers as a covariate. Because of the variation induced by these 405 snoRNAs, they may have been hidden in the background so far. In fact, it may generally become 406 necessary for behavioral studies to include the SNORD115 and SNORD116 copy numbers as 407 covariates, to reveal new patterns independent of this variation. 408 409

Evolutionary implications 410
One can raise the question of how and why such a hypervariable system could have evolved and how 411 it is maintained. This question has in fact been posed since a long time and there have been a number 412 of attempts to propose evolutionary models that could explain the large variability in behavioral traits. 413 These include drift effects in a neutral context (Tooby and Cosmides, 1990), mutation-selection 414 balance (Zhang and Hill, 2005) or balancing selection processes (Dingemanse and Wolf, 2010;Penke 415 and Jokela, 2016). But none of these models has taken the possibility into account that a hypervariable 416 locus could control this variation. Evidently, we need to caution that that the specific mechanism that 417 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint is revealed by our study is only applicable to mammals, while the question of variability of 418 personality traits is an issue across all animal taxa. Still the system found in mammals could provide a 419 guidance towards finding comparable systems also in other taxa. C57BL/6J inbred strain was purchased at the age of 3 weeks from Jackson Laboratory. 444 Mice were usually kept in type III cages (Bioscape, Germany) and were weaned at the age of 3 weeks. 445 Males were housed together with brothers or in individual cages. Females were housed in sister 446 groups to a maximum of 5 mice per cage. Enrichment, including wood wool, toilet paper, egg cartons 447 and a spinning wheel (Plexx, Netherland), was provided in each cage. Mice were fed standard diet 448 1324 (Altromin, Germany) and provided water ad libitum. Housing prior to experiments was at 20-449 24°C, 50-65% humidity and on a 12:12 light-dark schedule with lights on at 7 am. mg/mL) for 16 hours in Thermomixer (Eppendorf, Germany) at 55°C. 500 μL sodium chloride (4.5 457 M) was added to each sample and was incubated on ice for 10 minutes. Then chloroform was added, 458 mixed and spun for 10 minutes at 10,000 rpm. The upper aqueous phase was separated, mixed well 459 with Isopropanol (0.7 volume) and spun for 10 minutes at 13,000 rpm. The pellet was washed with 460 Ethanol (70 %), air dried and dissolved in TE-buffer (10 mM Tris, 0.1 mM EDTA). DNA 461 concentration was measured on the Nano Drop 3300 Fluorospectrometer using Quant-iT dsDNA BR 462 Assay kit (Invitrogen) reagent. 463 RNA extraction was done by using Trizol reagent. 1mL Trizol per 40mg tissue was added to each 464 sample. Then the samples were lysed by Tissue lyser II (QIAGEN, Germany) at 30 Hertz for 5 465 minutes. Homogenized samples were incubated at room temperature for 5 minutes. 200μL chloroform 466 (per 1 mL TRIzol) was added to each sample, shook vigorously by hand 15 seconds, followed by 3 467 minutes incubation at room temperature and spun at 12,000g for 15 minutes at 4°C. The aqueous 468 phase was transferred to a new tube and 0.5 volumes Isopropanol was added, incubated at room 469 temperature for 10 minutes and spun at 12,000g at 4°C. The supernatant was removed and the pellet 470 was washed with 75% EtOH (made with RNAse-free water). Samples were mixed by hand several 471 times and then spun at 7,500g for 5 minutes at 4°C. The supernatant was removed and the pellet dried 472 shortly at room temperature, dissolved in 200μl RNAse free water and stored at -20°C for overnight. 473 An equal volume of LiCL (5M) was added to the crude RNA extract, mixed by hand and incubated 474 for one hour at -20°C. Samples were spun at 16,000g for 30 minutes. The supernatant was removed; 475 samples were washed twice with EtOH 70% and spun at 10,000 at 4°C. The pellet was dried at room 476 temperature, dissolved in RNAse free water and kept for long-term storage at -70°C. 477 The quality of the RNA samples were measured with Bio-Analyzer chips and samples with RIN 478 values below 7.5 were discarded. cDNA was synthesized using the MMLV High Performance 479 Reverse Transcriptase kit according to the instructions of the supplier (epicenter, an Illumina 480 company). 481 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

RNAseq analysis 490
Poly-A + RNA was used for cDNA synthesis and Illumina library preparation by using the Truseq 491 stranded RNA HT kit. The libraries passing quality control were subjected to sequencing on an 492 Illumina NextSeq 500 sequencing system. Raw sequence reads were quality trimmed using 493 Trimmomatic (Bolger et al., 2014). The quality trimming was performed base wise, removing bases 494 below quality score of 20 (Q20), and keeping reads whose average quality was of at least Q60. Reads 495 were mapped to the mouse mm10 reference genome (Waterston et al., 2002)  used as input for SAMtools (Li et al., 2009) by using option -c for total read from each exon and 500 option -q 60 for total read of each sample. GO and KEGG pathway enrichment analyses were 501 performed using DAVID online tools (Version 6.8, https://david-d.ncifcrf.gov/), with the 502 classification stringency set to "medium" P value of <0.05 503 504

Droplet digital PCR 505
Digital PCR is a method enabling absolute quantification of DNA targets without the need to 506 construct a calibration curve as used commonly in qPCR (Zhao et al., 2016). It requires a reference to 507 calculate gene copy number and to normalize gene expression level. For the latter we used β-catenin 508 for mRNAs and SNORD66 for standardizing SNORD RNAs. SNORD 66 is a single copy gene 509 located in an intron of the eukaryotic translation initiation factor 4 and was therefore used also as 510 reference gene for copy number calculations. 511 Primers were designed with 50-70 bp amplicon length, GC content <50% and with very low potential 512 for The Elevated Plus Maze consisted of four arms, each 50 cm long and a 10x10 cm neutral area in the 544 middle. Two of the arms were made of clear Plexiglas, indicating the unsafe zone and two were made 545 of grey PVC, indicating a safe zone. The floor was made of white PVC. Mice were placed in the 546 center and the behavior of each mouse was monitored for 5 min (Holmes et al., 2000). During this 547 experiment, the time spent in the dark and light arms were measured, as well as the speed and distance 548 travelled. 549 For the Open Field test, mice were placed in a 60x60 cm arena with 60 cm high grey PVC walls and 550 they were allowed to explore it for 5 min (Reale, 2007;Wilson et al., 1994;Yuen et al., 2015). The 551 speed of the mouse, the distance travelled and time spent within 10 centimeters off the wall vs in the 552 central area were measured. 553 For the Dark/Light Box, the focal mouse was placed in a test apparatus containing a small dark shelter 554 with two exits. During the first five minutes, the time until the mouse pokes its nose out of the shelter 555 and the first time the tail is visible, was recorded. At five minutes, a set of keys was dropped next to 556 the test apparatus from 136 cm height, and the second part of the experiment began. The time it took 557 for the mouse to first look out and when the entire mouse was visible were measured. If mice did not 558 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint come out at all, the time was set to be 600 seconds. This test was adapted from tests in inbred mice 559 and common voles (Herde and Eccard, 2013;Young and Johnson, 1991). 560 The behavioral tests were filmed using a TSE camera (TSE system, Germany). To score the videos 561 from each test, all the videos were transferred to Videomot2 system (TSE system, Germany). Mice 562 were detected by the software in 3 points (head/center/tail base tracking) and then the software 563 automatically generates the numerical data of the time that each mouse spent at zones of interest. 564 Since personality is defined as consistent behavioral traits, we needed only those measurements which 565 were consistent over the course of the experiment. Hence, each behavioral test was repeated every 4 566 weeks for three times. 567 Statistical analysis were carried out using R 3.3.3 and R 3.3.2 . As the repeatability of a behavior is a 568 key component for the identification of personality trait, all single measurements assessed in the 569 behavioral tests were subjected to repeatability analysis. Repeatability was calculated using "rptR" 570 package (Nakagawa and Schielzeth, 2010). Normally distributed data were calculated using rpt (anova 571 based) and for count data rpt.poisGLMM was used. ID was used as a random effect in these models. 572 To determine whether individual behavioral measurements are correlated, a Spearman correlation 573 matrix was generated. P-values were corrected using the Holm method. Behaviors were clustered 574 using the protocol from (Herde and Eccard, 2013). An hierarchical cluster function was used from the 575 R package "cluster", specifically "agnes", to determine the relationship between the measurements. 576 All measurements were clustered using Manhattan clustering with complete linkage ( The behavioral tests on the wild cavies (Cavia aperea) were conducted as described in (Guenther and 580 Trillmich, 2015). In short, to measure struggle docility, the animal was turned on its back and held in 581 the hand of an observer. For 30 s, the time the animal struggled to actively escape that situation was 582 recorded. To measure open field anxiety, the distance moved (cm) when individuals were exposed to 583 an open field for 20 min, was scored. The first 10 min, a shelter was present in the arena under which 584 animals could hide. For the second 10 min, this shelter was lifted out of the arena. 585 586 Craniofacial shape analysis 587 The morphometric analysis was performed as described previously (Pallares et al., 2015). Briefly, 588 mouse heads were scanned using a computer tomograph (micro-CT-vivaCT 40; Scanco, 589 Bruettisellen, Switzerland) at a resolution of 48 cross-sections per millimeter. Using the TINA 590 landmarking tool (Schunke et al., 2012), 36 three-dimensional landmarks were positioned in the skull. 591 The raw 3D landmark coordinates obtained in TINA tool were exported to MorphoJ (Klingenberg,592 2011) for further morphometric analyses. 593 The symmetric component of the skull was obtained following (Klingenberg et al., 2002). In short, a 594 mirror image of the landmark configuration of each individual was generated, and a full GPA was 595 performed with the original and mirror configurations. Again, the resulting con-figurations were 596 averaged to obtain the symmetric component of shape variation. The new landmark coordinates 597 generated by the GPA are called "Procrustes coordinates". 598 Shape features were computed in MorphoJ principal components (PCs) from the n x 3k covariance 599 matrix of Procrustes coordinates, where n is the number of samples and k is the number of landmarks; 600 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint 3k represents the number of Procrustes coordinates. PC loadings computed in this analysis are defined 601 as morphological score in this study. 602 603 Human cohort and evaluation of personality traits. 604 541 healthy controls were randomly selected from the Munich registry of residents and interviewed 605 for the presence of DSM-IV anxiety, affective, somatoform, eating, alcohol dependence, drug abuse 606 or dependence, disorders using a modified version of the Munich Composite International Diagnostic 607 Interview (Wittchen and Pfister, 1997) at the MPIP. Only individuals negative for the above-named 608 disorders were included in the context of a large study on the genetics of major depression using 609 EDTA blood as the source for DNA (see (Heck et al., 2009) for more detail). Five hundred and 610 fortyone individuals also completed the Tridimensional Personality Questionnaire (TPQ), a validated 611 personality measure (Weyers et al., 1995). To match the behavioral assessments in mice, single items 612 of the TPQ reflecting anxiety and activity were used to select individuals on the extremes of this 613 distribution. Individuals can score from 0 to 6 on the activity and anxiety scale, respectively and their 614 sum was used to identify the extreme groups, while matching for age and sex. The high anxiety and 615 activity group (N = 48) had a mean additive score of 7.35 (SD 1.04). The low anxiety and activity 616 group (N = 48) had a mean additive score of 0.73 (SD 0.49). The mean age of the high anxiety group 617 was 42.3 year and 41.6 years for the low anxiety group with 33.3% and 62.5% females, respectively. 618 619 . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted November 26, 2018. ; https://doi.org/10.1101/476010 doi: bioRxiv preprint