Metanalysis of genome-wide association studies for panic disorder suggest pathways and mechanisms of pathogenesis

Panic disorder (PD) is characterized by abrupt surges of intense fear and distress. There is evidence for a genetic component in this disorder. We ran a meta-analysis of genome-wide association studies of patients with PD, and found 25 single-nucleotide polymorphisms that were associated with the disorder. Causal gene prediction based on these polymorphisms uncovered 20 hits. Exploratory analyses suggested that these genes formed interactor networks, which was enriched in signaling pathways associated with immune and inflammatory responses, as well as growth factors and other developmental mediators. A subset of genes is enriched in limbic regions of the human brain and in microglia and myelinating oligodendrocytes of mice. While these genes were not associated with relevant neurobehavioral phenotypes in mutant mice, expression levels of several causal genes in the amygdala, prefrontal cortex, hippocampus, hypothalamus, and adrenal gland of recombinant mouse strains was associated with endophenotypes of fear conditioning. Drug repositioning prediction was unsuccessful, but this does not discard these genes and pathways as targets for investigational drugs. In general, ASB3, EIF2S2, RASGRF2, and TRMT2B (and its coded proteins) emerged as interesting targets for mechanistic research on PD. These exploratory findings point towards hypotheses of pathogenesis and neuropharmacology that need to be further investigated.

Panic disorder (PD) is characterized by repeated and unpredictable panic attacks, usually associated 39 with the development of anticipatory anxiety and avoidance strategies that can culminate in 40 agoraphobia (American Psychiatric Association, 2013). Patients with PD have high rates of 41 medically unexplained symptoms that produce a burden on healthcare services (Coley et al., 2009). 42 While highly vulnerable to environmental and life-event causes, panic disorder have a significant 43 genetic component. There is evidence of familial aggregation and moderate heritability of PD 44 (~44%) (Merikangas and Pine, 2002). Many genetic studies have tried to identify linkage or 45 association of specific anxiety disorders with genomic regions or specific genes, but the success rate 46 of such studies is low (Gratacòs et al., 2007). Different approaches to understand the genetic 47

Brain and cell type expression 251
Expression patterns in the human brain were assessed via clustering of microarray data ( Figure 2A). 252 Clustering of brain regions revealed 3 clusters with |r²|

Neurobehavioral phenotypes in mice 278
Contrary to expectations, only two of the candidate genes were associated with behavioral or 279 neurological abnormalities in mouse models ( Figure 3). Asip was associated with abnormal food 280 intake and hypoactivity; Itch was associated with neurological abnormalities (tremors and decreased 281 grip strength). The lack of mouse models also demonstrate a literature gap, since only 12 of the 20 282 candidate genes were found in the MGI database. Moreover, it is probable that the published 283 literature studying these genes in mouse mutants did not investigate behavioral phenotypes which 284 are relevant to panic disorder. 285 To further explore the associations between PD candidate genes and neurobehavioral phenotypes,

Protein-protein interactions and functional associations 346
When mapping the protein-coding candidate genes to a merged and curated BioGRID and HPRD 347 protein-protein interaction database, only 109 proteins, with no interactions between them, were 348 found. Similarly to a cross-disorder set of candidate genes (including genes for ADHD, autistic 349 spectrum disorders, bipolar disorder, major depressive disorder, and schizophrenia; Lotan et al., UNC119: n = 84); moreover, these proteins have not yet been associated with neuropsychiatric 358 disorders. Therefore, interaction is unlikely to be specific for PD. 359 In spite of this lack of direct protein-protein interaction, important signaling pathways were found 360 to be enriched in the PD dataset and interactor network -including WNT, immune and 361 inflammatory pathways, neurotrophins, EGF/MAFPK, and Rho GTPase pathways. Nonetheless, 362 from the candidate genes only RASGRF2 (which was part of the MAPK, NGF, and Rho GTPase 363 pathways) and ITCH (which was part of Endocytosis, Ubiquitin, and NGF pathways) were part of 364 enriched signaling pathways. Both the MAPK and the Wnt pathways were implicated in threat 365 conditioning and aversive learning in animal models (Maguschak and Ressler, 2011;Schafe et al., 366 2001). NGF, on the other hand, has been implicated in stress and stress-related psychopathology 367 (Cirulli and Alleva, 2009). While most of these pathways have been implicated in neural 368 development, mRNA expression of PF-ranked genes in recombinant inbred mice was poorly 369 associated with morphological traits of the BLA; moreover, these associations were more common 370 between BLA morphology and mRNA expression of candidate genes in the HPA axis, and not in the 371 amygdala, as would be expected. Interestingly, Erlec1 and Psme4 expression in both the amygdala 372 and adrenal were associated with BLA morphology. 373 374

Tissue and cell expression 375
Expression patterns in the human brain suggested that most genes were at best moderately 376 expressed in "classical" limbic regions (frontal and cingulate cortices, hippocampus, amygdala); 377 higher expression was found for TRMT2B, ERLEC1, CENPI, CDK19, PSME4, EISF2S2, and ASB3. 378 In rodent PFC cell types, orthologues Trmt2b, Eisf2s2, Cdk19, Erlec1, and Asb3 show higher 379 expression in MOs, microglia, and endothelial cells. These results contrast with a recent GWAS of 380 major depression, in which genes were highly expressed in the frontal and cingulate cortices, 381 hippocampus, and basal ganglia, and significantly enriched in neurons (vs. oligodendrocytes and 382 astrocytes) (Wray et al., 2018), suggesting disorder-specific patterns. While a region-specific pattern 383 of expression cannot be discarded, these results suggest that MOs and microglial cells in limbic 384 regions are associated with PD. Among these genes only CDK19 and ASB3 appeared in interactor 385 networks (CDK19 was the core of N2), and none was part of the enriched signaling pathways; 386 overall, these results suggest that expression patterns, instead of general function in interactor 387 networks and signalling pathways, is the commonality between these PD-associated genes. 388 When mRNA expression across limbic regions is associated with behavioral traits in rodents, a 389 more specific pattern emerges. Some genes were associated with behavioral traits in fear 390 conditioning paradigms; in all cases, the association was with freezing to cue and/or to context. 391

13/23
Gene expression in the amygdala (Zcchc9, Rasgrf2, Sox10) was associated with freezing to cue 24 h 392 after conditioning, while gene expression in the hypothalamus (Trmt2b, Itch, Cntn4) was associated 393 with freezing to cue 48 h after conditioning. Genes expression in the PFC (Itch, Slc22a16, Sox21) 394 were negatively associated with freezing to context 48 h after conditioning. In the mouse PFC, 395 Zcchc9, Itch, and Trmt2b were expressed mainly in MOs, microglia, and endothelial cells, while 396 Rasgrf2 and Cntn4 were expressed mainly in astrocytes, neurons, OPCs, and NOFs. 397 The higher expression in MOs is of interest, since myelin is known to stabilize synaptic networks, 398 and oligodendrocyte-related proteins have been shown to be modified in the rat dentate gyrus 399 during contextual fear conditioning (Houyoux et al., 2017). Moreover, microglia has been shown to 400 participate in cued fear conditioning (Parkhurst et al., 2013), and microglial acid sensing has been 401 shown to participate in carbon dioxide-evoked fear (Vollmer et al., 2016). These results also make 402 sense in light of the enrichment of immune and inflammatory pathways in the interactor network. 403 404

Drug repositioning attempts 405
Top 20 drugs which perturb PF-ranked genes include interesting findings. Cyclosporin A has been 406 shown to inhibit calcineurin (Fakata et al., 1998), impairing aversive memory formation in chicks 407 (Bennett et al., 1996); and valproic acid enhances the extinction and habituation of fear conditioning 408 in healthy humans (Kuriyama et al., 2011). Nonetheless, evidence for their use in treating PD (or in 409 exacerbating PD symptoms) is scarce. The current results do not provide sufficient evidence for the 410 association of these drugs and their effects on the PF-ranked genes found for PD; in fact, no gene-411 drug association was found either at the individual drug level, or at the drug class level. Differently 412 from previous reports (So, 2017;So et al., 2016), which support drug repositioning from genes 413 extracted from GWAS for anxiety and mood disorders, our findings do not support such proposals. 414 Extensive methodological differences are probably responsible for this discrepancy. 415 416

Consequences for the pathophysiology and pharmacology of PD 417
One important limitation of the present results is that they are exploratory in nature, awaiting 418 confirmatory research on all findings. Nonetheless, the present results suggest interesting avenues 419 of investigation on the relationship between genetics and the pathophysiology and pharmacology of 420 PD. First, the observation of cross-species associations between genes discovered in the GWAS 421 metanalysis and neurobehavioral traits of fear prompts the idea of conservation across species of 422 specific endophenotypes (de Mooij-van Malsen et al., 2011;Kas et al., 2012), such as alterations in 423 fear conditioning (Battaglia and Ogliari, 2005;Bouton et al., 2001). Second, and perhaps a more 424 interesting point, is the observation that these endophenotypes are not only related to the expression 425 14/23 levels of risk genes in regions such as prefrontal cortex, amygdala, and hypothalamus, but also that 426 these genes are more expressed in myelinating oligodendrocytes and microglia than neurons and 427 astrocytes, the brain cells which are more studied and usually highly associated with psychiatric 428 disorders in GWAS ontology studies (Lotan et al., 2014;Wray et al., 2018). 429 This last observation ties well with the observation of enrichment in immune and inflammatory 430 pathways, given the role of microglia in these functions in the brain (Hanisch, 2002). Small studies 431 have suggested increased levels of proinflammatory cytokines in serum from patients with PD (e. 432 g., Hoge et al., 2009), and cytokines have been associated with acute mental stress in non-clinical 433 populations (Kunz-Ebrecht et al., 2003;Porterfield et al., 2011;Steptoe et al., 2001). Whether these 434 changes are generalized in PD, and whether genes associated with increased PD risk do so by 435 altering peripheral cytokine responses as well as microglia and oligodendrocyte function in the 436 brain is still unknown. 437 The present results also highlight putative targets for basic pharmacological research. While drug 438 repositioning analysis was inconclusive, experimental drugs acting on these pathways were not part 439 of the dataset. Therefore, the present results point towards an interesting a posteriori hypothesis that 440 targeting specific pathways (WNT, immune and inflammatory responses, neurotrophins, 441 EGF/MAPK, Rho GTPase) specifically in microglia and oligodendrocytes could elucidate the 442 pathophysiology of PD and suggest novel investigational drugs. Further confirmatory research will 443 allow that hypothesis to be tested. Gratacòs, M., Sahún, I., Gallego, X., Amador-Arjona, A., Estivill, X., Dierssen, M., 2007. 491 Candidate genes for panic disorder: Insight from human and mouse genetic studies. Genes, 492 Brain Behav. 6, 2-23. doi:10.1111/j.1601-183X.2007 Koefoed  Otowa, T., Kawamura, Y., Nishida, N., Sugaya, N., Koike, A., Yoshida, E., Inoue, K., Yasuda, S., 540 Nishimura, Y., Liu, X., Konishi, Y., Nishimura, F., Shimada, T., Kuwabara, H., Tochigi, M., 541 Kakiuchi, C., Umekage, T., Miyagawa, T., Miyashita, A., Shimizu, E., Akiyoshi, J., Someya, 542 T., Kato, T., Yoshikawa, T., Kuwano, R., Kasai, K., Kato, N., Kaiya, H., Tokunaga, K., 543 Okazaki, Y., Tanii