Bridging the gap: Short structural variants in the genetics of anorexia nervosa.

Anorexia nervosa (AN) is a devastating disorder with evidence of underexplored heritability. Twin and family studies estimate heritability (h2 ) to be 57%-64%, and genome-wide association studies (GWAS) reveal significant genetic correlations with psychiatric and anthropometric traits and a total of nine genome-wide significant loci. Whether significantly associated single nucleotide polymorphisms identified by GWAS are causal or tag true causal variants, remains to be elucidated. We propose a novel method for bridging this knowledge gap by fine-mapping short structural variants (SSVs) in and around GWAS-identified loci. SSV fine-mapping of loci associated with complex disorders such as schizophrenia, amyotrophic lateral sclerosis, and Alzheimer's disease has uncovered genetic risk markers, phenotypic variability between patients, new pathological mechanisms, and potential therapeutic targets. We analyze previous investigations' methods and propose utilizing an evaluation algorithm to prioritize 10 SSVs for each of the top two AN GWAS-identified loci followed by Sanger sequencing and fragment analysis via capillary electrophoresis to characterize these SSVs for case/control association studies. Success of previous SSV analyses in complex disorders and effective utilization of similar methodologies supports our proposed method. Furthermore, the structural and spatial properties of the 10 SSVs identified for each of the top two AN GWAS-associated loci, cell adhesion molecule 1 (CADM1) and NCK interacting protein with SH3 domain (NCKIPSD), are similar to previous studies. We propose SSV fine-mapping of AN-associated loci will identify causal genetic architecture. Deepening understandings of AN may lead to novel therapeutic targets and subsequently increase quality-of-life for individuals living with the illness. PUBLIC SIGNIFICANCE STATEMENT: Anorexia nervosa is a severe and complex illness, arising from a combination of environmental and genetic factors. Recent studies estimate the contribution of genetic variability; however, the specific DNA sequences and how they contribute remain unknown. We present a novel approach, arguing that the genetic variant class, short structural variants, could answer this knowledge gap and allow development of biologically targeted therapeutics, improving quality-of-life and patient outcomes for affected individuals.

INTRODUCTION

Anorexia nervosa (AN) is a complex metabo‐psychiatric disorder, and novel approaches are required to further elucidate its etiology. AN mortality is six times higher than the general population (Wonderlich et al., 2020). Standard treatment for adult AN patients combines renourishment, psychotherapy, and medications targeting related comorbidities (Kaye & Bulik, 2021). No biologically targeted treatments exist, and treatment efficacy is low (Kaye & Bulik, 2021). Risk factors for AN have been identified; however, mechanisms underlying heterogeneity in clinical presentation (e.g., restricting vs. binge eating/purging) remain to be clarified. Modern genetics uses several methods to estimate the contribution of genetics to a trait (i.e., heritability). Heritability estimates are derived from family and twin studies, which provide an estimate of the total contribution of genetics to the trait, and single nucleotide polymorphism (SNP) heritability estimates, which provide an estimation of the specific contribution of common single nucleotide variations to the trait. Family and twin studies have yielded and estimates of ~64% and ~57% for AN, respectively, indicating a notable genetic contribution to the disorder. Two separate genome‐wide association studies (GWAS) have identified a total of nine loci significantly associated with AN. The largest GWAS reported a of 11%–17% and identified eight loci associated with AN, with cellular adhesion molecule 1 (CADM1) and NCK interacting protein with SH3 domain (NCKIPSD) as the nearest genes to the SNPs in the top two hits (Watson et al., 2019). We address the considerable gap between estimates and (Manolio et al., 2009). The most likely explanation is that the variants that account for the heritability gap may be more informative than SNPs and are unable to be detected by GWAS (Wainschtein et al., 2021). Short structural variants (SSVs) are sequences of DNA 2–50 base pairs in length and are multiallelic, meaning that more than two variations exist within the population (Roses et al., 2016). SSVs have individual and synergistic effects on molecular biological functioning, including altering transcription rates of genes and affecting protein folding (Mis et al., 2017). Fine‐mapping SSVs in and around GWAS‐identified loci is a potential method of bridging this heritabilitfy gap. GWAS‐identified loci are viable candidates for identifying SSVs because informative SSVs are often located in regions surrounding the lead SNPs identified in GWAS (Gymrek et al., 2016). Accordingly, using GWAS data to select candidate SSVs is a time‐ and cost‐effective method, particularly for initial investigations; future investigations could utilize whole genome sequencing data to identify an increased number of SSVs in order to uncover greater heritability. The informative power of SSVs has been demonstrated in complex disorders including amyotrophic lateral sclerosis (ALS), late onset Alzheimer's disease and schizophrenia (Fotsing et al., 2019; Pytte, Anderton, et al., 2020; Pytte, Flynn, et al., 2020;Roses et al., 2016; Theunissen et al., 2021). Exploring SSVs is a viable approach to further explicate genetic contributions to AN.

GWAS AS A STARTING POINT

AN GWAS provide a starting point to initiate investigating SSVs in AN. GWAS examine evidence for association between a trait and common SNPs across the genome. Association with a given SNP indicates an association exists within the surrounding genomic region and does not suggest the SNP is a coding variant (Tam et al., 2019). GWAS are powerful tools for interrogating heritability in complex traits and are followed by downstream molecular interrogations. Two GWAS have identified a total of nine genetic loci associated with AN (Duncan et al., 2017; Watson et al., 2019). The first GWAS (N cases = 3495; N controls = 10,982) revealed a single genome‐wide significant locus on chromosome 12, associated with lead SNP rs4622308 (Duncan et al., 2017). The locus had been previously associated with rheumatoid arthritis and type 1 diabetes, with autoimmune‐associated loci reported in surrounding regions (Barrett et al., 2009; Okada et al., 2014). Increasing sample sizes yielded a second GWAS (N cases = 16,992; N controls = 55,525) that revealed eight additional significant loci (Table 1); however, the initial locus identified by Duncan et al. (2017) was not replicated (Watson et al., 2019). The lead SNPs for the first five genetic loci were intronic (located in noncoding regions of a gene) and the lead SNPs for  the last three genetic loci were intergenic (located in regions of the genome between genes; Table 1). No lead SNPs were located in exonic regions, the coding regions of the gene. Linkage disequilibrium analyses, the analysis of nonrandom co‐occurrence, revealed significant positive genetic correlations between AN and obsessive‐compulsive disorder, anxiety disorders, schizophrenia, and major depressive disorder (Watson et al., 2019), reinforcing the psychiatric nature of AN (Duncan et al., 2017; Hübel et al., 2021). The study also reaffirmed genetic predisposition to an AN‐prone metabolic profile (Duncan et al., 2017; Watson et al., 2019). Consequently, AN is hypothesized to include genetic predispositions to both psychiatric and metabolic traits (Hübel et al., 2021). The GWAS findings have identified likely informative regions within the genome containing causal genetic architecture (Table 1). Characterizing SSVs is a potential method to extend GWAS findings to further interrogate genetic factors contributing to AN (Chiba‐Falek, 2017; Roses et al., 2016; van Rheenen et al., 2016).

TABLE 1

The eight newly identified genetic loci associated with anorexia nervosa

CHR	Lead SNP	Nearest gene	Functions	pValue
3	rs9821797	NCKIPSD	Growth and cellular signaling in dendrites and sarcomeres; stress fiber formation (Cho et al., 2013).	6.99 × 10−15
11	rs6589488	CADM1	Cellular adhesion; neural network formation; synaptic formation and number (Jin et al., 2019).	6.31 × 10−11
2	rs2287348	ASB3and ERLEC1	ASB3: Phosphorylation and ubiquitination (Chung et al., 2005). ERLEC1: N‐glycan binding (Cruciat et al., 2006).	5.62 × 10−9
10	rs2008387	MGMT	Alkylating agent removal (Yu et al., 2020).	1.73 × 10−8
3	rs9874207	FOXP1	Transcription factor (Siper et al., 2017).	2.05 × 10−8
1	rs10747478	PTBP2	RNA splicing in neuronal cell maturation (Romanelli et al., 2013).	3.13 × 10−8
5	rs370838138	CDH10	Sodium dependent intercellular adhesion (Kools et al., 1999).	3.17 × 10−8
3	rs13100344	NSUN3	Catlysation of 5‐formylcytadine at position 34 of methionine transfer RNA (Nakano et al., 2016)	4.12 × 10−8

Note: Eight genetic loci were identified in the 2019 ANGI GWA by gene proximity to each lead SNP. Lead SNP was determined as the most strongly associated. p‐Value was considered significant (after Bonferroni adjustment) at ≤.05. The major functions for each of the nearest genes to the lead SNP have been described in column 4.

Abbreviations: ASB3, ankyrin repeat and SOCS box containing 3; CADM1, cell adhesion molecule 1; CDH10, cadherin 10; CHR, chromosome; ERLEC1, endoplasmic reticulum lectin 1; FOXP1, forkhead box P1; MGMT, O‐6‐methylguanine‐DNA methyltransferase; NCKIPSD, NCK interacting protein with SH3 domain; NSUN3, NOP2/Sun RNA methyltransferase 3; PTBP2, polypyrimidine tract binding protein 2; RNA, ribonucleic acid; rs, reference SNP accession number; SNP, single nucleotide polymorphism.

EXTENDING GWAS HITS BY STUDYING SSVs

SSVs are multiallelic and include a variety of sizes and sequence arrangements. We focus on short tandem repeats, which are sequence motifs of 1–6 nucleotides repeated numerous times, for example a dinucleotide repeat may contain a repeat of thymine (T) and adenine (A), such as TATATA, while a penta‐nucleotide repeat might comprise of a repeat of a sequence of T, A, guanine (G) and cytosine (C), such as TAGGCTAGGCTAGGC (Roses et al., 2016). Short tandem repeats are highly mutable in nature and accumulating evidence suggests that this class is the most variable of SSVs and consequently the most likely to be functionally relevant (Gymrek et al., 2016). SSVs can be located within a gene but are also frequently found within noncoding or regulatory regions (Gharesouran et al., 2021; Theunissen et al., 2020). SSVs can contribute to more biological variability than SNPs as their multiallelic nature engenders them with a greater likelihood of producing diverse results (Chaisson et al., 2019; Gymrek et al., 2016; Roses et al., 2016). Ribonucleic acid (RNA) expression studies demonstrate that the functional impact of SSVs is significantly greater than SNPs, even though SSVs are less frequent (Jakubosky, D'Antonio, et al., 2020; Jakubosky, Smith, et al., 2020). Growing evidence for informative power of SSVs motivates exploration of their contribution to AN.

SSV functional mechanisms include influencing gene expression, regulation of gene expression, and RNA splicing. Such influences may modify disorder presentation, indicate risk, and influence therapeutic response among patients (Gharesouran et al., 2021; Mahmoud et al., 2019; Pytte, Anderton, et al., 2020; Pytte, Flynn, et al., 2020; Roses et al., 2016; Theunissen et al., 2020). Technically, SSVs located within promoter regions can regulate gene expression by modifying histones, which alters accessibility of the region to transcriptional machinery, and influences promoter binding specificity (Fotsing et al., 2019; Gharesouran et al., 2021). SSVs can also alter transcription and splicing outside of the promoter region (Gharesouran et al., 2021). The variation of SSV length within intronic regions can alter secondary RNA structure, affecting the availability and accessibility of the region to splicing factor binding, altering transcription efficiency. Additionally, SSV polymorphisms can alter the binding sites of intronic and exonic splicing enhancers or silencers to favor one or the other, thus modifying splicing to alter the final messenger RNA (mRNA) transcript (Gymrek et al., 2016). Downstream effects of this have been observed by SSVs such as the G4C2 repeat expansion in chromosome 9 open reading frame 72‐Smith–Magenis chromosome region 8 complex subunit (C9orf72‐SMCR8 complex subunit;C9orf72) and CAG trinucleotide repeat expansion in Ataxin 2 (ATXN2) for ALS (Mis et al., 2017; Van Damme et al., 2011). Both SSVs alter the native structure of the respective protein product leading to aberrant binding, producing truncated protein products, or toxic protein aggregates, which has downstream effects on neurobiological functions that contribute to disease pathogenesis (Mis et al., 2017;   Van Damme et al., 2011). A second example of the informative power of SSVs in ALS is in the CA dinucleotide Stathmin‐2 (STMN2) intronic repeat (Theunissen et al., 2021). Here, the presence of two long alleles in a cohort of sporadic ALS (N cases = 321) was associated with increased disease risk, earlier age of onset, and decreased survival duration for cases of bulbar onset, and disease severity compared with controls (N controls = 332; Theunissen et al., 2021). In another sporadic ALS cohort (N cases = 67), the presence of two long alleles was associated with lower ALS functional rating scale scores and revealed variation in expression levels of Stathmin‐2 mRNA between sporadic ALS cases and control laser‐captured spinal motor neurons based on the CA genotype (Theunissen et al., 2021). With such effects on gene expression and regulation, and the ability to act as genetic markers, uncharacterized SSVs potentially possess considerable power in reducing the heritability gap in AN and may further our understanding of AN etiology, mechanisms, and heterogeneity.

SSVs ELUCIDATING UNDERLYING GENETIC MECHANISMS OF AN

Fine‐mapping poorly characterized regions in and around AN GWAS‐associated loci are likely to elucidate AN heritability, risk factors, and novel pathogenic mechanisms. We utilized an SSV bioinformatics algorithm, “SSV evaluation system” (Saul et al., 2016) to prioritize candidate SSVs in AN‐associated genetic loci NCKIPSD and CADM1 for an initial genetic investigation (Table 2). These loci were prioritized based on association strength reported in the latest AN GWAS (Table 1) and were considered valid for investigation. Both genes have functions that could be biologically relevant to the pathological mechanisms of AN. The NCKIPSD protein possesses several functions across numerous tissues, centered on its signal transduction abilities. Within the context of its role in the nervous system, its function appears to be related to the formation of dendritic spines and modulation of neuronal synaptic activity (Cho et al., 2013). CADM1 has many functions within the human nervous system, with notable roles in neuronal structure and activity, synaptic formation and number, and neuro‐immune cross talk (Jin et al., 2019; Magadmi et al., 2019).

TABLE 2

Initial short structural variants prioritized by the short structural variant evaluation algorithm for future characterization and investigation in anorexia nervosa case/control studies

Gene	rs Number	Symbol	Gene feature
CADM1	rs11358670	28T	Intronic variant
CADM1	rs58589028	29T	Intronic variant
CADM1	rs61694033	25A	Intronic variant
CADM1	rs72085573	20T	5′ Intergenic region
CADM1	rs140815983	15A	3′ Intergenic region
CADM1	rs147798460	32T	3′ Intergenic region
CADM1	rs148209064	33A	Intronic variant
CADM1	rs386374979	29T	3′ UTR downstream contiguous variant
CADM1	rs747352768	11TGG	Exonic variant (Coding exon 8)
CADM1	rs991408884	33T	3′ UTR variant
NCKIPSD	rs71074264	24T	Intronic variant (NCKIPSD and LINC02585)
NCKIPSD	rs71627345	21A	Intronic variant (NCKIPSD and LINC02585)
NCKIPSD	rs375474983	5ACAA	Intronic variant
NCKIPSD	rs377051084	9AGGG	Intronic variant
NCKIPSD	rs545029045	20AC	Intronic variant
NCKIPSD	rs757842104	31T	Downstream variant IP6K2
NCKIPSD	rs34837885	8AAAT	Intronic variant IP6K2; upstream variant NCKIPSD
NCKIPSD	rs35746542	24T	Intronic variant IP6K2; upstream variant NCKIPSD
NCKIPSD	rs67509214	28A	Intronic variant IP6K2; upstream variant NCKIPSD
NCKIPSD	rs71074266	22A	Intronic variant

Note: The 20 SSVs prioritized by the SSV evaluation algorithm (designed by Saul et al., 2016) as candidates for further investigation to elucidate potential roles in AN risk. Ten SSVs have been prioritized for each genetic candidate loci, NCKIPSD, and CADM1. The column titled “Gene” refers to which candidate loci the SSV was reported for. The column labeled “rs Number” refers to the unique identifier supplied by the current human reference genome for that variant. The column titled “Symbol” refers to the most frequently occurring variation of that SSV according to the Allele Frequency Aggregator Project. The final column, titled “Gene Feature,” refers to the functional property of the region of the genome in which SSV is situated. The symbol, rs number and gene feature listed here are as reported in the current human reference genome GRCh38.p13.

Abbreviations: A, adenine; C, cytosine; CADM1, cell adhesion molecule 1; G, guanine; IP6K2, inositol hexakisophosphate kinase 2; LINC02585, long intergenic non‐protein coding RNA 02585; NCKIPSD, NCK interacting protein with SH3 domain; rs, reference SNP accession number; T, thymine; UTR, untranslated region.

The SSV evaluation algorithm enters results from GWAS into a customizable workflow to rank SSVs that are likely to have significant biological effect (Saul et al., 2016). The ranking reduces search time for causal SSVs, enabling efficient prioritization of potential trait modifying SSVs in complex disorders (Saul et al., 2016). Implicated SSVs are prioritized and further interrogated via molecular biology techniques in case/control association studies (Saul et al., 2016). SSV polymorphisms are initially identified via Sanger sequencing and quantified via fragment analysis with end‐labeled fluorescent primers in small control cohorts. Sanger sequencing allows sequence visualization at a single base pair resolution—an effective method for initially determining polymorphisms of an SSV. Several characteristics make Sanger sequencing less suited to genotyping large cohorts as it is particularly prone to errors when sequencing repetitive stretches of identical nucleic acids, such as SSVs, and can be time consuming. Fragment analysis provides an empirical measure of the total size of the genomic region of interest in base pairs, thus catering for repeats of any length with high accuracy and can be performed in a time efficient and high‐throughput manner, rendering it suitable for genotyping large cohorts for confirmed SSV polymorphisms. This approach has been employed effectively in multiple studies, such as Theunissen et al., (2021), which identified and characterized polymorphisms for the CA dinucleotide repeat in the STMN2 gene prior to performing the association studies between the SSV and ALS, presenting it as an exciting future avenue in AN genetics research (Pytte, Anderton, et al., 2020; Pytte, Flynn, et al., 2020; Theunissen et al., 2021). The SSVs we select for the NCKIPSD and CADM1 genetic loci are predominantly noncoding intronic or intergenic variants, as functionally relevant SSVs occur more frequently in noncoding regions of the genome (Table 2; Fotsing et al., 2019). The variant rs747352768 is the only SSV, between both target loci, which falls in a coding exon region (Table 2). The outlined potential of these SSVs to have impact on the CADM1 and NCKIPSD genetic loci that are potentially functionally relevant to AN pathology makes this a valid direction of investigation.

CONCLUSION AND FUTURE DIRECTIONS

AN is severe and potentially fatal with an underexplored heterogeneous etiology. No medications exist that target the underlying biology of AN (Kaye & Bulik, 2021). Outcomes could be improved with increasing understanding of pathogenic mechanisms responsible for the development of AN (Kaye & Bulik, 2021). GWAS have been reported and larger studies are underway to expand upon the GWAS findings and deepen understandings of the heritability of eating disorders more broadly, including bulimia nervosa and binge‐eating disorder (Bulik et al., 2021). Accumulating evidence indicates that SSVs have diverse functional effects on genotypic variability (Fotsing et al., 2019; Pytte, Anderton, et al., 2020; Pytte, Flynn, et al., 2020; Theunissen et al., 2021). Successful SSV mapping has the potential to extend genomic discovery in AN, unveil undetected heritability, elucidate novel pathogenic mechanisms, and identify targets for new therapies, with the long‐term objective of reducing mortality and improving the quality of life for individuals with AN.

CONFLICT OF INTEREST

Cynthia M. Bulik reports: Shire (grant recipient, Scientific Advisory Board member); Idorsia (consultant); Lundbeckfonden (grant recipient); Pearson (author, royalty recipient); Equip Health Inc. (Clinical Advisory Board). The other authors report no conflicts.

AUTHOR CONTRIBUTIONS

Natasha Berthold:Conceptualization; investigation; methodology; writing – original draft; writing – review and editing. Julia Pytte: Methodology; supervision; writing – original draft; writing – review and editing. Cynthia M. Bulik: Supervision; writing – review and editing. Monika Tschochner: Supervision; writing – review and editing. Sarah E. Medland: Supervision; writing – review and editing. P. Anthony Akkari: Methodology; supervision; writing – original draft; writing – review and editing.