Genetics

TL;DR — AN is substantially heritable but not genetically deterministic. A 2019 GWAS of 16,992 cases and 55,525 controls identified eight loci and genetic correlations spanning psychiatric, activity, glycaemic, lipid and anthropometric traits that persisted after accounting for common BMI-associated variants (Watson 2019, PMID 31308545). A 2026 integrative meta-analysis added a novel locus near SOX5 and, using multi-trait analysis, 86 significant loci of which 25 were new — but it did so by borrowing power from correlated traits, not by enlarging the AN case series, so Watson 2019 remains the anchor case-control study (Song 2026, PMID 41927769). This supports a metabo-psychiatric research model, not clinical genetic testing or a proven metabolic treatment. Current common-variant scores are research tools with limited individual predictive value.

Evidence hierarchy

Evidence	What it can show	What it cannot show
Twin/family study	Familial aggregation; heritability under assumptions	Specific causal variants; immutability
GWAS	Common-variant associations at population scale	Mechanism without functional follow-up
Genetic correlation	Shared common-variant architecture	Direction, mediation or individual comorbidity
Mendelian randomization	Causal inference under instrument assumptions	Freedom from pleiotropy/model misspecification
Polygenic score	Relative liability in a target sample	Diagnostic certainty or equitable portability

Twin-based heritability summarized by Watson and colleagues was 50–60% (Watson 2019, PMID 31308545). A review of the twin and family literature quotes a wider band — genetic factors predisposing for approximately 33–84% of AN liability — which is a reminder that "heritability of AN" is a range of study-specific estimates rather than a constant (Donato 2022, PMID 36479493). Heritability is population- and environment-specific; it is not the proportion of one person's illness "caused by genes."

That same review notes eating disorders affect up to 5% of the population in industrialized countries while remaining under-detected and under-diagnosed, which matters genetically: ascertainment through specialist services selects the cases that enter discovery samples (Donato 2022, PMID 36479493).

Twin and family evidence

The heritability band quoted above is the summary of a small number of population-based twin analyses, each with wide intervals. The Swedish Twin Registry study that anchors most citations screened 31,406 twins born 1935–1958 and identified AN by clinical interview, hospital discharge diagnosis or death certificate. It reported lifetime prevalence of 1.20% in women and 0.29% in men, and estimated additive genetic effects at a² = 0.56 (95% CI 0.00–0.87), with shared environment c² = 0.05 (95% CI 0.00–0.64) and unique environment e² = 0.38 (95% CI 0.13–0.84) (Bulik 2006, PMID 16520436). The confidence interval on the genetic component spans essentially the whole possible range: the point estimate of 56% is the number that gets quoted, and the interval that makes it uninformative on its own is the number that does not. The same cohort found neuroticism measured roughly three decades before diagnostic assessment predicted later AN (OR 1.62, 95% CI 1.27–2.05, p<0.001), and that people with lifetime AN were less likely to become overweight (OR 0.29, 95% CI 0.16–0.54) — an observation that anticipated the negative genetic correlation with BMI found later by GWAS.

Twin work also bears on whether AN is the extreme of a continuum or a demarcated entity. In 1,481 female Swedish twin pairs assessed at age 18, heritability of the continuous Eating Disorder Inventory-2 score was 0.65 (95% CI 0.61–0.68) and DeFries-Fulker group heritabilities at successive extreme cut-offs were essentially unchanged (0.59–0.65) — consistent with dimensional aetiology. But the twin-based genetic correlation between EDI-2 score and a registered AN diagnosis was only 0.26 (95% CI 0.08–0.42), against 0.52 (95% CI 0.39–0.65) for other eating-disorder diagnoses. The authors' conclusion is the interesting one: some eating disorders behave like the tail of a continuous distribution, while AN "might be more distinctly genetically demarcated" from general-population eating features (Dinkler 2021, PMID 31843035). This is an open dispute, not a settled point — continuum and category models are both alive in the twin literature (Thornton 2011, PMID 21243474).

2019 GWAS landmark

The Psychiatric Genomics Consortium/ANGI analysis combined 16,992 cases and 55,525 controls and identified eight genome-wide significant loci. Genetic correlations involved other psychiatric disorders, physical activity, metabolic traits, lipids and anthropometric traits; correlations with metabolic traits persisted after accounting for common BMI-associated variants (Watson 2019, PMID 31308545).

Finding	Interpretation	Unresolved step
Eight loci	Reproducible common-variant signals at that sample size	Effector gene and tissue for each locus
Psychiatric correlations	Partly shared liability	Which pathways are causal
Physical-activity correlation	Shared architecture with activity traits	Volitional, neurobiological and metabolic mediation
Glycaemic/lipid correlations	Metabolic biology is relevant	State-independent mechanisms and direction
BMI-independent component	Not reducible to generic thinness genetics	Translation to target or biomarker
SOX5 locus (2026 meta-analysis of European and Finnish data)	An additional genome-wide significant signal beyond the 2019 eight	Replication in an independent AN case series (Song 2026, PMID 41927769)

What came before 2019

The eight-locus result is the third wave, and the earlier waves constrain how it should be read. The Psychiatric Genomics Consortium's 2017 meta-analysis of 12 case-control cohorts (3,495 cases, 10,982 controls) found a single genome-wide significant locus on chromosome 12 (rs4622308), in a region already implicated in type 1 diabetes and autoimmune disease, and estimated common-variant heritability at h²_SNP = 0.20 (SE 0.02). The metabolic signature that defines the current model was already visible: significant positive genetic correlations with schizophrenia, neuroticism, educational attainment and HDL cholesterol, and significant negative correlations with BMI, insulin, glucose and lipid phenotypes across 159 tested traits (Duncan 2017, PMID 28494655). Watson 2019 doubled the case count; the correlation structure did not change, which is why it is treated as reproducible rather than as a single-study finding.

An exome-chip GWAS in 2,158 cases and 15,485 ancestrally matched controls tested low-frequency and rare variation directly and returned nothing at genome-wide significance; its two most notable common signals were rs10791286 in OPCML (p = 9.89 × 10⁻⁶) and intergenic rs7700147 (p = 2.93 × 10⁻⁵). The study was powered to detect low-frequency variants of large effect, so the negative result is informative: within that search space, AN appears not to have them (Huckins 2018, PMID 29155802).

Age of onset is a partly separate genetic question

A secondary analysis of the PGC AN GWAS (9,335 cases, 31,981 screened controls) found that age of onset itself is barely heritable at the common-variant level (SNP-h² 0.01–0.04), while early-onset AN (<13 years) and typical-onset AN were each substantially so (0.16–0.25 and 0.17–0.25). The two onset groups had distinct genetic correlation profiles: early-onset AN correlated with younger age at menarche, typical-onset AN negatively with anthropometric traits, and Mendelian randomization supported a causal path from younger menarche to early-onset AN (Watson 2022, PMID 36324647). If replicated, this splits what is currently one diagnosis into at least two partially separable liabilities — with the implication that pooling onset ages in discovery samples dilutes both signals.

The 2026 integrative analysis

A meta-analysis of AN GWAS data across European and Finnish populations, combined with multi-trait analysis of GWAS (MTAG), local genetic correlation, Mendelian randomization, co-expression network analysis and single-cell transcriptomics, reported: one novel genome-wide significant locus near SOX5; 185 genomic regions of significant local genetic correlation between AN and correlated phenotypes, with 100 loci showing pleiotropy across multiple traits; 86 MTAG-significant loci (34 overlapping the local-correlation results) including 25 novel ones such as VAMP2 (17p13.1), LPL (8p21.3) and BDNF (11p14.1); gene modules spanning synaptic signalling and lipid metabolism; and concentration of genetic risk in limbic and striatal GABAergic neurons with extension to motor cortical regions (Song 2026, PMID 41927769).

Three cautions belong with that list. MTAG loci are not equivalent to loci discovered in AN cases — they borrow power from genetically correlated traits and inherit those traits' confounding. Local genetic correlation identifies shared architecture, not shared mechanism. And single-cell enrichment maps where the implicated genes are expressed, which is not the same as where the causal biology acts. The analysis nonetheless does something the 2019 paper could not: it moves the metabo-psychiatric claim from trait-level correlation toward named genes (LPL for lipid handling, BDNF and VAMP2 for synaptic function) and a cell type.

Pleiotropy and cross-disorder structure

Cross-disorder genomic analyses show widespread pleiotropy among psychiatric disorders, cautioning against one-locus/one-diagnosis narratives. Analysing 232,964 cases and 494,162 controls across anorexia nervosa, ADHD, autism, bipolar disorder, major depression, OCD, schizophrenia and Tourette syndrome, the Cross-Disorder Group found meaningful structure among the eight disorders (three groups of inter-related conditions) and detected 109 loci associated with at least two disorders, including 23 with pleiotropic effects on four or more and 11 with antagonistic effects across disorders (PMID 31835028). Genetic-correlation atlases similarly demonstrate that correlated traits can share common-variant architecture without being clinically interchangeable: cross-trait LD Score regression estimated 276 genetic correlations among 24 traits, including AN with schizophrenia and AN with obesity, at a time when no SNP had reached genome-wide significance for AN at all (Bulik-Sullivan 2015, PMID 26414676).

The pace of that last point is worth recording. In 2015 there were no genome-wide significant AN loci; a 2017 review counted three completed GWAS yielding one significant locus and noted candidate-gene meta-analyses implicating serotonin genes (Baker 2017, PMID 28940168); 2019 produced eight; and by 2026 a multi-trait analysis reports 86 (Watson 2019, PMID 31308545; Song 2026, PMID 41927769). The increase largely reflects larger samples and the borrowing of power across correlated traits; it should not be read as 86 independently validated AN mechanisms. The 2026 analysis reports pleiotropic effects across multiple traits at 100 of 185 locally correlated regions (Song 2026, PMID 41927769).

A related reconceptualization proposes that AN and obesity may represent "metabolic bookends" — the argument rests on significant negative genetic correlation between the two conditions — and raises, without answering, whether they are also "microbiome bookends", given reported gut dysbiosis in AN that may itself follow from a nutrient- and energy-deprived gut environment (Bulik 2019, PMID 31056797). The bookend framing is a hypothesis about extreme weight dysregulation at both ends of a spectrum, not an established mechanism.

Rare variants: small studies, unreplicated signals

The common-variant picture is well powered; the rare-variant picture is not. Three lines exist, none replicated at scale.

Finding	Sample	Result	Status
BBOX1 burden	183 AN patients with clinical whole-exome sequencing vs gnomAD	Rare damaging variants in 12/183 (6.6%) vs 4.4 expected (2.4%); OR 2.86, p = 0.0117 (Lutter 2025, PMID 40665398)	Single-centre, no independent replication; BBOX1 is required for carnitine synthesis and hence long-chain fatty-acid β-oxidation
NNAT (Neuronatin) variants	Two families plus 8 male and 144 female AN cases	Nonsense p.Trp33* and a 5'UTR variant in probands; variants in 40.0% of males and 6.25% of females screened (Lombardi 2019, PMID 30933048)	Screening study without a matched control cohort — carrier frequencies cannot be interpreted as excess
Structural/repeat variation at known loci	Targeted nanopore sequencing of 200 kb around each of the eight GWAS loci in 10 AN cases	20 prioritized non-coding variants, including a polymorphic SVA-D element between IP6K2 and PRKAR2A and a poly-T short tandem repeat in the FOXP1 3'UTR (Berthold 2024, PMID 39741260)	Explicitly not powered for functional effects; a methods demonstration

Exome sequencing in one multiplex family likewise supported a multigenic rather than monogenic architecture (Bienvenu 2019, PMID 31388831). The honest summary is that no rare variant has been established as an AN risk factor, and the negative low-frequency result from the exome-chip GWAS argues against large-effect variants being common in this space (Huckins 2018, PMID 29155802).

Polygenic scores and the BMI paradox

Polygenic scores are being tested as prognostic rather than diagnostic tools. In 2,843 Swedish register patients with lifetime AN followed a mean of 5.3 years (up to 16), polygenic scores for AN and for schizophrenia were not robustly associated with clinical impairment after multiple-testing correction; only the BMI polygenic score was, and it survived at every p-value threshold (β = 1.30, 95% CI 0.72–1.88, p = 1.2 × 10⁻⁵). Two features deserve attention. First, the association reversed sign relative to expectation: BMI and AN are negatively genetically correlated at the case-control level, yet higher polygenic liability to high BMI predicted greater severity within cases. Second, the result did not survive a change of method — PRS-CS scoring gave inconsistent results for every score tested (Johansson 2022, PMID 35173158). Both facts point the same way: current AN polygenic scores are underpowered for individual prediction, and any reported effect should be checked against an alternative scoring algorithm before it is believed.

Polygenic liability also interacts weakly, if at all, with the best-established environmental risk factor. In up to 63,989 women in the Norwegian MoBa cohort, all four types of self-reported childhood maltreatment were strongly associated with eating disorders (ORs 1.71–3.29) and eating-disorder polygenic scores much more weakly (ORs 1.05–1.31); there were no multiplicative interaction effects, and only small additive interactions in exploratory analyses (Bjørndal 2026, PMID 42471970). Gene–environment interaction is frequently invoked in AN aetiology; in the largest test available it was not found.

Epigenetics: reversible marks, small samples

Two systematic reviews frame the field the same way. The first identified only 18 articles and conference abstracts through 2017, mostly candidate-gene methylation studies in very small and overlapping samples, and judged the results inconclusive and exploratory (Hübel 2019, PMID 30353170). The second, through May 2023, found 23 papers and five epigenome-wide association studies, and drew the substantive conclusion that malnutrition-induced methylation changes appear at least partly reversible on recovery (Käver 2024, PMID 38849516).

The primary data behind that conclusion are worth stating precisely, because "reversible" is doing real work. A longitudinal EWAS in 75 actively ill, 31 remitted and 41 non-eating-disordered women found 58 differentially methylated sites between active illness and controls (Q < 0.01) and 265 probes distinguishing remitted from active cases, mapping to lipid and glucose metabolism, serotonin and insulin signalling and immune function — with the direction of effect in remitted participants tending to be opposite to that in active illness, and illness chronicity correlating inversely with methylation at 64 sites (Steiger 2019, PMID 30693739). A larger replication in 145 active, 49 one-year-remitted and 64 non-eating-disordered women found 205 sites separating active illness from controls and 162 separating active from remitted, again implicating psychiatric, metabolic and immune genes (SYNJ2, PRKAG2, STAT3, CSGALNACT1, NEGR1, NR1H3), with lower BMI and longer illness associated with more pronounced alteration and remission with normalization (Steiger 2023, PMID 35703085). An independent methylation study using two reference-free cell-composition methods and five monozygotic AN-discordant twin pairs confirmed hypermethylation at TNXB but found the direction of effect at NR1H3 opposite to a previous report (Kesselmeier 2018, PMID 27367046).

That last discrepancy is the field in miniature: the same gene, opposite directions, in samples of a few dozen. Methylation in AN is currently better evidence for starvation as an epigenetic exposure than for epigenetics as an aetiological mechanism, and the state–trait problem is not solved by measuring a mark that itself changes with weight.

Microbiome: transfer experiments and one human trial

Gut dysbiosis in AN is consistently reported, and the "microbiome bookends" question raised above (Bulik 2019, PMID 31056797) has since been tested by transferring human microbiota into germ-free animals. Such transfer can show that donor microbiota transmits phenotypes in an animal model; it cannot establish whether dysbiosis preceded or followed illness in the human donors.

Study	Design	Result
Hata 2019, PMID 31504398	Germ-free mice colonized with faeces from 4 restricting-type AN patients vs 4 healthy controls	Recipients of AN microbiota gained less weight, ate less, and had a lower food-efficiency ratio (weight gain per unit intake); increased anxiety-like and compulsive behaviour; lower brainstem serotonin. Bacteroides vulgatus administration reversed compulsive behaviour but not weight
Gabriel-Segard 2025, PMID 41239983	Germ-free BALB/c mice colonized from characterized AN patients vs controls	Transmission of food restriction, anxiety-like behaviour, physical hyperactivity and elevated inflammatory response, plus liver dysfunction and disrupted ovarian follicles
Kooij 2024, PMID 38934721	Faecal transplant from AN patients into rats, behavioural flexibility endpoint	Did not alter flexible behaviour — a negative result in the same paradigm
Panah 2026, PMID 41535289	Open-label single-FMT feasibility trial in adult women with AN (NCT05834010)	18/22 completed; 19/22 chose oral capsules; no serious adverse events; microbiota shifted significantly toward donor composition at one week; no significant change in psychopathology or appetite-related biomarkers

The animal studies show that transferred microbiota can alter some AN-relevant phenotypes, while a rat study was negative on behavioural flexibility; the different endpoints prevent treating it as a direct replication. The human trial was explicitly a feasibility study and was not designed to detect symptom change; the ClinicalTrials.gov record lists 17 actual enrolment against the 22 recruited in the report, so registry and publication do not fully agree. A placebo-controlled 20-participant adolescent FMT feasibility trial has since been published as a protocol, without results (Couturier 2026, PMID 42009386). Nothing here supports microbiome-directed treatment of AN outside trials.

Clinical non-readiness

No current genetic result in this evidence set establishes a diagnostic test, predicts treatment response sufficiently for routine selection, or justifies screening asymptomatic individuals. Translation requires ancestry-diverse replication, functional mapping, calibrated absolute risk, prospective validation and evidence that acting on the result improves outcomes (Bulik 2022, PMID 35524137; Baker 2017, PMID 28940168).

Starvation and biology

Genotype is fixed, but most molecular phenotypes measured during illness are not. Weight loss, endocrine adaptation, activity and medication can alter metabolites, hormones, epigenetic marks and expression. Studies must distinguish inherited liability from state-dependent consequences through longitudinal and family-based designs.

Open questions

• Which tissues and molecular pathways mediate the 2019 loci? Transcriptome- and single-cell-informed work now points to limbic and striatal GABAergic neurons and to synaptic-signalling and lipid-metabolism modules, but as of September 2026 no effector gene at any AN locus has been confirmed by functional experiment (Watson 2019, PMID 31308545; Song 2026, PMID 41927769).
• Are metabolic genetic correlations causal contributors, compensatory correlates, or horizontally pleiotropic (Bulik 2022, PMID 35524137)?
• How portable are polygenic findings across ancestry groups underrepresented in discovery samples? Both the 2019 GWAS and the 2026 meta-analysis draw on European and Finnish data; as of September 2026 no genome-wide significant AN locus has been reported from a non-European discovery sample (Watson 2019, PMID 31308545; Song 2026, PMID 41927769).
• Can genetics identify treatment modifiers rather than merely case-control liability?
• Is AN the extreme of a continuously distributed eating-disorder liability or a genetically demarcated entity? Twin data support the continuum model for other eating disorders but give AN a twin-based genetic correlation with general-population eating features of only 0.26 (95% CI 0.08–0.42) (Dinkler 2021, PMID 31843035).
• Do early-onset (<13 y) and typical-onset AN have separable genetic architectures, and does menarcheal timing sit on the causal path (Watson 2022, PMID 36324647)?
• Why does polygenic liability to higher BMI predict greater within-case severity when BMI and AN are negatively genetically correlated at case-control level (Johansson 2022, PMID 35173158; Duncan 2017, PMID 28494655)?
• Is the gut microbiome a contributor or a consequence? Two germ-free transfer experiments reproduce AN-like phenotypes and one rat study finds no behavioural effect; the only human FMT trial was a feasibility study with no psychopathology endpoint met (Hata 2019, PMID 31504398; Gabriel-Segard 2025, PMID 41239983; Kooij 2024, PMID 38934721; Panah 2026, PMID 41535289).
• Do any of the reported rare-variant signals (BBOX1, NNAT) survive replication in an independent, control-matched cohort (Lutter 2025, PMID 40665398; Lombardi 2019, PMID 30933048)?

Related pages

• Neurobiology and cognition — candidate intermediate phenotypes.
• Medical complications — state-dependent metabolic consequences.
• Neuromodulation and experimental therapy — unproven translation.

References

1. Watson HJ, et al. Genome-wide association study identifies eight risk loci and implicates metabo-psychiatric origins for anorexia nervosa. Nat Genet. 2019. PMID 31308545.
2. Bulik CM, et al. Genetics and neurobiology of eating disorders. Nat Neurosci. 2022. PMID 35524137.
3. Bulik CM, et al. Reconceptualizing anorexia nervosa. Psychiatry Clin Neurosci. 2019. PMID 31056797.
4. Baker JH, et al. Genetics of anorexia nervosa. Curr Psychiatry Rep. 2017. PMID 28940168.
5. Cross-Disorder Group of the Psychiatric Genomics Consortium. Genomic relationships, novel loci, and pleiotropic mechanisms across eight psychiatric disorders. Cell. 2019. PMID 31835028.
6. Bulik-Sullivan B, et al. An atlas of genetic correlations across human diseases and traits. Nat Genet. 2015. PMID 26414676.
7. Donato K, et al. Gene variants in eating disorders: focus on anorexia nervosa, bulimia nervosa, and binge-eating disorder. J Prev Med Hyg. 2022. PMID 36479493.
8. Song Y, et al. Integrative GWAS identifies novel loci and genetic links between psychiatric and metabolic factors in anorexia nervosa. Mol Psychiatry. 2026;31:4679-4689. PMID 41927769.
9. Bulik CM, et al. Prevalence, heritability, and prospective risk factors for anorexia nervosa. Arch Gen Psychiatry. 2006;63:305-312. PMID 16520436.
10. Thornton LM, et al. The heritability of eating disorders: methods and current findings. Curr Top Behav Neurosci. 2011;6:141-156. PMID 21243474.
11. Dinkler L, et al. Association of etiological factors across the extreme end and continuous variation in disordered eating in female Swedish twins. Psychol Med. 2021;51:750-760. PMID 31843035.
12. Duncan L, et al. Significant locus and metabolic genetic correlations revealed in genome-wide association study of anorexia nervosa. Am J Psychiatry. 2017;174:850-858. PMID 28494655.
13. Huckins LM, et al. Investigation of common, low-frequency and rare genome-wide variation in anorexia nervosa. Mol Psychiatry. 2018;23:1169-1180. PMID 29155802.
14. Watson HJ, et al. Common genetic variation and age of onset of anorexia nervosa. Biol Psychiatry Glob Open Sci. 2022;2:368-378. PMID 36324647.
15. Johansson T, et al. Polygenic association with severity and long-term outcome in eating disorder cases. Transl Psychiatry. 2022;12:61. PMID 35173158.
16. Bjørndal LD, et al. Investigating relationships between genetic risk, childhood maltreatment, and eating disorders in women. Biol Psychiatry Glob Open Sci. 2026;6:100761. PMID 42471970.
17. Lutter M. Patients with anorexia nervosa have an increased burden of rare, damaging mutations in the BBOX1 gene. J Eat Disord. 2025;13:140. PMID 40665398.
18. Lombardi L, et al. Anorexia nervosa is associated with Neuronatin variants. Psychiatr Genet. 2019;29:103-110. PMID 30933048.
19. Berthold N, et al. Nanopore sequencing as a novel method of characterising anorexia nervosa risk loci. BMC Genomics. 2024;25:1262. PMID 39741260.
20. Bienvenu T, et al. Exome sequencing in a familial form of anorexia nervosa supports multigenic etiology. J Neural Transm (Vienna). 2019;126:1505-1511. PMID 31388831.
21. Hübel C, et al. Epigenetics in eating disorders: a systematic review. Mol Psychiatry. 2019;24:901-915. PMID 30353170.
22. Käver L, et al. Epigenetic alterations in patients with anorexia nervosa — a systematic review. Mol Psychiatry. 2024;29:3900-3914. PMID 38849516.
23. Steiger H, et al. A longitudinal, epigenome-wide study of DNA methylation in anorexia nervosa. J Psychiatry Neurosci. 2019;44:205-213. PMID 30693739.
24. Steiger H, et al. DNA methylation in people with anorexia nervosa: epigenome-wide patterns in actively ill, long-term remitted, and healthy-eater women. World J Biol Psychiatry. 2023;24:254-259. PMID 35703085.
25. Kesselmeier M, et al. High-throughput DNA methylation analysis in anorexia nervosa confirms TNXB hypermethylation. World J Biol Psychiatry. 2018;19:187-199. PMID 27367046.
26. Hata T, et al. The gut microbiome derived from anorexia nervosa patients impairs weight gain and behavioral performance in female mice. Endocrinology. 2019;160:2441-2452. PMID 31504398.
27. Gabriel-Segard T, et al. Anorexia nervosa symptoms are induced after specific gut microbiota dysbiosis transfer in germ-free mice. Gut Microbes. 2025;17:2563701. PMID 41239983.
28. Kooij KL, et al. Fecal microbiota transplantation of patients with anorexia nervosa did not alter flexible behavior in rats. Int J Eat Disord. 2024;57:1868-1881. PMID 38934721.
29. Panah FM, et al. Impact of a single fecal microbiome transplantation in adult women with anorexia nervosa: an open-label feasibility pilot trial. Nat Commun. 2026;17:1747. PMID 41535289.
30. Couturier J, et al. Protocol for a pilot feasibility randomised controlled trial of fecal microbiota transplantation for adolescent anorexia nervosa. BMJ Open. 2026;16:e109115. PMID 42009386.