Genomics-based identification of a potential causal role for acylcarnitine metabolism in depression

Altered metabolism of acylcarnitines, transporting fatty acids to mitochondria, may represent a mechanism linking cellular energy dysfunction to depression. We examined the potential causal role of acylcarnitine metabolism in depression by leveraging genomics and Mendelian randomization. Summary statistics were obtained from large GWAS: the Fenland Study (N= 9,363), and the Psychiatric Genomics Consortium (246,363 depression cases and 344,901 controls). Two-sample Mendelian randomization analyses tested the potential causal link of 15 endogenous acylcarnitines with depression. In univariable analyses, genetically-predicted lower levels of short-chain acylcarnitines C2 (Odds Ratio [OR] 0.97, 95% Confidence Intervals [CIs] 0.95-1.00) and C3 (OR 0.97, 95%CIs 0.96-0.99) and higher levels of medium-chain acylcarnitines C8 (OR 1.04, 95%CIs 1.01-1.06) and C10 (OR 1.04, 95%CIs 1.02-1.06) were associated with increased depression risk. No reverse potential causal role of depression genetic liability on acylcarnitines levels was found. Multivariable analyses showed that the association with depression was driven by the medium-chain acylcarnitines C8 (OR 1.04, 95%CIs 1.02-1.06) and C10 (OR 1.04, 95%CIs 1.02-1.06), suggesting a potential causal role of these acylcarnitines in the risk of depression. Causal estimates for C8 (OR=1.05, 95%CIs=1.02-1.07) and C10 (OR=1.05, 95%CIs=1.02-1.08) were confirmed in follow-up analyses using genetic instruments derived from a GWAS meta-analysis including up to 16,841 samples. Accumulation of medium-chain acylcarnitines is a signature of inborn errors of fatty acid metabolism and age-related metabolic conditions. Our findings point to a link between altered mitochondrial energy production and depression pathogenesis. Acylcarnitine metabolism represents a promising access point for the development of novel therapeutic approaches for depression.


INTRODUCTION
Depression is a major public health burden, due to its high prevalence, early onset, chronic nature, and unsatisfactory levels of treatment response (1). Developing more effective treatments requires a better comprehension of depression pathophysiology, which so far remains substantially elusive.
Defects in cellular metabolism and mitochondrial function have been proposed as a potential pathophysiological mechanism in depression. (2)(3)(4)(5) Cell metabolism dysfunctions are involved in brain processes (neurotoxicity, impaired neuroplasticity) related to neuropsychiatric diseases including depression and schizophrenia, (6,7) and linked to inflammation and metabolic conditions (e.g. insulin resistance) commonly found in depression. (2,8) Acylcarnitines (ACs) are biogenic compounds formed from the transfer of the acyl group of a fatty acyl-Coenzyme A to carnitine, and serve as central cogs in the mitochondrial machinery. Dysregulation of AC metabolism has been linked to impaired β -oxidation of fatty acids, thereby reducing mitochondrial energy production.(9,10) Recently, depression has been linked to AC metabolism through studies (11,12) showing altered circulating AC levels in subjects with Major Depressive Disorder as compared to healthy controls. A recent metabolomics study (13) on >1,000 subjects identified lower levels of the medium-chain acylcarnitines decanoylcarnitine (C10) and dodecanoylcarnitine/laurylcarnitine (C12) in participants with elevated depressive symptoms as compared to controls. Furthermore, ACs profiles have been shown to correlate with depression severity and variation in response to treatment with SSRIs and Ketamine (14-16).
Despite this preliminary evidence, whether the observed alterations of AC levels have a role in depression onset or are a consequence of the disease is unknown. The few available and mainly cross-sectional observational studies preclude making any causal inferences because . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021. ; https://doi.org/10.1101/2021. 10.18.21265157 doi: medRxiv preprint observational associations may emerge as a result of (1) residual confounding from several biopsychosocial factors such as physical inactivity, poor dietary habits, or medication use, impacting AC levels and depression, or (2) reverse causation, with subclinical depression influencing AC levels via behavioral and lifestyle-related (e.g. reduction in physical activity, worsening of dietary habit) mechanisms secondary to the ensuing disorder.
Genomics provides unique opportunities to investigate causality between traits applying new statistical tools such as Mendelian randomization (MR)(17) to results from large genomewide association studies (GWAS). MR is a technique inferring causality by leveraging genetic variants as proxy instruments, which are less likely related to confounders due to random allele segregation and not reversely affected by the phenotype. Previous metabolite GWAS (mGWAS) (18) showed that a limited number of genetic variants explained a large proportion circulating levels of ACs, a pattern reflecting a sparse genetic architecture with few biologically relevant loci harboring key genes for AC synthesis, metabolism and transport. This set of genetic variants with clear biological connotation and large effect sizes provide valuable instruments to examine the potential causal role of AC metabolism in the development of depression. In the present study, we therefore used results from recent large GWAS and applied MR analyses exploring the potential bidirectional causality between AC levels and depression.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)  SNPs. For the present analyses, we retained data from 15 ACs with at least two independent associated SNPs (eTable 1).

Genetic architecture of selected acylcarnitines
Full GWAS summary statistics of ACs were used to derive two parameters related to their genetic architecture. SNP-heritability (h 2 SNP , the proportion of trait variance explained by the joint effect of all genotyped common SNPs) was estimated using linkage-disequilibrium score regression (LDSC).(21) Pairwise genetic correlations (rg, determined by the number of SNPs and their level of concordance shared between two traits) between ACs were estimated . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

Mendelian randomization analyses
Two-sample Mendelian randomization (2SMR) analyses (23) based on GWAS summary statistics were performed to test the potential causal role of ACs on depression risk. For each AC used as exposure, genome-wide significant independent SNPs were used as instruments.
Selected SNPs were aligned on the positive strand for exposures and outcome (DEP). Fstatistics (all F>10, eTable 1) indicated that the strength of selected genetic instruments was adequate.(24) Lack of sample overlap across the two discovery GWAS reduced the likelihood of causal estimates biased toward the observational correlation. (25) First, a series of univariable 2SMR analyses were performed based on the inverse variance weighted (IVW) estimator, (26) pooling SNP-exposure/SNP-outcome estimates inversely weighted by their standard error. Since IVW assumes that all SNPs are valid instruments or that the sum of the directional bias is zero, the robustness of significant results was tested in sensitivity analyses based on weighted median and MR-Egger estimators. The weighted median (27)  Furthermore, heterogeneity among included SNPs was tested via Cochran's Q test, single . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 19, 2021. ; https://doi.org/10.1101/2021.10.18.21265157 doi: medRxiv preprint SNP, and leave-one-out SNP analyses. The presence of potential horizontal pleiotropy (a genetic instrument for exposure influencing the outcome by mechanisms other than exposure) was tested using the MR-Egger intercept (29) and the MR-PRESSO (pleiotropy residual sum and outlier) method(30) (supplemental methods). Finally, we performed reversed univariable 2SMR analyses testing the potential causal impact of depression liability on ACs circulating levels.
ACs significantly associated at nominal level with DEP risk in univariable analyses were carried forward in multivariable MR (MVMR) analyses, (31) in which the association of each exposure instrument was conditioned on the others. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021. ; https://doi.org/10.1101/2021.10.18.21265157 doi: medRxiv preprint (Table 1). MR-Egger estimates were not statistically significant, likely due to the limited number of genetic instruments, which were lower than those recommended (N ≥ 10) to achieve adequate statistical power.(28) Nevertheless, MR-Egger causal estimates were highly convergent with those obtained by other 2SMR analyses. Furthermore, heterogeneity across SNPs was not statistically significant (eTable 5). Inspections of single-SNP and leave-oneout-SNP analyses plots for C8 (eFigure 1-2), which had a Cochran's Q p-value=0.05, did not reveal the presence of outlier SNPs. MR-Egger intercept estimates and results from MR-PRESSO (eTable 6) did not show statistically significant evidence of horizontal pleiotropy.
Finally, we performed reversed 2SMR analyses examining the potential causal role of DEP on the levels of the 15 ACs. Figure 3 depicts estimates representing change in SD of (log) AC levels per doubling (2-fold increase) in the prevalence of the exposure (full results eTable 7).
None of the causal estimates were statistically significant, indicating lack of evidence for a reversed causal effect of depression liability on AC levels.

Multivariable Mendelian randomization analyses
Genetic instruments for C2, C3, C8 and C10 were applied in MVMR analyses conditioning the effect of each exposure instrument on the others. Since C8-C10 genetic correlation was equal to 1, we alternatively included them in two models with C2 and C3 (

Follow-up analyses
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021.

DISCUSSION
This study examined the potential causal role of AC metabolism in depression using recent genetic data and Mendelian randomization analyses. We found that genetically-predicted higher levels of the short-chain acetylcarnitine (C2) and propionylcarnitine (C3) were inversely associated with the risk of depression, while genetically-predicted higher levels of the medium-chain octanoylcarnitine (C8) and decanoylcarnitine (C10) were associated with an increased depression risk. In multivariable analyses, the association with higher depression risk was mainly driven by genetic instruments indexing higher levels of medium-chain ACs.
Furthermore, no evidence was found for a potential reversed association between depression liability as exposure and AC levels as outcome.
Present findings showed that circulating ACs are substantially influenced by genetics, with common SNPs explaining 14-30% of the population variance in AC levels (SNP-heritability). This is consistent with previous studies (18) showing large heritability estimates for these metabolites. Furthermore, the different ACs shared a substantial proportion their genetic liability, as indicated by moderate (across chain lengths) to strong (within short, medium, and long chain lengths) genetic correlations between ACs. These patterns of genetic overlap are in line with genetic association results replicated across many mGWASs(18,19,32,34,35).
These studies reported chain length-unspecific associations at loci harboring ACs transporters, such as the organic cation transporters SLC22A4 and SLC22A5, or co-enzymes involved in beta-oxidation, such as ETFDH. In contrast, enzymes with narrower substrate specificity, such as the short-, medium-, and long-chain acyl-CoA dehydrogenases ACADS, ACADM, and ACADL that are involved in beta-oxidation, show more specific association patterns for the respective AC species, resulting in the stronger genetic correlations observed here.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The main findings of the present study suggest that higher levels of medium-chain ACs, or the mechanism translating genetic variation to higher levels of these ACs, are potentially causally involved in the development of depression. In a broader perspective, the present findings are consistent with the hypothesis (2)  Furthermore, no evidence was found for a potential reversed causal role of depression in influencing AC levels. Other implicit properties of MR need to be taken into account when interpreting the present findings. Genetic variants index the average lifetime exposure to AC levels and, consequently, MR estimates derived from these instruments describe a potential average lifetime causal effect of ACs on depression and thus are unable to identify specific critical windows or acute events. Furthermore, since genetic instruments for depression were derived from large case-control GWAS, MR results provide information on potential causal mechanisms of disease onset rather than progression, which could involve different pathophysiological pathways. Finally, the causal relationship between ACs and depression identified in the present study requires confirmation across multiple causally-oriented methodologies (triangulation(54)), including further experimental and mechanistic studies in animals and humans.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021. ; https://doi.org/10.1101/2021.10.18.21265157 doi: medRxiv preprint A major strength of the present study is the application of 2SMR analyses leveraging novel GWAS summary statistics obtained from the largest international consortia. Nevertheless, despite the use of the largest available input data, the lack of statistically significant causal estimates may have resulted from insufficiently powered MR analyses, which may be determined by a combination of related factors, including the genetic architecture of the trait of interest, the sample size of the discovery GWAS, the number of independently associated SNPs and the strength of their associations. Biologically meaningful genetic instruments for ACs had adequate strength; this was confirmed in follow-up analyses in which stronger instruments derived from larger discovery GWAS summary statistics provided the same results as those obtained in the main analyses. Nevertheless, the number of independently associated genetic variants across ACs was low. In contrast, a higher number of genetic instruments for depression were available but are consistent with a polygenic architecture of weak effects scattered across the genome typical of all behavioral traits. Another limitation is that genetic instruments were derived from GWAS based on samples of European ancestry so results cannot be generalized to different populations. Thus, the present findings should be reevaluated in the future when results of larger and more ancestry-diverse GWAS will become available. Finally, we could not properly deconvolute depression heterogeneity, which may aggregate different dimensions characterized by distinct pathophysiology. We previously showed (8,55) that inflammatory and metabolic alterations, potentially related to mitochondrial dysfunctions(56), maps more consistently to depressive "atypical" symptoms characterized by altered energy intake/expenditure balance (e.g hyperphagia, weight gain, hypersomnia, fatigue, leaden paralysis) and anhedonia. Future studies in large cohorts with deeper phenotypes should investigate whether the genetic signature of ACs is differentially associated with various depression clinical features and symptom profiles.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021. ; https://doi.org/10.1101/2021. 10.18.21265157 doi: medRxiv preprint In conclusion, we found evidence of the potential causal role of AC metabolism on the risk of depression, in particular of high levels of medium-chain C8 and C10. Accumulation of medium-chain ACs is a signature of inborn errors of metabolism and age-related metabolic conditions. This suggests that altered cellular energy production and mitochondrial dysfunctions may have a role in depression pathogenesis. Acylcarnitine metabolism may represent a promising access point to depression pathophysiology suggesting novel therapeutic approaches.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021. ; https://doi.org/10.1101/2021.10.18.21265157 doi: medRxiv preprint Alzheimer's Gut Microbiome Project that contributed to acylcarnitine discoveries. She is an inventor on key patents in the field of Metabolomics and hold equity in Metabolon, a biotech company in North Carolina. In addition, she holds patents licensed to Chymia LLC and PsyProtix with royalties and ownership. The funders listed above had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the paper; and decision to submit the paper for publication.
B.W.D has received research support from Acadia, Compass, Aptinyx, NIMH, Sage, Otsuka, and Takeda, and has served as a consultant to Greenwich Biosciences, Myriad Neuroscience, Otsuka, Sage, and Sophren Therapeutics.
All the other authors declare no conflict of interest.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.    . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021. ; Odds ratios (ORs) and 95% confidence intervals (lCI, lower bound; uCI, upper bound) per SD increase in genetically-predicted levels of (log) ACs .

CC-BY-NC-ND 4.0 International license
It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021.  Odds ratios [ORs] and 95% confidence intervals per SD increase in genetically-predicted levels of (log) ACs. Full results eTable 4.  Odds ratios [ORs] and 95% confidence intervals per SD increase in genetically-predicted levels of (log) ACs. Full results eTable 8.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted October 19, 2021.