Effects of BMI and inflammation on depression and treatment resistance
Depression is a highly prevalent mental health disorder, consistently ranked among the top three leading causes of disability worldwide [
1]. In people with a more severe presentation, prescription of an anti-depressant medication, coupled with psycho-social interventions, is recommended as first-line management [
2]. For research purposes, response can be quantified as a meaningful reduction in a symptom severity scale, such as the Composite International Diagnostic Interview (CIDI) or Patient Health Questionnaire-9 (PHQ9) and may differ among different antidepressant agents [
3]. It is recommended that, if a satisfactory symptom reduction cannot be achieved within the first 6 weeks, then treatment can be augmented with other agents, including other antidepressants, lithium and anti-psychotics [
4].
An inadequate response to at least two successive antidepressant medications, each administered for at least 6 weeks, is referred to as treatment resistant depression (TRD) and affects at least
\(7\%\) of those initially diagnosed with depression [
5]. Identifying contributors to treatment resistance early could potentially assist prompt management and guide appropriate interventions targeting other pathways that may predispose to treatment resistance. Recent advances in electronic health record analysis have allowed for the definition of a TRD phenotype in very large databases of routinely collected healthcare data, allowing linkage with existing genetic databases [
6].
Overweight and obesity have also been shown to predict the development of depression in multiple observational studies. This relationship could be explained by worsening physical health with obesity which may in turn affect mood. A meta-analysis of 15 prospective cohort studies estimated that being overweight was associated with a
\(27\%\) increase in the odds of subsequently developing depression. There is also evidence for a dose-response relationship, with obese individuals having higher risk for depression [
7]. This relationship could be partially explained by social stigma due to the negative perceptions of overweight/obesity in certain cultures [
8]. Another dimension of the effect is its potential appearance later in life. A meta-analysis reported a positive association only in adults older than 20 years of age but not in children and adolescents [
7]. Recent studies derived a binary classification of overweight (metabolically favourable and unfavourable adiposity) based on metabolic sequelae, namely hyperlipidemia, compromise in liver function, and sex hormone levels [
9,
10]. Whilst individuals with favourable adiposity face much less of the commonly described adverse effects of high adiposity, both phenotypes appeared to exert effects of similar magnitude on the risk of multiple depression outcomes [
10]. This was interpreted as a predominantly social, rather than biological, effect. Multiple studies have investigated the effects of weight loss on depressive symptoms in people with overweight or obesity. In general, caloric restriction, behavioural training, or supplements were used as interventions to encourage weight loss, and weight loss was found to reduce depressive symptoms in most studies, as collated in a systematic review [
11].
One aspect of the downstream metabolic consequences of overweight that is not explicitly captured by the phenotype of unfavourable adiposity is inflammation. Furthermore, evidence from genome wide association studies has suggested genetic variants important for cytokine and immune regulation predict major depressive disorder (MDD) [
12]. C-reactive protein (CRP) is a protein synthesised by the liver as part of the inflammatory response. Measurement of CRP in serum is a common part of investigations for inflammatory conditions, e.g. microbial infections and auto-immune conditions. Given a stable general medical status, CRP levels are largely stable and multiple observational studies have investigated its potential utility as a proxy for disease progression in infectious disease [
13] and autoimmune conditions. Despite its predictive value, its potentially causative role in driving the pathophysiological course of a disease has been disputed in multiple settings such as coronary heart disease [
14,
15]. In depression, recent work has indicated a higher CRP in 102 individuals with TRD compared with treatment-responsive patients and controls [
16].
Despite the advantage of large sample sizes and extensive phenotyping that UKB offers, additional care has to be taken to avoid the inherent limitations of observational data. As phenotypes may be correlated due to confounding rather than a true causal relationship, the measurement of observational associations alone may not reflect a causal mechanism [
17]. Mendelian randomisation (MR) is an epidemiological approach that employs genetic variants, most commonly single nucleotide polymorphisms (SNPs) as instrumental variables in order to circumvent environmental confounding [
18]. By genetically predicting the levels of an exposure such as CRP by a set of relevant SNPs, the proxied levels of CRP reflect a value of CRP that may not be affected by later-life influences that could distort the value (e.g. BMI, smoking, auto-immune conditions). Associations between genetically predicted CRP and an outcome of interest can then much more readily be interpreted as a causal effect [
19].
Previous work investigating the causal role of CRP, interleukin-6 (IL-6, major moderator of CRP), BMI and specific symptom dimensions of depression (sleep, appetite, suicidality) used LD score regression and a range of two-sample MR analyses. The results of this study did not find associations of CRP with any of the outcomes but report an association of IL-6 with suicidality [
20]. MR has also been used to investigate how BMI and fat mass affects mood outcomes [
12,
21‐
23].
The expansion of genome-wide association studies (GWAS) has lead to the discovery of multiple new significant causal associations. However, it is likely that many are false positives due to pleiotropy, the phenomenon whereby a SNP is an invalid IV due to exerting an effect on the outcome not through the exposure of interest. [
24]. Multivariable MR (MVMR) can be used to assess whether an exposure causally influences an outcome
conditional on a larger set of genetically instrumented exposures [
25,
26]. Incorporating additional exposures stops them from acting as pleiotropic pathways and because of this MVMR is seen to be more robust than univariable MR (UVMR). Indeed, a wealth of evidence exists CRP as a downstream consequence of high body mass. For example, a recent GWAS of serum CRP levels on 204,402 individuals found that adjusting for BMI significantly reduced the strength of association between CRP and well known obesity genes (
FTO [
27]
, TMEM18 [
28],
ABO, previously described genes for obesity) [
29].
In this paper, we aim to estimate the causal contributions of CRP and BMI on TRD and other depression phenotypes using a combination of UVMR, MVMR and causal mediation analyses. We further investigate whether these highly correlated exposures exert an independent effect on depression phenotypes or if their effect is mediated, and if these relationships are constant across age distributions.