Iron (Fe) and copper (Cu) are two essential mineral nutrients for human health through their vital roles in enzymatic reactions and cellular energy metabolism [1
]. Physiological processes that maintain the homeostasis of these minerals are often influenced by dietary loss, malabsorption, inflammation, infection, liver disease, and dysregulated erythropoiesis. Impaired systemic iron or copper homeostasis, including deficiency, excess, and even fluctuations in the normal ranges, could have clinical implications [3
]. Iron deficiency, the most widespread micronutrient deficiency worldwide, is well-established to cause anemia, while iron overload can lead to chronic liver disease, cirrhosis, and hepatocellular carcinoma [4
]. Copper is a cofactor of many redox enzymes, and it participates in iron metabolism either by competing with iron for binding ligands or through the action of various iron-regulating cuproenzymes [6
]. Copper is protective against iron deficiency anemia, and the mechanistic basis is relatively well-established [7
]. First, the efflux of iron from enterocytes, hepatocytes, and macrophages into the bloodstream requires the action of two copper-dependent ferroxidases (i.e., hephaestin and ceruloplasmin), which oxidize ferrous iron into the transferrin-binding ferric iron [8
]. In addition to enabling iron transport, copper is also required for hemoglobin biosynthesis, probably by assisting iron import into or utilization in mitochondria [9
]. These interactions of copper and iron likely underlie their shared associations with various diseases. For instance, epidemiological studies have reported that elevated blood levels of iron and copper are associated with a higher risk of type 2 diabetes, anemia, and osteoarthritis [10
]. Serum ferritin and plasma copper have been associated with an improved blood lipid profile, corresponding to a reduced risk of hyperlipidemia [12
]. However, conflicting associative patterns have also been reported [14
]. Since most existing studies were observational and often complicated by reverse causality and residual confounding, it is still unknown whether these associations indicate causal relationships.
Mendelian randomization (MR), a complementary approach to epidemiological observations, utilizes genetic variants as instrumental variables to approximate the lifetime status of an exposure (e.g., the blood level of a mineral) and evaluates its causal effect on a clinical outcome. The random allocation of alleles at conception and the natural direction of causality from genetic variants to phenotypes protect MR estimates from confounding and reverse causality [15
]. MR studies have been performed to evaluate the causality of specific mineral–outcome pairs, such as iron on stroke, coronary artery disease, and Parkinson’s disease [17
]; and copper on ischemic heart disease [20
]. Compared to these hypothesis-driven MR studies with an obvious bias toward cardiovascular diseases, a phenome-wide association study coupled with MR (PheWAS-MR) enables an unbiased and hypothesis-free scan through a wide range of phenotypes (i.e., phenome) and prioritized candidate clinical outcomes for MR causal inferences. PheWAS-MR has been conducted on iron, but not copper [21
]. Most importantly, few existing studies simultaneously examine multiple blood minerals at a phenome-wide scale to disentangle their confounded clinical effects [22
In this study, we systematically evaluated and compared the causal clinical effects of iron and copper. Genetic instruments for the blood levels of iron and copper were curated from existing genome-wide association studies (GWAS) with the largest sample sizes. Candidate clinical outcomes were identified based on a phenome-wide association study between these genetic instruments and a large number of disease outcomes from the UK Biobank (UKBB). All signals passing stringent correction for multiple testing were followed by MR analyses. Additionally, we examined the causal association between the iron or copper level and a subset of lipid profiles, which are essential biomarkers for lipid metabolism diseases, in a secondary analysis with an independent dataset.
Our study adopted a phenome-wide approach to systematically evaluate and compare the clinical effects of blood iron and copper. This is the first PheWAS-MR study for copper, and the first to perform systematic comparison across iron and copper. The MR approach reduces the biases from confounding and reverse causality, which affect most observational associations. This strategy utilizes genotype–exposure and genotype–disease associations to strengthen inferences between modifiable exposure and diseases, aiming to reduce disease risk in the population by modifying the exposure through lifestyle changes or clinical interventions. Our findings highlight the shared protective effects of iron and copper on lipid metabolism disorders and iron deficiency anemia. Some potential causal mineral–outcome relationships identified in this study have been well known and supported by existing mechanistic studies, while others are novel, awaiting further confirmation and future mechanistic exploration.
Most notably, we found that genetically predicted higher blood levels of iron and copper are both protective against lipid metabolism disorders and its two subcategories, hyperlipidemia and hypercholesterolemia. Consistently, they are also associated with lower blood TC and LDL cholesterol levels. This is the first MR study to establish the probable causal protective effect of copper on lipid metabolism disorders. The interplays among lipid, iron, and copper metabolism have been recognized, but they are complex and not fully elucidated. Epidemiological findings of the effect of copper on lipid metabolism are equivocal, with some studies showing negative associations of blood copper with both TC and LDL cholesterol [40
], and others showing positive associations [41
]. The relationship between iron status and blood lipids is similarly ambiguous. Some epidemiological studies found that blood iron is lower in hyperlipidemic patients [42
] and that blood ferritin is negatively associated with LDL cholesterol [13
]. Very likely related, elevated iron status is negatively associated with coronary artery disease in both observational and MR studies [43
]. Other studies have found that anemic and/or iron-deficient patients and animal models tend to have lower blood TC and LDL cholesterol [45
], while blood ferritin has often been found to be positively associated with an unhealthy lipid profile (i.e., elevated TC, LDL cholesterol, and triglyceride, but decreased HDL cholesterol) [46
]. These conflicting observations likely suffer from residual confounding and reverse causation, highlighting the importance of MR studies.
To gain mechanistic insights into how iron regulates lipid metabolism, we searched for iron-responsive elements in genes involved in cholesterol metabolism and fatty acid degradation using available databases [47
]. Multiple potential iron-responsive genes were identified, including LIPH
(Figures S1 and S2
). Additional mechanistic insights into the potential protective effects of blood iron on lipid metabolism disorders can be drawn from studies on rats. It has been shown that iron deficiency upregulates lipogenic genes but downregulates apolipoprotein H and genes involved in the mitochondrial beta-oxidation, resulting in increased circulating lipids [48
]. For copper, its deficiency has long been linked to increased risks of hyperlipidemia and cardiovascular diseases in both humans and animal models [49
], while copper supplementation in patients of hyperlipidemia was shown to improve the blood lipid profile [50
]. Multiple possible molecular mechanisms have been suggested for the effect of copper deficiency on lipid metabolism disorders, mainly from studies on copper-deficient rats. First, copper deficiency has been shown to increase the level of a key enzyme in the cholesterol synthesis pathway, hydroxymethylglutaryl-coenzyme A (HMG-CoA) reductase [49
]. An intestinal transcriptome analysis found that in copper-deficient rats, genes involved in mitochondrial and peroxisomal fatty acids beta-oxidation are down-regulated, and genes involved in plasma cholesterol transport are up-regulated [52
]. It was also observed that HDL apolipoprotein catabolism was increased, but HDL uptake did not change in the liver and adrenal gland, which are organs that can further metabolize cholesterol [53
]. Our study highlights the deficiency of iron and copper as likely causal risk factors for dyslipidemia and calls for future studies into their physiological mechanisms.
Well-known mineral–outcome relationships with established physiological mechanisms provide support for the power and validity of our study. Iron is an integral component of the oxygen-carrying hemoglobin and is required for the process of erythropoiesis [4
]. In agreement with this, we found that a genetically predicted higher level of blood iron is protective against anemia, while in women, it increases the risks of myeloproliferative disease, polycythemia vera, and secondary polycythemia. Myeloproliferative diseases are bone marrow and blood disorders featured by abnormal hematopoiesis, while its most common subtype, polycythemia vera, is characterized by erythrocytosis (i.e., excessive red blood cell production) [55
]. Very interestingly, iron deficiency without anemia is present in virtually all patients of polycythemia vera [56
]. Our findings suggest that iron deficiency is the result, rather than the cause, of erythrocytosis and reaffirm the current practice of not using iron supplements to treat patients of polycythemia vera. At the other extreme, iron overload can cause tissue damage, and its excessive accumulation in the liver leads to cirrhosis and hepatocellular carcinoma [3
]. Consistently, our study revealed that elevated blood iron in men is positively associated with risks of chronic liver disease and cirrhosis, and one of its subtypes, other chronic nonalcoholic liver disease.
Our study revealed a novel role of blood iron in increasing the risk of varicose veins. Varicose veins, a common venous disease of the lower extremity, is characterized by incompetent valves, reflux, and venous wall dilation. Its etiological process involves the hydrostatic-pressure-induced activation of matrix-degrading enzymes and inflammatory cascade [57
]. Iron overload and its causal HFE
genetic variations have been associated with the development of varicose veins [58
]. Mechanistically, iron overload induces oxidative stress and the hyperactivation of matrix-degrading enzymes [59
]. Other significant mineral–outcome relationships are also worth mentioning. We found that genetically predicted higher blood iron is associated with an increased risk of glossitis (i.e., tongue inflammation), which is consistent with what has been reported in a previous MR study [21
]. However, it is fairly well established that patients of atrophic glossitis frequently suffer from deficiencies of nutrients, including iron, and corresponding nutrient supplementation is able to resolve oral symptoms [60
]. The reconciliation of these disparate data is needed in future studies. Our findings open many new research avenues to elucidate the roles of blood iron and copper in these clinical conditions.
Our study has strengths and limitations. This is the first comparative PheWAS-MR of iron and copper, based on a large prospective cohort, to simultaneously evaluate their shared and unique causal clinical effects. Our approach is unbiased and hypothesis-free. In this study, only MR estimates with no indications of pleiotropy were reported. For selected outcomes (i.e., blood lipids), we replicated the results in an independent dataset. The sex-stratified analysis enabled the identification of sex-specific relationships. Our results confirmed some previous MR findings while making novel discoveries, most of which are supported by existing epidemiological and mechanistic studies. Our study has a number of limitations. The first assumption of MR is that the genetic instrument must be strongly associated with the exposure [61
]. We attempted to satisfy this assumption by using exposure-associated genetic variants at the genome-wide significance level. However, these instruments only explain a small portion of the phenotypic variance of these blood minerals. The instrumental variables we used explain ≈3.8% of the variance in blood iron status and 4.6% for blood copper in relatively small samples. This means that the analyses may be subject to weak instrument bias, which may reduce the statistical power to identify significant associations [62
]. However, our F and I2GX
statistics for all instruments suggest that our results were not substantially affected by weak instrument bias. Nevertheless, it would be essential to repeat these analyses using instruments from better powered GWAS with larger sample sizes. Additionally, we could not fully rule out the possibility that horizontal pleiotropy affected our results. Although the presence of horizontal pleiotropy can be examined or corrected using Cochran’s Q test, the MR-Egger tests, and WM MR, these methods usually require a large number of instrument SNPs [63
]. Without a large enough number of genetic instruments SNPs, some of our results need to be interpreted with caution. Still, we want to emphasize that our leave-one-out sensitivity analysis confirmed that our results were not driven by any specific variants.
Despite a large sample size (N = 310,999) of the overall cohort, the case numbers of specific outcomes are still small, limiting our statistical power for rare diseases and those with modest effects. Small sample sizes may also play a role in our sex-stratified analysis, with reduced statistical power in the male and female groups, compared to the combined group. It is possible that some negative results in one gender group are due to a lack of power. However, we focused on outcomes that are significant in one gender group but not significant in either the other gender group or the combined group. If the effects are consistent across the genders, they should have been detected in the combined analysis. Another related issue is that MR estimation assumes a linear relationship between the exposure and the outcome, and as a result, non-linear effects might have been missed. Given the well-known threshold effects of mineral deficiency and overload, future studies with a non-linear model will likely reveal more clinical effects. Genetic instruments approximate the average effects over the life course, while the physiological relevance of a blood mineral could vary by life stages and is not captured by our study. Finally, the UKBB is a middle-and old-aged cohort, and we restricted the analysis to those of European descent. Future studies in cohorts of younger ages or other ethnic backgrounds are needed to confirm our findings and to search for more clinical consequences.