Linkage disequilibrium score (LDSC) regression analyses enable the examination of SNV-associated heritability and coheritability between traits. MR analyses permit the evaluation of possible causal relationships between two traits based upon Mendel’s law of independent inheritance, offering an opportunity for a natural randomized control trial (RCT) (20). LDSC and MR approach complement one another as strategies for exploring how to traits are related to one another. A restrictive cubic spline (RCS) strategy was also used with appropriate multivariate regression analyses as a means of examining relationships between exposures and outcomes to define optimal threshold values for exposures of interest (21) (Supplementary Figure 1). As the analyses performed herein were based upon publically available datasets, no further ethical oversight or informed consent were necessary.

The primary outcomes for this analysis were CVDs and longevity, with outcome data sources being provided in detail in Table 1. AF-related data were derived from a study performed by Nielsen et al. (22), with paroxysmal or permanent AF and atrial flutter included in the definition of AF (60,620 cases, 970,216 controls), while summary statistics for HF were derived from the largest published Genome-Wide Association Studies (GWAS) meta-analysis performed by the HF Molecular Epidemiology for Therapeutic Targets (HERMES) Consortium analyzing individuals of European ancestry (47,309 cases, 930,014 controls) (23). Participants in this study were individuals diagnosed with HF of any etiological basis determined based upon left ventricular ejection fraction (LVEF) (24). Summary-level CAD data were derived from the CAS Genetics (CARDIoGRAMplusC4D) Consortium (122,733 cases, 424,528 controls) (24). Stroke summary statistics for individuals of European ancestry (67,162 cases, 454,450 controls) including 67,162 AS, 60,341 AIS, 6,688 LAS, 9,006 CES, and 11,710 SVS cases, were derived from the MEGASTROKE consortium aimed at reducing bias resulting from population stratification (25). Summary-level MI data were derived from the CARDIoGRAMplusC4D (60,801 cases, 123,504 controls), MIGen, and CARDIoGRAM Exome consortia (42,335 cases, 78,240 controls), and ESP EOMI (4,703 cases, 5,090 controls) datasets (26).

Longevity analyses were performed with summary statistics derived from a recent GWAS meta-analysis of individuals of European ancestry in ∼20 population- or family based cohorts in Europe and the USA (27). Cases (n = 11,262) were individuals who lived to an age above the 90 or 99th percentile age based on cohort life tables from census data for the appropriate country, sex, and birth cohort. Controls (n = 25,483) were individuals who died at or before the 60th percentile age or whose age at the last follow-up visit was at or before the 60th percentile age.

Data pertaining to serum Ca (28), VD (29), and VK (30) concentrations for the included sources of exposure data are summarized in Table 1. Serum Ca and VD levels were derived from the UK Biobank Resource (Project #73697). All participants in the UK Biobank had provided informed consent, with oversight from the North West Multi-Centre Research Ethics Committee (11/NW/0382). VK1-related data were derived from the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium Nutrition Working Group (30). The largest GWAS study focused mainly on individuals of European ancestry was used to select genetic variants related to modifiable risk factors.

Instrumental variables for modifiable risk factors were determined with the Plink software through the use of the clump procedure. Considering that VK did not screen out independent sites at the threshold p < 5E–08, single nucleotide polymorphisms (SNPs) linked to risk factors were selected at the selected threshold for possible genome-wide significance (p < 1E–05). While for serum Ca and VD, the causal correlation didn’t change much when we used more stringent threshold (Supplementary Table 2). When linkage disequilibrium (r² > 0.1) for SNPs was evident for a given trait, SNPs that were most strongly associated with the exposure of interest based on the smallest measured p-value were selected. SNPs not included in CVD- or longevity-focused GWAS datasets were not included in this study. For the selected outcomes, the number of SNPs chosen as instrumental variables ranged from 8 to 448. These variants explained from 0.021 to 0.529% of phenotypic variation (Supplementary Table 3). To synchronize data between exposure and outcome GWAS, estimates of SNP effects were flipped with unrelated alleles and effects.

Cross-trait LDSC analyses were used to evaluate genetic correlations between pairs of phenotypes and genome-wide SNPs (31). LD scores for individual SNPs were determined in accordance with genotypes for common SNPs [minor allele frequency (MAF) > 0.01, Hardy–Weinberg equilibrium p > 1 × 10^–5] over a 10 Mb window when evaluating data derived from 503 European individuals included in the 1000 Genomes Project. The exact number of SNPs used in the genetic correlation analyses in each pair were shown in the Supplementary Table 1. LDSC analyses were then performed through the use of a weighted linear model via regressing Z-statistic products for two traits on LD scores across all variants throughout the genome. The resultant regression slow should provide an unbiased tool for estimating genetic correlations even if some individuals overlap between two GWASs. Bonferroni correction was used to correct for multiple testing, with a two-sided significance level of 0.0056 being established (0.05 divided by the nine included outcomes). Those associations exhibiting a p-value between 0.05 and 0.0056 were thought to be suggesting of a possible association. LDSC packages (32) in R version 4.0.2 were used for all analyses.

Causal estimates for the impact of genetically predicted serum Ca, VD, and VK levels on outcome variables were assessed with an inverse-variance-weighting (IVW) approach using a fixed-effects model. Weighted median, MR-Egger regression, and MR Pleiotropy RESidual Sum and Outlier (MR-PRESSO) strategies were also employed to improve the reliability and robustness of study conclusions. The weighted median approach assumes that a minimum of 50% of available information is based upon valid Ivs (32). The MR-Egger approach offers validity despite permitting the presence of invalid variants, but can yield wide confidence interval (CI) values (33). The MR-PRESSO approach enables researchers to detect and correct for any analyses detected through IVW linear regression analyses (34). Bonferroni correction of these results was used as above, with a p-value < 0.0056 as the significance threshold and a p-value between 0.05 and 0.0056 being indicative of a possible association. Odds ratios (ORs) are given for every 1 standard deviation (SD) difference in serum levels of Ca, VD, and VK. The TwoSampleMR (35) and MRPRESSO (36) packages in R version 4.0.2 was used for all analyses.

Analyzing potential pleiotropy is vital given that pleiotropic Ivs have the potential to have an indirect impact on study outcomes, serving as confounders of MR analysis efforts. A range of strategies were utilized herein in an effort to detect possible pleiotropy. Initially, heterogeneity among Ivs when utilizing the fixed-effects IVW approach was detected through Cochrane’s Q test. Lower levels of heterogeneity are indicative of the possibility that estimates between Ivs vary based on random chance, which can only occur when pleiotropic effects are not evident. In cases where significant heterogeneity was detected, a multiplicative random-effects IVW model would be implemented. An MR-Egger intercept was additionally performed, with a zero intercept (p > 0.05) being indicative of a lack of any pleiotropic bias. The MR-PRESSO method was additionally used for global heterogeneity testing and to detect horizontal pleiotropy (36). To assess the extent to which these associations were under the influence of any one SNP, a leave-one-out sensitivity analysis was conducted. Moreover, all SNPs included in the GWAS catalog database (37) were searched, with the goal of determining the association between those SNPs and risk factors pertaining to CVD incidence and longevity outcomes. After removing pleiotropic SNPs, causal associations were also analyzed.

Restrictive cubic spline analyses entail the use of a piecewise polynomial function capable of examining non-linear relationships between predictors and outcomes in a flexible manner (38). Here, spline models were adjusted for covariates including age, sex, body mass index (BMI), genotype batch, assessment center, and Townsend deprivation index (TDI) (39). Multivariate logistic analyses were used to examine relationships between serum Ca or VD concentrations and CVD incidence at the 25, 50, 75, and 95th centiles. When less than 20% of covariate data was absent, these missing values were accounted for through multiple imputations based upon five replicates and a chained equation method using the R MI procedure. Baseline categorical data were summarized across serum Ca and VD concentrations as percentages, while continuous variables were summarized using means and SDs. A two-sided p < 0.05 was the threshold of significance. R version 4.0.2 was used for all analyses.

When examining genetic correlations pertaining to serum Ca levels, the positive genetic correlation following Bonferroni correction was detected for MI [r_g (SE) = 0.890 (0.012); p = 1.356E–04] and CAD [r_g (SE) = 0.868 (0.014); p = 3.601E–04]. Serum Ca levels also exhibited a negative genetic correlation with AF, much as VD levels did with CAD [r_g (SE) = −0.061 (0.029); p = 0.035; p_adj = 0.350] (Figure 1 and Supplementary Table 1). No genetic correlations were detected when examining the relationship between VK (VK1, circulating phylloquinone concentrations) and any CVD subtypes. Results from SNV-based heritability testing suggest that these three tested exposures were unrelated to longevity (p_range = 0.722–0.900).

When conducting IVW MR analyses, higher VK1 levels were found to be strongly related to the risk of HF (p = 0.003, OR per 1-SD increase: 1.044; CI: 1.015–1.074) and LAS (p = 0.003, OR per 1-SD increase: 1.172; CI: 1.054–1.302). Potential associations were also observed between increases in VK1 levels and the risk of AS (p = 0.031, OR per 1-SD increase: 1.041; CI: 1.004–1.080). MR analyses also indicated a suggestive association between higher serum Ca concentrations and reduced odds of longevity (p = 0.014, OR per 1-SD increase: 0.865; CI: 0.770–0.971), with VD levels being significantly associated with reduced odds of longevity (p = 4.620E–04, OR per 1-SD increase: 0.777; CI: 0.674–0.895) (Figure 2 and Supplementary Table 3). Fixed-effects IVW estimates failed to reveal any causal associations between VK levels and longevity outcomes. Serum Ca or VD levels were also not found to be causally related to all tested CVDs.

To ensure that the causal inferences drawn from MR analyses are valid, it is critical that it be established that SNV-outcome relationships are the result of a given exposure and not the consequence of horizontal pleiotropy or a similar mechanism. Pleiotropy-resistant sensitivity analyses including Cochrane’s Q test, as well as MR-Egger intercept, MR-PRESSO, and leave-one-out sensitivity analyses were thus performed (Supplementary Figures 2, 3). Observed relationships between VK levels and HF/LAS/AS remained robust in these analyses (Cochrane’s Q_HF = 13.088, p_HF = 0.159; MR-Egger intercept, p_HF = 0.197; and MR-PRESSO, p_HF = 0.181; Cochrane’s Q_LAS = 3.631, p_LAS = 0.889; MR-Egger intercept, p_LAS = 0.010; and MR-PRESSO, p_LAS = 0.900; and Cochrane’s Q_AS = 4.678, p_AS = 0.791; MR-Egger intercept, p_AS = 0.953; and MR-PRESSO, p_AS = 0.804) (Table 2 and Supplementary Table 4). Consistent directionality was also evident for the relationships between increases in serum Ca/VD levels and longevity across all of these sensitivity analyses (Cochrane’s Q_calcium = 372.154, p_calcium = 0.983; MR-Egger intercept, p_calcium = 0.197; and MR-PRESSO, p_calcium = 0.983 and Cochrane’s Q_{vitamin D} = 26.940, p_{vitamin D} = 0.956; MR-Egger intercept, p_{vitamin D} = 0.339; and MR-PRESSO, p_{vitamin D} = 0.961) (Table 2 and Supplementary Table 4).

In total, serum Ca and VD level data in the UK Biobank were available for 429,863 and 448,777 individuals, respectively. Mean ages and gender distributions for these individuals are summarized in Supplementary Table 5. Individual VK data was not available with respect to CVD incidence. Through the use of an RCS regression analysis, curvilinear associations were detected between CVDs and both serum Ca and VD levels. All CVDs (CAD, AF, HF, MI, stroke) at the serum Ca plateau point (2.376 mmol/L, p < 0.001) were significantly different, whereas four CVDs (MI, stroke, HF, CAD) differed significantly at the VD plateau point (46.8 nmol/L, p < 0.001) (Figure 3).

Calcium is a divalent cation that plays essential roles in diverse physiological processes such as nerve excitation, muscle contraction, the mineralization of the skeleton, and coagulatory function (41, 42). In observational analyses, serum Ca concentrations have been shown to be positively correlated with CVD risk (43, 44). RCT-derived evidence suggests that Ca supplementation, which can lead to acute or persistently elevated serum Ca concentrations (45, 46), can result in a modest increase in the risk of MI and other cardiovascular events (47). An MR analysis of 184,305 participants (including 60,801 CAD cases, of which ∼70% had experienced MI, and 123,504 non-cases) revealed an association between a genetic predisposition toward elevated serum Ca levels and a higher risk of MI and CAD (48). LDSC analyses in this study confirmed a significant positive genetic relationship between serum Ca levels and both CAD (p = 3.601E–04) and MI (p = 1.356E–04). Insufficient evidence is currently available regarding the association between serum Ca and AF incidence, in line with the weak negative genetic correlation between these variables observed in this study. The inverse genetic associations between serum Ca and both CAD and MI, and serum Ca and AF may be due to the non-linear relationship between Ca and CVDs itself. Both genetic correlation analysis and MR are based on the assumption of linear relationship between serum Ca and CVDs, which may need to be further explained by observational data. However, our results of RCS just confirmed that there is a U-shaped relationship between them through individual observational data analysis.

Recent epidemiological evidence further suggests that circulating Ca levels are associated with CVD-related mortality and with longevity (43, 44, 49, 50). Specifically, elevated levels of serum Ca were linked to an increase in the odds of non-fatal CVD (HR = 1.12, 95% CI 1.10–1.14, MI: 1.19, 1.14–1.25) and fatal CVD (HR = 1.41, 95% CI 1.35–1.47; MI: 1.41, 1.31–1.51) (44). The MR analysis conducted herein also revealed a causal relationship between serum Ca levels and longevity (p = 0.014). Higher serum Ca levels were significantly associated with longevity and were negatively correlated with the incidence of AS, SVS, AIS, MI, and CAD, although these latter relationships were not significant. The combination of MR and LDSC results revealed a correlation between serum Ca levels, CVDs, and longevity.

Vitamin D plays an essential role in regulating Ca homeostasis, but the expression of nuclear VD receptor (VDR) by cardiomyocytes and vascular endothelial cells suggests that VD may be directly involved in the development and progression of CVD (51). These data thus prompted a more in-depth analysis of VD in addition to Ca.

In published studies, a 1.41-fold greater risk of CVD mortality (95% CI: 1.18–1.68) for individuals in the lowest plasma VD quintile based on a meta-analysis of prospective cohort studies. Acute VD deficiencies can contribute to inflammation and impaired insulin secretion, thereby increasing the odds of plaque rupture and arterial thrombosis. Chronically insufficient VD levels can contribute to increased arterial stiffness (52). Overall, VD deficiencies are detrimental to cardiovascular or longevity outcomes over any time scale. Observational results suggest that low levels of serum 25-hydroxyVD [25(OH)D], with is the primary form in which VD is stored, are related to an elevated risk of CVD incidence and mortality (53). VD deficiency has also been found to be associated with a more severe cardiovascular risk profile and increased CAD prevalence (54). VD was also shown to suppress NF-κB pathway signaling within cells to inhibit the progression of CAD, highlighting a possible mechanism whereby VD may mitigate vascular inflammation and atherosclerosis (55). The genetic and causal association analyses conducted herein revealed VD levels to be genetically related to CAD (p = 0.035) and causally related to longevity (p = 4.620E–04), confirming the association between VD exposure and these endpoints. An inverse relationship was observed between VD and CES, AS, HF, and AF incidence, but these relationships did not attain the level of statistical significance.

As a fat-soluble vitamin, VK is required for the activation of certain proteins and has been suggested to play some role in CVD incidence. Through anti-inflammatory activity that has been observed in vitro and in vivo, VK can potentially protect against vascular calcification, thus lowering the odds of CVD development and all-cause mortality (17). One observational prospective analysis of 601 individuals found lower VD and VK levels to be related to adverse cardiac remodeling and greater all-cause mortality risk (56). Conversely, a meta-analysis of three cohorts in the USA found VK1 levels to be related to all-cause mortality risk but unrelated to CVD (57). Circulating VK1 levels were also found not to be causally associated with CHD in a prior two-sample MR study (RR = 1.00, 95% CI: 0.98–1.04) (58). Here, analyses of different CVD subtypes revealed VK1 levels to be causally associated with HF (p = 0.003), AS (p = 0.031), and LAS (p = 0.003). However, epidemiological data pertaining to correlations between VK1 and various CVD subtypes are lacking at present, underscoring the need for further research focused on this topic and the underlying mechanisms that link VK levels between CVD or other health outcomes.

Calcium supplementation is a common practice in the USA, and there is rising clinical interest with respect to the association between these supplements and CVD. Some work suggests that Ca supplements may lower blood pressure and contribute to better serum lipid profiles, yet they also have the potential to increase serum Ca levels, thereby elevating the risk of vascular calcification and concomitant CVD event incidence. Perhaps unsurprisingly, prior research has yielded conflicting results with respect to the relationships between CVD and Ca supplementation (13). This issue is made more complex by the fact that many adults seek to improve their bone health through the combined intake of Ca and VD supplements despite the inconclusive evidence suggesting possible relationship between Ca intake and the risk of CVD (53). Some adverse effects have been reported in individuals utilizing supplemental VD and Ca alone or in combination, with these effects likely being attributable to the dose of supplemental Ca utilized (59). The present results suggest a genetic relationship between serum Ca, VD concentrations, and CVD incidence such that these serum biomarkers may offer value for the selection of appropriate nutritional interventions designed to mitigate CVD-related risk. Importantly, this study enabled the establishment of threshold Ca and VD concentrations in CVD patients and healthy controls, revealing that serum Ca and VD levels of 2.376 mmol/L and 46.8 nmol/L, respectively, were related to the lowest risk of CVD development among individuals of European heritage. When these levels fall too far above or below these levels, they may contribute to CVD development. A study of 441,738 individuals during a median follow-up time of 21 years found that serum Ca concentrations greater than 2.40 nmol/L were associated with increased risk of non-fatal CVD (44). Another study discovered L-shaped associations between VD level and CVD mortality. When VD concentrations were less than 27.70 nmol/L, the risk of death from CVD was increased. When VD concentrations exceeded 54.40 nmol/L, there was no association with all-cause mortality in America (60). Although no studies have yet provided exact epidemiological data on serum Ca and VD concentrations, the range of concentrations given by previous studies supports our results. At the same time, the Ca (2.376 mmol/L) and VD (46.8 nmol/L) plateau points of CVD are within the standard the normal clinical reference range. Accordingly, the intake of Ca or VD from primarily dietary sources may be most appropriate, reserving the minimum necessary Ca/VD supplementation for individuals dealing with Ca/VD intake deficiencies following the exhaustion of other forms of dietary modification (53).

When analyzing supplemental VK1 intake, following adjustment for confounding lifestyle and demographic factors, moderate-to-high VK1 intake levels (87–192 μg/days) were related to a decrease in the odds of all-cause [HR (95% CI): 0.76 (0.72, 0.79)], and CVD-related [HR (95% CI): 0.72 (0.66, 0.79)] (61). These findings confirmed that VK1 levels were related to AS, LAS, and HF. However, individual VK1 data were unavailable such that it was not possible to estimate the threshold levels necessary to minimize the risk of CVD development. As such, further research will be needed to provide specific guidance regarding supplemental VK1 dosing in different populations.

A major strength of this study is that the analyses of serum Ca-associated exposures for the nine included CVD types and overall longevity were performed using the largest GWAS datasets available. Causal inferences should generally be based upon several study types given that MR analyses are based on three major assumptions that are not always met or fully testable (62, 63). Genetic correlation analyses were thus used herein in an effort to complement MR-related research design limitations. Furthermore, these outcomes included both specific analyses for nine CVD subtypes as well as longevity as a control outcome for co-analyses, thereby strengthening the overall reliability of these findings. An additional strength of this approach is that the genetic instruments employed herein were selected based on a recent European population GWAS dataset for individuals with accessible serum Ca and plasma VD/K1 levels together with summary-level information regarding CVDs and longevity. These results are not likely to have been affected by population stratification bias given that these were GWAS data for individuals who were primarily of European ancestry. Lastly, correlation analyses of the associations between serum Ca and VD levels and the CVD risk factors enabled the estimation of the serum Ca and VD levels associated with the minimum CVD risk, thus enabling the establishment of recommended threshold levels for these nutrients aimed at mitigating the odds of CVD development.

There are certain limitations to this analysis. For one, the GWAS study used for these analyses was derived from a public database pertaining to a European population, and the results may thus not be applicable to populations of Asian, African, or other ancestries. In addition, individual-level VK data were not available for these European CVD patients, precluding the establishment of optimal concentrations of this vitamin for cardiovascular health.