The elusive quest for causality in adiponectin’s bimodal relationship with cardiovascular disease: Mendelian randomization meets Janus

This editorial refers to ‘Plasma adiponectin levels and risk of heart failure, atrial fibrillation, aortic valve stenosis, and myocardial infarction: large scale observational and Mendelian randomization evidence’, by M. B. Nielsen et al., https://doi.org/10.1093/cvr/cvad162. The identification of adiponectin nearly three decades ago as the most copious secretory product of adipocytes was a landmark, spurring seminal experimental work that came to delineate the insulin-sensitizing, anti-inflammatory, and anti-atherogenic properties of this adipokine.1 The prospects for therapeutic manipulation by repleting the deficient levels seemed simple, clear, and, given the obesity epidemic overtaking the globe, momentous. Consistent with the pre-clinical findings, population studies found inverse associations between plasma adiponectin and diabetes, and, in a middle-aged cohort, coronary heart disease (CHD).2 Among cohorts with established heart disease, kidney disease, or advanced age, however, a positive relationship with cardiovascular disease (CVD) and mortality was instead documented.2 This ‘adiponectin paradox’ was characterized in elders by a U-shaped association with cardiovascular events.3 The divergent associations painted a complicated picture for the adipokine, one befitting its complex biology. Adiponectin circulates in multimeric forms whose activities vary, as does that of its cleaved carboxy terminus, globular adiponectin.1 Membrane receptors AdipoR1 and AdipoR2, which are near-ubiquitously expressed, were determined to activate various intracellular signalling cascades to elicit favourable cardiometabolic actions.1 AdipoR1 and AdipoR2 also possess hydrolase activity, effecting disposal of the metabolically deleterious compound ceramide.1 Yet, T-cadherin has emerged as a key binding partner for the protein colocalizing with AdipoRs in the heart, vasculature, skeletal muscle, and fat.4 Responsible for adiponectin localization in these tissues, T-cadherin is a crucial accessory to another identified role of the adipokine, fostering exosome formation for cellular detoxification.4 Still, adiponectin’s blood concentrations exceed 103–106-fold those of other bioactive molecules engaged in high-affinity receptor signalling.2 Recognition that adiponectin could promote macrophage clearance of apoptotic bodies by mass action pointed to a broader housekeeping role.2 More recently, adiponectin has also been shown to bind anionic phospholipids and sphingolipids.5 Such low-affinity interactions would enable the adipokine to bind such harmful lipids for opsonization by phagocytes or possibly AdipoR hydrolysis, preventing their injurious effects.5 Meanwhile, adiponectin’s precise significance to human health remained uncertain. Leveraging randomly allocated alleles at conception as unbiased instruments of lifelong biomarker exposure, Mendelian randomization (MR) analysis represents a powerful technique to probe causality, contingent on certain assumptions. Major studies using one- and/or two-sample MR, however, failed to detect a consistent causal association between adiponectin and insulin sensitivity or diabetes6–8 (Table 1). Similarly, two-sample MR analyses of CVD did not find compelling evidence of causal associations between adiponectin and CHD, heart failure (HF), or atrial fibrillation (AF).8–10 Summary of major Mendelian randomization studies evaluating linear associations of adiponectin with cardiometabolic outcomes Multi-cohort ADIPOQ n ≤ 31 000 Fasting insulin Diabetes β 0.02 SD (−0.07, 0.11) per ↓APN SD OR 0.94 (0.75, 1.19) per ↓APN SD One sample CGPS ADIPOQ, CDH13 n = 96 903 Diabetes RR 1.07 (1.06, 1.09) per ↓log APN Two sample ADIPOGen, GERA, DIAGRAM, UK Biobank ADIPOQ + others n = 659 316 RR 1.26 (1.01, 1.57)b per ↓log APN Two sample CARDIoGRAM CARDIoGRAMplusC4D ADIPOQ (conservative approach) n = 86 895 n = 194 427 CHD −2.3% ΔOR (−4.1, −0.4) per APN-raising allele −0.6% ΔOR (−1.9, 1.0) per APN-raising allele Two sample DIAGRAM UK Biobank + CARDIoGRAMplusC4D UK Biobank Multiple cohorts UK Biobank ADIPOQ + others n = 898 130 n = 547 261 n = 977 323 n = 587 446 n = 401 937 Diabetes CHD HF AF Cerebral ischaemia β 0.013 (−0.045, 0.060) per avg ↑APN β 0.006 (−0.009, 0.020) per avg ↑APN β 0.001 (−0.012, 0.014) per avg ↑APN β 0.001 (−0.012, 0.014) per avg ↑APN β −0.002 (−0.033, 0.00) per avg ↑APN Two sample UK Biobank + CARDIoGRAMplus4D CARDIoGRAMplus4D Multiple cohorts HERMES MEGASTROKE ADIPOQ + others n = 547 261 n = 171 875 n = 537 409 n = 977 323 n = 446 696 CHD MI AF HF Stroke OR 0.94 (0.87, 1.01) per ↑ log APN OR 0.96 (0.86, 1.08) per ↑ log APN OR 0.99 (0.92, 1.07) per ↑ log APN OR 0.96 (0.86, 1.08) per ↑ log APN OR 0.98 (0.88, 1.09) per ↑ log APN One sample CPGS ADIPOQ, CDH13 N = 96 903 n = 977 323 HF AF AS MI RR 0.92 (0.65, 1.29) per ↑ log APN RR 0.87 (0.68, 1.12) per ↑ log APN Two sample HERMES HUNT, deCODE, MGI, DiscovEHR, UK Biobank, AFGen UK Biobank UK Biobank ADIPOQ + others n = 1 030 836 n = 367 561 n = 367 561 HF AF AS MI RR 1.55 (0.87, 2.76) per ↑ log APN RR 0.93 (0.67, 1.30) per ↑ log APN RR 0.99 (0.89, 1.09) per ↑ log APN RR 1.00 (0.92, 1.08) per ↑ log APN RR 1.01 (0.79, 1.28) per ↑ log APN RR 0.99 (0.86, 1.13) per ↑ log APN Multi-cohort ADIPOQ n ≤ 31 000 Fasting insulin Diabetes β 0.02 SD (−0.07, 0.11) per ↓APN SD OR 0.94 (0.75, 1.19) per ↓APN SD One sample CGPS ADIPOQ, CDH13 n = 96 903 Diabetes RR 1.07 (1.06, 1.09) per ↓log APN Two sample ADIPOGen, GERA, DIAGRAM, UK Biobank ADIPOQ + others n = 659 316 RR 1.26 (1.01, 1.57)b per ↓log APN Two sample CARDIoGRAM CARDIoGRAMplusC4D ADIPOQ (conservative approach) n = 86 895 n = 194 427 CHD −2.3% ΔOR (−4.1, −0.4) per APN-raising allele −0.6% ΔOR (−1.9, 1.0) per APN-raising allele Two sample DIAGRAM UK Biobank + CARDIoGRAMplusC4D UK Biobank Multiple cohorts UK Biobank ADIPOQ + others n = 898 130 n = 547 261 n = 977 323 n = 587 446 n = 401 937 Diabetes CHD HF AF Cerebral ischaemia β 0.013 (−0.045, 0.060) per avg ↑APN β 0.006 (−0.009, 0.020) per avg ↑APN β 0.001 (−0.012, 0.014) per avg ↑APN β 0.001 (−0.012, 0.014) per avg ↑APN β −0.002 (−0.033, 0.00) per avg ↑APN Two sample UK Biobank + CARDIoGRAMplus4D CARDIoGRAMplus4D Multiple cohorts HERMES MEGASTROKE ADIPOQ + others n = 547 261 n = 171 875 n = 537 409 n = 977 323 n = 446 696 CHD MI AF HF Stroke OR 0.94 (0.87, 1.01) per ↑ log APN OR 0.96 (0.86, 1.08) per ↑ log APN OR 0.99 (0.92, 1.07) per ↑ log APN OR 0.96 (0.86, 1.08) per ↑ log APN OR 0.98 (0.88, 1.09) per ↑ log APN One sample CPGS ADIPOQ, CDH13 N = 96 903 n = 977 323 HF AF AS MI RR 0.92 (0.65, 1.29) per ↑ log APN RR 0.87 (0.68, 1.12) per ↑ log APN Two sample HERMES HUNT, deCODE, MGI, DiscovEHR, UK Biobank, AFGen UK Biobank UK Biobank ADIPOQ + others n = 1 030 836 n = 367 561 n = 367 561 HF AF AS MI RR 1.55 (0.87, 2.76) per ↑ log APN RR 0.93 (0.67, 1.30) per ↑ log APN RR 0.99 (0.89, 1.09) per ↑ log APN RR 1.00 (0.92, 1.08) per ↑ log APN RR 1.01 (0.79, 1.28) per ↑ log APN RR 0.99 (0.86, 1.13) per ↑ log APN AF, atrial fibrillation; APN, adiponectin; AS, aortic stenosis; CHD, coronary heart disease; HF, heart failure; MI, myocardial infarction; OR, odds ratio; RR, risk ratio; SD, standard deviation. aSNP–adiponectin associations for two-sample MR drawn from ADIPOGen consortium. bNon-significant in multivariable MR adjusting for waist–hip ratio. Summary of major Mendelian randomization studies evaluating linear associations of adiponectin with cardiometabolic outcomes Multi-cohort ADIPOQ n ≤ 31 000 Fasting insulin Diabetes β 0.02 SD (−0.07, 0.11) per ↓APN SD OR 0.94 (0.75, 1.19) per ↓APN SD One sample CGPS ADIPOQ, CDH13 n = 96 903 Diabetes RR 1.07 (1.06, 1.09) per ↓log APN Two sample ADIPOGen, GERA, DIAGRAM, UK Biobank ADIPOQ + others n = 659 316 RR 1.26 (1.01, 1.57)b per ↓log APN Two sample CARDIoGRAM CARDIoGRAMplusC4D ADIPOQ (conservative approach) n = 86 895 n = 194 427 CHD −2.3% ΔOR (−4.1, −0.4) per APN-raising allele −0.6% ΔOR (−1.9, 1.0) per APN-raising allele Two sample DIAGRAM UK Biobank + CARDIoGRAMplusC4D UK Biobank Multiple cohorts UK Biobank ADIPOQ + others n = 898 130 n = 547 261 n = 977 323 n = 587 446 n = 401 937 Diabetes CHD HF AF Cerebral ischaemia β 0.013 (−0.045, 0.060) per avg ↑APN β 0.006 (−0.009, 0.020) per avg ↑APN β 0.001 (−0.012, 0.014) per avg ↑APN β 0.001 (−0.012, 0.014) per avg ↑APN β −0.002 (−0.033, 0.00) per avg ↑APN Two sample UK Biobank + CARDIoGRAMplus4D CARDIoGRAMplus4D Multiple cohorts HERMES MEGASTROKE ADIPOQ + others n = 547 261 n = 171 875 n = 537 409 n = 977 323 n = 446 696 CHD MI AF HF Stroke OR 0.94 (0.87, 1.01) per ↑ log APN OR 0.96 (0.86, 1.08) per ↑ log APN OR 0.99 (0.92, 1.07) per ↑ log APN OR 0.96 (0.86, 1.08) per ↑ log APN OR 0.98 (0.88, 1.09) per ↑ log APN One sample CPGS ADIPOQ, CDH13 N = 96 903 n = 977 323 HF AF AS MI RR 0.92 (0.65, 1.29) per ↑ log APN RR 0.87 (0.68, 1.12) per ↑ log APN Two sample HERMES HUNT, deCODE, MGI, DiscovEHR, UK Biobank, AFGen UK Biobank UK Biobank ADIPOQ + others n = 1 030 836 n = 367 561 n = 367 561 HF AF AS MI RR 1.55 (0.87, 2.76) per ↑ log APN RR 0.93 (0.67, 1.30) per ↑ log APN RR 0.99 (0.89, 1.09) per ↑ log APN RR 1.00 (0.92, 1.08) per ↑ log APN RR 1.01 (0.79, 1.28) per ↑ log APN RR 0.99 (0.86, 1.13) per ↑ log APN Multi-cohort ADIPOQ n ≤ 31 000 Fasting insulin Diabetes β 0.02 SD (−0.07, 0.11) per ↓APN SD OR 0.94 (0.75, 1.19) per ↓APN SD One sample CGPS ADIPOQ, CDH13 n = 96 903 Diabetes RR 1.07 (1.06, 1.09) per ↓log APN Two sample ADIPOGen, GERA, DIAGRAM, UK Biobank ADIPOQ + others n = 659 316 RR 1.26 (1.01, 1.57)b per ↓log APN Two sample CARDIoGRAM CARDIoGRAMplusC4D ADIPOQ (conservative approach) n = 86 895 n = 194 427 CHD −2.3% ΔOR (−4.1, −0.4) per APN-raising allele −0.6% ΔOR (−1.9, 1.0) per APN-raising allele Two sample DIAGRAM UK Biobank + CARDIoGRAMplusC4D UK Biobank Multiple cohorts UK Biobank ADIPOQ + others n = 898 130 n = 547 261 n = 977 323 n = 587 446 n = 401 937 Diabetes CHD HF AF Cerebral ischaemia β 0.013 (−0.045, 0.060) per avg ↑APN β 0.006 (−0.009, 0.020) per avg ↑APN β 0.001 (−0.012, 0.014) per avg ↑APN β 0.001 (−0.012, 0.014) per avg ↑APN β −0.002 (−0.033, 0.00) per avg ↑APN Two sample UK Biobank + CARDIoGRAMplus4D CARDIoGRAMplus4D Multiple cohorts HERMES MEGASTROKE ADIPOQ + others n = 547 261 n = 171 875 n = 537 409 n = 977 323 n = 446 696 CHD MI AF HF Stroke OR 0.94 (0.87, 1.01) per ↑ log APN OR 0.96 (0.86, 1.08) per ↑ log APN OR 0.99 (0.92, 1.07) per ↑ log APN OR 0.96 (0.86, 1.08) per ↑ log APN OR 0.98 (0.88, 1.09) per ↑ log APN One sample CPGS ADIPOQ, CDH13 N = 96 903 n = 977 323 HF AF AS MI RR 0.92 (0.65, 1.29) per ↑ log APN RR 0.87 (0.68, 1.12) per ↑ log APN Two sample HERMES HUNT, deCODE, MGI, DiscovEHR, UK Biobank, AFGen UK Biobank UK Biobank ADIPOQ + others n = 1 030 836 n = 367 561 n = 367 561 HF AF AS MI RR 1.55 (0.87, 2.76) per ↑ log APN RR 0.93 (0.67, 1.30) per ↑ log APN RR 0.99 (0.89, 1.09) per ↑ log APN RR 1.00 (0.92, 1.08) per ↑ log APN RR 1.01 (0.79, 1.28) per ↑ log APN RR 0.99 (0.86, 1.13) per ↑ log APN AF, atrial fibrillation; APN, adiponectin; AS, aortic stenosis; CHD, coronary heart disease; HF, heart failure; MI, myocardial infarction; OR, odds ratio; RR, risk ratio; SD, standard deviation. aSNP–adiponectin associations for two-sample MR drawn from ADIPOGen consortium. bNon-significant in multivariable MR adjusting for waist–hip ratio. In this issue of Cardiovascular Research, Nielsen and colleagues11 seek to further illuminate the causal basis of adiponectin’s relationship with CVD. They conducted observational analyses of adiponectin in relation to incident HF, AF, aortic stenosis (AS), and myocardial infarction (MI) in the Copenhagen General Population Study (n = 30 045), followed by one-sample MR analysis in the same cohort using genetic variants in ADIPOQ and CDH13 (n = 96 903). The authors then performed two-sample MR expanding these variants to 12 distinct loci for evaluation of CVD in various cohorts (n = 361 194 to n = 1 030 836). Nielsen and colleagues found significant positive associations of plasma adiponectin modelled linearly with all CVD events except MI after extensive adjustment for covariates. One- and two-sample MR analyses were null, however, leading the authors to conclude that genetic evidence does not support causality for these associations. The strengths of the present work include its large population-based longitudinal study of plasma adiponectin and new evaluation of its relationship with AS, in conjunction with one- and two-sample MR analyses. Their two-sample MR analyses had ample power to detect linear associations for HF and AF, although this was limited for AS and MI. It bears noting that the increased risks of HF, AF, and AS reported for plasma adiponectin came in models that adjusted for putative intermediates. Such adjustment can obscure the true nature of an exposure–outcome association. Indeed, when such putative intermediates were removed, the associations became evidently U-shaped for MI and HF, consistent with earlier findings for CHD3 and mortality.12 Given that the true association of adiponectin with HF and MI was bimodal, the authors are commended for implementing a novel non-linear MR technique. This technique, however, could only be applied in the one-sample MR analysis, where individual-level data were available, limiting its power to capture non-linear causal effects. Thus, the present analyses cannot exclude that the associations of low or high adiponectin with HF or MI have a causal basis. What might explain the dual, Janus-like association of the adipokine with these outcomes? There is evidence that adiponectin secretion is not confined to adipose tissue, where its production decreases with dysfunctional fat. It is also produced by cardiomyocytes, skeletal myocytes, and other cell types.1 Such production appears to be up-regulated in the setting of disease, where resistance of AdipoR-mediated signalling occurs. Notably, T-cadherin can be cleaved by glycosyl-phosphatidylinositol phospholipase D, whose levels increase in conditions of metabolic stress, raising blood adiponectin levels but diminishing their tissue availability.13 Natriuretic peptides can also stimulate adipocyte production, elevating adiponectin levels in the setting of cardiac strain.2 Directly harmful effects of hyperadiponectinaemia are also possible, although the posited ability of the protein to activate complement has not been supported.5 Hence, the conceptual model is one where elevated adiponectin concentrations seen with illness, and notably with high-grade inflammatory disorders, are predominantly markers of disease and tissue adiponectin resistance.2 Absent these adiponectin-raising conditions, low circulating levels deprive target tissues of the adipokine’s favourable anti-lipotoxic, anti-inflammatory, and anti-apoptotic effects, leading to heightened CVD risk. So where might the answer to the causality question lie? There is evidence that hyperinsulinaemia suppresses adiponectin,1 such that a causal effect of the adipokine on glucose homeostasis would reflect a bidirectional relationship. But available pre-clinical evidence certainly bolsters the plausibility of a causally protective role for CHD and HF. This would likely be most identifiable, and clinically actionable, in the setting of hypoadiponectinaemia, where the adipokine’s functions would be less confounded by tissue resistance to its effects. Pursuit of non-linear MR analysis, however, would best be focused on non-pleiotropic variants limited to ADIPOQ. (Consideration of CDH13, in particular, can be muddled by the reciprocal relation of adiponectin with T-cadherin.4) This would face obstacles of very large sample size and the need to combine individual-level data across studies, currently made infeasible by restrictions to genetic data sharing. Ultimately, development of adiponectin receptor agonists in the mould of AdipoRon,1 with profiles suitable for clinical testing in individuals with hypoadiponectinaemia, may offer the best demonstration of the adipokine’s properties and therapeutic efficacy. MR analysis has squared off against adiponectin’s Janus, but the contest so far is locked up by the double-visaged god. There are no new data associated with this article.