APOE ε4 carriage associates with improved myocardial performance from adolescence to older age

The ALSPAC is a birth cohort that recruited 14,541 pregnant women with an expected date of delivery in 1991–1992 [16].

The MRC NSHD is the world’s longest-running birth cohort with continuous follow-up. In 1946 in Britain, 5362 individuals (2547 males and 2815 females) born in the same week in March were enrolled. Participants were invited for periodic follow-ups in which health and socio-economic assessments were performed which have been described elsewhere [17].

The SABRE study is a tri-ethnic cohort of European, South Asian, and African Caribbean participants living in North and West London. Between 1988 and 1981, participants aged 40–69 years were randomly selected from 5-year age and sex stratified primary care lists (n = 4063) and workplaces (n = 795). Full details have been described elsewhere [18].

The UK Biobank is a large prospective cohort study with more than half a million individuals recruited between 2006 and 2010 when study participants were aged 40–69 years old, and features demographic, genetic, health outcome and imaging data for participants [19]. Details of subjects’ comorbidities were obtained through self-reported diagnoses and International Classification of Disease (ICD-9 and ICD-10) codes from linked medical records This project was conducted using the UK Biobank (UKBB) resource under application numbers 40,616 and 46,696.

In ALSPAC, echocardiography was performed when study participants were 17 and 24 years by 1–2 experienced echocardiographers in accordance with the American Society of Echocardiography (ASE) guidelines with good reproducibility (both intraobserver and interobserver correlation coefficients ranged between 0.75 to 0.93 [20]). Since non-attendance to clinic visits is especially relevant within this group [21], echocardiography measurements were either averaged if more than one scan was available, or the one available scan was used to reduce the bias associated with data missingness.

In NSHD, when study members were 60–64 years (2006–2010), British-based NSHD participants who had not been lost to follow-up or withdrawn, were invited to attend a clinic-based assessment that included resting transthoracic echocardiography using General Electric (GE) Vivid I machines. The echocardiographic protocol included long and short axis (LAX and SAX), apical 5-, 4-, 3- and 2- chamber, and aortic SAX views [17]. In NSHD, echocardiography quality assurance was evaluated based on blind duplicate readings showing excellent inter- and intrareader variability (coefficients > 0.80) [22].

In SABRE, study members were invited between 2008 and 2012 to a clinic visit in which echocardiographic data was acquired using a Phillips iE33 ultrasound machine S5–1 phased array and a X3–1 matrix transducer and analyzed in line with the with the ASE guidelines. For structural and volumetric metrics, the inter- and intra-observer agreement was also high in SABRE (coefficients > 0.71) [23].

In all three cohorts, echocardiographic data provided left ventricular (LV) ejection fraction (EF), E/e’, systolic and diastolic LV posterior wall and interventricular septal thickness (LVPWTs/d, IVSs/d), LV mass (LVmass) and the stroke volume (SV). Myocardial contraction fraction (MCF) was calculated as the ratio between stroke volume and myocardial volume. Although indexation to body surface area (BSA), is commonly done in clinical practice, BSA is a poor indexation metric as it creates a bias for overweight individuals [24]. Although indexation to allometric height is a better alternative [24], indexation might lead to spurious associations, as the exposure might be associated with height/weight rather than with the outcome itself. Therefore, we used unindexed echocardiographic outcomes in all subsequent analyses.

Participants in the UK Biobank were randomly invited for a CMR scan on a 1.5 T Siemens Aera scanner from 2014. Briefly, the CMR imaging protocol consisted of three long-axis views and a complete short axis stack of balanced steady state free precession cines [25]. Grey-scale short axis cine stacks were automatically segmented using a deep learning neural network that has optimised for UKBB scan images, with human expert level performance [26]. The short-axis segmentations underwent post-processing to compute end-systolic, end-diastolic and stroke volumes in both ventricles [27]. Left ventricular mass (LVM) was computed from left ventricular volume (assuming a density of 1.05 g/ml). Left ventricular wall thickness was computed as the perpendicular radial-line distance between endocardial and epicardial surfaces at end-diastole for each of the 17 myocardial segments as defined by the American Heart Association (AHA) [28]. MCF was derived as above. Thickness of the IVS was calculated as the mean wall thickness of segments 2, 3, 8, 9 and 14, while PWT was taken as the mean of segments 5,6, 11, 12, and 16. To compute longitudinal and radial peak diastolic strain rates, non-rigid image co-registration was performed between successive frames to enable dynamic motion tracking of the heart during the cardiac cycle [29]. Unindexed CMR metrics were used in all subsequent analyses as discussed above.

In ALSPAC, genetic samples were available for 2009 children. APOE ε genotype was appraised using integrated single-label liquid phase assay in 2011 [30].

In NSHD, blood samples were collected at age 53 by a trained research nurse, and DNA was extracted [31]. Genetic analysis of stored samples took place in in 1999 and 2006–2010. In SABRE, blood samples were collected during baseline studies in 1988–1991 and during follow-up from 2007 to 2012 [18]. Genotyping of rs439358 and rs7412 was conducted at the Exeter University for SABRE and by LGC, Huddleston, UK for NSHD [32].

Genotyping of UK Biobank participants is detailed elsewhere [33], however in brief, genotyping for 488,252 subjects was performed using the UK BiLEVE or UK Biobank Axiom arrays and imputation based on the HaplotypeReference Consortium and UK10K + 1000 Genomes panels. Imputation V3 (in GRCh37 coordinates) was used for the current study. Genotypes in their released PLINK-format files were used on the DNANexus platform (https://​www.​dnanexus.​com/​).

Based on the presence or absence of APOE ε4, genotypes were categorically defined as: non-APOE ε4 carriers (ε2ε2, ε2ε3, ε3ε3), heterozygous-APOE ε4 (ε2ε4 and ε3ε4) or homozygous-APOE ε4 (ε4ε4). Heterozygous-APOE ε4 and homozygous-APOE ε4 were further grouped into APOE ε4 carriers.

Sex was recorded as male or female. The age, weight, and height at the time of the imaging were used to compute the body mass index (BMI) in all 3 cohorts. In NSHD, participants’ socioeconomic position (SEP) was evaluated at the time of echocardiography according to UK Surveys Registrar General’s social class, dichotomized as manual or non-manual. In ALSPAC, father’s SEP was available in the same format as the NSHD. In UK Biobank, we used the Townsend deprivation index scores derived from national data about ownership and unemployment aggregated by postcodes [34]. The presence of cardiovascular disease (CVD), diabetes or high cholesterol was recorded as 1 = present or 0 = absent. In ALSPAC, congenital heart disease was used instead of CVD. Congenital heart disease and CVD were self-reported or GP-based diagnoses, while diabetes was defined based on doctor diagnosis and the use of diabetes medications. High cholesterol was defined based on the use of lipid-lowering drugs or as a total cholesterol higher than 240 mg/dl.

All analyses were performed in R 4.0 [35]. For all analyses, a two-tailed p-value < 0.05 was considered statistically significant.

Distribution of data were assessed on histograms and using Shapiro-Wilk test. Continuous variables are expressed as mean ± 1 standard deviation (SD) or median (interquartile range) as appropriate; categorical variables, as counts and percent.

In the main analysis, we compared non-APOE ε4 carriers with APOE ε4 carriers. Given the skewed distributions of echocardiographic and CMR data, generalized linear models with gamma distribution and log link were used to investigate the association of APOE ε4 genotypes as the exposures to predict the continuous echocardiographic and CMR variables as the outcomes. As the longitudinal and radial PDSR also spanned negative values, generalized linear models with Gaussian distribution and identity link were used instead. Being a combination of gene variants, APOE ε genotype is expected to be an instrumental variable and therefore unconfounded. Thus, Model 1 was unadjusted. To obtain more precise regression estimates, Model 2 was adjusted for factors associated with the outcome, namely age, sex, and SEP. To explore the mechanistic pathway downstream of APOE ε genotype but upstream of the echocardiographic outcomes, subsequent models were adjusted for mediators as follows: Model 3 for BMI; Model 4 for the presence of CVD; Model 5 for diabetes; Model 6 for high cholesterol; and Model 7 for hypertension (Fig. 1). Model assumptions were verified with regression diagnostics and found to be satisfied.

For all the models, regression estimates were obtained separately for ALSPAC, NSHD, SABRE and UK Biobank (i.e., cohort specific analyses). Since both NSHD and SABRE participants had echocardiography and were of a similar age (i.e., > 60 years on average), random-effects meta-analyses were performed across these 2 cohorts. Heterogeneity was evaluated using the Cochran Q test and Higgins I² statistic. Although ALSPAC had echocardiographic data, participants were < 24 years of age and thus were not included in the meta-analysis given the heterogeneity with the older cohorts. Since UK Biobank had CMR data, it was not included in the meta-analysis.

To explore dose responses, APOE ε4 genotypes were recoded as an ordered category based on the number of ε4 possessed. Thus, class 0 = ε2ε2, ε2ε3, ε2ε3; class 1 = ε2ε4 and ε3 ε4; and class 2 = ε4ε4. Given the existence of 3 classes, generalized linear models with gamma distribution (or Gaussian distribution for longitudinal and radial PDSR) and orthogonal polynomial contrasts with 2 equally spaced levels (i.e., linear and quadratic) were employed to look for a dose response by ε4 variants. Then, we filtered significant results correcting for multiple testing at a false discovery rate (FDR) of 0.15.

As a sensitivity analyses, APOE ε4 carriers were split into heterozygous-APOE ε4 (ε2ε4 and ε3ε4) and homozygous-APOE ε4 (ε4ε4), and all the analyses were replicated as above.

Participants with available APOE ε4 genotype and at least one cardiac imaging metric were included, yielding a total of 37,023 participants (n = 1397 from ALSPAC, n = 1467 from NSHD, n = 1187 from SABRE and n = 32,972 from UK Biobank). Their characteristics are shown in Table 1. In total, there were 843 homozygous-APOE ε4 and 9460 heterozygous-APOE ε4 individuals, with a similar prevalence across ALSPAC, NSHD, SABRE, and UK Biobank. ALSPAC participants were younger (< 24 years), had a lower BMI (median 23), and were less likely to have diabetes, cardiovascular diseases, high cholesterol, or hypertension (< 5%). SABRE participants were more likely to be males (76.75%), have a higher BMI (median 27 years) or suffer from hypertension (58.98%) compared to NSHD and UK Biobank participants. On the other hand, UK Biobank participants were least likely to suffer from CVD (6.53%), diabetes (18.64%), or hypertension (27.62%).

In NSHD, when compared to the non-APOE ε4 group, APOE ε4 carriers had a 6% higher LV MCF (95% confidence interval [CI] 0–12%, p = 0.050) which persisted unattenuated after adjusting for sex and SEP (p = 0.038) and diabetes (p = 0.056), was attenuated to 5% after adjusting for BMI (95% CI 0–11%, p = 0.064), CVD (95% CI 0–12%, p = 0.112) or hypertension (95% CI 1–11%, p = 0.081), and increased to 8% after adjusting for high cholesterol (95% CI 1–14%, p = 0.020, Supplementary Table S1). Similarly, APOE ε4 carriers had a 5% higher LVmass p = 0.057 which was increased to 6% after adjusting for CVD (p = 0.040) or hypertension (p = 0.040), and to 7% after adjusting for diabetes (p = 0.024). No significant associations were found in SABRE (Supplementary Table S2). Moreover, in NSHD APOE ε4 carriers had an 8% higher SV 95%CI 3–12% p = 0.001 (Supplementary Table S5).

In the NSHD + SABRE meta-analyses, compared to the non-APOE ε4 group, APOE ε4 carriers had similar cardiac phenotypes in terms of EF, E/e’, LVPWT_s/d, IVS_s/d and LVmass, but had a 4% higher MCF (95% CI 1–7%, p = 0.016) which persisted after adjustment for age, sex and SEP (95% CI 1–7%, p = 0.008). This was attenuated to 3% after adjustment for CVD, diabetes or hypertension (all 95% CI 0–6%, all p < 0.070, Table 2, Fig. 1). However, no significant dose response for the number of APOE ε4 alleles was found in relationship with LV MCF (Table 3, Supplementary Table S3). Moreover, in NSHD + SABRE meta-analysis, APOE ε4 carriers had a 6% higher SV.

In ALSPAC, APOE ε4 carriers had a 2% higher MCF (95% CI 0–5%) albeit it was not statistically significant p = 0.059 (Table 4). In addition, ε4 carriers had a 2% lower IVSd (p = 0.057) and LVPWT_d (p = 0.064), although these results were also not significant.

In the sensitivity analysis, only heterozygous-APOE ε4 carriers had a 4% higher MCF (95% CI 1–7%, p = 0.016) which persisted after adjusting for sex and SEP (95% CI 1–7%, p = 0.013), and BMI (95% CI 1–7%, p = 0.018), but was attenuated to 3% after adjusting for CVD (95% CI 0–6%, p = 0.043, diabetes (95 CI % 0–7%, p = 0.060), or hypertension (95% CI 0–6%, p = 0.028, Table 5, Supplementary Table S4) in the meta-analysis. Similarly, in ALSPAC only heterozygous ε4 carriers had a higher MCF when compared to non-carriers.

In ALSPAC, NSHD, or SABRE, neither a linear nor a quadratic dose effect based on the number of ε4 alleles was observed. The association between ε4 and MCF in the SABRE + NSHD meta-analysis persisted at an FDR of 0.15.

In UK Biobank, when compared to the non-APOE ε4 group, APOE ε4 carriers had a 1% higher MCF 95% (CI 0–1%, p = 0.020) which persisted after adjusting for age, sex and SEP (Model 2, p = 0.080), CVD (Model 4, p = 0.006), high cholesterol (Model 5, p = 0.0001) or hypertension (Model 7, p = 0.034), but was attenuated to 0% (95% CI 0–1%) after adjusting for BMI (Model 3, p = 0.079) or diabetes (p = 0.058, Table 6, Fig. 1). There was a dose-response relationship based on the number of ε4 alleles, especially when adjusting for CVD in Model 4 (p = 0.036) and high cholesterol in Model 6 (p = 0.006, Table 3). However, although heterozygous-APOE ε4 carriers had a higher MCF, the association was not significant for homozygous-APOE ε4 carriers (Table 5).

In addition, APOE ε4 carriers had a 2% higher longitudinal PDSR (95% CI 0–3%, p = 0.045), which persisted after adjusting for CVD and diabetes, but was attenuated to 0% in Model 2 and to 1% after adjusting for diabetes (Model 5). Conversely, they had a 5% lower radial PDSR (95% CI 0.90–1.00, p = 0.05) which behaved similar to longitudinal PDSR on adjustment (Table 6).