Low LDL cholesterol and risk of bacterial and viral infections: observational and Mendelian randomization studies

Abstract

Aims

Low levels of LDL cholesterol may be associated with risk of infectious disease. We tested the hypothesis that low LDL cholesterol due to genetic variation in the LDLR, PCSK9, and HMGCR genes and a polygenic LDL cholesterol score is associated with risk of infectious diseases in the general population.

Methods and results

Using observational and Mendelian randomization designs, we examined associations of low plasma LDL cholesterol with risk of bacterial and viral infections in 119 805 individuals from the Copenhagen General Population Study/Copenhagen City Heart Study, 468 701 from the UK Biobank, and up to 376 773 from the FinnGen Research Project. Observationally, low LDL cholesterol concentrations were associated with risk of hospitalization for both bacterial and viral infections. In genetic analyses, a 1 mmol/L lower LDL cholesterol was associated with lower plasma PCSK9 {−0.55 nmol/L [95% confidence interval (CI): −1.06 to −0.05]; P = 0.03}, leucocyte count [−0.42 × 10⁹/L (−0.61 to −0.24); P < 0.001], and high-sensitivity C-reactive protein [−0.44 mg/L (−0.79 to −0.09); P = 0.014]. Using an LDLR, HMGCR, and PCSK9 score, a 1 mmol/L lower LDL cholesterol was associated with risk ratios of 0.91 (95% CI: 0.86–0.97; P = 0.002) for unspecified bacterial infection, of 0.92 (0.87–0.97; P = 0.004) for diarrhoeal disease, and of 1.15 (1.03–1.29; P = 0.012) for unspecified viral infections and 1.64 (1.13–2.39; P = 0.009) for HIV/AIDS. Using a polygenic LDL cholesterol score largely showed similar results and in addition a lower risk of 0.85 (0.76–0.96; P = 0.006) for bacterial pneumonia and 0.91 (0.82–0.99; P = 0.035) for sepsis.

Conclusion

Genetically low LDL cholesterol concentrations were associated with lower concentration of markers of inflammation; lower risk of hospitalization for unspecified bacterial infections, infectious diarrhoeal diseases, bacterial pneumonia, and sepsis; and higher risk of viral infections and HIV/AIDS.

Graphical abstract

Open in new tabDownload slide

Introduction

During bacterial infections, lipopolysaccharides are shredded from bacterial cell membranes, activating the innate immune response and causing inflammation. Lipopolysaccharides are integrated into lipoprotein particles and removed via the LDL and very LDL (VLDL) receptor pathways. Studies have suggested that increasing the number of LDL receptors with lipid-lowering drugs during septic shock reduces short-term risk of death.^1–3 On the other hand, some viruses can bind directly to these receptors and be taken up into the cells and facilitate virus replication.^4,5

Theoretically, this means that low LDL cholesterol, i.e. by increasing the number of LDL receptors by statins or the number of LDL and VLDL receptors by proprotein convertase subtilisin kexin type 9 (PCSK9) inhibition, may be beneficial during bacterial infections clearing lipopolysaccharides and harmful during viral infections facilitating virus uptake and replication.^1,2,4–6

Genetic variation in the LDLR, PCSK9, and HMGCR genes encoding, respectively, the LDL receptor, PCSK9, and the 3-hydroxy-3-methylglutaryl-CoA reductase mimic the effect of lipid lowering by statins and PCSK9 inhibitors.⁷ We hypothesized that low LDL cholesterol concentration due to genetic variation in these specific genes, potentially improving clearance of inflammatory lipopolysaccharides and potentially increasing uptake of virus, is associated with low risk of bacterial infections and high risk of viral infections. We tested this hypothesis in observational and Mendelian randomization designs with hospitalization for specific infectious diseases as outcomes/endpoints. The Mendelian randomization design utilizes the random assortment of alleles at conception, and used in a general population setting, the design is largely unaffected by survivor bias, reverse causation, and confounding.^8,9 To examine whether potential changes in risk were due to specific involvement of the LDL receptor and PCSK9 protein or simply due to lower LDL cholesterol in general, we used both a target LDLR, PCSK9, and HMGCR genetic score using variants in the LDLR, PCSK9, and HMGCR genes and a polygenic LDL cholesterol score, not including these genes. This has not been examined previously.

The hypotheses were tested in three steps: (i) the observational association of LDL cholesterol concentration with risk of bacterial and viral infections in two cohorts from the Danish general population, the Copenhagen General Population Study (CGPS) and the Copenhagen City Heart Study (CCHS), and the UK Biobank cohort; (ii) the association of a genetic score of LDL lowering alleles (LDLR, PCSK9, and HMGCR genetic variants) and a polygenic LDL cholesterol score excluding these variants, with LDL cholesterol, PCSK9, leucocyte count, and C-reactive protein; and (iii) the causal genetic effect of a 1 mmol/L lower plasma LDL cholesterol on risk of bacterial and viral infections in the two Copenhagen cohorts and the UK Biobank cohort¹⁰ using individual data information, while using summary data from the Global Lipids Genetics Consortium (GLGC)¹¹ and FinnGen Research Project in a two-sample design and, finally, combined results from all cohorts in meta-analyses.

Methods

Study populations

We included individuals from two similar prospective studies of the Danish general population, the CGPS and the CCHS.¹² Combining the two studies for observational examinations yielded a total of 119 805 individuals with information on both LDL cholesterol concentration and genotype data. All individuals were white and of Danish descent, none was included in more than one study, and no individuals were lost to follow-up due to the complete Danish health registries. Information on the UK Biobank,¹⁰ GLGC,¹³ and FinnGen Research Project¹⁴ is summarized in Supplementary Appendix 1.

Ethical reporting

The study was conducted in accordance with the Declaration of Helsinki, and CGPS and CCHS were approved by the Danish ethical committees (KF-100.2039/91, KF-01-144/01, and H-KF-01-144/01), the UK Biobank study by the North West Haydock Research Ethics Committee (16/NW/0274), and the FinnGen by the Coordinating Ethics Committee of the Helsinki and Uusimaa Hospital District. Written informed consent was obtained from all participants.

Study endpoints/outcomes

In the Copenhagen studies, we used the Danish National Patient Register¹⁵ which covers all Danish hospitals with information on hospitalizations, outpatient, and/or emergency room visits with a primary discharge diagnosis of an infectious disease. In the Copenhagen studies, infectious disease diagnoses were classified according to the World Health Organization’s International Statistical Classification of Diseases 8th and 10th Revision (ICD-8 and ICD-10) and in the UK Biobank and the FinnGen Research Project using ICD-8, ICD9, and ICD-10. In all studies, codes were harmonized across the three revisions and collapsed into the following categories of bacterial infections: any bacterial infection including unspecified bacterial infections, infectious diarrhoeal disease, bacterial pneumonia, sepsis, urinary tract infections/cystitis, endocarditis, skin infections, and bacterial meningitis and viral diseases: any viral infection including unspecified viral infection, human immunodeficiency virus infection/acquired immunodeficiency syndrome (HIV/AIDS), hepatitis, viral meningitis, and viral pneumonia (see Supplementary material online, Table S1). Categories of infectious diseases in the CCHS have been validated and found to correspond well to discharge diagnoses and positive culture findings.¹⁶ In the Copenhagen studies and the UK Biobank, follow-up time began at the first inclusion into a study and ended with censoring at date of death, occurrence of an event, emigration, or end of follow-up on 13 December 2018 in the Copenhagen studies and 21 March 2023 in the UK Biobank, whichever came first. In the FinnGen Research Project, data from Release 11 was used (released on 27 June 2024).

Covariates

In the Copenhagen studies, plasma LDL cholesterol was calculated using the Friedewald equation if plasma triglycerides were ≤4.0 mmol/L and measured by a direct enzymatic method at higher triglyceride concentrations. Plasma total cholesterol, HDL cholesterol, triglycerides, and high-sensitive C-reactive protein and blood leucocytes were measured using standard hospital assays. Plasma PCSK9 was measured by BG Medicine (Waltham, MA, USA) with a sandwich ELISA assay using antibodies supplied by Merck (Whitehouse Station, NJ, USA).

Participants reported on smoking (current smokers and non-smokers); work and leisure time physical activity (‘low’ for 0–2 h moderate activity per week and ‘intermediate and high’ for more than 2 h moderate or vigorous activity per week during either work or leisure time); alcohol consumption in units per week (1 unit ∼12 g alcohol); education (<8 and ≥8 years of completed education); use of cholesterol-lowering medication with more than 97% using statins; and for women, menopausal status. Body mass index was weight/height squared. Hypertension was defined as systolic blood pressure ≥ 140 mmHg (≥135 mmHg for individuals with diabetes), diastolic blood pressure ≥ 90 mmHg (≥85 mmHg for individuals with diabetes), or use of antihypertensive medication prescribed specifically for hypertension. Diabetes mellitus was self-reported disease, a non-fasting plasma glucose > 11.0 mmol/L, medication prescribed for diabetes and/or hospitalization due to diabetes (ICD-8 249–250; ICD-10 E10–11, E13–14). Non-skin cancer was defined as any cancer—basal cell carcinoma not included. Ischaemic heart disease was defined as hospitalization due to ischaemic heart disease (ICD-8 410–414; ICD-10 I20-I25).

Genetic instruments: targeted approach and polygenic LDL cholesterol score

In the Copenhagen studies, an ABI PRISM 7900HT (Applied Biosystems, USA) and TaqMan-based assays were used to genotype for LDL cholesterol associated variants in the LDLR, PCSK9, and HMGCR genes. Variants for the polygenic LDL cholesterol score were genotyped using the customized Metabochip¹⁷ and the Infinium HumanExome BeadChip (Illumina, San Diego, CA, USA). Variants were selected based on their association with plasma LDL cholesterol in genome-wide association studies, either as variants in the LDLR, HMGCR, or PCSK9 genes for the target approach or genome-wide, excluding the three genes above (±100 000 base pairs up- and downstream from the three genes) for the polygenic risk score. For LDLR, HMGCR, or PCSK9 variants, their functions are well-known and pleiotropic effects have not been observed.¹² Also, none of the variants should not be in linkage disequilibrium (LD) using a threshold of R² < 0.3. If possible, the same variants were selected in the UK Biobank and GLGC/FinnGen Research Project, or else proxies (LD > 0.90) or other LDL cholesterol associated variants were selected (see Supplementary material online, Tables S2 and S3). In the Copenhagen studies and the UK Biobank, weighted allele scores were constructed by summation of LDL cholesterol-decreasing alleles, weighted by allele frequency and effect size on plasma LDL cholesterol, as done previously (see Supplementary material online, Figure S1).¹⁸ The LDLR, HMGCR, or PCSK9 variants explained 0.5% of the variation in plasma LDL cholesterol in the Copenhagen studies and 0.8% in the UK Biobank; and the polygenic LDL cholesterol score explained 0.6% of the variation in plasma LDL cholesterol in the Copenhagen studies and 1.8% in the UK Biobank.

Statistical analysis

We used Stata SE 17.0 (StataCorp). Deviation from the Hardy–Weinberg expectations was tested using Pearson’s χ² test. To test whether plasma LDL cholesterol was observationally associated with risk of bacterial and viral diseases, we used Cox regression and restricted cubic splines with three knots, selected using the Akaike’s information criterion.¹⁹ Adjustments were done for potential confounders, including age (as underlying time scale), sex, year of birth, body mass index, smoking, physical activity, alcohol consumption, education, hypertension, type 2 diabetes mellitus, and non-skin cancer. Adjustment for year of birth was included to accommodate changes in diagnostic criteria and treatment over calendar time.

To test whether LDL cholesterol alleles were associated with plasma LDL cholesterol, PCSK9, leucocyte count, and C-reactive protein, analyses of variance was used. Tests for trend across ordered categories of concentrations and the weighted allele score were by the non-parametric Cuzick’s extension of a Wilcoxon rank sum test. Logistic regression was used to assess whether observational lower LDL cholesterol or the allele score was associated with the potential confounders of age, sex, body mass index, smoking, physical activity, alcohol consumption, education, hypertension, type 2 diabetes, and non-skin cancer.

The effect of genetically lower plasma LDL cholesterol concentration on risk of infectious diseases was estimated using instrumental variable analysis by two-stage least-squares regression using the user-written ivreg2 and ivpois commands.²⁰ The statistical strength of the genetic instrument was confirmed by F statistics for the weighted allele score of F = 144.8 and F = 981.4 in the Copenhagen studies and the UK Biobank, respectively, where F > 10 is considered acceptable.²¹ Sargan’s statistic for the weighted allele score indicated that the genetic variants were valid instruments, with no indication of heterogeneity/pleiotropy (all P > 0.36).²²

Estimates of the causal effect of plasma LDL cholesterol on risk of infectious disease from summary-level data from the GLGC²³ and the FinnGen Research Project¹⁴ were obtained by regression analyses using inverse-variance-weighted and Egger regression,²⁰ in the user-written mregger command in Stata.^20,24 Because Egger regression did not indicate pleiotropy of genetic variants (largest intercept = 0.01; P = 0.33), results were reported as inverse-variance-weighted estimates. The Bonferroni method was used to correct for multiple testing with an overall α=0.05 and a Bonferroni corrected α=0.05/(number of hypotheses tested, i.e. endpoints tested).²⁵

A causal effect summary estimate for the Copenhagen studies, UK Biobank, and GLGC/FinnGen studies was obtained by meta-analysis using the metan command with a random effects model.

Results

Participant characteristics

In the Copenhagen studies, up to 119 805 individuals were included for observational and Mendelian randomization analyses, and in the UK Biobank, 468 701 individuals were included for observational and genetic Mendelian randomization analyses. The GLGC and FinnGen Research Project included 188 577 and up to 451 201 individuals for genetic Mendelian randomization analyses, respectively. Genotype distributions did not deviate from Hardy–Weinberg expectations in any of the studies (all P > 0.05). Baseline characteristics of the Copenhagen studies and UK Biobank participants by study are shown in Table 1. Such data were not available for the FinnGen Research Project. Individuals in the Copenhagen studies and the UK Biobank were of a similar age at study entry, with largely similar percentages of women and men. The Copenhagen studies included only white individuals and the UK Biobank 94% white, 2.4% Asian, 1.5% black, 0.4% mixed, and 1.2% other individuals. Individuals in the UK Biobank had slightly higher body mass index, and a larger fraction had hypertension and diabetes and received lipid-lowering treatment, compared to the Copenhagen studies. In the Copenhagen studies, there was a larger fraction of smokers, individuals with ischaemic heart disease, and individuals having had bacterial infections, compared to the UK Biobank.

LDL cholesterol and observational risk of bacterial and viral infections

Compared with the population median for plasma LDL cholesterol concentration of 3.3 mmol/L, a lower LDL cholesterol concentration was during a median of 9.7 years follow-up (range 0–42.8) associated with higher observational prospective risks of hospitalization for any bacterial or viral infection, while a high concentration did not associate with risk (P < 0.001) (Figure 1, upper panel). Similar results were found in the UK Biobank study (see Supplementary material online, Figure S2). An LDL cholesterol concentration below 2 mmol/L was associated with a hazard ratio for any bacterial infection of 1.32 [95% confidence interval (CI): 1.24–1.41; P < 0.001] and for any viral infection of 1.60 (1.17–2.19; P < 0.001), compared to individuals with an LDL cholesterol ≥ 3 mmol/L (Figure 1, lower panel).

Confounding factors in observational and genetic analyses

Observationally, high age, male sex, high body mass index, smoking, high physical activity, high alcohol consumption, education ≥ 8 years, hypertension, type 2 diabetes, non-skin cancer, and ischaemic heart disease were all associated with both plasma LDL cholesterol and bacterial and viral infections and may therefore have confounded the observational associations (see Supplementary material online, Figure S3). However, the genotypes were either not or only modestly associated with any of the potential confounders, suggesting that a pleiotropic effect through any of the above factors is limited.

Genetic score, plasma LDL cholesterol, PCSK9, hsCRP, and leukocyte count

A weighted allele score consisting of genetic variants in the LDLR, PCSK9, and HMGCR genes was associated with stepwise lower plasma LDL cholesterol (P < 0.0001) and PCSK9 concentrations (P < 0.0004) and blood leucocyte count (P < 0.0001) (Figure 2, upper panels). In instrumental variable analysis, a 1 mmol/L lower LDL cholesterol reduced plasma PCSK9 by 0.55 nmol/L (−1.06 to −0.05; P = 0.033), leucocyte count by 0.42 × 10⁹/L (−0.61 to −0.24; P < 0.001), and hsCRP by 0.44 mg/L (−0.79 to −0.09; P = 0.014) (Figure 2, lower panel).

Genetic low LDL cholesterol and risk of bacterial infections

For the LDLR, HMGCR, and PCSK9 genetic score, a 1 mmol/L lower LDL cholesterol was associated with age- and sex-adjusted causal risk ratios of 0.91 (95% CI: 0.86–0.97; P = 0.002) for unspecified bacterial infection and sex-adjusted causal risk ratios of 0.92 (0.87–0.97; P = 0.004) for diarrhoeal disease in studies combined (Figure 3). Individual study estimates are shown in Supplementary material online, Figure S4. Using a polygenic LDL cholesterol score, a 1 mmol/L lower LDL cholesterol was associated with age- and sex-adjusted causal risk ratios of 0.88 (0.81–0.95; P = 0.001) for unspecified bacterial infection, of 0.94 (0.89–1.00; P = 0.053) for diarrhoeal disease, of 0.85 (0.76–0.96; P = 0.006) for bacterial pneumonia, of 0.91 (0.82–0.99; P = 0.035) for sepsis, and of 0.91 (0.85–0.98; P = 0.008) for skin infections in studies combined (Figure 3). Individual study estimates are shown in Supplementary material online, Figure S5. A 1 mmol/L lower LDL cholesterol did not have a causal effect on risk of urinary tract infections/cystitis, endocarditis, or bacterial meningitis.

Genetic low LDL cholesterol and risk of viral infections

For the LDLR, HMGCR, and PCSK9 genetic score, a 1 mmol/L lower LDL cholesterol was associated with age- and sex-adjusted causal risk ratios of 1.15 (95% CI: 1.03–1.29; P = 0.012) for an unspecified viral infection and of 1.64 (1.13–2.39; P = 0.009) for HIV/AIDS in studies combined (Figure 4). Individual study estimates are shown in Supplementary material online, Figure S6. A 1 mmol/L lower LDL cholesterol did not have a causal effect on risk of hepatitis, viral meningitis, or viral pneumonia.

In genetic analyses using a polygenic LDL cholesterol score, a 1 mmol/L lower LDL cholesterol did not have a causal effect on risk of viral infections, HIV/AIDS hepatitis, viral meningitis, or viral pneumonia in studies combined. Individual study estimates are shown in Supplementary material online, Figure S7.

Genetic low LDL cholesterol and risk of ischaemic heart disease and all-cause death

As a positive control of the validity of the LDLR, HMGCR, and PCSK9 genetic score and the polygenic LDL cholesterol score and for comparison of magnitude of risk, the causal association of the allele scores on risk of ischaemic heart disease and all-cause death²⁶ was included. For the LDLR, HMGCR, and PCSK9 genetic score, a 1 mmol/L lower LDL cholesterol was associated with age- and sex-adjusted causal risk ratios of 0.60 (95% CI: 0.56–0.65; 7.4 × 10⁻⁴⁶) for ischaemic heart disease and 0.74 (0.57–0.95; P = 0.017) for all-cause death in studies combined (Figure 4). In genetic analyses using a polygenic LDL cholesterol score, a 1 mmol/L lower LDL cholesterol was associated with age- and sex-adjusted causal risk ratios of 0.65 (0.55–0.74; 5.0 × 10⁻¹²) for ischaemic heart disease and 0.65 (0.58–0.74; P = 1.5 × 10⁻⁵) for all-cause death in studies combined. Individual study estimates are shown in Supplementary material online, Figures S6 and S7.

Discussion

In the present study, we found that in observational prospective analyses, low plasma LDL cholesterol concentrations were associated with high risk of bacterial and viral infections. Similar observational associations have been reported in other studies. It has been suggested that confounding by other diseases, i.e. an existing cancer, and reverse causation, as infectious diseases in the acute phase lower plasma LDL cholesterol concentrations, may contribute to these associations.^27,28 For these reasons, we examined the genetic causal effect of low LDL cholesterol on risk of infectious diseases using Mendelian randomization designs⁹ in a target approach using a LDLR, HMGCR, and PCSK9 genetic score mimicking low LDL cholesterol via statins or PCSK9 inhibitors and a polygenic LDL cholesterol score representing low LDL cholesterol in general. In causal genetic analyses, a 1 mmol/L lower LDL cholesterol resulted in a 9–12% lower risk of unspecified bacterial infections and a 6–8% lower risk of infectious diarrhoeal diseases both via the LDLR, HMGCR, and PCSK9 genetic score and the polygenic LDL cholesterol score. The LDLR, HMGCR, and PCSK9 genetic score also associated with a 64% higher risk of HIV/AIDS in studies combined and the polygenic LDL cholesterol score with a 15% lower risk of bacterial pneumonia, 9% lower risk of sepsis, and 9% lower risk of skin infections in studies combined. The results suggest that the effects are caused by lower LDL cholesterol in general, rather than by a specific pathway.

For comparison, a 1 mmol/L lower LDL cholesterol resulted in 35–40% lower risk of ischaemic heart disease and 26–35% lower risk of all-cause mortality.

Bacterial infections

Mechanistically, animal studies have shown that LDL particles take up toxic bacterial products such as lipopolysaccharides shredded from Gram-negative bacteria, ameliorating manifestations of experimental sepsis and infections.^3,29,30 Overexpression of PCSK9 in mice resulting in lower LDL receptor expression, reduced removal of LDL particles, and increased organ damage and inflammation during experimentally induced bacterial infections (sham surgery or caecal ligation/puncture),³¹ while PCSK9 knock-out mice with increased LDL receptor expression and LDL removal, and low LDL cholesterol concentrations had reduced organ damage and inflammation.³² Jointly, this suggests that removal of LDL particles and expression of receptors normally degraded by PCSK9, e.g. the LDL receptor, potentially improves the overall response to bacterial infections.⁶

In human studies, PCSK9 loss-of-function variants have been associated with improved survival in patients with septic shock and with a reduced inflammatory cytokine response in patients with septic shock and in healthy volunteers after lipopolysaccharide administration.¹ In contrast to this, low plasma LDL cholesterol concentrations in observational studies have been associated with a high risk of infection and critical disease in studies of patient cohorts and prospective general population cohorts.^33–36 Some of these studies have reported a higher frequency of pre-existing co-morbidities among individuals with low plasma LDL cholesterol, indicating potential reverse causation where an infection causes the low LDL cholesterol, and confounding where a critical disease, i.e. cancer, causes both a low LDL cholesterol concentration and a high risk of infection.^36–40 This is in line with the observational findings in the present study, showing a higher risk of infection in individuals with low plasma LDL cholesterol.

Very few randomized controlled trials of lipid-lowering drugs have reported on risk of infectious disease. A meta-analysis combining eight randomized controlled trials of statins did not find an association of statin treatment with risk of infections [25 770 participants, 4035 events, relative risk 1.00 (95% CI: 0.96–1.05)] or infection-related mortality [22 260 participants, 620 events, relative risk 0.97 (0.83–1.13)] compared to placebo.⁴¹ These studies did not include information on type of infection.⁴¹ The lack of an association might be due to the short treatment duration in clinical trials compared to the lifelong effect of lower LDL cholesterol in genetic studies. A meta-analysis of 14 PCSK9 inhibitor randomized controlled trials have shown a higher rate of upper respiratory tract infection signs and symptoms in patients receiving alirocumab compared to placebo [5234 participants, hazard ratio 1.87 (1.04–3.38)]⁴²; however, it is not known if this is a side effect of alirocumab rather than caused by a lowering of LDL cholesterol per se. A planned randomized placebo-controlled double-blind trial will assess the difference in survival after 28 days in patients with sepsis or septic shock, treated with alirocumab or placebo in an intensive care unit (NCT03634293).

A Mendelian randomization study using the UK Biobank examined a polygenic risk score for LDL cholesterol and found no effect of higher concentrations of LDL cholesterol on risk of hospitalization for any infection (bacterial and viral combined), antibiotic use, or 28 days survival after sepsis [407 558 participants, 29 600 events, relative risk 1.01 (95% CI: 0.97–1.05)].⁴⁰ Similar results were found in a Mendelian randomization study of a patient cohort [7804 participants; odds ratio (OR) for sepsis 1.02 (0.97–1.07); for intensive care unit admission 1.01 (0.95–1.07); and for in-hospital death 0.92 (0.81–1.05)].³⁶ In contrast to this, two-sample Mendelian randomization studies using the GLGC and UK Biobank data found a reduced risk of sepsis with an OR of 0.72 (0.57–0.90) for a 1 mmol/L lower LDL cholesterol in individuals younger than 75 years⁴³ and borderline increased risk of sepsis with an OR of 1.10 (0.99–1.24) for a 1 mmol/L higher LDL cholesterol.⁴⁴ These studies did not examine other specific types of infections and are not directly comparable to our results.

Taken together, both animal and human experimental studies support a beneficial effect of low LDL cholesterol on risk of bacterial infections in line with the causal genetic results of the present study. Randomized controlled trials and previous Mendelian randomization studies pooling bacterial and viral infections have shown no causal effect of LDL cholesterol on infection risk. The reduced risk of unspecified bacterial infection and infectious diarrhoeal disease observed in the present study may be caused by increased removal of lipopolysaccharides as suggested in experimental studies; however, a lower inflammatory burden as seen by the lower concentrations of C-reactive protein and leucocyte count in the present study and that well-perfused (non-atherosclerotic) tissues and organs form better barriers against bacteria compared to an intestine with a reduced perfusion may also contribute.

Viral infections

Virus requires several molecules for virus attachment, cell entry, replication, and virion release. Some viruses can bind directly to the LDL and VLDL receptors⁶ and be taken up into host cells and thus facilitate replication.^4,5 In cell studies, hepatitis B and C virus, rhinovirus, Japanese encephalitis virus, and vesicular stomatitis virus have been shown to use LDL receptors and other LDL receptor family members for cell binding and entry.^45–47 There is some evidence from cell studies that PCSK9 might influence hepatitis C virus infectivity, as it down-regulates the LDL and VLDL receptors.^5,48 Theoretically, this means that PCSK9 inhibition might facilitate hepatitis C virus uptake into hepatocytes, although results are conflicting.⁴⁹

Human immunodeficiency virus is a retrovirus with surface envelope glycoproteins facilitating viral attachment and entry into T helper cells, monocytes, macrophages, and dendritic cells. During binding to host cells, the virus envelope fuses with the host cell membrane, incorporating CD81 and CD63 into the virus membrane and expressing these proteins on the cell surface after the fusion.⁵⁰ Both CD81 and CD63 are targets for degradation by PCSK9^4,5; however, it is unclear if this might increase susceptibility to HIV infection and how such a susceptibility may be mediated.

Analysis of 14 observational studies pooling data from 19 988 patients with COVID-19 showed that use of statins reduced risk of adverse outcomes by 49% [OR 0.51 (95% CI: 0.41–0.63)].⁵¹ Several Mendelian randomization studies have examined the effect of LDL cholesterol on risk of hospitalization for COVID-19. One study found that a one standard deviation (0.87 mmol/L) higher LDL cholesterol increased risk of COVID-19 by 58% [387 079 participants and 1211 events; OR 1.58 (1.21–2.06)].⁵² Another study found that a 1 mmol/L higher LDL cholesterol increased risk of COVID-19 with an OR of 1.11 (1.0–1.23).⁵³

Taken together, cell studies show that many viruses—including human immunodeficiency virus—use LDL and VLDL receptors for cell entry; however, it is unclear whether increasing the number of receptors by lipid-lowering treatment with statins or PCSK9 inhibitors might increase risk of virus uptake and propagation. The present study suggests that genetically low LDL cholesterol may increase risk of hospitalization for an HIV infection. This risk contrasts with the large reduction in risk of ischaemic heart disease and all-cause death also shown in the present study. People with HIV in anti-viral treatment often have dyslipidaemia caused by the treatment. In these individuals, both statins⁵⁴ and PCSK9 inhibitors⁵⁵ have been shown to lower risk of cardiovascular disease.

Limitations

Infection with bacteria or viruses requires exposure to the specific agents and may be associated with a specific geographical location and potentially also a certain behaviour for transmission. Hospitalization for an infectious disease also requires the disease to be of some severity and potentially that co-morbidities are present. This may be a limitation of the present study, using hospital diagnoses in populations from Western and Northern European countries. This also limits the spectrum of infectious diseases examined to diseases present in these areas. Individuals with less severe disease may not have been hospitalized, and some diseases cannot be examined due to a low number of cases. Also, using register data to define infectious disease, endpoints have limitations, as not all cases of endocarditis, infectious diarrhoeal disease, and sepsis are caused by bacterial infections; and many disease agents are not easily adjudicated. The Mendelian randomization design has three assumptions,⁹ all addressed in the present study: (i) the genotypes used as instruments were associated with low LDL cholesterol, as shown in Figure 2; (ii) the genotypes should be associated with the risk of infectious diseases through LDL cholesterol, as verified using the MR Egger test for pleiotropic effects; and (iii) the genotypes were independent of other factors affecting the outcome, as shown in Supplementary material online, Figure S3. These findings support the validity of the Mendelian randomization results.

Conclusion

Genetically low LDL cholesterol concentrations were associated with lower concentration of markers of inflammation; lower risk of hospitalization for unspecified bacterial infections, infectious diarrhoeal diseases, bacterial pneumonia, and sepsis; and higher risk of viral infections and HIV/AIDS.

Data availability

Due to Danish legislation, the data that support this study cannot be made publicly available. Additional analyses can be performed upon a reasonable request to the corresponding author.

Supplementary material

Supplementary material is available at European Heart Journal Open online.

Acknowledgements

We are indebted to the participants and staff of the Copenhagen General Population Study and the Copenhagen City Heart Study. We also thank the participants of the UK Biobank study,¹⁰ the Global Lipids Genetics Consortium (GLGC),¹¹ the FinnGen Research Project, and the consortia for making their data publicly available.

Funding

This work was supported by the Copenhagen University Hospital—Rigshospitalet and Copenhagen University Hospital—Herlev and Gentofte. The funding sources had no role in the study design, collection of data, writing of the report, or decision to submit the article for publication.

References

Author notes