Participants
Plasma metabolic biomarkers were quantified in 118,461 baseline plasma samples. Table 1 delineates the characteristics of participants in the UK Biobank. In a single assay, each specimen was subjected to quantitative analysis of 249 metabolic measurements, which included 168 absolute level measurements and 81 ratio-based assessments. The metabolic profile was classified into various categories including amino acids, fatty acids, glycolysis metabolites, markers of lipoprotein, cholesterol, free cholesterol, cholesteryl esters, triglycerides, phospholipids, total lipids, size and Apo-lp, fatty acids, cholines, glycolysis and amino acids, ketone bodies, inflammation and fluid balance subclasses (Fig. 1).
Table 1
Baseline characteristics of study participants by incident sepsis
| Total population | Without incident sepsis | Incident sepsis | P-value |
No. of participants | 106,533 | 103,047 | 3486 | |
Age at recruitment, years | 56.56 ± 8.09 | 56.42 ± 8.09 | 60.67 ± 6.87 | < 0.001 |
Sex, n(%) | | | | < 0.001 |
Male | 50,027 (46.96) | 47,988 (46.57) | 2039 (58.49) | |
Female | 56,506 (53.04) | 55,059 (53.43) | 1447 (41.51) | |
Ethnicity, n (%) | | | | 0.036 |
White | 101,042 (95.25) | 97,710 (95.22) | 3332 (96.00) | |
Non-white | 5040 (4.75) | 4901 (4.78) | 139 (4.00) | |
Unknow | 451 | 436 | 15 | |
Qualifications, n (%) | | | | < 0.001 |
College/university degree | 34,143 (32.45) | 33,334 (32.74) | 809 (23.59) | |
Others | 71,090 (67.55) | 68,469 (67.26) | 2621 (76.41) | |
Unknow | 1300 | 1244 | 56 | |
Socio-economic status (Townsend deprivation index) | | | | < 0.001 |
Q1 | 21,030 (19.77) | 20,458 (19.85) | 572 (16.41) | |
Q2 | 21,440 (20.15) | 20,805 (20.19) | 635 (18.22) | |
Q3 | 21,276 (20.00) | 20,621 (20.01) | 655 (18.79) | |
Q4 | 21,313 (20.03) | 20,595 (19.99) | 718 (20.60) | |
Q5 | 21,333 (20.05) | 20,428 (19.82) | 905 (25.96) | |
Unknow | 141 | 140 | 1 | |
Body mass index, kg/m2 | 27.53 ± 4.77 | 27.48 ± 4.74 | 29.08 ± 5.46 | < 0.001 |
Unknow | 383 | 352 | 31 | |
Smoking status, n (%) | | | | < 0.001 |
Never | 57,877 (54.61) | 56,411 (55.02) | 1466 (42.39) | |
Past | 36,889 (34.80) | 35,425 (34.55) | 1464 (42.34) | |
Current | 11,224 (10.59) | 10,696 (10.43) | 528 (15.27) | |
Unknow | 543 | 515 | 28 | |
Alcohol drinking, n (%) | | | | < 0.001 |
Never | 4693 (4.42) | 4495 (4.37) | 198 (5.70) | |
Past | 3902 (3.67) | 3693 (3.59) | 209 (6.02) | |
Current | 97,686 (91.91) | 94,622 (92.04) | 3064 (88.27) | |
Unknow | 252 | 237 | 15 | |
C-reactive protein level, mg/dl | 2.61 ± 4.37 | 2.57 ± 4.29 | 3.97 ± 6.07 | < 0.001 |
Unknow | 4961 | 4779 | 182 | |
Diabetes mellitus, n (%) | 5764 (5.41) | 5269 (5.11) | 495 (14.20) | < 0.001 |
Hypertension, n (%) | 30,776 (28.89) | 29,197 (28.33) | 1579 (45.30) | < 0.001 |
Myocardial infarction, n (%) | 2526 (2.37) | 2301 (2.23) | 225 (6.45) | < 0.001 |
Stroke, n (%) | 1848 (1.73) | 1710 (1.66) | 138 (3.96) | < 0.001 |
Renal failure, n (%) | 1487 (1.40) | 1353 (1.31) | 134 (3.84) | < 0.001 |
Liver disease, n (%) | 780 (0.73) | 743 (0.72) | 37 (1.06) | 0.020 |
Cancer, n (%) | 7977 (7.49) | 7464 (7.24) | 513 (14.72) | < 0.001 |
Table 2
Associations of the first 13 metabolic biomarker principal components with incident sepsis and sepsis death
Principal component | Incident sepsis | Sepsis death |
Hazard ratio (95% CI)* | P-value | Hazard ratio (95% CI)* | P-value |
PC1 | 1.04 (1.00, 1.08) | 0.049 | 1.06 (0.98, 1.15) | 0.273 |
PC2 | 0.91 (0.88, 0.95) | 6.04×10− 06 | 0.96 (0.88, 1.04) | 0.482 |
PC3 | 0.99 (0.95, 1.02) | 0.420 | 0.99 (0.91, 1.07) | 0.825 |
PC4 | 0.89 (0.86, 0.91) | 3.38×10− 13 | 0.88 (0.82, 0.95) | 3.68×10− 03 |
PC5 | 0.96 (0.92, 0.99) | 0.020 | 1.00 (0.92, 1.08) | 0.933 |
PC6 | 1.09 (1.06, 1.13) | 1.46×10− 06 | 1.28 (1.18, 1.38) | 4.23×10− 8 |
PC7 | 0.94 (0.92, 0.98) | 1.07×10− 03 | 0.94 (0.87, 1.01) | 0.144 |
PC8 | 1.03 (1.02, 1.04) | 4.08×10− 06 | 1.03 (1.00, 1.06) | 0.115 |
PC9 | 1.04 (1.00, 1.08) | 0.052 | 1.11 (1.02, 1.21) | 0.069 |
PC10 | 0.94 (0.91, 0.97) | 4.39×10− 04 | 0.89 (0.83, 0.95) | 3.68×10− 03 |
PC11 | 0.96 (0.92, 0.99) | 0.016 | 0.99 (0.92, 1.06) | 0.813 |
PC12 | 1.01 (0.98, 1.04) | 0.402 | 0.98 (0.91, 1.05) | 0.733 |
PC13 | 1.03 (0.99, 1.06) | 0.161 | 1.08 (1.00, 1.17) | 0.115 |
Models were adjusted for age, sex, ethnicity, qualifications, socio-economic status, body mass index, smoking status, alcohol drinking, C-reactive protein level, diabetes mellitus, hypertension, myocardial infarction, stroke, renal failure, liver disease, cancer. |
In the NMR cohort comprising 106,533 individuals, 3,486 cases of sepsis as defined by the study were identified, and among these, 635 instances of sepsis-related mortality occurred (Table 1 and sup table1). In a study of 3,486 patients diagnosed with sepsis, 8.49% were male, average age was 60.67 years and a vast majority (96%) were Caucasian. Factors more commonly observed in these patients included lower education levels, lower socio-economic status, higher body mass index, habits of smoking and alcohol consumption, and prevalent health conditions such as diabetes, hypertension, myocardial infarction, strokes, renal and liver diseases, and cancers. Additionally, individuals with sepsis exhibited higher baseline inflammatory states, evidenced by elevated C-reactive protein levels (Table 1). The characteristics of this sepsis population are consistent with those reported in previous studies.[3, 18, 20]
Odds of sepsis incidence and death
The results showed that cholesterol/triglycerides in L-, XL-, IDL-HDL and L-LDL fractions (at the 3 to 5 o'clock position) and components lipid of plasma membrane (total cholines, phosphatidylcholines, sphingomyelins, phosphoglycerid) are protective factors for the onset sepsis. Moreover, metabolites of glycolysis, lipolysis (ketone bodies), inflammation and fluid balance are risk factors for the death from sepsis (Fig. 2). In details, the HR of incident sepsis of cholesterol concentrations was 0.92 in IDL (95% CI, 0.89–0.95), 0.90 in large LDL(95% CI, 0.87–0.93), 0.90 in medium LDL(95% CI, 0.87–0.93), and 0.91 in small LDL particles(95% CI, 0.88–0.94) (P < 0.001) (Fig. 2 and Sup Table 2).
For traditional healthy lipid (DHA, polyunsaturated fatty acids), the HR of incident sepsis was 0.90 in docosahexaenoic acid (95% CI, 0.87–0.94), 0.92 in Linoleic acid (95% CI, 0.89–0.96), 0.90 in Omega-3 (95% CI, 0.87–0.93), 0.92 in Omega-6 (95% CI, 0.89–0.95). (P < 0.001). For plasma membrane lipid components, the HR for incidence sepsis ranged from 0.93 to 0.95. In the case of metabolites related to glycolysis and lipolysis (ketone bodies), the HR for incident sepsis was between 1.02 and 1.07. Meanwhile, metabolites associated with inflammation exhibited the highest HR, at 1.10 (95% CI, 1.06–1.14). (Sup Table 2).
Further age-stratified analyses (aged ≥ 60 and < 60 years) indicated that the effects of lipolysis (acetone, acetoacetate, acetate) and triglycerides on the incidence and mortality of sepsis were not statistically significant in the population aged < 60 years, while the remaining inflammation and fluid balance metabolites were statistically significant in both age strata (Fig. 3A and B).
In the subgroup analysis by gender, the effects of lipolysis metabolites (acetone, acetate), the inflammatory metabolite (3-hydroxybutyrate), and triglycerides on the incidence and mortality of sepsis were not statistically significant in females (Fig. 4).
Given the critical role of chronic inflammation in the development of sepsis [21–23] and the notable elevation of CRP levels in sepsis patients, we conducted a subgroup analysis on the occurrence of sepsis using CRP levels. Protective lipids remained protective factors in both low and high CRP level groups. Similarly, metabolites related to glycolysis, as well as those associated with inflammation, were also identified as risk factors for the onset of sepsis (Fig. 5).
For sepsis mortality, certain lipids may also serve as protective factors, such as cholesterol in medium (HR: 0.89, 95%CI, 0.82–0.97), P < 0.001) and Large LDL particles(HR:0.9, 95%CI, 0.82–0.97), docosahexaenoic acid (HR: 0.83, 95%CI, 0.76–0.91)), Omega-3 (HR:0.81, 95%CI, 0.75–0.89, ). Glycolysis-related metabolites (lactate) was risk factor (HR: 1.04, 95%CI, 0.96–1.1). The metabolites related lipolysis were risk factors (HR:1.09–1.14). (Fig. 2 and Sup Table 2).
In metabolomics research, albumin is typically categorized as a protein or a marker of nutritional status. However, considering its crucial role in regulating fluid volume in sepsis, we have classified it under fluid balance. Albumin serves as a protective factor for both the incidence (HR: 0.87, 95%CI, 0.84–0.90,) and death of sepsis (HR: 0.83, 95%CI, 0.77–0.90, P < 0.001). (Fig. 2 and Sup Table 2).
Since the NMR batch results were derived from a two-year testing process, we further excluded the results from the first 2 years of follow-up to ensure the robustness of the findings. The results remained consistent with this adjustment (sup Table 2).
To investigate the dose-response relationship between metabolites and the incidence and mortality of sepsis, participants were stratified into baseline categories based on the quartiles of their metabolite levels.
The concentrations of lipoprotein and cholesterol in IDL, large LDL, medium LDL, medium VLDL and small HDL, LDL particles, metabolites refer to glycolysis and ketone bodies were positively associated with incident and death sepsis (P for trend < 0.05). Conversely, and fluid balance were inversely associated with incident and death sepsis. Similar trends are observable in components lipid of plasma membrane and unsaturated fatty acids (Figure S1). Regrettably, no statistically significant trend was observed between metabolites and sepsis death except unsaturated fatty acids (Figure S1).
The first 13 principal components (PCs) of the NMR biomarkers accounted for 90% of the total variance observed in all metabolites. The loadings for these 13 PCs are depicted in (Figure S3, 4), and the associations of these PCs with incident sepsis are presented in Figure S4. Notably, the major contributors to PC1 included cholesterol and triglyceride particle concentrations, relative lipoprotein particles, and unsaturated fatty acids.
The C-statistics of the prediction model that included a set of established risk factors were 0.729 (95% CI: 0.721, 0.737) for sepsis incidence and 0.782 (95% CI: 0.764, 0.799) for sepsis mortality. After adding the first 13 PCs of the NMR metabolites biomarkers to the conventional prediction model, the c-statistic increased to 0.737 (95% CI: 0.729, 0.744) for sepsis incidence and to 0.792 (95% CI: 0.774, 0.807) for sepsis mortality.