Association of Plasma Fatty Acid Profile With Trajectory of Multimorbidity and Mortality: A Community-Based Longitudinal Study

Abstract

Background

Plasma fatty acids have been linked to various chronic diseases and mortality, but the extent to which fatty acids are associated with the trajectory of multimorbidity remains unclear. We investigated the association of fatty acid profile with multimorbidity trajectories and event-free survival.

Methods

Within the UK Biobank, 138,685 chronic disease-free participants were followed for up to 16 years. Seventeen plasma fatty acids were measured by nuclear magnetic resonance. A comprehensive healthy fatty acid score (HFAS) was constructed using LASSO regression. Incidence of chronic diseases and death were ascertained through linkages to medical and death records. Event-free survival was defined as survival without chronic diseases or death. Data were analyzed using a linear mixed-effects model, Cox regression, and Laplace regression.

Results

High HFAS was associated with lower risk of chronic diseases/death (hazard ratio [HR]: 0.907, 95% confidence interval [CI]: 0.888–0.925) and prolonged event-free survival time by 0.636 (95% CI: 0.500–0.774) years compared with low HFAS. High HFAS was also associated with a slower accumulation trajectory of multimorbidity (β: −0.042, 95% CI: −0.045 to −0.038). There was a significant multiplicative interaction between moderate-to-high HFAS and healthy lifestyle on chronic disease/death (p for interaction = .002) and multimorbidity accumulation trajectories (p for interaction < .001).

Conclusions

A healthier plasma fatty acid metabolic profile is associated with a slower accumulation of multimorbidity and prolonged event-free survival time. A healthy lifestyle may strengthen the protective association of HFAS with the risk of chronic diseases/death and the accumulation trajectory of multimorbidity.

Multimorbidity, defined as the co-existence of 2 or more chronic diseases (1), represents a significant challenge in the field of global public health. One-third of the adult population is estimated to suffer from multiple chronic diseases (2). The concurrence of multiple chronic diseases leads to an exponential increase in mortality and a substantial decrease in life expectancy in contrast to a single chronic disease (3,4). Therefore, the identification of potentially modifiable risk factors for chronic disease, death, or multimorbidity has significant implications for reducing the global health burden and promoting healthy aging. Among these factors, metabolic biomarkers, including fatty acids, have been shown to influence various health outcomes and may serve as important indicators for assessing the risk of multiple chronic conditions.

Hydrocarbon chains with varying lengths and desaturation are the composition of fatty acids, which are involved in energy metabolism and storage and play a crucial role as signaling molecules (5). Fatty acids are often classified based on their saturation into saturated fatty acid (SFA), monounsaturated fatty acid (MUFA), and polyunsaturated fatty acid (PUFA). Fatty acids are currently being considered an attractive preventive and therapeutic target for the management of certain metabolic diseases and cancer (6). Several longitudinal studies have reported that high levels of omega-3 (ω-3) fatty acids such as eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), docosapentaenoic acid (DPA), alpha-chidonic acid (ALA), and arachidonic acid (AA) in omega-6 (ω-6), as well as a low (ω-6)/(ω-3) ratio, are associated with a decreased risk of several chronic diseases (ie, type 2 diabetes [T2D], stroke, dementia, and psychiatric disorder) (7–10) and mortality (11–13). Some other cohort studies have suggested that ω-3, ω-6, ω-6/ω-3 ratio, EPA, and ALA are not related to the occurrence of ischemic stroke, cardiovascular disease (CVD), T2D (14–16), and mortality (17,18). Additionally, several observational studies have shown that fatty acids are associated with multimorbidity (19–21). However, most of the previous studies have only focused on individual fatty acids, limited chronic diseases, or a single mortality outcome. Life expectancy has increased, but the high prevalence of many chronic diseases and the lack of a corresponding increase in disease-free life could pose a substantial burden on the individual, society, and healthcare systems. In light of this, composite indicators of chronic disease-free survival possess notable advantages in the domain of comprehensive health assessment. By amalgamating disease occurrence and survival time, it not only mirrors the holistic health status but also has the capacity to delineate the disparities in quality of life among diverse individuals (22). However, no studies have yet explored the association of fatty acid profile with multimorbidity trajectories and disease-free survival.

It is well accepted that healthy lifestyle factors (i.e., normal body mass index [BMI], non-smoking, not drinking heavily, and regular physical activity) can reduce the onset of many chronic diseases such as T2D (23), CVD (24), dementia (25), and mortality (26). Meanwhile, some studies have shown that a healthy lifestyle is related to fatty acid metabolism, including high DHA (27) and AA (28), as well as low SFA (29) and ω-6/ω-3 ratio (30). However, to the best of our knowledge, no studies have explored the interaction of lifestyle and fatty acid profiles on chronic disease-free survival.

Based on the previous knowledge, we hypothesized that fatty acids profile might have an impact on event-free survival. In this community-based longitudinal study from the UK Biobank, we aimed to: (a) explore the association of plasma fatty acids profile, quantified using a healthy fatty acid score (HFAS), with event-free survival and multimorbidity accumulation; and (b) examine the interplay between HFAS and lifestyle on chronic disease/death and multimorbidity.

Methods and Materials

Study Design and Population

The UK Biobank recruited 502,412 participants aged 37–73 years between 2006 and 2010 from 22 assessment centers. Upon study entry, participants to provided sociodemographic and health-related information via touchscreen questionnaires, physical assessments, and blood draw. The UK Biobank study received ethical approval from the North West Multi-Centre Research Ethics Committee (Ref 11/NW/0382), and all registered participants provided informed and written consent. This research was conducted using the UK Biobank resource (Application Number 67048).

Of the 502,412 subjects, after excluding 228,312 without baseline plasma fatty acid information and 135,415 with chronic diseases at baseline, a total of 138,685 participants were included in the current study (Supplementary Figure 1).

Assessment of Fatty Acids

Detailed information on the nuclear magnetic resonance (NMR) platforms and experimentation can be obtained in the UK Biobank research document (https://biobank.ctsu.ox.ac.uk/ukb/ukb/docs/nmrm_companion_doc). Briefly, EDTA plasma samples were stored in a freezer at −80°C. Before preparation, frozen samples were slowly thawed at +4°C overnight, and then mixed gently and centrifuged (3 minutes, 3400g, +4°C) to remove possible precipitate, and finally measured by NMR for fatty acid levels including total fatty acids, DHA, linoleic acid (LA), SFA, MUFA, PUFA, ω-3, ω-6, DHA/FA, LA/FA, SFA/FA, MUFA/FA, PUFA/FA, PUFA/MUFA, ω-3/FA, ω-6/FA, and ω-6/ω-3 in plasma samples collected from 2007 to 2010 at baseline. Outliers were detected using a triple interquartile range (IQR). Outliers below Q1 − 1.5 × IQR were reassigned as Q1 − 1.5 × IQR and outliers above Q3 + 1.5 × IQR were reassigned as Q3 + 1.5 × IQR. All fatty acids were converted to z-scores before analysis.

Calculation of HFAS

HFAS was constructed using the least absolute shrinkage and selection operator (LASSO) regression with chronic diseases/death. LASSO regression could handle the strong collinearity among variables, thus, we incorporated fatty acids and fatty acid ratios simultaneously when constructing HFAS. First, the data set was divided into a training set (selecting variables and training the HFAS model) and testing model (validating the HFAS model) based on the ratio of 4:1. Supplementary Figure 2 and Supplementary Table 2 show the weights for identifying individual fatty acids in the training set by LASSO regression. The coefficients on linoleic acid (LA) and the ratios of LA and MUFA to total fatty acid (FA) were zero and thus were not involved in the calculation of HFAS. Second, in order to ascertain the consistency in the direction of the association between various fatty acids and fatty acid ratios and chronic disease/death, within the framework of this study, a momentous step was undertaken. The fatty acid coefficients that demonstrated a positive correlation with chronic disease/death were reversed and reassigned. Thereby, it could be ensured that all the fatty acids and fatty acid ratios would exhibit a negative correlation with chronic disease/death. That is to say, the higher the score of each item was, the lower the incidence rate of chronic diseases or the mortality rate could be. Only in this way could the score constructed be called the health fatty acid score. Finally, the weighted sum of fatty acids calculated by the β coefficients from the LASSO regression yielded the final score of HFAS, with a higher score indicating higher levels of healthy fatty acids in the plasma circulatory system. Finally, the HFAS ranged from −1.78 to 1.72. We further tertiled HFAS into low [as a reference], moderate, and high levels. In this study, we determined the minimum λ = 0.00013 and mean square error = 17.16059 for the model using 10 cross-validation iterations of LASSO regression. The model will generate predicted values for the target variable based on the input features of the test samples. Then, C-Index (reflecting the explanatory power of the model) and the Brier Score (assessing the predictive accuracy of the model) were used to validate the HFAS model. The final model has a C-Index = 0.545 and the Brier Score = 0.256.

Assessment of Event-Free Survival and Multimorbidity

A total of 60 chronic conditions were ascertained by ICD-10 codes (Supplementary Table 3) and the death registry (31). The composite event was defined as the occurrence of any chronic disease or death, whereas the nonoccurrence of the event was event-free survival (that is, neither any chronic disease nor death occurred). Given the residual damage, metabolic memory, and persistent nature of chronic diseases, once detected, chronic diseases were considered present in all successive follow-up waves. Multimorbidity was defined as the number of chronic conditions listed above per year.

Assessment of Covariates

At baseline, information on age, sex, education, race, socioeconomic status, smoking status, alcohol drinking status, physical activity, and social activity was collected through the computerized touch-screen questionnaire and interview.

Education was categorized as college or noncollege. Race was dichotomized as white or non-White (including Asian, Black, and mixed background). Townsend Deprivation Index (TDI), which combined information on housing, employment, and car availability, was calculated based on census and postcode before participant recruitment, reflecting the subject’s socioeconomic status. Smoking and alcohol drinking status were both categorized as never, former, or current. BMI was calculated as weight (kg) divided by squared height (m²). Regular physical activity was defined as at least 150 minutes of moderate-intensity activity per week, and at least 75 minutes of vigorous-intensity activity per week or its equivalent (32). A social connection was evaluated based on responses to the question “How often do you visit friends or have them visit you” and classified as high (“almost daily,” “2-4 times a week,” and “about once a week”) and low levels (“about once a month,” “once every few months,” “never or almost never,” and “no friends/family outside household”).

In the current study, we considered 4 healthy lifestyle-related factors: being a nonsmoker, nondrinker, regular physical activity, and having a normal BMI. A healthy lifestyle was defined as having more than 2 healthy lifestyle factors (33).

Statistical Analysis

Baseline characteristics of the study population by HFAS category were compared using analysis of variance (ANOVA) for continuous variables and chi-square tests for categorical variables. Two-tailed p-value < 0.05 was considered statistically significant. All statistical analyses were performed using R Version.4.1.0 (R Institute for Statistical Computing, Vienna, Austria) and Stata SE 17.0 for Windows (StataCorp LLC, College Station, TX). A value of p < .05 (2-tailed) was considered statistically significant.

Cox proportional hazard models were used to examine the hazard ratio (HR) and 95% confidence interval (CI) for HFAS (continuous or categorical) associated with event-free survival. The follow-up time was calculated as the time from baseline (2006–2010) until the occurrence of any chronic disease, death, or January 2022, whichever occurred first. The proportional hazards assumption was assessed using the Schoenfeld residual method, and violations of proportionality were observed for HFAS, BMI, sex, and physical activity. To deal with this, we constructed time-dependent coefficients for these variables in the Cox proportional hazard model. To assess multiplicative interaction, we included the cross-product term of HFAS (low vs moderate-to-high) and lifestyle (unhealthy vs healthy) in the model. The 50th percentile difference (PD) and 95% CI at the onset time of developing event-free survival in relation to HFAS were estimated using Laplace regression.

Linear mixed-effects models were performed to determine multimorbidity accumulation trajectories. The fixed effects included HFAS, follow-up time (year), and their interaction to test the differences in multimorbidity accumulation trajectories by HFAS level. The random effects included random intercept and slope, allowing for individual differences to be reflected at baseline and across follow-up. The assumptions of linear mixed-effect model, including linearity, homoscedasticity for the residuals, and normality of the residuals, were tested and no violations were observed. We additionally assessed the joint effect of HFAS (low vs moderate-to-high) and lifestyle (unhealthy vs healthy) on the accumulation trajectory of multimorbidity. Then, the statistical interactions were tested by examining the cross-product term in the multivariable-adjusted model.

In sensitivity analysis, we repeated the analyses after (a) using multiple imputations with 5 replications for randomly missing covariates based on the chained equation method (including race, education, TDI, BMI, smoking status, drinking status, physical activity, and social connection); (b) excluding incident chronic diseases/death cases occurring in the first 3 years from baseline; (c) stratifying by age; (d) using the competing risks model (the Fine–Gray model) with death as the competing event; (e) using the generalized linear mixed models with or negative binomial distribution to examine the association between HFAS and the accumulation trajectory of multimorbidity; and (f) evaluating the associations of HFAS with chronic diseases and mortality, respectively.

Results

Baseline Characteristics of the Study Population

Among all participants, the mean age was 54.83 (standard deviation: 8.06) years, and 53.59% were female at baseline. Compared with participants with low HFAS, those with high HFAS were more likely to be older, female, college-educated, non-White, nonsmoker, nondrinker, to have lower TDI, lower BMI, regular physical activity, high social connection, and healthy lifestyle (Table 1). The demographic baseline characteristics of the training and testing sets are indicated in Supplementary Table 1.

Association of HFAS with Event-Free Survival

During the follow-up (median: 11.53 years, IQR: 6.47–12.98 years), 74,444 (53.68%) participants had incident chronic diseases or death. Among 73,855 participants with incident chronic diseases, 4,373 (5.92%) died. In multivariable-adjusted Cox regressions, higher HFAS (as a continuous variable) was associated with a lower risk of chronic diseases/death (HR: 0.796, 95% CI: [0.758–0.836]). The HRs (95% CIs) of chronic diseases/death were 0.882 (95% CI: 0.841–0.926) for moderate HFAS and 0.807 (95% CI: 0.766–0.849) for high HFAS, compared to low HFAS (Table 2). In multivariable-adjusted Laplace regressions, participants with moderate/high HFAS had longer event-free survival 0.349 (50th PD, 95% CI: 0.220–0.478) and 0.636 (50th PD, 95% CI: 0.500–0.774) years, respectively (Table 2).

Supplementary Table 4 shows that DHA, LA, ω-6, PUFA, PUFA/MUFA ratio, the ratios of DHA, LA, ω-6, and PUFA to total FA were inversely associated with the risk of chronic diseases/death, whereas MUFA, MUFA/total FA ratio, and SFA/total FA ratio were positively related to the occurrence of chronic diseases/death among the whole population. The associations between HFAS and chronic diseases/death were significant in both the training set and the testing set (Supplementary Table 6). The higher HFAS had longer event-free survival among training and testing sets ( Supplementary Table 7).

Association Between HFAS and Multimorbidity Accumulation Trajectories

Table 3 and Figure 1a show the association between HFAS and the accumulation trajectory of multimorbidity over follow-up. Participants with moderate HFAS (β: −0.037, 95% CI: −0.040 to −0.034) and high HFAS (β: −0.042, 95% CI: −0.045 to −0.038) had a slower multimorbidity accumulation trajectory, compared with those with low HFAS. A higher level of HFAS remained associated with a slower accumulation trajectory of multimorbidity in both the training and testing sets (Supplementary Table 7 and Supplementary Figure 2).

The Joint Effect of HFAS and Lifestyle on Chronic Disease/Death and Multimorbidity

In the joint effect analyses (Figure 2a and Supplementary Table 8), compared with individuals with low HFAS and an unhealthy lifestyle, those with moderate-to-high HFAS and a healthy lifestyle (HR: 0.664, 95% CI: 0.628–0.701) had the lowest risk of chronic diseases/death, with a significant multiplicative interaction (p for interaction = .002).

Individuals with moderate-to-high HFAS and a healthy lifestyle (β: −0.080, 95% CI: −0.084 to −0.076) had the slowest accumulation trajectory of multimorbidity, compared with those with low HFAS and an unhealthy lifestyle. There was a significant multiplicative interaction between moderate-to-high HFAS and a healthy lifestyle on slower multimorbidity accumulation trajectories (p for interaction < .001; Figure 1b, 2b, andSupplementary Table 9).

Sensitivity Analysis

When we performed the following sensitivity analyses, the study results did not substantially change: (a) performing multiple imputation to account for missing values of some covariates (including race, education, TDI, BMI, smoking status, drinking status, physical activity, and social connection; Supplementary Table 10); (b) excluding incident cases of chronic diseases/death occurring in the first 3 years of follow-up (n = 132 600; Supplementary Table 11); (c) stratifying by age (Supplementary Table 12); (d) taking into account the competing risk of death using Fine and Gray regression (Supplementary Table 13); (e) using the generalized linear mixed models with or negative binomial distribution to examine the association between HFAS and the accumulation trajectory of multimorbidity (Supplementary Table 14); assessing the associations of HFAS with chronic diseases and mortality, respectively (Supplementary Table 15).

Discussion

In this community-based longitudinal study from the UK Biobank, we combined individual fatty acids and fatty acid ratios to construct a comprehensive HFAS using LASSO regression. We found that (a) higher HFAS was associated with decreased risk of chronic diseases/death and prolonged event-free survival; (b) higher HFAS was related to a slower accumulation trajectory of multimorbidity; and (c) there was a significant interaction between moderate-to-high HFAS and a healthy lifestyle on decreased risk of chronic diseases/death and slower accumulation trajectory of multimorbidity.

Several longitudinal studies have shown that high plasma levels of EPA, DHA, DPA, and ALA as well as low ω-6/ω-3 ratio is related to a reduced risk of T2D (7), stroke (8), dementia (9), and psychiatric disorders (10); however, other studies have failed to find such associations (14–16). Additionally, some studies have suggested that low plasma ω-6/ω-3 and high ω-6, ω-3, and ALA were associated with lower mortality (11–13), but inconsistencies also exist (17,18). The composite endpoint of 60 chronic diseases and death, that is, chronic disease-free survival, enables a comprehensive assessment of the corresponding loss of healthy lifespan, as it takes into account both morbidity and mortality and thus captures a summary measure of the “quantity” and “quality” of life. Furthermore, our construction of an overall HFAS using a combination of fatty acids and fatty acid ratios provides a more comprehensive characterization of an individual’s fatty acid profile. In contrast to examinations of individual fatty acids which have been reported in prior studies, the HFAS takes into account the intricate associations within fatty acids, which may provide new insights into lengthening healthspan and identifying modifiable factors for the prevention and management of multiple chronic diseases. In the present study, we found that a healthy plasma fatty acid profile was associated with longer event-free survival.

Observational studies have reported that PUFA, DHA, LA, and ALA are associated with a lower risk of multimorbidity (20,21,34), but some inconsistent findings have also been observed (35,36). Only one cohort study of older adults from Spain has examined the relationship between individual plasma fatty acids and multimorbidity, (comprising 60 mutually exclusive chronic diseases) reporting that higher plasma levels of ω-6 were linked to a lower incidence of multimorbidity among the elderly (34). Interestingly, a recent cross-sectional study of the same population has found that lipid metabolism factors (including PUFA, ω-3, ω-6, LA, MUFA, and SFA, etc.) are associated with a higher risk of cardiac metabolic multimorbidity (36). However, several studies have reported an association of lipid metabolites with an increased risk of several chronic diseases (37,38), and thus grouping fatty acids with these molecules may mask their true effect on multimorbidity. Individuals with multimorbidity may require multiple therapeutic interventions to treat different diseases, but interactions between some therapies can be detrimental, thus complicating the management of multimorbidity (39). Fatty acids are considered an effective strategy for intervening in multiple diseases and multimorbidity (39,40). Although previous studies have provided evidence regarding the possible association of fatty acid metabolites with multimorbidity, they have mostly been limited to specific populations (usually older populations) and multimorbidity based on limited amounts of chronic diseases. Latent class analyses focus on identifying latent classes in the data, and growth curve mixed models, whereas capable of identifying subgroups with different developmental trajectories, have relatively weak parameter explanatory power at the individual level. Thus, in our study, differences in multimorbidity trajectories across HFAS levels were tested using linear mixed-effects models which can comprehensively and directly explain both fixed and random effects at the individual level. Both fatty acid concentrations and their ratios provide complementary information about the fatty acid profile. Concentrations reflect the absolute levels of individual fatty acids in plasma, whereas ratios capture the relative balance between specific fatty acids, which is critical for understanding metabolic pathways and health outcomes. Incorporating both concentrations and ratios allowed us to account for both absolute and relative effects, enhancing the explanatory power of the HFAS. Moreover, the LASSO regression method can also effectively handle the potential collinearity among these data to obtain reliable results. In the current study, we leveraged data on 60 different chronic diseases in a large-scale community-based longitudinal study and found that higher HFAS was related to a significantly slower accumulation of multimorbidity trajectories.

Lifestyle factors (BMI, smoking, alcohol consumption, and physical activity) are well-established modifiable risk factors for a wide range of chronic diseases and mortality (23–26). A healthy lifestyle, such as maintaining a normal BMI, nonsmoking, avoiding heavy alcohol consumption, and engaging in regular physical activity, can modify an individual’s fatty acid profile by, for example, decreasing circulatory SFA and ω-6/ω-3 ratio (29,30), and increasing DHA and AA levels (27,28). We therefore investigated the multiplicative interaction of HFAS with lifestyle and found that a healthy lifestyle may reinforce the beneficial effects of high HFAS on chronic diseases/death and the accumulation trajectory of multimorbidity.

Several mechanisms may underlie the association of fatty acids with chronic diseases and death. First, DHA may exert its antioxidant activity by enhancing mitochondrial function and biogenesis, thereby increasing the body’s resistance to chronic diseases and alleviating their symptoms (41). Second, SFA leads to an increased secretion of tumor necrosis factor-alpha (TNF-α) as a component of the inflammatory response (42). The inflammatory response is recognized as a crucial factor in the development of various chronic diseases and mortality (43). Third, PUFA-derived oxylipins regulate various inflammation-related outcomes, including immune cell chemotaxis and cytokine production (44). Finally, a low ω-6/ω-3 inhibits the nuclear factor kappa-B (NF-κB) signalingly (45), which once activated enhances the synthesis and secretion of chemokines (ie, monocyte chemotactic protein-1) from adipocytes or hepatocytes, resulting in pro-inflammatory macrophage infiltration (46).

The strengths of this study include the construction of a comprehensive HFAS capturing the complex associations among various fatty acids, as well as the exploration of event-free survival and multimorbidity in a large-scale community-based longitudinal study. However, the limitations of the study should also be acknowledged. First, the majority of subjects in the UK Biobank are likely to be healthier than the wider population of the United Kingdom, making it difficult to generalize our results to populations other than Europeans, and the association between HFAS and chronic diseases/death may have been overestimated. Second, we identified a limited variety of baseline fatty acids using NMR, which made it impossible to assess their time-varying relationship with the risk of chronic disease/death, in addition to requiring further external validation of the constructed HFAS in this study. Third, because data on incident chronic diseases and death were based on electronic health records, some cases may have been missed, which may have overestimated the relationship between HFAS and chronic diseases/death. Finally, our classification of the 60 included diseases as persistent throughout follow-up after each initial diagnosis does not account for instances of disease reversal, which could have led to an overestimate of multimorbidity.

In conclusion, this study demonstrated that a healthier fatty acid profile is associated with a decreased risk of chronic diseases/death and a slower accumulation of multimorbidity trajectories. Our findings also provide evidence that a healthy lifestyle may strengthen the protective association of HFAS with the risk of chronic diseases/death and the accumulation of multimorbidity trajectories.

Funding

The study was supported by the National Key Research and Development Program of China (2021YFA1301202), the National Natural Science Foundation of China (82473677, 82273676, 32400963, 92357305), the Natural Science Foundation of Chongqing Municipality (2024NSCQ-MSX3537), Swedish Research Council (No. 2021-01647), Swedish Council for Health Working Life and Welfare (2021-01826), Lindhés Advokatbyrå AB (2021-0134), and Stiftelsen för Gamla Tjänarinnor (2021-2022).

Conflict of interest

None.

Data Availability

The data that support the findings of this study are available from the UK Biobank. The UK Biobank resources are available and can be accessed through applications on their website (https://www.ukbiobank.ac.uk/).

Acknowledgments

The authors thank all participants of the UK Biobank study who have dedicated their valuable time toward this project and the UK Biobank team for collecting, preparing, and providing data used in this work. The research has been done using the UK Biobank resource as a part of the approved Research Application 67048.

Author Contributions

The authors’ responsibilities were as follows—W.X. and J.W.: conceived and designed the study; Y.L.: performed the statistical analyses and drafted the first draft of the manuscript; J.W., M.D., S.M., and Y.M.: revised the manuscript; J.W.: supervised the data analysis and interpretation; Z.F. and Q.Z.: primary responsibility for the final content; and all authors: read and approved the final manuscript.

References

Author notes