Mediating role of white blood cells in the relationship between the advanced lung cancer inflammation index and serum neurofilament light chain levels: A study based on NHANES 2013–2014 data

Abstract

This study aimed to explore the mediating role of white blood cell (WBC) count in the relationship between the advanced lung cancer inflammation index (ALI) and serum neurofilament light chain (sNfL) levels. We conducted a cross-sectional analysis of 1725 adult participants from the 2013 to 2014 National Health and Nutrition Examination Survey. ALI was calculated as body mass index × serum albumin/neutrophil-to-lymphocyte ratio, and participants were categorized into quartiles based on ALI distribution. Associations among ALI, WBC count, and sNfL levels were examined using weighted multivariable linear regression, nonlinear curve fitting, and mediation analysis. Higher ALI values were significantly associated with lower sNfL concentrations (β = −0.04 per unit increase; 95% confidence interval [CI]: −0.06–−0.02; P = .0007), demonstrating a clear dose-response pattern across ALI quartiles. A nonlinear relationship was identified, with a threshold effect at an ALI value of 71.64. Below this threshold, ALI showed a stronger inverse association with sNfL (β = −0.14; 95% CI: −0.21–−0.08; P < .0001), which attenuated above the threshold (β = −0.01; P = .63). WBC count partially mediated the association between ALI and sNfL, accounting for 21.04% of the total effect (95% CI: 6.85–43.18%; P = .006). ALI is inversely associated with serum sNfL levels in the general United States adult population, with WBC count acting as a significant mediator. Clinically, these findings suggest that ALI, a readily available composite marker of systemic inflammation and nutritional status, may help identify individuals at higher risk of neuroaxonal injury, thereby supporting its potential utility in population-level neurodegenerative risk stratification and early preventive assessment. Further longitudinal studies are warranted to confirm these associations and clarify their prognostic implications.

1. Introduction

Serum neurofilament light chain (sNfL), a core structural component of the neuronal cytoskeleton, exists as a heteropolymeric complex comprising light, medium, and heavy chains.^[1] This protein is predominantly localized in mature neurons and is released into cerebrospinal fluid and peripheral circulation following axonal injury.^[2] Accumulating evidence indicates that sNfL concentrations are significantly elevated in individuals with central or peripheral nervous system damage and closely reflect the extent of neuroaxonal injury.^[3] Moreover, longitudinal changes in sNfL levels have been shown to correlate strongly with disease activity, progression, and clinical outcomes in a range of neurodegenerative disorders, including multiple sclerosis,^[4] Parkinson disease, and Alzheimer disease.^[5] As such, sNfL has emerged as a sensitive and reliable biomarker of neurodegeneration and axonal integrity.

Systemic inflammation and nutritional status are increasingly recognized as important contributors to neurodegenerative processes. Chronic low-grade inflammation may disrupt the blood-brain barrier, promote microglial activation, and accelerate axonal injury, while malnutrition and metabolic dysregulation can further exacerbate neural vulnerability. The advanced lung cancer inflammation index (ALI), which integrates inflammatory burden (neutrophil-to-lymphocyte ratio [NLR]) and nutritional parameters (serum albumin and body mass index [BMI]), has been widely applied as a composite marker of systemic inflammatory and nutritional status.^[6] Beyond its original oncological context, ALI has demonstrated prognostic and clinical relevance in a variety of nonmalignant conditions, including hypertension, heart failure, and Crohn disease,^[7] suggesting its broader utility as an indicator of systemic physiological stress.^[8]

Despite growing interest in inflammation-related biomarkers and neurodegeneration, evidence directly linking ALI to neuroaxonal injury, as reflected by sNfL levels, remains limited. In particular, the potential role of peripheral leukocyte-related inflammatory processes in mediating the relationship between ALI and sNfL has not been adequately explored in population-based settings. Addressing this gap may provide important insights into how systemic inflammatory and nutritional states jointly influence neuroaxonal health.

Therefore, using nationally representative data from the 2013 to 2014 National Health and Nutrition Examination Survey (NHANES), the present study aimed to investigate the association between ALI and serum sNfL levels in United States (U.S.) adults, with a specific focus on examining the mediating role of white blood cell (WBC) count. By clarifying these relationships, this study seeks to contribute to a more integrated understanding of systemic inflammation, nutritional status, and neurodegenerative risk.

2. Methods

2.1. Study design and population

From the initial cohort of 10,175 participants in the 2013 to 2014 NHANES cycle, individuals lacking sNfL data, those with sNfL levels outside the detection range, and participants without ALI measurements were systematically excluded. Furthermore, participants with incomplete covariate data were omitted from the analysis. The final sample comprised 1725 individuals, reflecting the adult demographic of the U.S. (Fig. 1). All study protocols were approved by the National Center for Health Statistics Institutional Review Board, and written informed consent was obtained from all participants. Detailed information on NHANES methodology is provided in Supplementary Methods 1, Supplemental Digital Content 1.

Figure 1.

Flowchart of sample selection. ALI = advanced lung cancer inflammation index, N = number of participants, NHANES = National Health and Nutrition Examination Survey, sNfL = serum neurofilament light chain.

2.2. ALI and WBC measurements

Hematological indices were obtained according to NHANES protocols. The ALI was computed as a composite indicator integrating nutritional and inflammatory status, based on BMI, serum albumin, and the NLR. ALI values were then stratified into quartiles (Q1–Q4) for analysis. Detailed calculation steps and component definitions are provided in Supplementary Methods 2, Supplemental Digital Content 2.

2.3. SNfL measurements

SNfL was designated as the dependent variable. Consistent with previous research protocols (PMID: 40208889), serum samples collected between 2013 and 2014 were analyzed for neurofilament light chain (NfL) using an immunoassay developed by Siemens Healthineers, characterized by exceptional sensitivity. The detection range was 3.9 to 500 pg/mL. A detailed description of the sNfL quantification procedure and analytical methods can be found on the following URL: https://wwwn.cdc.gov/Nchs/Nhanes/2013–2014/SSSNFL_H.htm.

2.4. Covariate definitions

Sociodemographic and clinical covariates were defined according to NHANES standards and included age, sex, race/ethnicity, education level, marital status, smoking, alcohol use, diabetes, and hypertension. Definitions and diagnostic criteria are detailed in Supplementary Methods 3, Supplemental Digital Content 3.

2.5. Statistical analysis

The sampling weights (WTSSNH2Y) were employed to ensure population representativeness of NHANES data. Continuous variables were analyzed using survey-weighted linear regression (svyglm), with P values calculated and presented as weighted means ± standard error. Categorical variables were examined using survey-weighted chi-square test (svytable), with results reported as weighted percentages.

ALI was stratified based on quartile distribution. We conducted generalized linear model regression to assess the relationship between sNfL and ALI levels, calculating β coefficients and 95% confidence intervals. The model-building process commenced with an unadjusted model, progressively incorporating adjustments for age, gender, and race, with the final model additionally controlling for covariates including smoking status, alcohol consumption, marital status, education level, poverty-to-income ratio, hypertension, and diabetes.

To verify potential nonlinear relationships between variables, smooth curve fitting techniques were applied, and threshold effect analysis with a 2-piece linear model was used to identify potential inflection points. We further analyzed the relationships between WBC, ALI, and sNfL, and explored the mediating effect of WBCs on the ALI-sNfL relationship using mediation analysis.

Statistical analyses were performed using R software (R Foundation for Statistical Computing, Vienna, Austria, http://www.r-project.org) and EmpowerStats software (X&Y Solutions, Inc., Boston, http://www.empowerstats.com), with statistical significance defined as a two-sided P value < .05.

3. Results

3.1. Baseline characteristics

The study cohort comprised 1725 participants (mean age 45.5 ± 0.5 years; 858 males [49.7%], 867 females [50.3%]) with significant heterogeneity in ALI across quartiles (Q1–Q4: 35.09 ± 0.61–120.00 ± 3.58; P < .001;Table 1). SNfL levels demonstrated a graded inverse association with ALI quartiles (Q1: 19.38 ± 1.61 vs Q4: 14.47 ± 0.83 pg/mL; P = .012), paralleled by elevated WBC counts in lower ALI groups (Q1: 8.14 ± 0.21 vs Q4: 6.35 ± 0.10 × 10³ cells/μL; P < .001). Significant racial composition shifts occurred across quartiles (P < .001), with non-Hispanic Black representation rising from 6.7% (Q1) to 21.0% (Q4). Current smoking prevalence decreased progressively from 29.0% (Q1) to 17.6% (Q4) (P = .015), while borderline diabetes increased from 33.6% (Q1) to 46.7% (Q4) (P = .005). No significant interquartile differences were observed in age (P = .065), gender distribution (P = .167), socioeconomic (P = .957) or marital status (P = .455), educational attainment (P = .245), alcohol consumption (P = .373), or hypertension prevalence (P = .110).

Table 1.

Demographic characteristics of ALI.

Characteristics	Overall (n = 1725)	Q1 (n = 431)	Q2 (n = 431)	Q3 (n = 431)	Q4 (n = 432)	P value
ALI	70.23 ± 1.31	35.09 ± 0.61	55.66 ± 0.39	74.25 ± 0.46	120.00 ± 3.58	< .0001
sNfL(pg/ml)	17.06 ± 1.22	19.38 ± 1.61	18.49 ± 2.17	15.73 ± 1.40	14.47 ± 0.83	.0122
Age (yrs)	45.50 ± 0.51	47.11 ± 0.64	45.83 ± 0.64	45.28 ± 0.83	43.60 ± 1.11	.0647
Income-to-poverty ratio	2.96 ± 0.15	3.00 ± 0.21	2.95 ± 0.15	2.97 ± 0.17	2.92 ± 0.14	.9578
WBC (103 cells/μL)	6.99 ± 0.10	8.14 ± 0.21	6.70 ± 0.14	6.70 ± 0.12	6.35 ± 0.10	< .0001
Gender, %						.1668
Male	49.74	51.03	44.45	50.94	52.58
Female	50.26	48.97	55.55	49.06	47.42
Race/Ethnicity, %						< .0001
Mexican American	8.74	6.93	9.39	9.3	9.43
Other Hispanic	5.09	4.19	5.39	6.51	4.18
Non-Hispanic White	67.86	73.75	68.97	69.15	58.72
Non-Hispanic Black	11.24	6.73	8.04	10.01	20.99
Other Races	7.07	8.4	8.21	5.03	6.67
Education level, %						.2445
Less than High School	3.39	1.81	4.68	3.15	4.04
High School or GED	31.25	31.52	32.19	29.35	32.1
Above High School	65.35	66.66	63.13	67.5	63.86
Marital Status, %						.4547
Married/Living with Partner	64.95	60.92	66.73	66.76	65.47
Widowed/Divorced/Separated	13.98	16.52	12.33	11.98	15.15
Never Married	21.07	22.56	20.94	21.26	19.38
Smoking Status, %						.0145
Current	20.62	28.95	19.58	16.07	17.62
Former	23.10	22.82	21.42	24.63	23.49
Never	56.28	48.23	59	59.3	58.89
Drinking Status, %						.3726
Yes	79.01	82.98	78.75	76.74	77.47
No	20.99	17.02	21.25	23.26	22.53
Diabetes, %						.0053
Yes	14.43	13.67	15.74	13.31	15.15
No	45.66	52.72	45.36	45.66	38.18
Borderline	39.91	33.61	38.9	41.03	46.67
Hypertension, %						.1099
Yes	32.55	32.31	26.09	35.91	35.93
No	67.45	67.69	73.91	64.09	64.07

ALI = advanced lung cancer inflammation index, GED = General Educational Development, n = number of participants, sNfL = serum neurofilament light chain, WBC = white blood cells.

3.2. Relationship between ALI and sNfL

In multivariable-adjusted models, each 1-unit increase in ALI was associated with a decrease in WBC count: β = −4.48 (95% CI: −5.64–−3.91) in Model 1, β = −4.48 (95% CI: −5.33–−3.62) in Model 2, and β = −5.08 (95% CI: −5.97–−4.18) in Model 3 (P < .0001 for all). Meanwhile, a 1-unit increase in WBC count corresponded to an increase in sNfL: β = 0.88 (95% CI: 0.44–1.32) in Model 1, β = 1.06 (95% CI: 0.63–1.49) in Model 2, and β = 0.81 (95% CI: 0.35–1.27) in Model 3 (P ≤ .0006). The total effect of ALI on sNfL was estimated as β = −1.54 (95% CI: −2.40–−0.83; P < .0001), as shown in Table 2.

Table 2.

Association between ALI and sNfL.

ALI	Model 1 [β (95% CI) P value]	Model 2 [β (95% CI) P value]	Model 3 [β (95% CI) P value]
Per 1 increment	−0.04 (−0.06, −0.02) .0007	−0.04 (−0.06, −0.01) .0018	−0.04 (−0.06, −0.02) .0007
Q1	Reference	Reference	Reference
Q2	−2.86 (−5.59, −0.12) .0410	−2.44 (−5.08, 0.20) .0706	−2.25 (−4.89, 0.39) .0954
Q3	−4.81 (−7.55, −2.07) .0006	−4.37 (−7.02, −1.72) .0013	−4.61 (−7.27, −1.96) .0007
Q4	−5.29 (−8.03, −2.56) .0002	−4.59 (−7.27, −1.90) .0008	−5.00 (−7.71, −2.29) .0003
P for trend	.0001	.0006	.0002

ALI = advanced lung cancer inflammation index, CI = confidence interval, sNfL = serum neurofilament light chain.

3.3. Nonlinear relationship between ALI and sNfL

Our investigation uncovered a complex association pattern between ALI and sNfL concentrations. While conventional linear regression demonstrated a significant inverse correlation (β = −0.04, 95% CI: −0.06–−0.02, P = .0007;Table 3), segmented linear regression coupled with scatterplot visualization revealed substantial nonlinear characteristics in this relationship. Likelihood ratio testing confirmed superior model fit for the 2-segment linear model compared to the standard linear model (Fig. 2), identifying an optimal inflection point at ALI = 71.64. Below this threshold (ALI < 71.64), we observed a pronounced inverse association (β = −0.14, 95% CI: −0.21–−0.08, P < .0001), suggesting marked sNfL reduction per ALI unit increase at lower index ranges. Beyond the inflection point (ALI > 71.64), this association attenuated to non-significance (β = −0.01, 95% CI: −0.04–0.02, P = .6262), indicative of a plateau effect. Quantile analyses reinforced this nonlinear pattern, demonstrating incremental effect magnitudes across ascending ALI quartiles (Q2-Q4 vs Q1: β = −2.25, −4.61, −5.00) with a robust linear trend (P = .0002).

Table 3.

The nonlinear relationship between ALI and sNfL.

Variable	sNfL
Fitting by the standard linear model	−0.04 (−0.06, −0.02) .0007
Fitting by the 2-piecewise linear model
Inflection point	71.64
< Inflection point	−0.14 (−0.21, −0.08) < .0001
> Inflection point	−0.01 (−0.04, 0.02) .6262
P for Log-likelihood ratio	.001

ALI = advanced lung cancer inflammation index, sNfL = serum neurofilament light chain.

Figure 2.

The nonlinear relationship was characterized using smooth curve fitting. ALI = advanced lung cancer inflammation index, sNfL = serum neurofilament light chain.

3.4. Subgroup analyses

Subgroup analyses were performed to assess potential effect modifications across key demographic and clinical variables (Fig. 3). A statistically significant interaction was observed for hypertension status (P_interaction = .0447). Specifically, the protective effect of ALI on serum sNfL concentrations was attenuated in hypertensive patients (β = −0.02, 95% CI: −0.05–0.01) compared to non-hypertensive individuals (β = −0.06, 95% CI: −0.10–−0.03). Notably, diabetes status showed a borderline significant interaction (P_interaction = .0729), with stronger inverse associations observed in diabetic participants (β = −0.08, 95% CI: −0.13–−0.04) versus nondiabetic counterparts. No significant effect modifications were detected for other subgroups including sex (P = .743), age (P = .952), race/ethnicity (P = .849), socioeconomic status (P = .626), or lifestyle factors (all P_interaction > .30). The consistency of effect estimates across most subgroups (overlapping confidence intervals for male vs female, smoking status categories) suggests robustness of the primary association.

Figure 3.

Subgroup analyses of the association between ALI and sNfL. ALI = advanced lung cancer inflammation index, CI = confidence interval, OR = odds ratio, sNfL = serum neurofilament light chain, PIR = poverty-to-income ratio.

3.5. Associations Among ALI, WBC, and sNfL

In multivariable-adjusted models, higher ALI levels were significantly associated with reduced WBC counts across all models. Specifically, each 1-unit increase in ALI was linked to a decrease in WBC by β = −4.78 (95% CI: −5.64–−3.91; P < .0001) in Model 1, β = −4.48 (95% CI: −5.33–−3.62; P < .0001) in Model 2, and β = −5.08 (95% CI: −5.97–−4.18; P < .0001) in Model 3 (Table 1).Elevated WBC levels showed a strong positive association with sNfL concentrations. A 1-unit increase in WBC corresponded to a rise in sNfL by β = 0.88 (95% CI: 0.44–1.32; P = .0001) in Model 1, β = 1.06 (95% CI: 0.63–1.49; P < .0001) in Model 2, and β = 0.81 (95% CI: 0.35–1.27; P = .0006) in Model 3, after adjusting for covariates. The total effect of ALI on sNfL was β = −1.54 (95% CI: −2.40–−0.83; P < .0001), indicating that higher ALI levels were associated with reduced sNfL concentrations in Table 4.

Table 4.

Associations Among ALI, WBC, and sNfL.

Outcome	Model 1 [β (95% CI) P value]	Model 2 [β (95% CI) P value]	Model 3 [β (95% CI) P value]
sNfL	0.88 (0.44, 1.32) .0001	1.06 (0.63, 1.49) < .0001	0.81 (0.35, 1.27) .0006
ALI	−4.78 (−5.64, −3.91) < .0001	−4.48 (−5.33, −3.62) < .0001	−5.08 (−5.97, −4.18) < .0001

ALI = advanced lung cancer inflammation index, CI = confidence interval, sNfL = serum neurofilament light chain, WBC = white blood cells.

3.6. Mediation analysis of WBC in the ALI-sNfL pathway

WBC count partially mediated the ALI-sNfL association, accounting for 21.04% (95% CI: 6.85–43.18%; P = .006;Table 5) of the total effect (β = −1.54, 95% CI: −2.40–−0.83; P < .001). The indirect effect through WBC was β = −0.32 (95% CI: −0.58–−0.10; P = .006), while the direct effect of ALI on sNfL independent of WBC remained significant (β = −1.22, 95% CI: −2.01–−0.54; P < .001). This aligns with the protective role of ALI, where reduced systemic inflammation (reflected by lower WBC) contributes to decreased neuroaxonal damage (lower sNfL) (Figure 4).

Table 5.

Mediation analysis of WBC in the ALI-sNfL pathway.

Characteristics	β (%) (95% CI)	P value
Total effect	−1.54 (−2.40, −0.83)	< .0001
Mediation effect	−0.32 (−0.58, −0.10)	.0060
Direct effect	−1.22 (−2.01, −0.54)	< .0001
Proportion mediated	21.04 (6.85, 43.18)	.0060

ALI = advanced lung cancer inflammation index, CI = confidence interval, sNfL = serum neurofilament light chain, WBC = white blood cells.

Figure 4.

Mediating effect analysis of ALI on sNfL levels. ALI = advanced lung cancer inflammation index, sNfL = serum neurofilament light chain.

4. Discussion

This study, based on 1725 participants from the NHANES database, revealed an inverse association between the ALI and sNfL levels in the general adult population. This association remained consistent across subgroups after multivariable adjustment. Further mediation analysis showed that WBC count partially mediated this relationship. These findings suggest that systemic inflammation, as indicated by WBC count, may be 1 pathway linking poor nutritional and inflammatory status to neuroaxonal damage.

ALI, as a comprehensive marker of systemic inflammation, nutritional, and immune function, is negatively associated with the neuronal damage biomarker sNfL. ALI incorporates multiple physiological parameters, including BMI, serum albumin, and the NLR.^[9] A higher ALI value indicates better nutrition, lower inflammation, and stronger immune function. The significant decrease in sNfL levels may reflect reduced neuronal and axonal damage in this favorable physiological state. Regression models at 3 adjustment levels show consistent association strength and statistical significance, indicating a robust association resistant to potential confounding factors.^[10]

Non-neurological determinants (age, cardiovascular risk, and renal function) significantly enhance the predictive power for sNfL concentrations. The positive correlations between inflammatory mediators (interleukin [IL]-1β/IL-6) and sNfL validate neuroinflammatory mechanisms, as evidenced by animal studies demonstrating that cytokines (IL-2, IL-4, IL-6, tumor necrosis factor-α) drive cerebrospinal fluid NfL elevation through chronic immune activation.^[11–13] Midlife obesity (40–60 years), an independent risk factor for Alzheimer disease,^[14] involves pathogenic neuroinflammatory cascades mediated by microglial activation: a process also implicated in Parkinson disease progression.^[15–17] Oxidative stress exacerbates these pathologies via pathways such as mitochondrial dysfunction.^[18,19] Longitudinal sNfL monitoring quantifies axonal damage intensity, providing critical decision-making support for disease staging,^[20,21] therapeutic response evaluation (immunotherapies), and prognostic prediction (poststroke recovery), thereby advancing precision neurology management.^[22,23]

This study unveiled a nonlinear dose-response link with a threshold between the ALI and neuronal integrity.^[24–26] In the low ALI bracket (< 72), denoting systemic hyperinflammation, malnutrition, and immune dysregulation, sNfL levels rose exponentially (β = −0.14 per ALI unit) as ALI declined, implying considerable neuroprotective gains with each ALI unit increase. Once ALI surpassed 72, the sNfL decline plateaued (β = −0.01 per ALI unit), possibly due to a biological ceiling in neuroprotection or compensatory mechanisms. This nonlinear pattern gives rise to a dual-tier clinical intervention paradigm. For those with ALI < 72, a multidimensional approach, including anti-inflammatory, nutritional, and immune-modulating therapies, can optimize ALI parameters. This could yield a substantial sNfL reduction of 5.0 units, which is a marker of neuroprotection. Conversely, patients with ALI > 72 may need advanced neuroprotective strategies like blood-brain barrier repair and oxidative stress regulation. This study provides a quantitative assessment of the associations within the inflammation–nutrition–immunity axis in relation to neurodegeneration, and proposes a biomarker-stratified framework that may inform precision neurology.^[27,28]

4.1. Study strengths and limitations

Our study has several strengths. Using nationally representative data, we systematically examined the associations between leukocyte-mediated processes, the ALI index, and sNfL levels. This study provides initial evidence supporting a potential relationship between the ALI index and sNfL levels. Given the temporal variability of ALI and the inherent limitations of single time-point measurements in reflecting long-term inflammatory status, our findings underscore the importance of considering longitudinal ALI patterns in future investigations of neural axonal injury.^[29] Moreover, the observed associations between ALI fluctuations and sNfL levels suggest that repeated biomarker assessments may offer additional information for risk stratification in neurodegenerative conditions, although this hypothesis requires further validation in prospective studies.^[30] Future research should aim to explore standardized approaches for characterizing ALI trajectories and to assess their potential clinical relevance through longitudinal and interventional designs.

However, some limitations must be considered. The limited number of participants with NHANES 2013 to 2014 sNfL data restricts comprehensive analysis and result credibility. Despite adjusting for many potential covariates, we can’t rule out the impact of other confounding factors, such as occupational exposure history and genetic susceptibility. Moreover, the cross-sectional design prevents us from making causal inferences about the ALI index and NfL relationship. Future research using longitudinal methods and incorporating a broader range of neurobiological indicators would help examine these complex links.

5. Conclusion

In summary, our findings indicate a significant association between an elevated ALI index and higher sNfL concentrations in a nationally representative sample of U.S. adults. Given the cross-sectional design of this study, the directionality and clinical implications of this association remain to be fully elucidated. Nonetheless, these results highlight the importance of further longitudinal and mechanistic studies to validate these observations and to clarify the potential biological pathways and neurological consequences associated with an elevated ALI index in adults.

Acknowledgments

Our team would like to thank all the staff and all the subjects for their participation in the NHANES data collection.

Author contributions

Conceptualization: Shuang Wang, Hui Ma, Qimei Shi, Shoujian Zong, Guangjie Wang.

Data curation: Xiaojie Hu, Haifang Su, Guizhi Sun.

Formal analysis: Shuang Wang, Hui Ma, Shoujian Zong, Guangjie Wang.

Funding acquisition: Qimei Shi.

Investigation: Haifang Su, Guizhi Sun.

Methodology: Shuang Wang, Hui Ma, Qimei Shi, Shoujian Zong, Guangjie Wang.

Validation: Chunmei Yun.

Writing – original draft: Shuang Wang, Hui Ma, Qimei Shi, Shoujian Zong, Guangjie Wang, Xiaojie Hu.

Writing – review & editing: Shuang Wang, Hui Ma, Qimei Shi, Shoujian Zong, Guangjie Wang.