LC-MS-Based Urine Metabolomics for the Identication of Biomarkers Related to The Occurrence and Severity of Flight Fatigue

Background: Aircrew fatigue is a major contributor of operational errors in civil and military aviation, which translates a comprehensive performance associated with both neuromuscular and cognitive states. Here we have used untargeted and non-invasive urinary metabolomics to explore ight fatigue-related aberrations. In this sense, we aimed to identify biomarkers that could better monitor pilot fatigue and also assess its severity to prevent ‘nonfunctional over-reached’ state, thus promoting ight safety. Methods: In this study, 22 active-duty male pilots, who conducted different ight hour duties, were recruited to mimic different levels of fatigue. For this, respective urine samples were collected, before and after ight, and analyzed by liquid chromatography/mass spectrometry (LC/MS). Results: Except for the fatty acids and some amino acids, signicant changes on metabolite levels were observed during the progression of ight fatigue. Most of these metabolites corresponded to acyl carnitines, carbohydrates, purines and indoles. The majority of amino acids were downregulated after the ight mission. A total of 61 metabolites were found to be signicantly changed along with the extent of ight fatigue. To eciently discriminate the occurrence of ight fatigue, three candidate biomarkers (beta-guanidinopropionic acid, 3-dehydro-L-gulonate and 2-propylpent-3-enoic acid) were further characterized. Lastly, Bayes discriminant function models were established to stratify pilots with severity of fatigue and, therefore, to aid in ight risk management. Conclusion: To our knowledge, this study inaugurally provides a metabolic proling in response to ight fatigue, thus offering a novel and effective way to monitor and manage this physiological condition.


Introduction
Contemporary military aviation is stressful and complex in nature. As such, some substantial demand on attention and cognitive ability are required in order to provide safe operations (1), which potentially make pilots prone to mental fatigue, accompanied by certain physical stress. Fatigue can largely impact the success of an air mission, mostly due to its effects on a number of performance variables including reaction time, accuracy, attention, and executive decision (2,3). In fact, fatigue has been recognized as major contributing factors to operational errors (4). This condition is considered a common problem in military aviation (5), and growing evidence have indicated that fatigue can degrade neurobehavioral performance and fundamental piloting skills. The National Aeronautics and Space Administration's aviation safety report has disclosed that almost 20% of ight accidents are directly or indirectly related to fatigue (6). Flight fatigue is the most common physiological factor that leads to military accidents, costing hundreds of millions of dollars in lost equipment in addition to the incalculable value of welltrained pilots (7).
Although ight fatigue is one of the most prominent issues faced by the aviation health protection labor, there is still a lack of fast and straightforward evaluation methods as well as effective countermeasures.
The standard practice to determine whether a pilot is t to y is mainly based on the commander's personal assessment and experience, or on responses to subjective self-report measures (8, 9). As a result, the fatigue severity is evaluated according to a performance status scale (3). Still, subjective indicators of fatigue are problematic, especially in the military (10,11), due to their limited sensitivity to small variations and the susceptibility to biases arising from personal and motivational factors, such as social acceptance (12,13). In this context, health care professionals are frequently frustrated by the lack of an objective technology that may assist with a more precise and faster diagnosis, so the development of non-invasive methods to monitor and detect fatigue in pilots is an area of great interest (14,15).
Flight fatigue is a comprehensive state caused by many factors (16), which result in changes not only in a psychological level, but also in regard to the biochemistry of the human body as an integrated system. Nevertheless, the underlying biochemical mechanisms correlated with this physiological condition are poorly understood. Metabolomics represents a comprehensive area of study, characterized by the high sensitivity and selectivity of metabolites that may correlate with biochemical perturbations in different organisms. This scienti c area has been able to enhance the understanding of pathophysiological mechanisms (17)(18)(19) and, upon the establishment of novel diagnostic markers (20)(21)(22), it has been already used for exploring the underlying mechanisms of physical (23)(24)(25) and mental fatigue (22).
However, few metabolomics studies have speci cally aimed the issue of ight fatigue, which is mainly related to mental fatigue but accompanied by certain physical exhaustion. Understanding and monitoring these particular processes, in order to manage exposure to the continuum of fatigue, may play a potentially important role in ight safety. Hence, here we have evaluated the potential differences in the metabolic phenotypes of a representative number of pilots, based on the collection of urine samples before and after long-term ight missions. These samples were presently utilized to develop a novel, highly sensitive and non-invasive methodology to monitor pilot fatigue, linking the metabolic phenotype with ight fatigue state to the discovery of potential biomarkers. In addition, we further explored the molecular mechanism of ight fatigue, thus providing new insights for the development of molecularly targeted drugs that could accelerate fatigue elimination.

Study Populations
Twenty-two active-duty man pilots, with ages ranging from 26 to 45 (34.36 ± 4.55) years old, were recruited for this study. Eight pilots ew for 3 hours during the day, while seven others ew for 4 hours and the remaining (n = 7) had a single ight time of three hours during the day and four hours during the night. Urine samples from respective pilots were collected, before and after the long-term ight missions, and then stored at -80°C until analysis. This study was approved by the ethics committee of the Air Force Medical Center in accordance with the Declaration of Helsinki.

Sample Preparation and Metabolomics Analysis
After thawing the urine samples on ice, a 100 µl aliquot (per sample) was extracted with 300 µl methanol by vigorous vortexing. Subsequently, extracts were centrifuged at 13,000 × g for 15 mins to pellet each protein precipitate. A total of 100 µL supernatant per sample was transferred to a 200 µl vial insert for further analysis. A 1 µL aliquot of each supernatant was injected into a Thermo Scienti c™ Dionex™ UltiMate™ 3000 Rapid Separation LC (RSLC) system, to further perform an ultra-high performance liquid chromatography separation with a reversed phase C18 column (2.1 x 100 mm, 1.7 µm, waters), operated at 45°C. The gradient mobile phase was composed of methanol (A) and water (B) with 0.1% formic acid and 10mmol/L ammonium acetate. Gradient elution was performed at the ow rate of 0.3 mL/minute. For this, an initial condition of 80% of mobile phase A were maintained for 1 min, followed by a linear gradient to 0% A for 12 mins. After a 0.5-minute washing step, the column was equilibrated to the initial condition for 1.5 mins.
The eluate was then introduced by electrospray ionization into the Q Exactive™ hybrid quadrupole Orbitrap mass spectrometer (Thermo Fisher Scienti c, Bremen, Germany) that operated in positive (ESI+) and negative (ESI−) electrospray ionization modes (one run for each mode) with a spray voltage of 3.7 kV and 3.5kV, respectively. The HESI (heated electrospray ionization) source utilized, for both modes, a capillary temperature of 320°C, heater temperature of 300°C, sheath gas pressure of 30 psi, and auxiliary gas pressure of 10 psi. During the full-scan acquisition (ranging from 50 to 1500 m/z), the instrument operated at 70,000 resolution (m/z = 200), with an automatic gain control (AGC) target of 1 × 10 6 charges and a maximum injection time (IT) of 50 ms.
To avoid artifacts due to the order of injected samples, these were randomly processed. Quality control (QC) samples were prepared by pooling equal volumes of each serum sample and injected every 6-8 samples throughout the analytical run to monitor the stability and reproducibility of the system.

Data processing
The peak picking, identi cation, alignment and normalization of the acquired data were all conducted by the Progenesis QI data analysis software (Nonlinear Dynamics, Newcastle, UK). The processed data were imported into SIMCA-P 13.0 software (Umetrics, Umeå, Sweden) for multivariate pattern recognition analysis. PCA was performed to detect outliers, and the distribution of deferent groups. OPLS-DA was carried out to obtain an overview of the complete data set after centering mean values and scaling unit variance (UV) (18).

Statistical Analysis
Respective data were expressed as the mean ± standard deviation (SD) or standard error of the mean (SEM). Paired two-tailed Student's t-tests were conducted using GraphPad Prism 5.0 software.
Metabolites with variable importance in projection (VIP) scores that were greater than 1.5 and P < 0.05 were considered statistically signi cant. Hierarchical cluster analysis (HCA) was conducted using the MeV software package (version 4.9.0). A correlation network was constructed using the Cytoscape software package. Receiver operating characteristic curve (ROC), binary logistic regression and Bayes discriminant analysis were also conducted with the SPSS software. The correlation analysis between variables were performed in R version 3.6.2.

Results
The metabolic pro ling of the urine samples was performed in a random order and the representative total ion current (TIC) chromatograms of pilots before and after performing long-term ight are shown in  1B and 1E). These models exhibited apparent metabolic variations in metabolome pro ling when comparing before and after ight groups (BF vs. AF).
Paired two-tailed Student's t-tests were performed, where 70 and 72 metabolites exhibited signi cant changes (VIP > 1.0, P < 0.05 and CV < 30%) after long time ight in ESI + and ESI-modes, respectively. The relative normalized quantities of the differential metabolites identi ed in AF (as compared to those in BF group) were further visualized in a heat map, according to their Pearson correlation coe cients (Figs. 2A and 2B). As shown, almost two thirds of these metabolites were decreased in response to long time ight ( Figs. 2A and 2B). According to a Spearman correlation analysis, we further examined how the metabolite abnormalities could be related to the status of ight fatigue (Figs. 2C and 2D). A strong correlation between the severity of fatigue with several metabolite classes were identi ed. Fatty acids and nine amino acids were in general positively correlated with ight fatigue, while acyl carnitines, carbohydrates, purines, indoles and the majority of amino acids were negatively correlated. Overall, the levels of these important metabolites were signi cantly altered in urine samples after long-term ight, and then considered to be associated with ight fatigue.
The biological pathways involved in the metabolism of these differential metabolites as well as their biological roles were determined by enrichment analysis using MetaboAnalyst (Fig. 2E). All matched pathways were shown according to (i) the p-values from the pathway enrichment analysis (y-axis) and (ii) the pathway impact values from pathway topology analysis (x-axis) (26, 27) (most impacted pathways are colored in red, Fig. 2E). As a result, nine biochemical pathways including (i) phenylalanine, tyrosine and tryptophan biosynthesis, (ii) pyrimidine metabolism, (iii) tryptophan metabolism, (iv) lipoic acid metabolism, (v) phenylalanine metabolism, (vi) ascorbate and aldarate metabolism, (vii) purine metabolism, (viii) pentose and glucuronate interconversions, and (ix) valine, leucine and isoleucine biosynthesis, were considered closely related to ight fatigue.

Differential Metabolites Related to Flight Fatigue
Our cohort of active-duty male pilots were further divided into three groups, according to their ight time: (i) pilots who ew for three hours during the day (before ight group: BFL, after ight group: AFL), (ii) pilots who ew for four hours during the day (BFM and AFM groups), and (iii) pilots who had a single ight time of three hours during the day and four hours during the night (BFH and AFH groups). Since ight time is frequently corelated with the extent of fatigue, we hypothesized that changes in the levels of certain metabolites could emerge during the early stages of ight fatigue and then evolve according to the extent of fatigue. Consequently, OPLS-DA models for both ESI + and ESI-have shown a clear separation of each pilot group before and after the ight, as well as among the three distinct groups (See Supplementary Figure S2), indicating that metabolic alterations occurred during ight fatigue progression. Thus, pair-wise comparisons were carried out based on OPLS-DA models. As a result, the levels of 29 metabolites in ESI + mode (25 signi cantly reduced and 4 increased) and of 32 metabolites in ESI-mode (26 metabolites signi cantly reduced and 6 increased) were consistently altered (Fig. 3). These metabolites gradually changed according to the extent of fatigue, suggesting some strong and direct correlation between certain metabolites and ight fatigue (See Supplementary Figure S3).

Biomarkers of Flight Fatigue
To screen metabolites that could serve as potential biomarkers in ight fatigue states, we validated the differential metabolites in both AF and BF pilot groups. Based on the signi cantly changed metabolites in ESI+ (n = 70) and ESI-(n = 72) modes, a binary logistic regression was conducted to identify an optimal combination of metabolites as the potential biomarkers for AF. As a result, one metabolite for ESI+ (i.e. beta-guanidinopropionic acid) and two metabolites for ESI-(i.e. 3-dehydro-L-gulonate and 2-propylpent-3enoic acid) were identi ed as potential biomarkers for ight fatigue.

Multiple Discriminant Analysis of the Extent of Flight Fatigue
Although ight fatigue is a prominent issue that impacts mission success and ight safety, there is still a lack of direct evaluation methods to distinguish the fatigue severity and, therefore, to determine whether a pilot is prepared to y. To address this issue, a Bayes discriminant function model was established by stepwise discriminant analysis of differential metabolites. According to this analysis, 10 and 7 statistically signi cant variables (related to ESI + and ESI-modes, respectively) were included in the nal discriminant function models. A retrospective discrimination was conducted among the individuals presently tested (Figs. 5A and 5B), which resulted into an excellent discriminant performance, with an accuracy of 95.5% and 97.7%, respectively (in Supplementary Tables S1-S4).

Discussion
Flight fatigue is the most frequently cited physiological factor affecting aircraft pilots, which contributes to the occurrence of accidents in the military aviation. As such, higher levels of attention and cognitive ability are required due to the stressful and complex nature of this condition. Besides, ight timing is expected to be an important determinant of pilot fatigue and fatigue level. Thus, the discovery of straightforward and sensitive indices that may detect fatigue and also monitor fatigue levels is critical to assess any associated safety risk and prevent air-related catastrophes (28, 29).
Urine is a non-invasive uid to collect which contains many biomolecules of valuable diagnostic information (30). In this study, we retrieved urine samples from a cohort of 22 military pilots, before and after performing different periods of ight missions, for a comprehensive metabolomic investigation to particularly explore the metabolic characteristics of ight fatigue. To our knowledge, this is the rst report involving the metabolic pro ling of aircraft pilots exposed to different levels of fatigue severity.
In the present study, some signi cant metabolic alterations were observed during the progression of ight fatigue. Apart of fatty acids and a number of amino acids, most of the metabolites were downregulated after conclusion of ight mission. Carnitines play an important role in the transport of fatty acids across the inner mitochondrial membrane. According to the current literature, carnitines can also improve energy levels and physical function by reducing fatigue and improving cognitive functions (31). In our study, carnitines are signi cantly decreased in fatigue pilots, suggesting that ight-related stress may alter energy metabolism. We may speculate that, while performing stressful ight missions for a long period of time, pilots may consume more lipids to acquire su cient energy. Furthermore, carbohydrate levels were expectedly decreased after ight. In contrast, the levels of isocitrate (intermediate involved in energy conversion via the Krebs cycle) were obviously elevated, highlighting the activation of aerobic pathways during long-time ight. Consistent with the ndings in mental fatigue (32), exhaustive exercise (33) and chronic fatigue syndrome (34), most of the amino acids presently identi ed (75.68%) were negatively correlated with ight fatigue, implicating that amino acid consumption were accelerated in subjects performing extensive mental or physical tasks. Besides their utilization in protein synthesis, amino acids have also been involved in various metabolic activities in the brain. The serotonin precursor 5-Hydroxy-Ltryptophan (5-HTP) can easily cross the blood-brain barrier (BBB) and effectively increase the synthesis of serotonin in the central nervous system (CNS) (35). According to our data, the reducing urinary levels of 5-HTP suggest that mental activity may accelerate the uptake of this amino acid derivative into the brain. Interestingly, we have also observed a signi cant decrease of purines and purine derivatives in the urine samples of pilots after ight. The decreased excretion of purines may be consistent with the increased synthesis and/or turnover ( ux) of ATP and GTP under high tension state. As the end product of purine metabolism and also an important antioxidant, uric acid has also increased signi cantly in our studies. In general, ight fatigue can cause prominent alterations in the urinary metabolome.
Lastly, potential biomarkers of ight fatigue were presently revealed by multivariate statistical analysis, combined with the binary logistic regression analysis (Fig. 4). Speci cally, three metabolites were selected as candidate biomarkers, with remarkable sensitivity and speci city to accurately identify fatigue state. Furthermore, Bayes discriminant function models were established for the diagnosis of fatigue severity. Such insights can be potentially applied for the early detection of fatigue conditions, thus providing some valuable information about whether aircrew would be authorized to y. As such, this information serves as a critical tool for primary prevention and evaluation of ight safety. Monitoring some of these biomarkers levels may help optimizing ight conditions and/or pilot training to achieve a 'functional over-reached' state, in order to promote positive adaptation and increased performance following rest. Accordingly, we may prevent the development of a 'nonfunctional over-reached' state, which is a consequence of intense training leading to decrement in ability even with adequate rest (36).

Conclusion
Pilots fatigue is the most frequently cited physiological factor contributing to the occurrence of accidents in the military aviation, which requires high attention and cognitive ability due to its stressful and complex nature. Therefore, the discovery of objective and sensitive indices to detect fatigue and monitor fatigue levels is critical to assess the associated safety risk and prevent accidents and catastrophes. In this study, we collected the urine samples from recruit 22 military pilots before and after performing different time of ight missions for comprehensive metabolomics investigations to explore the metabolic characteristics of ight fatigue, and to identify objective and sensitive potential biomarkers for monitoring fatigue levels.
Monitoring the recovery of severe or chronic ight fatigue is of particular importance for commanders to effectively program training regimes/schedules to better emphasize positive adaptation. Given that these metabolomic markers have been primarily observed in a relatively small cohort, additional research is required using larger sample sizes to validate our current ndings and then identify stronger relationships among the multiple variables involved.

Declarations
Author contributions R.G. and F.J. were responsible for the execution of experiments, data analysis and preparation of the manuscript. F.Z.J. and M.Y.Z. helped with the statistical analysis of the data and contributed to the interpretation of the results. W.J.C. and Y.H.L. designed and supervised the study. All authors critically commented on and approved the nal submitted version of the paper.

Funding Sources
This work was partially supported by the National Natural Science Foundation of China (grant no. 81903680) and Air Force Medical University (2020ZTB04).

Availability of data and materials
The data and materials used in the current study are all available from the corresponding author upon reasonable request.

Ethics approval
This study was approved by the ethics committee of the Air Force Medical Center.

Con ict of Interest
The authors have declared that they have no con ict of interests.

Consent for publication
Not applicable Differential metabolites related to ight fatigue. a and b, the hierarchical cluster analysis of the differential metabolites in ESI+ and ESI-modes. The relative normalized quantities of the identi ed differential metabolites in AF compared to those in BF group were further visualized in a heat map according to their Pearson correlation coe cients. Coloured squares indicate the changes of metabolite content, and positive and negative values indicate upregulation and downregulation of metabolites, respectively. c and d, Spearman's correlation coe cients between differential metabolites in ESI+ and ESImodes and ight fatigue. Positive correlations are marked in red and negative correlations in blue. e, the summary of aberrant pathways caused by ight fatigue. All matched pathways were shown according to p values from the pathway enrichment analysis (y-axis) and pathway impact values from pathway topology analysis (x-axis), with the most impacted pathways colored in red.

Figure 3
Page 15/16 Please see the Manuscript le for the complete gure caption.

Figure 4
The diagnostic performance of the biomarkers screened out in ESI+(a) and ESI-(b) modes. The area under the curve AUC of the ROC was calculated to evaluate the ability of the potential biomarkers to distinguish the occurring of ight fatigue.