Abstract

Background

   Bariatric surgery induces significant weight loss, increases insulin
   sensitivity, and improves dyslipidemia. As one of the most widely
   performed bariatric surgeries, laparoscopic sleeve gastrectomy (LSG) is
   thought to improve the metabolic profile along with weight loss. The
   objective of this study was to evaluate longitudinal changes in the
   serum metabolite levels after LSG and elucidate the underlying
   mechanisms of metabolic improvement.

Methods

   Clinical metabolic parameters and serum samples were collected
   preoperatively and at 1, 3, and 6 months postoperatively from nine
   patients with obesity undergoing LSG. Serum metabolites were measured
   using a non-targeted metabolic liquid chromatography–mass spectrometry
   method.

Results

   During the 1, 3, and 6 months postoperative follow-up, the body mass
   index, HOMA-IR, and liver fat content showed a gradual descending
   trend. A total of 328 serum metabolites were detected, and 38 were
   differentially expressed. The up-regulated metabolites were mainly
   enriched in ketone body metabolism, alpha-linolenic acid and linoleic
   acid metabolism, pantothenate and CoA biosynthesis, glycerolipid
   metabolism, and fructose and mannose degradation, while the
   down-regulated metabolites were closely related to caffeine metabolism,
   oxidation of branched-chain fatty acids, glutamate metabolism, and
   homocysteine degradation. Notably, nine metabolites (oxoglutarate,
   2-ketobutyric acid, succinic acid semialdehyde, phthalic acid,
   pantetheine, eicosapentaenoate, 3-hydroxybutanoate, oxamic acid, and
   dihydroxyfumarate) showed persistent differential expression at 1, 3,
   and 6 months follow-up. Some were found to be significantly associated
   with weight loss, insulin resistance improvement, and liver fat content
   reduction.

Conclusions

   This finding may provide a new perspective for revealing novel
   biomarkers and mechanisms of metabolic improvement in obesity and
   related comorbidities.

   Keywords: bariatric surgery, metabolomics, obesity, metabolic
   improvement

Introduction

   The obesity epidemic is a major global health challenge in modern
   societies, with a global prevalence of 14% by 2020, and it is even
   expected to reach 24% by 2035
   ([46]https://data.worldobesity.org/publications/?cat=19). Compared to
   behavioral therapy and pharmacotherapy, bariatric surgery is currently
   the most effective and durable treatment for morbid obesity ([47]1,
   [48]2). Laparoscopic sleeve gastrectomy (LSG) is one of the most widely
   performed bariatric surgeries at present ([49]3). In addition to the
   primary objective of weight loss, LSG has also been shown to provide
   long-term solutions to related health problems, such as relief from
   type 2 diabetes mellitus (T2DM) ([50]4), dyslipidemia ([51]5),
   cardiovascular disease ([52]6), and non-alcoholic fatty liver disease
   (NAFLD) ([53]7), as well as a reduction in overall mortality ([54]8).
   However, the exact mechanisms underlying these effects remain unclear.

   Morbidly obese patients undergoing bariatric surgery often share a
   common metabolic background and similar environmental factors ([55]9).
   Several attempts have been made to describe the mechanisms of metabolic
   improvement in obesity after bariatric surgery, while the effect is far
   from being mechanistic and includes modulation of neural circuits,
   alterations in the intestinal microbiome, variations in bile acid
   excretion, restoration of the intestinal and adipose tissue hormonal
   milieu, and recruitment of intestinal glucose transport molecules
   ([56]10, [57]11, [58]12). Previous studies on these mechanisms mainly
   focused on the exploration of upstream genomics, transcriptomics, and
   proteomics. The recent application of metabolomics in medical research
   has enabled the development of metabolic biomarkers and the elucidation
   of physiological mechanisms underlying various diseases ([59]13).
   Elucidation of systemic metabolic changes after bariatric surgery is
   important to gain insights into new management strategies for obesity
   and related comorbidities ([60]14, [61]15).

   Because a short-term single comparison is underrepresented,
   longitudinal follow-up may be a more effective way to elucidate the
   underlying mechanisms of metabolic improvement after bariatric surgery.
   Therefore, this study set up a cohort that focused on longitudinal
   changes in the serum metabolomic profile after LSG.

Materials and methods

Subjects

   The inclusion criteria were in accordance with the guidelines of the
   Chinese Society for Metabolic and Bariatric Surgery. Patients aged
   18–65 years with body mass index (BMI) ≥ 32.5 kg/m^2 or 27.5 ≤ BMI <
   32.5 kg/m^2 plus one or more comorbidities (T2DM, dyslipidemia,
   hypertension, and obstructive sleep apnea) and non-surgical treatments
   that failed to achieve the desired weight loss or manage complications
   were invited to participate in this study. Finally, nine individuals
   (four males and five females) who underwent LSG at Zhongshan Hospital,
   Fudan University, between December 2014 and December 2016, were
   recruited into our cohort and completed the entire follow-up.
   Initially, all patients were obese, and all surgeries were performed by
   the same surgeon. Follow-up assessments were conducted at baseline
   (pre-surgery) and 1-, 3-, and 6-month post surgery. Anthropometric
   measurements, clinical biochemical parameters, and blood serum
   metabolome data were collected during each follow-up period.

   This study was approved by the Ethical Review Committees of Zhongshan
   Hospital, Fudan University (No.: B2021-157R), and the study protocol
   adhered to the Declaration of Helsinki. Oral and written informed
   consent was obtained from all patients.

Clinical indicators and biochemical measurements

   Body height and weight were measured using a height-weight scale, with
   participants removing their shoes and outer clothing. Blood pressure
   was measured in the right arm three times using the standard brachial
   cuff technique after at least 5 min of rest in a seated position, with
   an interval of 1 min between each measurement. The three readings were
   averaged for analysis.

   Venous blood samples for biochemical measurements were obtained after a
   fasting period of at least 12 h. Serum samples were centrifuged and
   stored at −80°C until analysis. The serum total cholesterol (TC),
   high-density lipoprotein cholesterol (HDL-C), and triglyceride (TG)
   levels were measured using the oxidase method on a 7600 automated
   bioanalyzer (Hitachi). Low-density lipoprotein cholesterol (LDL-C) was
   calculated using the Friedewald equation. Glycosylated hemoglobin
   (HbA1c) levels were determined using high-pressure liquid
   chromatography on the Variant™ II machine (Bio-Rad). Plasma glucose
   levels were measured using the glucose oxidase method. Serum insulin
   concentration was determined by Auto DELFIA fluoroimmunoassay. The
   homeostatic model assessment of insulin resistance (HOMA-IR) was
   calculated as fasting glucose (mmol/L) × fasting insulin (mU/mL)/22.5.
   Liver fat content was defined as the ratio of area lipid (AL) to AL
   plus area water (AW) and was measured using hydrogen proton magnetic
   resonance spectroscopy (^1H-MRS) ([62]16).

Metabolomics profiling of serum samples

Sample preparation

   The serum (100 µL) was mixed with 400 µL methanol/acetonitrile (1/1,
   v/v) by vortexing for 30 s. The mixture was then sonicated for 10 min
   at 4°C. After 1 h of incubation at −20°C, the samples were centrifuged
   at 18,832 g for 15 min at 4 °C. The supernatant was transferred to a
   new tube and evaporated to dryness at 4°C using a vacuum concentrator.
   The samples were reconstituted in 100 µL of acetonitrile/H[2]O solution
   (1/1, v/v) for injection analysis. The injection volume used was 3 µL.

Chromatographic and mass spectrometry conditions

   Serum metabolites were analyzed using a Shimadzu Prominence UPLC system
   (Nexera UHPLC LC-30A, Kyoto, Japan) interfaced with an AB SCIEX Triple
   TOF 5600+ system (AB Sciex, Singapore) equipped with an ESI source.
   Metabolites were separated using a HILIC column (ZIC-pHILIC, 5 μm, 2.1
   × 100 mm, PN: 1.50462.0001, Millipore) with the column temperature
   maintained at 40°C. The mobile phase consisted of 25 mM ammonium
   acetate in 25 mM ammonia water (mobile phase A) and acetonitrile
   (90/10, v/v) (mobile phase B) and was run at a flow rate of 0.3 mL/min.
   The gradient was as follows: 95% B for 1 min, 65% B for 14 min, 40% B
   for 16 min, held for 18 min, then increased to 95% B for 18.1 min, then
   back to 9% B, and held for another 23 min. The flow rate for mobile
   phases was set at 0.3 mL/min. The mass spectrometer was run in positive
   information-dependent acquisition (IDA) mode, with the source
   temperature at 550°C, ion source gases 1 and 2 at 55 psi, curtain gas
   at 35 psi, collision energy at +30 eV or −30 eV, ion spray voltage
   floating at 5.5  kV or −4.5 kV, and a mass range at 60–1250 m/z. The
   accumulation time for the full scan was set at 150 ms, for each IDA
   scan, it was 45 ms. Peaks of metabolites with intensities greater than
   100 c.p.s., after adding up the signal from ten rounds of IDA scans,
   were chosen for further analysis.

Data processing and metabolite identification

   After the data were acquired, the SCIEX OS software was used to perform
   peak quantification and identification, integration, and retention time
   correction. MetaboAnalyst 5.0 ([63]https://www.metaboanalyst.ca/) was
   used to process the data, including alignment, deconvolution,
   normalization, and data scaling. Internal standard peaks and known
   false-positive peaks were removed from the data matrix to reduce
   redundancy and pool similar peaks. In the data matrix, a minimum of 80%
   of the metabolic features detected across any set of samples were
   preserved. Following the exclusion of samples with metabolite levels
   falling below the quantification threshold, the lowest metabolite value
   was approximated. Each metabolic signature was then normalized based on
   its total sum. To mitigate errors arising from sample preparation and
   instrument fluctuations, the intensity of mass spectrometry peaks was
   standardized using the sum-normalization approach. Z-score
   normalization was applied for data scaling. The metabolite
   identification process commenced by matching the retention times (RTs)
   of detected peaks against the RTs of known metabolites. Concurrently, a
   comparison was made between the mass-to-charge ratios (m/z) of the
   detected peaks and those of the known metabolites. Peaks that met the
   initial RT and m/z matching criteria underwent further analysis through
   secondary mass spectrometry. Peaks exhibiting high similarity scores in
   this analysis were regarded as providing robust evidence for the
   identity of the corresponding metabolites. For this process, we used
   the AB SCIEX commercial library and the local library to confirm the
   identity of the metabolites.

Statistical analysis

   SPSS software (version 25.0) was used for statistical analysis of
   anthropometric and biochemical measurements. Normality tests were first
   performed for quantitative variables. Normally distributed variables
   are expressed as mean ± s.d., and paired t-tests were used to compare
   continuous variables between groups. Multiple groups of quantitative
   data were compared using analysis of variance. Next, partial least
   squares discriminant analysis (PLS-DA) was used for differential
   analysis of metabolomics, and a five-fold cross-validation method was
   performed to evaluate model stability. Metabolites with a variable
   importance (VIP) score >1 in the PLS-DA model, P < 0.05, and fold
   change (FC) values ≥2 or ≤0.5 were defined as differentially expressed
   metabolites (DEMs). The P-value was generated using paired t-test in
   the R language, and the post-hoc tests were corrected using the
   Benjamini–Hochberg false discovery rate test. Time-series cluster
   analysis and heat maps were generated using the bioinformatics website
   ([64]https://www.bioinformatics.com.cn/). Pathway enrichment analysis
   was performed using MetaboAnalyst 5.0, and the Small Molecule Pathway
   Database (SMPDB) was used for metabolic pathway annotation. Line charts
   depicting the variation trends of the continuously changing DEMs were
   generated using GraphPad Prism 8. P < 0.05 was considered statistically
   significant.

Results

Demographic and clinical characteristics

   The demographic, clinical characteristics, and metabolic parameters of
   the patients pre-surgery and 1-, 3-, 6-month post-surgery are shown in
   [65]Table 1. Compared with baseline, the BMI, HOMA-IR, and liver fat
   content of the patients decreased significantly at the three
   postoperative follow-up points (P < 0.05). Compared to 1 and 3 months
   after surgery, the HDL-C level of the patients increased significantly
   at 6 months after surgery (P < 0.05).

Table 1.

   Clinical characteristics and metabolic parameters for the patients pre-
   and post-surgery.
   Pre-surgery Post-surgery
   Baseline 1 month later 3 months later 6 months later
   Sample size 9 9 9 9
   Sex (male/female) 4/5 – – –
   Age (years) 40.33 ± 4.55 – – –
   BMI (kg/m^2) 43.60 ± 2.41 37.75 ± 2.15 34.09 ± 2.21^b 30.75 ± 2.04^b,c
   SBP (mm Hg) 131.11 ± 6.38 137.33 ± 5.41 122.44 ± 5.19 129.33 ± 6.22
   DBP (mm Hg) 84.78 ± 4.61 81.22 ± 4.26 76.22 ± 3.90 80.44 ± 2.86
   FBG (mmol/L) 6.74 ± 1.39 5.04 ± 0.83 4.53 ± 0.23 4.58 ± 0.32
   HbA1c (%) 6.13 ± 0.45 5.70 ± 0.37 5.24 ± 0.20 5.37 ± 0.21
   HOMA-IR 6.25 ± 1.22 1.86 ± 0.46^b 1.73 ± 0.29^b 1.25 ± 0.25^b
   TC (mmol/L) 3.82 ± 0.34 3.71 ± 0.27 3.88 ± 0.19 4.21 ± 0.19
   TG (mmol/L) 1.67 ± 0.35 0.97 ± 0.07 1.06 ± 0.12 0.91 ± 0.10
   LDL-C (mmol/L) 2.24 ± 0.19 2.29 ± 0.26 2.28 ± 0.13 2.46 ± 0.17
   HDL-C (mmol/L) 1.15 ± 0.05 0.97 ± 0.05 1.11 ± 0.08 1.33 ± 0.08^d,e
   Liver fat content (%) 35.9% ± 7.7% 18.7% ± 3.5%^a 14.6% ± 2.1%^b 8.3% ±
   1.9%^b
   [66]Open in a new tab

   ^aP < 0.05 compared with baseline; ^bP < 0.01 compared with baseline;
   ^cP < 0.05 compared with 1 month; ^dP < 0.01 compared with 1 month; ^eP
   < 0.05 compared with 3 months.

   BMI, body mass index; DBP, diastolic blood pressure; FBG, fasting blood
   glucose; HDL-C, high-density lipoprotein cholesterol; HOMA-IR,
   homeostasis model assessment-insulin Resistance; LDL-C, low-density
   lipoprotein cholesterol; SBP, systolic blood pressure; TC, total
   cholesterol; TG, triglyceride.

Characteristics of metabolome profile pre- and post-LSG

   Liquid chromatography–mass spectrometry (LC–MS) metabolomics was
   conducted on 36 serum samples from nine patients at four time points:
   pre-surgery, 1-, 3-, and 6-month post surgery. In total, 328
   metabolites were identified in serum metabolomes. After normalization
   by the baseline group and data scaling (mean-centered and divided by
   the s.d. of each variable), the PLS-DA plot showed that these four
   groups (pre-surgery, 1-, 3-, and 6-month post surgery) clustered in
   different directions, indicating that there were significant
   differences in metabolite profiles among the four groups ([67]Fig. 1).

Figure 1.

   Figure 1
   [68]Open in a new tab

   PLS-DA plot of baseline (pre-surgery), 1-month post surgery, 3-month
   post surgery, and 6-month post surgery. The X-axis and Y-axis are
   labeled with PC1 (the first principal component) and PC2 (the second
   principal component).

Identification of differentially expressed metabolites

   Through differential expression analysis, metabolites with P-values
   <0.05, VIP values >1, and FC values ≥2 or ≤0.5 were defined as DEMs.
   Finally, 38 metabolites were found to have significantly changed
   compared with baseline throughout the entire postoperative follow-up,
   mainly including amino acids, lipids, energy-related metabolites, and
   gut microbiota-related metabolites. Among them, 19 metabolites
   exhibited statistically significant changes at 1-month post surgery
   (ten up-regulated and nine down-regulated), 21 metabolites exhibited
   statistically significant changes at 3 months post surgery (15
   up-regulated and six down-regulated), and 11 metabolites exhibited
   statistically significant changes at 6 months post surgery (eight
   up-regulated and three down-regulated) ([69]Fig. 2).

Figure 2.

   Figure 2
   [70]Open in a new tab

   Volcano plot of metabolite abundance changes at 1-month post surgery
   (A), 3-month post surgery (B), and 6-month post surgery (C). Red points
   mean P < 0.05 and fold change (FC) ≥2 compared with baseline
   (up-regulated metabolites), blue points mean P < 0.05 and FC ≤ 0.5
   compared with baseline (down-regulated metabolites), gray points mean
   not significant.

Time-series cluster and pathway enrichment analysis of DEMs

   According to the time-series cluster analysis, all 38 DEMs were
   classified into four main clusters based on their response after
   surgery. Cluster 1, consisting of 12 metabolites, showed a gradual
   decline throughout the follow-up period, whereas cluster 4, containing
   eight metabolites, exhibited a consistent increasing trend. Cluster 2,
   containing eight metabolites, initially increased in the first month
   before returning to near-baseline levels over the entire 6-month
   period. Cluster 3, containing ten metabolites, gradually increased from
   baseline to 3 months after surgery and then returned to near-baseline
   levels at 6 months ([71]Fig. 3A). The heat maps of the four clusters
   visually showed different variation trends during follow-up ([72]Fig.
   3B). To explore the metabolic pathways, enrichment analysis was
   conducted on the four clusters, and the metabolic pathways were
   presented according to the enrichment degrees from high to low
   ([73]Fig. 3C). The most significantly enriched metabolic pathways in
   cluster 1 were caffeine metabolism, oxidation of branched-chain fatty
   acids, glutamate metabolism, and homocysteine degradation. Then, ketone
   body metabolism, alpha-linolenic acid, and linoleic acid metabolism
   were the main pathways enriched in cluster 2. Glycerolipid metabolism,
   fructose and mannose degradation, and gluconeogenesis were the main
   pathways enriched in cluster 3. Cluster 4 was mainly enriched in
   pantothenate and CoA biosynthesis, and propanoate metabolism. The
   specific metabolites in each cluster are shown in [74]Fig. 3D.

Figure 3.

   [75]Figure 3
   [76]Open in a new tab

   Time-series clustering, heat map, and pathway enrichment analysis. (A)
   Time-series clustering. Based on the core algorithm, fuzzy C-means
   custering, the metabolites with similar expression patterns were
   classified into four clusters. (B) Heat maps of metabolite
   concentration variations in the four clusters. (C) Metabolite
   enrichment analysis for each cluster based on the Small Molecule
   Pathway Database (SMPDB), and the top terms enriched with the lowest
   P-value are labeled. (D) The specific metabolite names of each cluster.

The characteristics of continuously changing DEMs and the correlation with
metabolic improvement

   In all DEMs, compared with baseline, two metabolites
   (3-hydroxybutanoate and eicosapentaenoate) showed significant
   alterations at both 1- and 3-month post surgery points, one metabolite
   (dihydroxyfumarate) changed at both the 1- and 6-month points, two
   metabolites (pantetheine and oxamic acid) changed at both 3- and
   6-month points, and four metabolites (phthalic acid, oxoglutarate,
   2-ketobutyric acid, and succinic acid semialdehyde) changed at all
   three postoperative time points. These metabolites were called
   continuously changing DEMs ([77]Fig. 4). The variation trends of the
   nine continuously changing DEMs are shown in [78]Fig. 5. To explore the
   potential link between these nine DEMs and metabolic improvement, a
   correlation analysis was performed between the nine DEMs and clinical
   parameters. The results showed that phthalic acid levels were
   significantly negatively correlated with BMI. Oxoglutarate,
   2-ketobutyric acid, and succinic acid semialdehyde levels were
   significantly positively correlated with HOMA-IR and liver fat content.
   Additionally, 3-hydroxybutanoate was negatively correlated with
   HOMA-IR. Pantetheine levels were significantly negatively correlated
   with BMI, HOMA-IR, and liver fat content. Oxamic acid levels were
   significantly negatively correlated with HOMA-IR and liver fat content.
   Dihydroxyfumarate was negatively correlated with liver fat content
   ([79]Fig. 6). The correlations between the other 29 DEMs and clinical
   parameters during the entire postoperative follow-up period are shown
   in Supplementary Table 1 (see section on [80]supplementary materials
   given at the end of this article).

Figure 4.

   Figure 4
   [81]Open in a new tab

   Venn plot of differentially expressed metabolites (DEMs) in three
   postoperative time periods.

Figure 5.

   [82]Figure 5
   [83]Open in a new tab

   Variation trends of the nine continuously changing DEMs at all three
   postoperative time points.

Figure 6.

   [84]Figure 6
   [85]Open in a new tab

   Correlation coefficient matrix diagram of the nine continuously
   changing DEMs and clinical metabolic parameters. *P < 0.05, **P < 0.01.

Discussion

   Through a non-targeted metabolomic approach, our longitudinal study of
   patients with obesity post-LSG revealed significant changes in
   metabolomic profiles at 1-, 3-, and 6-month postoperative time points.
   Some of the changed metabolites in this process were associated with
   different degrees of clinical metabolic improvement according to
   anthropometric and clinical parameters. Notably, we further analyzed
   the metabolic pathways involved in these changes and found some
   potential biomarkers that were intimately related to clinical metabolic
   improvements.

   Our results confirmed significant clinical metabolic improvement after
   LSG during the 6-month follow-up period, including weight loss,
   remission of insulin resistance, improvement of dyslipidemia, and a
   decrease in liver fat content. HOMA-IR, as an indicator of insulin
   resistance, showed a gradual downward trend at the three follow-up
   points after LSG, with significant differences compared with baseline.
   HbA1c, an indicator of blood glucose control, showed no significant
   decrease. This implies that the improvement in diabetes after LSG may
   begin with the remission of insulin resistance (IR). In addition, an
   increase in HDL-C levels was observed at the 6-month postoperative time
   point, but no statistically significant change was observed in TC, TG,
   and LDL-C levels, which means that the improvement of blood lipid
   profiles may require a longer process and may be greatly affected by
   postoperative dietary habits and lifestyle. Notably, liver fat content
   showed a gradually decreasing trend during the postoperative follow-up,
   suggesting that LSG can improve liver lipid metabolism; however, the
   underlying mechanisms remain unclear.

   According to the principles of P-value <0.05, VIP values >1, and FC ≥ 2
   or ≤ 0.5, we screened 38 DEMs that showed significant variations during
   follow-up compared to baseline. These metabolites were categorized into
   four clusters through time-series analysis, including upregulation
   types (continuous ascending, early ascending, and late ascending) and a
   downregulation type. Through metabolite enrichment analysis based on
   the SMPDB, in the upregulation type, we identified pantothenate and CoA
   biosynthesis, which are closely related to energy metabolism, fat
   synthesis and oxidation, and insulin resistance, as the most enriched
   pathway in the continuous ascending type (cluster 4). Pantothenate,
   also called vitamin B5, is a water-soluble vitamin required to sustain
   life. A recent study has shown that pantothenate protects against
   obesity via brown adipose tissue activation ([86]17). Pantothenate
   derivatives have a good hypolipidemic effect in mice with hypothalamic
   obesity induced by aurothioglucose ([87]18). Pantetheine, as the active
   form of pantothenate in the body, was found to be continuously elevated
   in our results, which was consistent with the metabolic improvement
   after LSG. Ketone body metabolism and alpha-linolenic acid and linoleic
   acid metabolism were the main metabolic pathways enriched in the early
   ascending type (cluster 2). Previous studies suggested that after
   bariatric surgery, a strong and early upregulation of catabolism and
   lipolytic activity occurred, and the end products of β-oxidation and
   ketogenesis increased, especially in the short-term period after
   surgery ([88]19, [89]20, [90]21). Another study showed that the highest
   weight loss group had greater increases in serum ketone bodies after
   LSG than the lowest weight loss group ([91]22). Our results were
   consistent with the above studies, which showed a rapid early rise of
   ketone body (3-hydroxybutanoate), followed by a slow decline after LSG,
   but consistently above the baseline. Eicosapentaenoate (EPA) is
   involved in alpha-linolenic acid and linoleic acid metabolism, which is
   an ω-3 polyunsaturated fatty acid that plays an important role in human
   health, such as in reducing the risk of cardiovascular disease,
   anti-inflammatory effects, and protection against lipemic-oxidative
   injury ([92]23). In our study, although no significant correlation was
   found between EPA and clinical metabolic indicators, the early increase
   in EPA levels reflected the metabolic benefits of obesity after LSG.
   Moreover, the enrichment of glycerolipid metabolism, fructose and
   mannose degradation, and gluconeogenesis in the late ascending type
   (cluster 3) suggested that LSG altered carbohydrate and lipid
   metabolism. We observed an increase in glyceraldehyde (a key
   intermediate in glycolipid metabolism), but only in the first 3 months
   post-surgery. This can be explained by lipolysis, which has been
   reported to increase in the short term post surgery ([93]24). Possible
   explanations for the gradual decline during further months after
   surgery can be a reduction in total lipids and their uptake by the
   liver, where they are used as a precursor of gluconeogenesis ([94]25).
   In addition, fructose is converted to glucose in the liver ([95]26).
   The plasma mannose levels were significantly up-regulated in obesity
   and potentially played a supplementary role in the development of IR
   ([96]27). The degradation of fructose and mannose implied that the
   disorder of glucose metabolism in obesity was improved after LSG.

   In the down-regulated metabolites, we identified caffeine metabolism,
   oxidation of branched-chain fatty acids, glutamate metabolism, and
   homocysteine degradation as the main enriched pathways (cluster 1). Two
   important applications of caffeine are metabolic phenotyping of
   cytochrome P450 1A2 (CYP1A2) and liver function testing ([97]28).
   Caffeine metabolites, such as 1-methylxanthine and paraxanthine, can
   differentiate the severity of early-stage liver disease ([98]29). Total
   overnight salivary caffeine assessment in cirrhotic patients was
   significantly higher than in controls ([99]30). The alteration in
   caffeine metabolism reflects the possible pathway of the effect of LSG
   on liver metabolism. Studies have shown that propionylcarnitine, an
   important molecule in branched-chain fatty acid oxidation, increases
   obesity and metabolic unwellness ([100]31). In our study,
   propionylcarnitine (C3), an intermediate product of fatty acid
   metabolism, showed a decreasing trend after an improvement in obesity,
   which was consistent with previous reports. Moreover, to our knowledge,
   elevated plasma homocysteine is an independent risk factor for
   atherosclerosis ([101]32). The enrichment of homocysteine degradation
   suggests that LSG may be significantly associated with an improvement
   in cardiovascular events.

   From the DEM profiles and enriched pathways above, it was evident that
   the up-regulated metabolites post-LSG were mainly enriched in the
   pathways related to carbohydrate and lipid metabolism, while the
   down-regulated metabolites primarily correlated with amino acid
   metabolic pathways. Prior studies have reported that circulating amino
   acid levels are elevated in obesity and undergo a rapid decline
   following bariatric surgery due to reduced protein intake and amino
   acid absorption ([102]33). Our results suggested that in the early
   postoperative period, the body's energy demand paradoxically increased
   due to stress responses and wound healing. To maintain energy
   homeostasis, the body significantly enhanced fat mobilization, thereby
   increasing lipolysis to provide energy. This process led to a massive
   accumulation of ketone bodies and β-oxidation metabolites. In the later
   stages of postoperative recovery, the lipolysis pathway tended to
   stabilize, leading to a decrease in ketone bodies and a relative
   increase in carbohydrate utilization, which promoted the production of
   glycolipid conversion intermediates (such as glyceraldehyde, as
   observed in our results). Overall, from the perspective of changes in
   metabolic pathways, the combined effects of diminished nutrient
   absorption and augmented mobilization of glucose and lipid metabolism
   may jointly contribute to the weight loss and metabolic benefits after
   LSG.

   In all DEMs, compared with baseline, nine metabolites showed persistent
   differential expression at least at two postoperative time points. In
   addition to pantetheine, 3-hydroxybutanoate, and EPA have already
   mentioned above, oxoglutarate, 2-ketobutyric acid, succinic acid
   semialdehyde, phthalic acid, oxamic acid, and dihydroxyfumarate were
   included. Correlation analysis revealed that some of these metabolites
   were significantly correlated with BMI, HOMA-IR, and liver fat content.
   Therefore, we analyzed the possible roles and potential pathways of
   these metabolites in obesity, glycometabolism, and lipid metabolism.

   Oxoglutarate, also known as α-ketoglutarate (α-KG), is an important
   metabolic intermediate that plays a key role in the tricarboxylic acid
   (TCA) cycle. Diabetes can cause significant increases in the levels of
   α-KG ([103]34). The study of the circulating metabolome reveals α-KG as
   a predictor of morbid obesity associated with NAFLD, and plasma α-KG
   levels are significantly decreased in lean controls compared with obese
   patients ([104]35). A nuclear magnetic resonance study also indicated
   increased hepatic TCA cycle flux in human NAFLD ([105]36). A study in
   the livers of high-fat diet mice demonstrated that increased TCA cycle
   flux was associated with hepatic steatosis ([106]37). In addition, the
   concentrations of glutamine, α-KG, citrate, and pyruvate were higher in
   patients with non-alcoholic steatohepatitis (NASH) than in patients
   without NASH, which may be related to the impairment of the TCA cycle.
   Activation of glutaminolysis may be the source of increased α-KG levels
   in patients with NASH ([107]38, [108]39). These metabolic responses may
   be remodeled following the resolution of liver damage through massive
   weight loss ([109]40). Our study found that α-KG decreased after LSG,
   which showed a positive correlation with liver fat content and HOMA-IR.
   One reason may be the reduced absorption of nutrients by the
   gastrointestinal tract after LSG, and another important reason may be
   related to the improvement of metabolic state after LSG, leading to the
   return of TCA circulation from high flux to normal with a reduction of
   α-KG. As recent studies reported, α-KG can be used as a signal molecule
   to control hepatic gluconeogenesis via the AKG-serpina1e pathway, which
   decreased blood glucose in mice ([110]41). It can also stimulate muscle
   hypertrophy and fat loss through oxoglutarate receptor 1
   (OXGR1)-dependent adrenal activation ([111]42). These findings suggest
   that α-KG holds promising potential as a novel biomarker and
   therapeutic target for combating obesity and its associated
   comorbidities.

   2-Ketobutyric acid, also known as 2-ketobutyrate or 2-oxobutyrate, is
   involved in the metabolism of various amino acids. It is an
   intermediate in the TCA cycle and is involved in the process of glucose
   oxidation for energy. Studies have shown that 2-ketobutyric acid can
   affect the level of mitochondrial substrate-level phosphorylation
   ([112]43). Although only a few studies have reported the anti-obesity
   effect of 2-ketobutyrate, a longitudinal cohort in children aged 5–16
   years showed that reduced concentrations of 2-ketobutyrate were
   associated with insulin resistance ([113]44). In addition,
   α-hydroxybutyrate, converted from 2-ketobutyrate in adipocytes,
   modulates lipid metabolism both in vitro and in vivo ([114]45). In
   high-fat diet-fed mice, the metabolomic analysis of lactis IDCC 4301
   intracellular metabolite profiles suggested that 2-ketobutyrate could
   be a potential target compound against obesity, mediating the bacterial
   anti-obesity effects ([115]46).

   At present, there are limited studies on succinic acid semialdehyde
   (SSA), mainly focusing on its synthetic metabolites succinic acid (SA)
   and gamma-aminobutyric acid (GABA). SSA is an extremely efficient
   substrate that energizes mitochondria during normoxia ([116]47). SA is
   produced from SSA by the catalysis of succinic semialdehyde
   dehydrogenase, which can participate in the TCA cycle and combine with
   the electron transport chain to produce a large amount of free energy
   ([117]47, [118]48). A previous genome-wide association study (GWAS) on
   obese/overweight women reported that SA was positively associated with
   obesity ([119]49). Our results showed a decline in SSA when obesity
   improved after SG surgery, which supports this claim. In addition, GABA
   is produced from SSA through a series of enzymatic reactions ([120]50).
   Treatment with GABA mitigated high-fat diet-induced hyperglycemia by
   repairing muscular oxidative stress and plasma free amino acid
   disorders in mice ([121]51). In high glucose concentrations, SSA acts
   as a cell-permeant ‘GABA-shunt’ metabolite that promotes islet GABA
   metabolism, increases ATP and the ATP/ADP ratio, depolarizes β-cells,
   and stimulates insulin secretion ([122]52). However, whether there is a
   causal relationship between SSA and insulin resistance improvement
   remains uncertain.

   Oxamic acid is a pyruvate analog that can competitively inhibit lactate
   dehydrogenase (LDH)-catalyzed conversion of pyruvate into lactate.
   Oxamate (OXA), its salt form, acts as an LDH inhibitor and can limit
   cancer cell migration induced by inflammatory processes ([123]53). As a
   key enzyme in glycolysis, LDH plays an important role in the regulation
   of glycometabolism. In db/db mice, OXA can improve glycemic control and
   insulin sensitivity via inhibition of tissue lactate production
   ([124]54). In our study, OXA showed a continuous increasing trend after
   LSG, but its direct contribution to metabolic improvements, such as the
   reduction of liver fat content, requires further investigation.

   Phthalic acid (PA) is a synthetic organic compound, which is mainly
   obtained from the external environment. As external precursors of PA,
   phthalates are known endocrine disruptors and oxidant stressors, and
   exposure has been associated with obesity and insulin resistance
   ([125]55). Their accumulation in adipose tissue is due to their
   lipophilic nature, which further increases the retention of other
   lipophilic chemicals ([126]56). A 10-year prospective cohort study
   showed that certain phthalate metabolites in urine were associated with
   modestly greater weight gain in a dose–response manner ([127]57).
   However, in our study, the concentration of PA in serum did not
   decrease but increased after LSG, which was not consistent with the
   weight loss and metabolic improvement status. Possible reasons are as
   follows. First, phthalates are widely used in medical devices, capsule
   drugs, food packaging, cosmetics, textiles, etc. ([128]58). Patients
   after bariatric surgery often need to take acid-suppressing and
   stomach-protecting drugs. The increased levels of PA in the bloodstream
   of these postoperative patients may be related to the drug capsules or
   other external exposures to phthalates, rather than the result of
   surgery. In addition, rapid weight loss is always accompanied by the
   cleavage of a large number of fat cells owing to the lipophilic nature
   of PA, resulting in the release of accumulated PA into the blood. We
   only detected PA in the bloodstream, and further tests of the liver and
   urine are necessary.

   As previous studies mainly focused on European and American
   populations, our study elucidated the metabolome characteristics in
   Chinese obese populations after LSG. The longitudinal follow-up at
   three time points enhanced the reliability of the results. However, in
   human samples, food or drug intake and other environmental exposures
   may affect serum metabolite concentrations. Considering the limitation
   of sample size, it is difficult to conduct subgroup analysis. Although
   untargeted metabolomics offers significant advantages in revealing
   novel metabolic signatures and discovering potential biomarkers, its
   quantitative accuracy is relatively limited compared to targeted
   metabolomics approaches. The targeted metabolomics validation in a
   larger sample population is needed in the future. Moreover, the role of
   these metabolites in metabolic improvement and their specific
   mechanisms require more in-depth in vivo and in vitro experiments to
   prove.

Conclusions

   In summary, this study identified significant changes in the serum
   metabolomic profile at 1-, 3-, and 6-month post surgery compared to
   pre-surgery, which opened up new insights into the biomarkers and
   mechanisms of metabolic improvement in obesity and related diseases
   after LSG.

Supplementary Materials

   Supplementary Table 1 Correlation analysis of other 29 DEMs and
   clinical parameters
   [129]supplementary_table_1.pdf^ (189.9KB, pdf)
   Supplementary Table 2
   [130]supplementary_table_2.xlsx^ (26.9KB, xlsx)
   Supplementary Table 3
   [131]supplementary_table_3.xlsx^ (53.7KB, xlsx)

Declaration of interest

   The authors declare that there is no conflict of interest that could be
   perceived as prejudicing the impartiality of the research reported.

Funding

   This research was supported by the National Science and Technology
   Major Project (Grant No. 2023ZD0508703 to HB); National Natural Science
   Foundation of China (Grant no. 82370870 to HB); Science and Technology
   Commission of Shanghai Municipality (Grant no. 22Y31900302 to HB);
   Shanghai Municipal Health Commission (Grant no. 202240295 to HB); and
   the Youth Fund of the National Natural Science Foundation of China
   (Grant no. 82100827 to YH).

Acknowledgements