Blood metabolomic profiling predicts postoperative gastrointestinal function of colorectal surgical patients under the guidance of goal-directed fluid therapy

Basic characteristics of patients in the delayed and enhanced postoperative recovery groups

The demographic data for all patients were shown in Table 1. Among 48 patients with colorectal cancer, 24 showed enhanced postoperative recovery after surgery. There were no significant differences between the enhanced group and the delayed group in age, sex, BMI, ASA grade, operation duration, preoperative SVV, postoperative SVV, preoperative mean blood pressure (MBP), postoperative MBP, intra-operative fluid administration, perioperative 24-hour fluid administration and other aspects. In the enhanced group, the time of flatus, defecation, and feeding was advanced, the incidence of anemia was reduced, and the incidence of postoperative nausea and vomiting was decreased.

Overall metabolomics analysis of blood samples

The metabolomics of 96 blood samples from 48 patients who underwent surgery for colorectal cancer in the enhanced and delayed groups were characterized and compared. Blood samples were collected when patients were in the post anesthesia care unit (PACU) and at 24 hours after surgery for all 48 patients. Delayed PACU samples were separated as group A, enhanced PACU samples were separated as group B, delayed 24 hours samples were separated as group C, and enhanced 24 hours samples were separated as group D. Representative ESI+ and ESI- ion chromatograms for each group are presented in Supplementary Figure 1. In total, 4,652 molecular features in ESI+ ion mode and 1,898 molecular features in ESI- ion mode were obtained and subjected to MVA. Principal component analysis extracted 16 principal components for ESI+ ion mode with a Q² of 0.255 and R²X of 0.553. For ESI- ion mode (Figure 1A), there were 15 principal components, with a Q² of 0.284 and R²X of 0.584 (Figure 1B). From the overall point of view, enhanced and delayed group distributions were overlapped in both ESI+ ion mode and ESI- ion mode, and a significant difference existed between PACU and 24 hours after surgery groups.

The partial least squares discriminant analysis (PLS-DA) results of enhanced and delayed postoperative recovery samples in the PACU revealed differences in their distributions, with a Q² of 0.343 and R²Y of 0.97 in ESI+ ion mode (Figure 2A) and a Q² of 0.433 and R²Y of 0.95 in ESI- ion mode (Figure 2B). Although the R²Y reflected a not over lifting model, the Q² had low prediction rate, suggesting that this model may be unreliable. Features with variable importance in projection (VIP) scores greater than 1.0 in multivariate statistical analysis and p values of less than 0.05 in univariate analysis were considered as the most significant metabolites. Twenty metabolites for ESI+ ion mode and eight metabolites for ESI- ion mode were selected as the most significant differentiating metabolites (Supplementary Table 1).

The PLS-DA results for blood samples collected 24 hours after surgery in the enhanced and delayed postoperative recovery patients revealed differences in their distributions, with a Q² of 0.691 and R²Y of 0.985 for ESI+ ion mode (Figure 2C) and a Q² of 0.453 and R²Y of 0.811 for ESI- mode (Figure 2D). This model also appeared to have a low prediction rate and to be not over lifting. Features with VIP scores greater than 1.0 in multivariate statistical analysis and p values less than 0.05 in univariate analysis were considered as the most significant metabolites. In total, 27 metabolites for ESI+ ion mode and seven metabolites for ESI- ion mode were selected as the most significant differentiating metabolites (Supplementary Table 2).

In a comparison of differential metabolites in groups A–B and C–D, we found similar variation tendencies. The differential metabolites are shown in Supplementary Tables 1, 2. Most of the metabolites showing similar trends were not identified as significant differential metabolites. However, phosphohydroxypyruvic acid rates were high in groups A–B and C–D in ESI+ ion mode.

After screening differential metabolites, the enriched pathways were further analyzed (Figure 3). In groups A–B, glyoxylic acid and dicarboxylic acid metabolism, vitamin B6 metabolism, and pyruvate metabolism were identified as significantly enriched pathways. Two of these three pathways showed high occupancy percentages (>90%). Notably, glycerophospholipid metabolism was the most significantly enriched pathway in groups C–D and showed the highest occupancy percentage (>90%). The phosphohydroxypyruvic acid selected only in groups C–D was enriched in glycine, serine, and threonine metabolism.

SVM analysis of blood samples delayed and enhanced postoperative recovery patients

Identification of differential blood metabolites between the enhanced and delayed postoperative recovery groups suggested that some of the identified metabolites may have applications as biomarkers to predict whether a patient would suffer from poor postoperative recovery after surgery. Therefore, an SVM model was applied to identify the patterns among metabolites. By using these differentiated blood metabolites between group A and B as input features, an SVM model with 22 features was obtained (Figure 4A). An AUC score of 0.983 indicated an effective model, and the prediction accuracy reached 92.8% (Figure 4B). For samples belonging to groups C and D, the optimal SVM model had 10 features (Figure 4D). An AUC score of 0.941 indicated an effective model, and the prediction accuracy reached 88.8% (Figure 4E).

In both the A–B and C–D SVM models, there were four metabolites calculated as important features (Figure 4C and 4F). Oleamide, ubiquinone-1, acetylcholine, and oleic acid could be used to separate patients exhibiting enhanced and delayed postoperative recovery both in the PACU and at 24 hours after surgery. Thus, these metabolites could be used as potential biomarkers to predict patients’ postoperative recovery state according to the SVM results. Among all these features, oleamide had the highest selected frequency, and ubiquinone-1 also ranked among the top 3 in both models.

There were 28 differential metabolites between groups A and B; 22 of these metabolites could be used as important features in the A–B SVM model. Among the nine metabolites showing similar variation trends compared with those in the C–D model, six were identified as SVM features. For the C–B model, 10 of 34 differential metabolites were used as important features in the SVM model, whereas five feature metabolites had similar variation trends compared with groups A–B among the 21 total metabolites.

Study design and participants

This was a single-center, prospective, observational cohort study. Two parallel groups (enhanced and delayed recovery) were enrolled according to the postoperative time of flatus. Patients who had undergone elective surgery for colorectal cancer resection were enrolled from December 2018 to October 2019 at Shanghai Changhai Hospital, Second Military Medical University. Written informed consent was obtained from all participants on the day before surgery. Patients who met any of the following criteria were excluded: ASA IV or V; age < 18 years or > 70 years; fasting time < 8 hours or > 20 hours; pre-operative ileus; body mass index (BMI) < 18 kg/m² or > 25.0 kg/m²; presence of cardiovascular diseases, such as coronary artery or valvular diseases, New York Heart Association Functional Classification class II or higher heart failure, and arrhythmia; presence of metabolic diseases, such as severe diabetes, ketoacidosis, hyperosmolar syndrome, or long-term hypoglycemia; presence of systemic immune disease; presence of liver and kidney dysfunction; presence of chronic obstructive pulmonary disease; and refusal to participate in the study or presence of mental illness and inability to communicate correctly. In addition, because of the pneumoperitoneum in laparoscopic surgery may affect the change of chest pressure, and the accuracy of SVV monitoring may be influenced, thereby the laparoscopic colorectal surgeries were also excluded.

Anesthesia

General anesthesia was induced by midazolam (0.04 mg/kg), propofol (1–2.5 mg/kg), fentanyl (2.5–6 μg/kg), cis-atracurium (0.2–0.3 mg/kg), and dexamethasone (10 mg). Anesthesia was maintained by inhalation of sevoflurane (0.8–1.3 MAC) with intermittent supplementation using other auxiliary drugs, such as remifentanyl (0.1–0.2 μg/kg/min), dexmedetomidine (0.2–0.5 μg/kg/h), cis-atracurium, and fentanyl. To accelerate postoperative extubation, inhalation was discontinued, and total intravenous anesthesia by propofol (100–200 μg/kg/min) was applied before abdominal closure. Patients were ventilated in a volume-controlled mode with a tidal volume of 8 mL/kg (air: 40–100% oxygen). The respiratory rate was set to maintain the PaCO₂ between 35 and 45 mmHg to improve the operating conditions, and the positive end-expiratory pressure was set at 6–8 mL H₂O when necessary. For intra-operative analgesia, each patient received bilateral transverse abdominal fascia nerve blocks containing 20–30 mL ropivacaine (0.375%) after induction.

Fluid management and hemodynamic monitoring

On arrival at the operating room, patients received intravenous ringer lactate (5 mL/kg), and hemodynamic parameters were recorded as the baseline. After induction of anesthesia, a 20-G radial arterial line was inserted and connected to an electrocardiogram monitor and fourth-generation Vigileo/Flotrac system (Edwards Lifesciences, Irvine, CA, USA) to determine changes in arterial blood pressure waveforms, SVVs, and cardiac indexes. A double-lumen 5.5-Fr catheter (arrow central venous catheter; Teleflex Life Sciences Ltd., Athlone, Ireland) was inserted into the right internal jugular vein for colloid supplementation. A continuous infusion of ringer lactate was given through peripheral veins during the operation at a rate of 3 mL/kg/h. Colloid (hydroxyethyl starch 130/0.4 sodium chloride injection; Fresenius Kabi, USA) management was conducted under the guidance of the Vigileo/Flotrac system. If the SVV changed by more than 10% 2–3 consecutive times, the colloid was injected via the central vein at 4 mL/kg. After 5 min, SVV changes were observed again. Vasoactive agents were given according to changes in intra-operative hemodynamics based on the recommendations of the anesthesiologist. Transfusions were applied according to recent guidelines [22].

Data collection

The study was designed as a two-arm, parallel, prospective, observational cohort. The primary endpoint of the study was the timepoint of flatus and differences in blood metabolomics between the two groups. Hemodynamic parameters at different timepoints, surgical resection methods, total intra-operative time, and perioperative fluid volume were recorded. Additionally, the number of hemodynamic fluctuations, such as hypotension or hypertension, was collected. The timepoints of flatus and feeding were recorded, and the time of flatus was used as an indicator of postoperative recovery. The occurrence of flatus 72 hours after surgery was identified as delayed postoperative recovery. The length of hospital stay and duration of postoperative complications were calculated.

Blood sample collection

Blood samples were collected from all patients at PACU admission and 24 hours after surgery. Blood was collected into heparinized tubes, and plasma was immediately separated by centrifugation at 3000 rpm for 10 min and stored at –80° C until metabolomics analysis.

Liquid chromatography (LC)/mass spectrometry (MS) analysis

LC/MS analysis was carried on an Ultimate 3000LC Q Exactive platform (Thermo, USA). The parameters for electron spray ionization (ESI)+ and ESI- ion modes were as follows: heater temperature, 300° C; sheath gas flow rate, 45 arb; auxiliary gas flow rate, 15 arb; sweep gas flow rate, 1 arb; spray voltage, 3.0 kV; capillary temperature, 350° C; S-Lens RF level, 30%. The data were analyzed using feature extraction and preprocessed with Compound Discoverer software (Thermo). In total, 2015 features in ESI+ ion mode and 1601 features in ESI- ion mode were evaluated, and editing of data was performed with the Multivariate Analysis (MVA) feature of SIMCA-P software (Umetrics AB, Umea, Sweden).

Biomarker identification

A supervised learning model was built by a support vector machine (SVM) to classify the delayed and enhanced postoperative recovery groups. Finally, we used radial as SVM kernel function under type C classification and selected 22 features (A–B) and 10 features (C–D) to construct the two SVM models separately. Under these parameters, the predictive accuracies reached 92.8% and 88.8%, and the area under the curve (AUC) values were 0.983 and 0.941 for groups A–B and C–D, respectively, indicating that the model was effective.

Statistical analysis

Statistical analysis was performed using GraphPad Prism version 8.0 (San Diego, CA, USA). Baseline demographics and postoperative complications were evaluated using Pearson’s chi-square tests for categorical variables, with Fisher’s exact test for small counts. Analysis of variance with Tukey’s correction was used for continuous data. Results with P values less than 0.05 were considered statistically significant.

Ethics approval and consent to participate

The study was approved by the appropriate Institutional Review Board (IRB) and Changhai Hospital Ethics Committee (Shanghai, China; approval no. CHEC2018-080) and was registered at the Chinese Clinical Trial Registry (http://www.chictr.org.cn/index.aspx; identifier: ChiCTR1800015899). Written informed consent was obtained from all patients or legal guardians.

Consent for publication

Patients gave consent for their personal or clinical details to be published in this study. A copy of the written consent is available for review by the Editor-in-Chief of this journal.

Availability of data and materials

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.