Metabolomic Profiling for Diagnosis and Prognostication in Surgery: A Scoping Review

STRUCTURED ABSTRACT

Objective:

This review assimilates and critically evaluates available literature regarding the use of metabolomic profiling in surgical decision-making.

Summary Background Data:

Metabolomic profiling is performed by nuclear magnetic resonance spectroscopy or mass spectrometry of biofluids and tissues to quantify biomarkers (i.e. sugars, amino acids, and lipids), producing diagnostic and prognostic information that has been applied among patients with cardiovascular disease, inflammatory bowel disease, cancer, and solid organ transplants.

Methods:

PubMed was searched from 1995–2019 to identify studies investigating metabolomic profiling of surgical patients. Articles were included and assimilated into relevant categories per PRISMA-ScR guidelines. Results were summarized with descriptive analytical methods.

Results:

Forty-seven studies were included, most of which were retrospective studies with small sample sizes using various combinations of analytic techniques and types of biofluids and tissues. Results suggest that metabolomic profiling has potential to effectively screen for surgical diseases, suggest diagnoses, and predict outcomes such as postoperative complications and disease recurrence. Major barriers to clinical adoption include a lack of high-level evidence from prospective studies, heterogeneity in study design regarding tissue and biofluid procurement and analytical methods, and the absence of large, multicenter metabolome databases to facilitate systematic investigation of the efficacy, reproducibility, and generalizability of metabolomic profiling diagnoses and prognoses.

Conclusions:

Metabolomic profiling research would benefit from standardization of study design and analytic approaches. As technologies improve and knowledge garnered from research accumulates, metabolomic profiling has potential to provide personalized diagnostic and prognostic information to support surgical decision-making from preoperative to post-discharge phases of care.

MINI-ABSTRACT

Diagnosis and prognostication of surgical diseases can benefit from metabolomic profiling of biofluids and tissues. This review assimilates and critically evaluates available literature regarding the use of metabolomic profiling in surgical decision-making, focusing on potential applications to screening, suggesting diagnoses, and predicting outcomes such as postoperative complications and disease recurrence.

INTRODUCTION

Metabolomics is an emerging discipline that examines interactions of metabolites within complex biological systems¹. These metabolites, such as sugars, amino acids, nucleotides, and lipids, are intermediates and products of cellular metabolism, can be influenced by exogenous factors and xenobiotics, and contribute to cellular function and dysfunction². Therefore, the metabolome provides a phenotypic evaluation of cellular and systemic health, with important implications for the pathogenesis and progression of disease, biomarker discovery, drug effectiveness, and personalized medicine³.

Metabolomic profiling can provide a comprehensive profile of thousands of small molecules in a biological sample, such as plasma, serum, saliva, urine, or feces, among others⁴. Multiple platforms have been developed for metabolomic research, with mass spectrometry (MS)-based instruments gaining popularity due to their sensitivity and mass accuracy, while others such as nuclear magnetic resonance (NMR) have the ability to provide detailed structural information. In addition, complementary front-end platforms, such as gas chromatography (GC) and liquid chromatography (LC), differentiate small molecules based on their chemical properties. Furthermore, LC tandem and triple quadrupole MS instruments can provide the increased sensitivity necessary for clinical applications, with the ability to accurately quantify a single analyte or a combination of analytes in a multiplex fashion^5–9.

As metabolomic profiling technologies become more precise and efficient, their potential for clinical application improves as well. Theoretically, metabolome data can screen for the insidious onset of surgical diseases that often remain subclinical until the disease process has advanced to a point that is difficult to treat and cure, such as peripheral arterial disease or pancreatic cancer. In addition, metabolic profiles predicting postoperative complications could augment decisions regarding resource use, e.g., renal protection bundles for patients at high risk for postoperative acute kidney injury (AKI) or continuous postoperative cardiorespiratory monitoring for patients at high risk for decompensation. Among surgical oncology patients in the post-discharge phase, metabolite profiles could surveil for disease recurrence.

This review will assimilate and critically evaluate available literature describing the use of metabolomic profiling in the pre-, peri-, and post-operative settings for patients with cardiovascular disease, inflammatory bowel disease, cancer, and solid organ transplants, investigating the potential for metabolomic profiling to augment surgical decision-making in each of these settings.

METHODS

Prior to initiation of research, the University of Florida Institutional Review Board (IRB #201400127) approved this as an exempt study with waiver of informed consent.

Identification of Studies for Inclusion

PubMed was searched from January 1, 2009 to October 25, 2019, following Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) criteria, as illustrated in Supplemental Figure 1 and listed in Supplemental Table 1¹⁰. The design, population, sample size, analytic methods, and major findings of included studies are summarized in Table 1. The authors performed automated and web-based literature searches using text mining, as illustrated in Supplemental Figure 2. The initial search included “metabolomics, metabolomics” and “surgery, perioperative, postoperative” and was subsequently combined with terms for specific conditions “coronary artery disease, abdominal aortic aneurysm, aortic dissection, Crohn’s disease, pancreatic cancer, breast cancer, colon cancer, lung cancer, esophageal cancer, congenital heart disease, kidney transplant, liver transplant” and specific metabolites “alanine, glutamate, glutamine, arginine, proline, aspartate, histidine, beta-alanine, mannose, ascorbate, dicarboxylate, glycoxylate, porphyrin, aldarate, ATP binding cassette, isoleucine”. These diseases and metabolites were selected based on relevance to review objectives and a text mining function in R (Version 3.4.2) run on RStudio (1.0.153), which queried PubMed for specified terms using library(rentrez), with function entrez_search. In addition, bibliographies of studies identified through this search were manually searched to identify additional studies relevant to review objectives and pertaining to the same surgical diseases identified in the primary search. In total, 1,344 records were identified through the primary database search and an additional 47 were identified from secondary searches. Following removal of duplicates, 1,085 records remained. After reviewing titles and abstracts to exclude non-human studies, non-surgical studies, and studies involving tissues other than blood, urine, or feces, 112 articles remained. Full texts were then assessed for eligibility. Another 65 articles were excluded for being present in a review already included in the study, non-human studies, non-surgical studies, studies involving tissues other than blood, urine, or feces, and not involving a surgical disease included in this review (e.g., bariatrics, trauma). Forty-seven studies were included.

Table 1.

Summary of included studies using metabolomic profiling for diagnosis and prognostication in surgery.

Diagnostic studies
Primary Author	Study Design	Study Population and Sample Size	Analytic Methods	Major Findings Pertinent to this Review
Rizza12	Prospective, Observational	67 elderly Italian patients (mean age 85) with history of CVD	Principal Component Analysis (PCA), Random Survival Forest analysis, and Cox proportional hazards regression modeling to evaluate associations among metabolites and major adverse cardiac events	Metabolite factor 1, composed by medium- and long-chain acylcarnitines, and factor 7 (alanine) were independently associated with major adverse cardiac events, after adjustment for potential confounders
Zagura13	Observational, Cross-sectional	42 males with symptomatic PAD 46 healthy controls	Univariate analysis for demographics and clinical characteristics of PAD and healthy patients; multiple regression models to evaluate independent determinants of arterial stiffness	Arterial stiffness, as measured by aortic pulse wave velocity, is independently associated with serum levels of tyrosine and oxLDL in patients with PAD and is related to pyruvate and oxLDL levels in the control group
Hernandez-Aguilera11	Observational, Cross-sectional	201 males with PAD 48 age-matched healthy men	Univariate analysis of PAD patients and control group; multivariate analyses to identify association between metabolites and comorbidities; ROC to assess diagnostic accuracy of variables	Association between several metabolites (3-hydroxybutirate, aconitate, (iso)citrate, glutamate, and serine) with markers of oxidative stress and inflammation; (iso)citrate and glutamate can discriminate between healthy participants and PAD patients without symptoms
Wang14	Observational, Cross-sectional	33 Chinese patients: 11 with acute AD, 11 with chronic AD, 11 with CHD but no evidence of dissection	Univariate analysis to evaluate biochemical analysis of each group; PCA was used to evaluate plasma AA profile between groups	Plasma aminograms were significantly altered in patients with AD compared with CHD, especially in acute AD
Zhou15	Observational, Cross-sectional	35 patients with acute AD (20 Stanford Type A, 15 Stanford Type B) 20 healthy controls with no AD	Univariate analysis to evaluate characteristics of control group, Type A and Type B dissections; major latent variables were modeled using PCA and partial least squares discriminant analysis (PLS-DA)	Compared with those in the healthy control group, LPC levels were significantly lower in both the Stanford type A and type B AAD groups
Ruperez17	Observational, Cross-sectional	11 patients with large AAA 11 patients with small AAA 11 controls patients with no AAA	Metabolites of the three groups were analyzed using PCA, PLS-DA and orthogonal partial least-squares discriminant analysis (OPLS-DA)	All fatty acids except laurate and several indicators of carbohydrate metabolism (e.g., acetoacetate, 3-hydroxybutyrate, and acetone) were significantly higher among AAA patients compared with controls; aminomalonate was 800% higher among patients with large AAA
Ciborowski18	Observational, Cross-sectional	15 patients with large AAA 15 patients with small AAA 11 control patients with no AAA	Differences between plasma metabolites were assessed with univariate analysis; PLS-DA was used to discriminate samples between groups and allocate new objects in the model	Guanidosuccinic acid was increased and hippuric acid was decreased in plasma of patients with large AAA compared with controls
Moxon19	Observational, Cross-sectional	161 patients with AAA 168 healthy controls	Univariate analysis to assess differences between patients with AAA and controls with PAD; binary logistic regression to assess associations of lipid species with AAA while adjusting for covariates	Serum levels of 3 diacylglycerol and 7 triacylglycerol species were associated with AAA
Ahmed22	Observational, Cross-sectional	62 patients with active CD 55 patients with inactive CD 48 patients with active UC 52 patients with inactive UC 109 healthy controls	Univariate analysis was used to identify discriminatory metabolites among the three groups; PCA and PLS-DA were performed to discern differences between groups and determine class membership	Heptanal, 1-octen-3-ol, 2-piperidinone and 6-methyl-2-heptanone were up-regulated in active CD group, while methanethiol, 3-methyl-phenol, short-chain fatty acids, and ester derivatives were less abundant in patients with active CD compared with inactive CD and healthy controls
Fathi25	Observational, Cross-sectional	26 patients with CD 29 healthy controls	Random forests were used for classification and regression models	CD group valine and isoleucine levels were lower and higher than those of the healthy cohort, respectively
Dawiskaba23	Observational, Cross-sectional	24 patients with UC 19 patients with CD 17 healthy controls	Differences in metabolites were assessed using PLS-DA; metabolites responsible for separation in models were tested using univariate analysis	No difference in urine and serum metabolomes between CD and UC patients
Long29	Systematic Review	Review article: included 25 studies focused on the identification rather than the validation of predictive capacity of potential biomarkers	PubMed, Scopus, and Web of Science were searched for relevant studies published before September 2017	External validation of the biomarker panels was observed in nine studies; the most reported candidates were glutamic acid and histidine in seven studies, and glutamine and isoleucine in five studies, leading to the predominant enrichment of amino acid-related pathways
Tumas30	Observational, Cross-sectional	74 patients (pancreatic adenocarcinoma (PDAC) n=50, other pancreatic cancers (OPC) n=17, and chronic pancreatitis (CP) n=7.	Univariate analysis to assess characteristics of study participants as well as to detect differences in AA concentrations between study groups	Concentrations of ornithine, threonine, phenylalanine, glycine, arginine, histidine, glutamine, 3-Mhis, and citrulline differed in PDAC vs. OPC and CP; differences in the concentrations of ornithine, threonine, phenylalanine, valine arginine, histidine, asparagine, glutamine, 3-Mhis, and citrulline were identified in PDAC vs. OPC; in patients with PDAC, there was a significant inverse correlation between plasma histidine concentrations and PDAC stage
Davis28	Observational, Cross-sectional	32 patients with adenocarcinoma 25 patients with benign pancreatic conditions 32 healthy controls	Univariate analysis to assess patient characteristics and metabolite differences between groups; PCA and OPLS-DA multivariate pattern recognition were used to discriminate between metabolites	22 metabolites showed significantly different levels of expression when comparing PDAC patients and controls
Bathe26	Observational, Cross-sectional	56 patients with pancreatic adenocarcinoma 43 patients with benign hepatobiliary disease	Univariate analysis to assess clinical and patient features in two groups; PCA and OPLS-DA analysis used to assess variance between groups and metabolite concentrations	Elevated glutamate, acetone, and 3-hydroxybutyrate are metabolites that are strongly related to disease alone and elevated glucose, phenylalanine, formate, and mannose are related to disease and age
Zhang32	Systematic review	Review article: 23 studies	A literature search was done through three databases (PubMed, Web of Science, and Embase) with the combinations of keywords from 1998 to January 2016	Histidine, methionine and tryptophan were decreased among colorectal cancer patients, and glutamic acid, proline, iso-glutamine, and putrescine were increased
Farshidfar33	Observational, Cross-sectional	62 patients with colorectal cancer (CRC) 31 patients with colon adenoma 81 healthy controls	Univariate analysis to assess clinical and patient features in two groups; PCA and OPLS-DA analysis used to assess variance between groups and metabolites	Ornithine and serine had the strongest positive correlation with CRC, and decenoylcarnitine had the strongest negative correlation with CRC
Abbasi-Ghadi35	Systematic review	Review article: 20 studies investigating the metabolomics profile of human biological samples from patients with esophago-gastric cancer compared to control group	A literature search (title and abstract) of Ovid Medline(R) (1948–2012), Embase (1974–2012), Web of Science and PubMed electronic databases was conducted up to and including 9th November 2012 for studies of metabolomic profiling of oesophago-gastric cancer	Lactate and fumurate were the most commonly recognized biomarkers of esophago-gastric cancer; valine, glutamine, and glutamate were the most commonly identified amino acid biomarkers
Davis36	Observational, Cross-sectional	44 patients with esophageal carcinoma 31 patients with Barrett’s Esophagus 75 healthy controls	Univariate analysis to assess clinical and patient features; PCA and OPLS-DA multivariate pattern recognition applied to discriminate metabolites between samples	9 metabolites distinguished cancer patients; sucrose, 2-aminobutyrate, cis-aconitate, pyroglutamate, hypoxanthine, and fucose were increased in the cancer cohort, and pantothenate, urea, and methanol were decreased
Dougan37	Observational, Cross-sectional	45 patients with breast cancer 45 healthy controls	Univariate analysis to assess clinical and patient features; PCA to discriminate metabolites between samples	24 metabolites differentiated cases from controls: 3-(cysteine-S-yl) acetaminophen and 4-acetylphenol sulfate as well as cysteine s-sulfate were increased among women with cancer, whereas indoleacetylglutamine, 2-ethylphenylsulfate, and sphingosine were increased among controls; estrogen receptor-positive patients had higher serum laurylcarnitine levels than estrogen receptor-negative patients, and progesterone receptor-positive patients had lower serum asparagine than progesterone receptor-negative patients
Cala39	Observational, Cross-sectional	31 women with breast cancer 29 women without breast cancer	Univariate analysis to assess clinical and patient features; PCA and OPLS-DA multivariate pattern recognition applied to discriminate metabolites between samples	Tricarboxylic acid cycle intermediates were decreased compared with healthy controls
Fan38	Observational, Cross-sectional	96 patients with breast cancer 79 healthy controls	Univariate analysis to assess clinical and patient features; PLS-DA to discriminate metabolites between samples	Eight potential small molecule biomarkers for the diagnosing BC subtypes; ER-positive patients had elevated alanine, aspartate, glutamate, and purine metabolism and decreased glycerolipid catabolism; HER2- positive patients had elevated carnitine, lysoPC, and proline metabolites
Kumar40	Observational, Cross-sectional	41 patients with lung cancer 41 healthy controls	Univariate analysis to identify significant metabolites from lung cancer samples; pathway analysis plot and correlation network plot to identify the plasma and serum biomarkers for lung cancer	Plasma taurine, aspartic acid, and pyruvic acid were increased in lung cancer patients compared with healthy controls, and glutamine, aspartic acid and inosine were decreased compared with healthy controls
Chen42	Observational, Cross-sectional	30 patients with lung cancer 30 healthy controls	Univariate analysis to assess clinical and patient features; PCA and PLS-DA to discriminate metabolites between samples	Sphingosine, which was lower in lung cancer patients, and glycerophospho-N-arachidonoyl ethanolamine, which was higher in lung cancer patients, were sensitive and specific in distinguishing lung cancer from healthy controls
Yu41	Mini-review	Review article: 71 articles	Retrieved articles in PubMed using the retrieval terms of “lung cancer AND metabolomics” and “lung cancer AND metabolic profiling”, a total of 71 papers on the lung cancer metabolomics published in the period of 2009 - August, 2017	Identified several early serum (benzaldehyde, urea, isoleucine, glycolic, phenylalanine, carnitine, propionylcarnitine, tyrosine, glycerol, malic acid, histidine, and valine) and urine (hippuric acid, hydroxy-isovaleric acid, creatinine, tryptophan, hippuric acid, carnitine, and threonine) markers for lung cancer
Prognostic studies
Primary Author	Study Design	Study Population and Sample Size	Analytic Methods	Major Findings Pertinent to this Review
Shah44	Prospective, observational study	2023 adult patients undergoing cardiac catheterization	Univariate analysis to assess clinical and patient features; PCA to assess metabolites with Cox modeling and log-likelihood and reclassification analyses	Five of 13 metabolite factors were independently associated with mortality, including medium-chain acylcarnitines, short-chain dicarboxylacylcarnitines, long-chain dicarboxylacylcarnitines, branched-chain amino acids and fatty acids; three factors independently predicted death/MI, including short-chain dicarboxylacylcarnitines, long-chain dicarboxylacylcarnitines and fatty acids
Davidson45	Prospective, observational study	82 infants age <120 days with heart defects undergoing open heart surgery	Univariate analysis to assess clinical and patient features; PLS-DA, pathway analysis to evaluate metabolome changes	Aspartate levels were significantly lower in survivors, while methylnicotinamide levels were higher; aspartate and methylnicotinamide were also predictive of intensive care unit length of stay
Correia46	Subset analysis of a prospective, randomized clinical trial	28 children with average age 180 days with heart defects undergoing open heart surgery, 15 underwent tight glycemic control and 13 received standard care	Univariate analysis to assess clinical and patient features; PCA to evaluate metabolites	Eight metabolites (3-d-hydroxybutyrate, acetone, acetoacetate, citrate, lactate, creatine, creatinine, and alanine) were associated with surgical and disease severity
Elmariah52	Prospective, observational study	44 patients undergoing TAVR 9 with post-operative AKI 35 without post-operative AKI	Univariate and multivariate analysis to evaluate the associations between metabolites and clinical outcomes	Elevated levels of 5-adenosylhomocysteine were predictive of both AKI and all-cause mortality
Kirov47	Prospective, observational study	20 adult patients undergoing CABG (10 on-pump, 10 off-pump)	Univariate analysis to assess patient features as well as metabolite concentrations	No differences in plasma metabolomes for on-pump versus off-pump CABG, but a significant inverse correlation between serum arginine at the end of surgery and postoperative vasopressor requirements
Maltesen 48	Prospective, observational study	50 adults undergoing cardiac surgery	Univariate analysis to assess clinical and patient features; PCA and PLS analysis to assess metabolome response to surgery	Longer ischemia time was associated with increased pyruvate, acetate, glycine, alanine, and glutamine, as well as decreased arginine, isoleucine, and 3-methylhistidine
Lee49	Prospective, observational study	100 adults undergoing cardiac surgery (49 received cardioplegia solution)	Univariate analysis to assess clinical and patient features; PCA to assess metabolome response to surgery	Cardioplegia was associated with up-regulated histidine metabolism with subsequently increased glutamine and glutamate metabolism and enhanced vitamin B6 metabolism
Zacharias53	Prospective, observational study	85 adults undergoing cardiac surgery (33 patients with post-operative AKI, 52 without post-operative AKI)	Prognostication with a random forest classifier; PCA to assess metabolites in patients with and without AKI	Serum magnesium, lactate, and the glucuronide conjugate of Propofol were predictive of AKI
Shah51	Prospective, observational study	478 patients referred for coronary artery bypass 352 with post-operative adverse events (MI, need for repeat revascularization, death) 126 without adverse events	PCA and Cox proportional hazards regression modeling to assess the associations between metabolites and clinical outcomes	Elevated serum levels of short-chain dicarboxylacylcarnities, ketone-related metabolites, and short-chain acylcarnitines were predictive of myocardial infarction, the need for revascularization, or death
Buter50	Prospective, observational study	90 adults undergoing cardiac surgery	Univariate and multivariate analysis to evaluate associations between plasma glutamine and clinical outcomes	Low preoperative glutamine was associated with increased risk for infectious complications after elective cardiac surgery
Keshteli54	Observational, Cross-sectional study	38 patients with or without CD recurrence after ileocolonic resection (10 without recurrence, 28 with recurrence)	Univariate analysis to assess clinical and patient features; multivariate analysis to assess metabolomic profiles among CD groups	Higher levels of urinary levoglucosan, L-DOPA, and ethylmalonate and lower levels of urinary propylene glycol were associated with disease recurrence
Tenori55	Retrospective study	95 women with metastatic breast cancer 80 women with early stage breast cancer with no relapse	Multivariate statistics and a random forest classifier to create a prognostic model for disease relapse for early stage breast cancer	Women who later developed metastatic disease had lower serum levels of histidine and higher serum levels of glucose, lactate, tyrosine, and lipids compared with women who did not develop recurrent disease
Hart56	Multicenter retrospective study	675 patients with early breast cancer 125 patients with metastatic breast cancer	PCA was used first to assess the presence of any clusters or outliers; a random forest classifier separated early and metastatic patients	Several metabolites (choline, acetate, formate, lactate, glutamate, 3-hydroxybutyrate, phenylalanine, glycine, leucine, alanine, tyrosine, isoleucine, histidine, creatine, methionine, and proline) were increased in the serum of women with metastatic disease
Asiago58	Retrospective study	56 breast cancer patients (20 with recurrence)	Univariate analysis to asses clinical and patient features as well as ranked set of markers; PLS-DA modeling to assess metabolomic profiles	Patients with recurrent cancer displayed lower concentrations of formate, histidine, proline, N-acetyl-glycine, and 3-hydroxy-2-methyl-butanoic acid
Jobard59	Prospective, observational study	85 patients (46 early stage breast cancer, 39 with metastatic disease)	Univariate analysis to assess clinical and patient features; PCA and OPLS to classify and derive group-specific metabolic phenotypes	Metastatic disease patients have higher serum concentrations of acetoacetate, 3-hydroxybutyrate (beta-hydroxybutyrate), glycerol, pyruvate, N-acetylglycoproteins, lipid (CCCH2CC), mannose, glutamate and phenylalanine, and lower concentration of histidine, alanine and betaine
Oakman57	Prospective, observational study	44 patients with early stage breast cancer 51 patients with metastatic breast cancer	Univariate analysis to assess clinical and patient features; OPLS differentiated of early and metastatic patients	Metastatic disease was characterized by elevated serum levels of phenylalanine, glucose, proline, lysine, and N-acetylcysteine, and lower lipid levels
Bassi60	Observational, Cross-sectional	40 renal transplant patients with chronic renal dysfunction	Univariate analysis to assess clinical and patient features; multivariate analysis to assess metabolite profiles	Serum trends in tryptophan, glutamine, and dimethylarginine as well as urine concentration of histidine, DOPA, dopamine, carnosine, SDMA, and ADMA were associated with increased estimated glomerular filtration rate
Blydt-Hansen61	Prospective, observational study	396 children with kidney transplant 40 with antibody-mediated rejection 278 without antibody-mediated rejection	Univariate analysis to asses clinical and patient features as well as ranked set of markers; PLS modeling to assess metabolomic profiles of patients	Children with antibody-mediated rejection (n=40) had higher urinary proline and citrulline and lower phosphatidylcholine, tetradecanoylcarnitine, and C10.2 than other children
Mugge62	Prospective observational study	86 heart transplant patients	Univariate analysis to assess difference in metabolites among classes of rejection	Elevated urinary excretion of nitrate was associated with acute rejection
Zhao63	Prospective observational study	45 heart transplant patients	Univariate analysis to assess difference in metabolites among classes of rejection	Elevated urinary excretion of thromboxane A2 was associated with acute rejection
Singh64	Case report	1 pediatric liver transplant patient (case report)	Metabolite analysis of a single case	High levels of serum and urine urea and glutamine are associated with urea cycle abnormalities signifying graft dysfunction
Tripathi65	Observational study	9 liver transplant patients	Univariate analysis to assess clinical and patient features and compare metabolites	Elevated concentrations of nine metabolites (lactate, alanine, lysine, glutamine, methionine, asparagine, tyrosine, histidine and phenylalanine) in nonsurvivors was attributed to graft dysfunction

Study Selection and Data Collection

Three authors (TK, EL, DA) independently screened titles and abstracts for eligibility. Discrepancies were resolved by consensus or a third party (AB). Full manuscripts were retrieved for studies that met selection criteria and for which relevance to the objectives of this review remained unclear after evaluating the title and abstract. All authors reviewed and interpreted data from included articles and disagreements were resolved by group consensus.

Mapping and Coding Categories

Included studies were assimilated in two categories: 1) diagnostic studies and 2) prognostic studies estimating the probability of future events. Diagnostic studies included those examining screening methodologies and comparing the use of metabolomics to standard diagnostic modalities such as imaging. Prognostication studies included those examining postoperative complications as well as surveillance for disease recurrence. Studies were further organized to discuss metabolomic profiling of patients with cardiovascular disease, inflammatory bowel disease, cancer, and solid organ transplants. Data extraction was performed manually.

RESULTS

Among the 47 studies included, most were retrospective studies with small sample sizes using various combinations of analytic techniques and types of biofluids and tissues, as listed in Table 1. Heterogeneity among patient populations, types of biolfuids and tissues, and analytic methods precluded a pooled analysis of all results. However, global assessment of results from these 47 studies suggests that personalized metabolomic profiling can provide two essential pieces of information to support surgical decision-making: diagnosis and prognosis, each with potential clinical applications in surgery, as illustrated in Figure 1. Preoperatively, this information can help determine whether predicted outcomes match patient values and expectations and can identify modifiable risk factors that may be amenable to prehabilitation risk reduction strategies. In the perioperative phase, analysis of new biofluids and tissues obtained during procedures can update preoperative predictions and potentially identify high-risk patients who would benefit from close postoperative monitoring in an intensive care unit or continuous cardiorespiratory monitoring on a surgical ward. Postoperatively, metabolomic profiling risk assessments can identify patients who are likely to benefit from targeted care bundles, such as renal protection bundles for patients at high risk for postoperative AKI, and can raise suspicion for time-sensitive postoperative complications, such as sepsis. In the post-discharge phase, changes in metabolites over time can suggest disease recurrence, and surveillance strategies can be tailored to favor frequent, intense surveillance for high-risk patients and infrequent, judicious surveillance for low-risk patients.

Figure 1:

Diagnostic and prognostic information derived from metabolomic profiling has potential applications to personalized surgical decision-making from preoperative to post-discharge phases of care.

Metabolomics in Diagnosis

At present, most surgical metabolomics research focuses on diagnostic biomarkers (31 included studies). This section summarizes all available evidence identified by review search criteria that biomarkers have diagnostic value for cardiovascular disease, as illustrated in Figure 2, as well as inflammatory bowel disease and some cancers, as illustrated in Figure 3.

Figure 2:

Summary of preoperative metabolite changes suggesting the presence and nature of aortic and peripheral vascular disease. (AAA: abdominal aortic aneurysm, PAD: peripheral arterial disease, AD: aortic dissection, *Medium- and long-chain acylcarnitines)

Figure 3:

Summary of postoperative metabolite changes suggesting recurrence of malignancies. (PC: pancreatic cancer, CRC: colorectal cancer, LC: lung cancer, EGC: esophago-gastric cancer)

Metabolomic profiling has several important applications to cardiovascular diseases, including peripheral arterial disease (PAD), or atherosclerosis of the aortoiliac segment and lower extremity arteries. Early lifestyle modifications and medical treatments can delay or avoid the need for revascularization, but early diagnosis of PAD is difficult because symptoms often remain subclinical until atherosclerotic disease has become severe and flow limiting. Hernandez-Aguilera et al.¹¹ examined 201 men with PAD and compared their randomly timed serum metabolic profiles to 48 age-matched healthy controls. Glutamate and glutamine were increased in patients with PAD, suggesting decreased glutaminolysis. In addition, changes in amino acid metabolism (increased serine, valine, isoleucine, and leucine) and a disrupted TCA cycle (increased isocitrate, aconitate, a-ketoglutarate, succinyl-CoA, and malate, and decreased fumarate and succinate) were observed in the PAD cohort, suggesting that quantification of these metabolites may facilitate early diagnosis and assessment for progression of PAD. Rizza examined the ability of serum biomarkers to predict major adverse cardiac events—including the need for peripheral arterial bypass—among 67 elderly adults with mean age 85 years¹². Higher serum levels of alanine and medium- and long-chain acylcarnitines, suggesting mitochondrial dysfunction, were associated with an increased rate of major adverse cardiac events (HR 1.77 (95% CI 1.11–2.81, p=0.016) and HR 2.18 (95% CI 1.17 – 4.07, p=0.014))¹². Zagura et al.¹³ examined the relationship between metabolites and arterial stiffness—a primary determinant of cardiovascular risk—among 42 men with symptomatic PAD and 46 healthy controls, finding that pyruvate and oxidized low-density lipoprotein independently predicted arterial stiffness. These findings suggest that metabolomic profiling can identify patients with PAD and those who may require revascularization in the future, providing information to patients and providers regarding the utility of early lifestyle modifications and medical treatments.

Metabolomic profiling has also been applied to diagnosing aortic dissection (AD). Wang et al.¹⁴ compared metabolite profiles among patients with acute AD (n=11), chronic AD (n=11), and controls with coronary heart disease but no AD (n=11). These three cohorts had distinct amino acid profiles, indicating that metabolite profiles can identify AD as well as its chronicity. Zhou et al.¹⁵ examined the metabolite profiles of 35 patients with acute aortic dissection (20 Stanford Type A, 15 Stanford Type B) and compared them with 20 healthy controls, finding increased serum sphingomyelin and decreased lysophosphatidylcholine in both Type A and Type B patients compared with controls, and lower sphingosine, phytosphingosine, and ceramide among patients with Type A compared with Type B dissections. Therefore, differences in metabolomes can suggest not only the presence and chronicity of AD, but also its anatomic classification. Imaging is the standard approach to diagnosing both acute and chronic AD. Given the lack of pertinent clinical implementation trials, it is unclear whether real-time metabolomic profiling for AD is realistic in an acute clinical setting. Given the speed, efficacy, and anatomic detail of modern imaging modalities, it seems unlikely that biomarkers will replace imaging for diagnosing AD. However, metabolomic profiling may have utility when applied to chronic AD over time. Chronic AD is usually managed medically, but despite optimal medical therapy, patients often suffer aneurysmal degeneration requiring surgical intervention. Although no studies have examined biomarker changes during this progression, it may be possible to use metabolomics to identify patients at high risk for future aneurysmal degeneration.

Metabolite profiles also differ by the presence and size of abdominal aortic aneurysms (AAA). Qureshi et al.¹⁶ examined seven studies regarding metabolomics in AAA patients, three of which used biofluids pertinent to this review. Ruperez et al.¹⁷ compared serum amino acids across three cohorts: patients with small (3–5 cm) AAAs, large (>5 cm) AAAs, and healthy controls (n=11 per group). All fatty acids except laurate and several indicators of carbohydrate metabolism (e.g., acetoacetate, 3-hydroxybutyrate, and acetone) were significantly higher among AAA patients compared with controls, and aminomalonate was 800% higher among patients with large AAA. Similarly, Ciborowski et al.¹⁸ performed metabolomic profiling on 30 AAA patients (15 large, 15 small) and compared them with 10 healthy controls, demonstrating that guanidosuccinic acid was associated with large AAA. Finally, Moxon et al.¹⁹ compared 161 patients with AAA to 168 patients with PAD, finding that linoleic acid was associated with AAA. Together, these results suggest that serum biomarker evidence of metabolic and oxidative stress is associated with the presence and size of AAA.

Metabolomic profiling could be used to support surgical decision-making regarding the optimal timing of procedural intervention, which is hindered by variation in AAA expansion patterns. Current guidelines indicate that AAA diameter greater than 5.5 cm or growth rate more than 1 cm annually justifies elective repair²⁰. However, there may be utility in moving beyond diameter thresholds to identify biomarkers that portend aneurysm rupture, justifying elective repair. In addition, patients with small aneurysms currently require annual surveillance imaging, which increases cumulative radiation exposure and overall healthcare expenses. An algorithm using metabolite profiles to predict aneurysm growth could individualize AAA surveillance programs, increasing intervals between imaging studies for low-risk patients and decreasing intervals between imaging studies for high-risk patients.

Metabolomic profiling also offers several potential advantages in diagnosing and managing inflammatory bowel disease. A major challenge in inflammatory bowel disease (IBD) is differentiating between ulcerative colitis (UC) and Crohn’s disease (CD). UC patients are candidates for total proctocolectomy with ileal-pouch anal anastomosis, whereas CD is a contraindication to this procedure, emphasizing the importance of accurate diagnosis. By current standards of care, approximately 5% (range 1–20%) of IBD patients are difficult to classify as UC or CD²¹. To improve differentiation between CD and UC, Ahmed et al.²² examined fecal metabolomes among 117 patients with CD, 100 patients with UC, and 109 healthy controls, finding that heptanal, 1-octen-3-ol, 2-piperidinone and 6-methyl-2-heptanone were up-regulated in CD, while methanethiol, 3-methyl-phenol, short-chain fatty acids, and ester derivatives were less abundant in patients with active CD compared with inactive CD and healthy controls. These results were not recapitulated in a study by Dawiskiba et al.²³, which assessed for differences in urine and serum metabolomes between CD and UC patients, and found none. The negative findings in this study may be attributable to a small sample size rendering the study underpowered, but this is difficult to assess due to the exploratory nature of this work. Beyond the potential to differentiate between IBD subtypes, metabolite profiles may aid in determining whether surgical intervention is appropriate in patients with CD. Seventy-five percent of all CD patients undergo surgery during their lifetimes for bowel abscesses, perforations, fistulae, bleeding, or obstruction, often requiring intestinal resection, risking development of short gut syndrome²⁴. Therefore, it is important to avoid surgery in CD patients whenever possible. Fathi et al.²⁵ examined metabolomes of 26 CD patients, comparing them with 29 healthy controls, finding that serum valine and isoleucine were lower and higher, respectively, in patients with CD. Considered in isolation, these data have little value in surgical planning. However, it remains plausible that metabolite profile trends can identify CD patients who would benefit from elective surgery performed prior to the development of severe complications requiring urgent or emergent surgery.

Several solid tumors are associated with distinct metabolite profiles that have may have utility for screening and diagnostic purposes. Pancreatic cancer is the fourth most common cause of cancer mortality in North America, with 5.1% five-year survival ²⁶. Effective screening methods are lacking. As a result, less than 10% of all patients have resectable disease at the time of diagnosis²⁷. Metabolite profiles have potential to aid in pancreatic cancer screening.^{26, 28} Long et al.²⁹ reviewed 25 studies examining the ability of metabolomics to diagnose pancreatic cancer by differentiating among patients with pancreatic cancer, pancreatitis, non-pancreatic cancer, and healthy controls, and found that amino acid-related pathways (e.g., alanine, aspartate, and glutamate metabolism, glycine, threonine, and serine metabolism, and taurine and hypotaurine metabolism) yielded perfect or near-perfect discrimination and sensitivity in several studies. Data from Tumas et al.³⁰ suggest that plasma histidine levels also correspond to the stage of pancreatic cancer. Therefore, as technologies improve, metabolomic profiling has the potential to provide sensitive screening for pancreatic cancer, which could shift treatment paradigms toward early-stage disease with curative intent.

Colorectal cancer is the second most commonly diagnosed cancer in women and the third most common in men³¹. Zhang et al. ³² reviewed 23 studies regarding metabolomics in diagnosis, recurrence, and survival among colorectal cancer patients. Overall, histidine, methionine and tryptophan were decreased among colorectal cancer patients, and glutamic acid, proline, iso-glutamine, and putrescine were increased, but included studies also had some contradictory results, with cancer patients having glycine, alanine, and taurine levels that were increased in some studies and decreased in others. Differences in results among studies were likely attributable to differences in patient populations and methods for biofluid and tissue procurement and analysis. Farshidfar et al.³³ examined serum from patients with colorectal cancer, colonic adenomas, and healthy controls, finding that ornithine and serine positively correlated with colorectal cancer, and decenoylcarnitine negatively correlated with colorectal cancer, suggesting that these metabolite changes have potential to aid in screening and diagnosis of colon cancer in a manner that is less invasive than colonoscopy, the current standard of care.

Esophago-gastric cancer accounts for 15% of all cancer-related deaths worldwide^{34, 35}. Similar to pancreatic cancer, effective screening tests are lacking, and many patients present with unresectable disease. Abbassi-Ghadi et al.³⁵ analyzed 20 studies—12 gastric cancer studies, six esophageal cancer studies, and two studies including both diseases—to identify serum and urine biomarkers implicated in the development of esophago-gastric cancer. Patients with cancer had higher serum levels of glutamate, glucose, and lactic acid, consistent with fast-growing tumor cell behavior; pyruvic acid, fumarate, and citrate were present in higher levels in some cancer cohorts, and lower levels in others. Davis et al.³⁶ compared urine metabolite profiles among patients with esophageal cancer, Barrett’s esophagus, and healthy controls, finding that nine metabolites distinguished cancer patients; sucrose, 2-aminobutyrate, cis-aconitate, pyroglutamate, hypoxanthine, and fucose were increased in the cancer cohort, and pantothenate, urea, and methanol were decreased. Together, these findings suggest that metabolite profiles may have utility as esophago-gastric cancer screening tests, facilitating early diagnosis before disease progresses to unresectable stages.

Breast cancer is the most common cancer in women, accounting for approximately 500,000 deaths worldwide annually³⁴. Screening and diagnosis are primarily performed with imaging and biopsy, respectively; plasma and urine biomarkers also have potential for screening and diagnosis, including hormone receptor status. Dougan et al.³⁷ compared metabolomes among 45 women with breast cancer to healthy control metabolomes, identifying 24 metabolites that differentiated between cases from controls: 3-(cysteine-S-yl)acetaminophen and 4-acetylphenol sulfate (xenobiotic pathway) as well as cysteine s-sulfate (amino acid pathway) were increased among women with cancer, whereas indoleacetylglutamine, (amino acid pathway), 2-ethylphenylsulfate (xenobiotic pathway), and sphingosine (lipid pathway) were increased among controls. In addition, estrogen receptor-positive patients had higher serum laurylcarnitine levels than estrogen receptor-negative patients, and progesterone receptor-positive patients had lower serum asparagine than progesterone receptor-negative patients. Similarly, Fan et al.³⁸ found that estrogen receptor-positive patients had elevated alanine, aspartate, glutamate, and purine metabolism and decreased glycerolipid catabolism. These differences in hormone receptor status indicate that metabolite profiles differ not only by cancer stages, but also biologic sub-types. Urine metabolites may also have utility in diagnosing breast cancer. Cala et al.³⁹ analyzed urinary biomarkers among 31 women with breast cancer, finding that tricarboxylic acid cycle intermediates were decreased compared with healthy controls, consistent with anaerobic tumor cell biology. These findings suggest that metabolite profiles have potential as breast cancer screening and diagnostic tools, in concert with current standards of care.

Lung cancer is the leading cause of cancer mortality worldwide. Imaging studies can identify lung lesions that may represent malignancy, but often require invasive aspiration and biopsy sampling of suspicious-appearing tissue to confirm the diagnosis. Biofluid metabolomic profiling can perform similar tasks without radiation exposure or invasive procedures. Kumar et al.⁴⁰ found that plasma taurine, aspartic acid, and pyruvic acid were increased in lung cancer patients compared with healthy controls, and glutamine, aspartic acid and inosine were decreased compared with healthy controls, consistent with tumor cell biology⁴⁰. Yu et al.⁴¹ reviewed available literature and identified several early serum (benzaldehyde, urea, isoleucine, glycolic, phenylalanine, carnitine, propionylcarnitine, tyrosine, glycerol, malic acid, histidine, and valine) and urine (hippuric acid, hydroxy-isovaleric acid, creatinine, tryptophan, hippuric acid, carnitine, and threonine) markers for lung cancer, similarly identifying metabolite changes that are consistent with high cell turnover in lung cancer cells. Chen et al.⁴² examined serum metabolomes in 30 patients with lung cancer and found that sphingosine, which was lower in lung cancer patients, and GpAEA, which was higher in lung cancer patients, were sensitive and specific in distinguishing a disease state from healthy controls. Like other solid organ malignancies, lung cancer metabolic profiles are not yet sensitive enough to function as screening tools in isolation, and are not yet accurate enough to be used as diagnostic tools in isolation. However, as technologies improve, metabolomic profiling has potential to aid in screening and diagnosis of lung cancer.

Metabolomics in Disease Prognostication

While most surgical metabolomic profiling research focuses on disease detection and diagnosis, recent efforts have investigated the prognostic value of blood, urine, and fecal biomarkers in identifying patients at increased risk for postoperative complications, tracking disease recurrence, and monitoring solid organ transplant function.

Approximately 45% of all surgeries are associated with a postoperative complication⁴³. Some of the most common and morbid postoperative complications are sepsis, AKI, and respiratory insufficiency, each of which are germane to the cardiovascular, gastrointestinal, oncologic, and transplantation surgeries discussed in this review. Predicting complications begins in the preoperative phase, informing shared decisions among patients, caregivers, and surgeons regarding treatment options and likely outcomes, and identifying potentially beneficial preoperative risk reduction strategies (e.g., prehabilitation) to mitigate the risk of specific complications. Shah et al.⁴⁴ compared metabolomes of 2,023 patients before and after cardiac catheterization, identifying 8 metabolite changes predictive of post-procedural death or myocardial infarction. With advances in metabolomic profiling to enable efficient identification of metabolite changes between pre-, intra-, and postoperative phases of care, similar approaches could be used for patients being considered for high-risk surgery to predict clinical outcomes and identify patients who may benefit from close postoperative surveillance for early detection of complications.

Congenital heart malformations often require high-risk surgery. Davidson et al.⁴⁵ evaluated pre- and postoperative metabolites in 82 neonates and infants (age <120 days) undergoing cardiac surgery with cardiopulmonary bypass, using metabolite profiles to predict survival and intensive care unit length of stay. Aspartate levels were significantly lower in survivors, while methylnicotinamide levels were higher, suggesting neuroendocrine dysregulation. Aspartate and methylnicotinamide were also predictive of intensive care unit length of stay. Correia et al.⁴⁶ examined postoperative metabolic changes among 28 children (up to age 16 years, median age 7 months) undergoing heart surgery, finding different directionality of amino acid concentration postoperatively when compared with neonates and infants in the Davidson study, a phenomenon which cannot be explained by methods and results from these studies alone, and merits further study.

Metabolite profiles have also been used to predict postoperative complications among adult cardiac surgery patients. Kirov et al.⁴⁷ found no differences in plasma metabolomes for on-pump versus off-pump coronary artery bypass grafting, but found a significant inverse correlation between serum arginine at the end of surgery and postoperative vasopressor requirements. The authors noted that this finding could be explained by decreased inducible nitric oxide synthase activity, arginine catabolism, and vasodilation, resulting in higher arginine levels and decreased vasopressor requirements. Maltesen et al.⁴⁸ collected blood from 50 cardiac surgery patients at nine time-points during pre-, intra-, and post-operative phases of care, finding that longer ischemia time was associated with increased pyruvate, acetate, glycine, alanine, and glutamine, as well as decreased arginine, isoleucine, and 3-methylhistidine, suggesting that these profiles may have utility in predicting the magnitude of ischemia-reperfusion injury. Lee et al.⁴⁹ examined urinary metabolome changes following cardiac surgeries with (n=49) and without (n=51) cardioplegia solution, finding that cardioplegia was associated with upregulated histidine metabolism with subsequently increased glutamine and glutamate metabolism and enhanced vitamin B₆ metabolism. These metabolic shifts provide insight regarding mechanisms by which cardioplegia solutions affect cardiac output and dysrhythmias, and with further investigation, may have prognostic value as well.

Other studies have directly predicted outcomes following cardiac surgery. Buter et al.⁵⁰ found that low preoperative glutamine is associated with increased risk for infectious complications after elective cardiac surgery, identifying patients who may benefit from high postoperative suspicion and vigilance for sepsis. In a cohort of 478 patients referred for coronary artery bypass grafting, Shah et al.⁵¹ found that elevated serum levels of short-chain dicarboxylacylcarnities, ketone-related metabolites, and short-chain acylcarnitines were predictive of myocardial infarction, the need for revascularization, or death. These prognostic data can inform the shared decision-making process among patients, caregivers, and providers, and estimate appropriate levels of postoperative monitoring.

Metabolomics can also predict postoperative AKI. Zacharias et al.⁵² collected plasma from 85 patients 24 hours following cardiopulmonary bypass. Serum magnesium, lactate, and the glucuronide conjugate of Propofol were predictive of AKI—which occurred in 39% of the cohort—suggesting that anaerobic metabolism due to ischemia likely contributed to AKI, and that serum magnesium and the glucuronide conjugate of Propofol were elevated as a result of AKI⁵³. Elmariah et al.⁵² examined serum metabolomes among patients undergoing transcatheter aortic valve replacement, and found that elevated levels of 5-adenosylhomocysteine were predictive of both AKI and all-cause mortality, possibly because 5-adenosylhomocysteine served as a sensitive indicator of AKI, because DNA methylation changes secondary to 5-adenosylhomocysteine were directly toxic to kidneys, or both. Regardless of underlying pathophysiology, predicting postoperative AKI has the theoretical advantage of identifying patients who may benefit from targeted application of renal protection bundles. For example, patients at high risk for pre-renal AKI may benefit from judicious intravenous fluid resuscitation, even before overt symptoms of volume depletion are manifest clinically.

Metabolomic profiling can also monitor for disease recurrence and estimate the need for reintervention, as demonstrated for CD and breast cancer. Keshteli et al.⁵⁴ obtained postoperative fecal and urinary samples from 38 CD patients who underwent ileocolic resection, finding that higher levels of urinary levoglucosan, L-DOPA, and ethylmalonate and lower levels of urinary propylene glycol were associated with disease recurrence, likely representing dysregulation of inflammatory and autoimmune pathways associated with CD⁵⁴. Metabolomic profiling to predict CD recurrence could be used to optimize the timing of endoscopic surveillance. Following surgical resection for breast cancer, many patients suffer from relapse and require additional therapy. Standard surveillance focuses on imaging and clinical examination to monitor for disease recurrence; metabolomics can also predict recurrent disease. Tenori et al.⁵⁵ analyzed metabolomes of 95 women with metastatic disease and 80 women with early-stage breast cancer who had blood drawn for metabolomic profiling immediately following oncologic resection with curative intent. Immediately after resection of early-stage cancer, women who later developed metastatic disease had lower serum levels of histidine and higher serum levels of glucose, lactate, tyrosine, and lipids compared with women who did not develop recurrent disease, suggesting long-term prognostic value for these metabolites⁵⁵. Hart et al. ⁵⁶ examined 675 women with early-stage breast cancer and 125 with metastatic disease and found that several metabolites (choline, acetate, formate, lactate, glutamate, 3-hydroxybutyrate, phenylalanine, glycine, leucine, alanine, tyrosine, isoleucine, histidine, creatine, methionine, and proline) were increased in the serum of women with metastatic disease, supporting the hypothesis that tumor cell biology can be manifest as metabolite changes⁵⁶. Oakman et al.⁵⁷ analyzed serum metabolomes in a similar population, finding that metastatic disease was characterized by elevated serum levels of phenylalanine, glucose, proline, lysine, and N-acetylcysteine, and lower lipid levels. Asiago et al.⁵⁸ and Jobard et al.⁵⁹ each found that lower serum histidine was associated with recurrent breast cancer. Together, these findings suggest that metabolomic profiling has potential to perform prognostic and surveillance functions for women with breast cancer, informing personalized surveillance strategies to ensure early identification of recurrent disease and avoid unnecessary imaging studies and biopsies for women at low risk for recurrence.

Solid organ transplantation is the gold standard for treatment for end-stage heart, liver, and kidney disease. Following transplantation, patients are closely monitored for allograft dysfunction or rejection, which often require biopsy for definitive diagnosis, and influence decisions regarding immunosuppressive medication regiments. Metabolomic data have the potential to identify allograft dysfunction and rejection by non-invasive methods, avoiding allograft biopsies. Bassi et al.⁶⁰ examined 40 renal transplant patients with chronic allograft dysfunction, finding that serum trends in tryptophan, glutamine, and dimethylarginine as well as urine concentration of histidine, DOPA, dopamine, carnosine, SDMA, and ADMA, surrogates for decreasing immunomodulation, hypertension, micro-ischemic events, fibrosis, and cytotoxicity, were associated with increased estimated glomerular filtration rate (eGFR). Blydt-Hansen et al.⁶¹ performed metabolomic profiling on 396 children who underwent kidney transplantation, finding that children with antibody-mediated rejection (n=40) had higher urinary proline and citrulline and lower phosphatidylcholine, tetradecanoylcarnitine, and C10.2 than other children, suggesting that metabolites have the potential to phenotype rejection with urine samples rather than kidney biopsies. Similar to renal allograft surveillance, urine metabolomic profiling has been successful in identifying patients with acute heart transplant rejection, with elevated urinary excretion of nitrate and thromboxane A2, a pro-thrombotic vasoconstrictor, being associated with acute rejection^62,63. Among patients who have undergone liver transplantation, high levels of serum and urine urea and glutamine are associated with urea cycle abnormalities signifying graft dysfunction^64,65. These metabolome characteristics have the potential to mitigate diagnostic difficulties in distinguishing between rejection and dysfunction while avoiding allograft biopsy.

DISCUSSION

This scoping review assimilates and critically evaluates available evidence regarding blood, urine, and fecal metabolomic profiling in surgery and identifies opportunities to apply metabolome diagnostic and prognostic information to surgical decision-making throughout preoperative, intraoperative, postoperative, and post-discharge phases of care. Despite the fact that most evidence regarding metabolomics in surgery are from retrospective studies with small sample sizes using various combinations of analytic techniques and types of biofluids and tissues, the weight of evidence suggests potential for clinical utility as technologies improve.

In the preoperative phase of care, accurate, precise predictions of postoperative outcomes can inform the shared decision-making process among patients, caregivers, and clinicians regarding whether predicted outcomes match patient values and patient-centered outcomes, and which risk reduction strategies (i.e., prehabilitation) might improve postoperative outcomes. The perioperative phase often provides access to biofluids and tissues, offering opportunities to reanalyze metabolomes, revise preoperative predictions, and use this information to update the plan of care for time-sensitive decisions such as determining the optimal postoperative triage destination and frequency of vital sign monitoring. Postoperatively, personalized risk assessments can identify patients who may benefit from care bundles targeting specific complications (e.g., renal protection bundles for patients at high risk for AKI) and raise suspicion and vigilance for the development of time-sensitive complications (e.g., sepsis). During post-discharge surveillance, serial metabolomic profiling can suggest the likelihood of recurrent disease and differentiate between solid organ transplant dysfunction and rejection, identifying patients who may benefit from confirmatory imaging and biopsy or empiric changes in immunosuppression regiments. Most of these advantages remain theoretical. Clinical integration of accurate, efficient metabolomic profiling technologies and high-level evidence from prospective studies are required before metabolomic profiling can be recommended as an effective supplement or alternative to current standards of care.

Metabolic changes occurring secondary to surgical diseases, surgery itself, general anesthetics, and commonly prescribed perioperative medications serve as the foundation for the diagnostic and prognostic value of metabolomic profiling⁴⁹. These metabolic changes primarily concern catabolism, anabolism, inflammation, and the neuroendocrine stress response ⁶⁶. Prior to the application of metabolomic profiling as a diagnostic or prognostic tool, several criteria must be met. First, the optimal approach to biomarker selection should be standardized. There are approximately 7,000 known metabolites, and it is not cost-effective or feasible to test each one⁶⁷. There is no consensus regarding which metabolites are associated with specific diseases. For example, this review describes two studies using metabolomes to differentiate between large and small AAA, and each identified distinct metabolites^18,16. There were also several contradictions among studies investigating metabolomic profiling for patients with malignancies. This is partially attributable to methodologic differences among studies using GC-MS, LC-MS, and H1-NMR to quantify metabolites, and may also be partially attributable to differences in age, gender, race, and ethnic origin among small, homogenous study populations. In surgical populations, it is also important to account for nutritional status. Surgical stress induces a catabolic state through the activation of neuroendocrine and inflammatory pathways. Perioperative nutritional supplementation with amino acids and omega-3 fatty acids may mitigate the effects of postoperative catabolism and affect both baseline and postoperative metabolomes, and most studies do not account or adjust for nutritional status⁶⁸. Therefore, consensus regarding the optimal approach to biomarker selection, analysis, and clinical application will likely require multi-center validation of generalizable metabolomic diagnostic and prognostic studies performed among heterogeneous patient populations with standardized approaches to accounting or adjusting for nutritional status, with comparison to normal baseline levels from healthy controls.

Most literature concerning the role of metabolomics in surgery pertains to diagnostic applications. Metabolomic screening tools can theoretically promote early detection by non-invasive methods. This approach seems particularly useful for surgical diseases in which standard screening tools are lacking or ineffective and delayed diagnosis limits therapeutic options and portends worse clinical outcomes. Metabolomic profiling has the potential to non-invasively monitor for recurrent breast cancer, but, as stated above, there is contradictory evidence regarding metabolomes among pancreatic, lung, breast, esophago-gastric, and colorectal cancer patients. Despite these inconsistencies, it is apparent that higher serum glutamate and glucose were observed in both pancreatic and esophago-gastric cancer patients, and could potentially be a pathway to early diagnosis in these diseases for which effective screening tools are lacking³⁵.

Metabolomic profiling also has potentially useful applications in identifying patients at increased risk for postoperative complications. Most evidence on this topic comes from cardiovascular literature. Low serum arginine levels were observed in patients requiring postoperative vasopressor therapy as well as in patients with prolonged aortic cross-clamp time, identifying patients for whom ischemic-reperfusion injury may be manifest as ongoing postoperative vasopressor requirements^{47, 48}. Similar alterations in the arginine-proline pathway follow congenital heart surgery, evident by increased serum putrescine, a byproduct of arginine catabolism⁴⁵. Therefore, it seems plausible that serial measurement of arginine could play a role in guiding resuscitation efforts in the immediate postoperative period, along with traditional methods for optimizing volume status, inotropy, and vascular tone. As cardiac and vascular surgery shift toward minimally invasive operative approaches and endovascular procedures, it may be interesting to compare the postoperative metabolomic changes among patients managed with and without aortic cross-clamping, especially in comparing urine metabolomics, as blood flow to the kidneys is altered during aortic cross-clamping.

The diagnostic and prognostic value of the fecal metabolome requires further investigation. The fecal microbiome may affect the pathogenesis of several diseases, including aneurysmal disease, coronary artery disease, end-stage renal disease, and sepsis, but there have been no large-scale studies investigating fecal metabolites among surgery patients in these populations. Investigating the impact of fecal diversity on metabolite production could inform decisions regarding pre- and probiotic therapies.

The pathway toward clinical adoption of metabolomics can be understood by examining a the clinical adoption of proteomics, which has already begun the transition from the laboratory to clinical settings. This process has been facilitated by the NCI Proteomic Tumor Analysis Consortium, which compiled a fully integrated list of DNA, RNA, and protein abnormalities in individual tumors⁶⁹. Similar to metabolomics, proteomics may yield diagnostic and prognostic advantages for cardiovascular diseases. Specific serum and plasma proteins altered in conditions such as dilated cardiomyopathy and valvular disease have been identified, and these data are being used to ascertain the exact timing of the transition from adequate to inadequate myocardial perfusion ⁷⁰. Urinary proteomics are promising as well. An online database of 2,206 urinary compounds and 2,651 metabolites has been constructed by Bouatra et al.,⁷¹ facilitating individual and collaborative research efforts by standardizing nomenclature and classification schemes. Metabolomic research could benefit from similar advances.

Future Directions in Diagnostic Metabolomics

If metabolic profiling can achieve high sensitivity for detecting pancreatic, esophago-gastric, and other cancers, these malignancies could be identified earlier, when they remain resectable and curable. Similarly, expanding metabolomics research to promote early diagnosis of PAD, AAA, and chronic AD could help prevent complications associated with the natural progression of these diseases through early interventions.

In the diagnostic setting, metabolomic profiling can apply in two ways: generalized screening and precision medicine. Currently, there are no accurate blood biomarkers for pancreatic cancer. Carbohydrate antigen 19–9 (CA 19–9) is often measured to suggest the presence of pancreatic cancer and other intra-abdominal malignancies, but it is a non-specific marker for biliary disease and has a positive predictive value of only 0.5 to 0.9%⁷². Identifying changes in blood and urine metabolite profiles associated with the development of pancreatic cancer could apply to screening protocols. However, several metabolic pathways are upregulated in pancreatic cancer, and further investigation is necessary to determine the optimal screening approach. Esophago-gastric cancer is similar to pancreatic cancer in that there are currently no effective screening methods. The studies reviewed herein found that esophago-gastric cancer patients have elevated serum glucose and lactic acid, but the non-specific nature of these metabolites hinders their clinical utility. Riekeberg et al.² suggest that obtaining metabolite profiles at regular intervals is the best approach to screening and diagnosis, providing insight regarding the trajectory of metabolites over time and capturing small changes in metabolite profiles that may otherwise remain subclinical, especially in early stage disease. To do so, more research is necessary, as this would require characterizing metabolomes in both healthy controls and patients with established disease, as well as serial profiling of healthy individuals who develop disease.

Similarly, PAD is underdiagnosed due to limited screening. Some studies estimate a detection rate of only 20% within the general public, and 30% among high-risk populations⁷³. Imaging studies provide anatomic detail necessary for operative planning, and will likely remain the diagnostic test of choice in the foreseeable future. However, metabolomic profiling may have a role in determining the optimal timing and frequency of surveillance and diagnostic imaging. As previously mentioned, patients with chronic AD are at risk for aneurysmal degeneration. Biomarkers elevated in patients with AAA such as guanidosuccinic acid and aminomalonate can suggest aneurysmal degeneration of a patient with chronic AD, prompting further investigation with imaging studies, and potentially facilitating early treatment prior to rupture. However, testing of this hypothesis has not reported, and further investigation is necessary prior to clinical application and adoption.

Future Directions in Disease Prognostication

One major challenge in postoperative surgical management is early identification and management non-technical complications, including prolonged mechanical ventilation, renal dysfunction, and sepsis. Identifying patients at increased risk for these complications could raise suspicion and vigilance for them among clinicians, facilitating early identification and treatment. Perhaps more importantly, this approach could identify patients who benefit from targeted care bundles, such as renal protection bundles for patients at high risk for postoperative AKI. Although increased surveillance and diagnostic and therapeutic interventions are associated with increased costs, these costs may be offset by savings attributable to prevention or early treatment of complications and avoiding progression to organ dysfunction and critical illness. A recent analysis of nearly eight thousand ICU patients demonstrated that postoperative complications are a major driver of prolonged ICU stays, which result in higher healthcare costs, especially when resource-intensive interventions such as mechanical ventilation are performed⁷⁴. Dasta et al.⁷⁵ reported that the average cost of an ICU stay involving mechanical ventilation is $31,574, representing potential cost savings for cases in which severe postoperative complications such as respiratory insufficiency can be prevented. However, high-level evidence supporting the accuracy and generalizability of metabolomic profiling for prognostic purposes would be required prior to clinical adoption.

Currently, monitoring for recurrence of many surgically treated diseases involves serial imaging and invasive procedures such as endoscopies. As discussed above, serial imaging can be costly and has associated risks, including radiation exposure and contrast-induced nephropathy. Frequent endoscopy subjects patients to discomfort, time away from work, and risk for perforation. Metabolite profiles that suggest very low or very high probability of recurrence could be used to defer imaging and invasive tests for low-risk patients and promote early imaging and invasive tests for high-risk patients, personalizing surveillance schedules and increasing the pre-test probability and diagnostic yield of confirmatory tests. Similarly, metabolomics could monitor allograft function in the solid organ transplant population and alert clinicians to possible failure or rejection in a timely manner, reserving allograft biopsy for cases in which organ dysfunction and rejection remain difficult to distinguish.

Limitations

Given the scoping nature of this review, there is an element of selection bias in identifying articles for inclusion. Although most articles were identified by a computer program, additional sources were identified based on titles and abstracts alone, which may have skewed results toward studies in which significant metabolite changes were included in the abstract, introducing reporting bias into the results of this review. In addition, many studies included in this review are likely underpowered and focus on specific, homogenous populations, limiting their generalizability. Due to the exploratory nature of most metabolomic research, power analyses are rarely performed, and it is difficult to ascertain if studies were adequately powered in a post-hoc manner.

CONCLUSIONS

Metabolomic profiling is performed by NMR spectroscopy or mass spectrometry on biofluids and tissues such as blood, urine, and feces to quantify biomarkers (i.e. sugars, amino acids, and lipids) that can function as screening tools, suggest diagnoses, and predict outcomes such as postoperative complications and disease recurrence. Major barriers to clinical adoption include a lack of high-level evidence from prospective studies, heterogeneity in study design regarding tissue and biofluid procurement and analytical methods, and the absence of large, multicenter metabolome databases to facilitate systematic investigation of the efficacy, reproducibility, and generalizability of metabolomic profiling diagnoses and prognoses. These limitations could be addressed by consensus regarding standardization of study design and analytic approaches, but this consensus is currently lacking. Despite these limitations, as technologies improve and knowledge garnered from research accumulates, metabolomic profiling has potential to provide personalized diagnostic and prognostic information to support surgical decision-making from preoperative to post-discharge phases of care.

Supplementary Material

Supplemental Figure 1

Supplemental Figure 1: Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) flow diagram depicting the article selection process.

Supplemental Figure 2

Supplemental Figure 2: Automated and web-based literature search terms.

Supplemental Table 1

Supplemental Table 1: Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) checklist.