Abstract

   Rationale: Gastric cancer (GC) is preceded by a stepwise progression of
   precancerous gastric lesions. Distinguishing individuals with
   precancerous gastric lesions that have progression potential to GC is
   an important need. Perturbated lipid metabolism, particularly the
   dysregulation of de novo lipogenesis, is involved in gastric
   carcinogenesis. We conducted the first prospective lipidomics study
   exploring lipidomic signatures for the risk of gastric lesion
   progression and early GC.

   Methods: Our two-stage study of targeted lipidomics enrolled 400
   subjects from the National Upper Gastrointestinal Cancer Early
   Detection Program in China, including 200 subjects of GC and different
   gastric lesions in the discovery and validation stages. Of validation
   stage, 152 cases with gastric lesions were prospectively followed for
   the progression of gastric lesions for a median follow-up of 580 days
   (interquartile range 390-806 days). We examined the lipidomic
   signatures associated with the risk of advanced gastric lesions and
   their progression to GC. Our published tissue proteomic data were
   referred to further investigate highlighted lipids with their
   biologically related protein expression in gastric mucosa.

   Results: We identified 11 plasma lipids significantly inversely
   associated with the risk of gastric lesion progression and GC
   occurrence. These lipids were integrated as latent profiles to identify
   5 clusters of lipid expression that had distinct risk of gastric lesion
   progression. The latent profiles significantly improved the ability to
   predict the progression potential of gastric lesions (AUC: 0.82 vs
   0.68, Delong's P = 4.6×10^-4) and risk of early GC (AUC: 0.81 vs 0.55,
   P = 6.3×10^-5). Significant associations were found between highlighted
   lipids, their biologically correlated proteins and the risk of GC,
   supporting the role of the pathways involving monocarboxylic acid
   metabolism and lipid transport and catabolic process in GC.

   Conclusions: Our study revealed the lipidomic signatures associated
   with the risk of gastric lesion progression and GC occurrence,
   exhibiting translational implications for GC prevention.

   Keywords: Gastric cancer, Lipidomics, Precancerous gastric lesion,
   Biomarker

Introduction

   Gastric cancer (GC) is one major public health threat with high
   morbidity and mortality worldwide [51]^1. GC of the intestinal type
   predominates in high-risk geographic areas [52]^2, and its occurrence
   experiences multistep cascade progression of gastric lesions, which
   evolve from superficial gastritis (SG), chronic atrophic gastritis
   (CAG), intestinal metaplasia (IM), and low-grade intraepithelial
   neoplasia (LGIN) to high-grade intraepithelial neoplasia (HGIN) and
   invasive GC [53]^3^,[54]^4. Studies of TCGA and other data have
   examined the molecular subtypes of GC, aiming to provide a roadmap for
   patient stratification and targeted therapies [55]^5^,[56]^6. However,
   while most GCs are diagnosed at locally advanced or advanced stages
   with unfavorable prognosis[57]^7, efforts are warranted to identify
   populations at particularly high-risk for progression of gastric
   lesions and development of GC, essential for improving the primary
   prevention and early detection of GC. Efficient biomarkers are
   therefore highly needed.

   Lipids play essential roles in cellular functions related to the
   carcinogenesis process [58]^8. Perturbated lipid metabolism, including
   increased lipid uptake, endogenous de novo fatty acid synthesis, fatty
   acid oxidation, and cholesterol accumulation, has been reported to
   promote tumor growth and progression [59]^9^-[60]^11. In addition,
   lipid content of phospholipids could compromise membrane fluidity and
   signal transduction which may in turn affect GC tumorigenesis and
   progression [61]^12^,[62]^13. In our recent study based on untargeted
   metabolomics covering carbohydrates, amino acids, nucleotides, polar
   lipids, and other metabolites; six lipids, including α-linolenic acid,
   linoleic acid, palmitic acid, arachidonic acid, sn-1
   lysophosphatidylcholine (LysoPC)18:3, and sn-2 LysoPC20:3 stood out to
   have the most robust associations with the risk of early GC, with the
   first three also significantly associated with the risk of gastric
   lesion progression in a prospective analysis [63]^14. These highlight
   the potential importance of the overall lipidomic profile underlying GC
   carcinogenesis [64]^15. However, previous metabolomics studies of GC
   were restricted to water-soluble compounds and volatile metabolites
   [65]^16, which lacked coverage and in-depth investigation for a wide
   range of lipids with potentially pivotal functions, thus leaving a
   knowledge gap on the full spectrum of lipidomic signatures associated
   with the development of GC.

   Based on a total of 400 subjects from Linqu county, a well-recognized
   high-risk area in eastern China [66]^4^,[67]^17, we conducted the first
   comprehensive lipidomics study for GC and delineated a plasma
   lipidomics profile for a sequence of gastric lesions and GC in two
   stages. We took advantage of our prospectively followed participants
   and longitudinally investigated the lipidomic signatures underlying the
   progression of gastric lesions and development of GC.

Methods

Study participants

   Our study involved a total of 400 subjects in two stages from Linqu
   County, Shandong Province of China, an established high-risk area for
   GC, where most GCs are of the intestinal type [68]^4^,[69]^17. All
   subjects were enrolled from those attending the National Upper
   Gastrointestinal Cancer Early Detection (UGCED) Program for rural
   areas, in which residents aged 40 to 69 years received upper
   gastroendoscopy examinations free of charge. Individuals with
   cardiovascular, liver and spleen disorder and other major chronic
   diseases are ineligible for gastroendoscopy and were therefore excluded
   from the program. Gastroendoscopy was performed by two experienced
   gastroenterologists using video endoscopes (Olympus). For each
   individual, biopsies were taken at five standardized sites and other
   sites with suspicious lesion detected by endoscopy, if any [70]^18.
   Formalin-fixed, paraffin-embedded tissue samples for biopsy were
   reviewed blindly by two pathologists. Each subject was given a global
   diagnosis of normal, SG, CAG, IM, LGIN, HGIN, or invasive GC, defined
   as the most severe gastric histology among all biopsies, following the
   criteria of the Updated Sydney System [71]^18 and the Chinese
   Association of Gastric Cancer [72]^19. Subjects were surveyed using
   standard questionnaires and had a 5ml blood sample collected following
   standardized collection process. H.pylori infection status was
   determined by enzyme-linked immunosorbent assay for plasma IgG [73]^20.

   The study consisted of two independent stages involving a total of 400
   subjects. The discovery set included a total of 200 subjects with
   gastric lesions of different stages (n = 169) and GC (n = 31, including
   22 HGINs and 9 invasive GCs) diagnosed in 2018. The validation set
   further independently enrolled 200 subjects, including 48 cases of GC
   and 152 cases with different gastric lesions diagnosed in 2017. We did
   not include any subjects with normal gastric mucosa as few of the adult
   residents had completely normal histology [74]^17^,[75]^19. We
   prospectively followed the subjects of gastric lesions in the
   validation stage (n = 152, “prospective cohort”) until May 31, 2021,
   for a median follow-up of 580 days (interquartile range 390 to 806
   days), with endoscopic examinations conducted at the endpoint for each
   individual. Among them, we had a multi-time point longitudinal
   sub-cohort of 76 participants who undertook further gastroendoscopy
   examinations in the middle of follow-up and thus had three or more
   measurement of gastric lesions during the follow-up. The progression of
   gastric lesions during the follow-up for the prospective cohort, or
   during a time window for the multi-time point longitudinal sub-cohort
   was assessed based on the global diagnosis of gastric lesions, defined
   as the most severe gastric histology among all biopsies (SG, CAG, IM,
   LGIN, HGIN or invasive GC). Subjects were considered to have
   progression of gastric lesions, if the severity of gastric lesion at
   follow-up endpoint is higher than that at baseline. Details of the
   participants in each cohort are presented in Figure [76]1 and Table S1.

Figure 1.

   [77]Figure 1
   [78]Open in a new tab

   General workflow of the study. Targeted lipidomics analysis involved a
   total of 200 subjects in two stages respectively. In the validation
   stage, 152 non-GC subjects were prospectively followed for the
   progression of gastric lesions (“prospective follow-up cohort”). For 11
   validated lipids significantly associated with risk of gastric lesion
   progression and GC occurrence, latent profiles were extracted using
   VAEN, representing the refined molecular pattern of lipids. Latent
   profiles of lipids were used to define lipidomic-based clusters of the
   prospective cohort subjects and the time-varying trajectories of
   gastric lesion progression were delineated by the clusters. XGBoost
   models were constructed to predict the risk of gastric lesion
   progression and GC occurrence. CAG, chronic atrophic gastritis; FDR:
   false discovery rate; GC, gastric cancer; HGIN, high-grade
   intraepithelial neoplasia; IM, intestinal metaplasia; LGIN, low-grade
   intraepithelial neoplasia; ROC, receiver operating characteristic; SG,
   superficial gastritis; VAEN, variational auto-encoder followed by the
   elastic net regression model; VIP, variable importance in projection;
   XGBoost, extreme gradient boosting.

   The study was approved by the Institutional Review Board of Peking
   University Cancer Hospital. Informed consent was waived as study
   subjects were selected within the framework of the National UGCED
   Program.

Targeted lipidomics profiling

   Targeted lipidomics profiling was performed on plasma samples using
   ultra high-performance liquid chromatography-mass spectrometry (LC-MS)
   [79]^21. Methods on sample preparation and LC-MS assays are detailed in
   the Supplementary Methods. Quality control (QC) samples were prepared
   using mixed plasma samples, with 1 QC sample inserted between every 20
   tested samples. A total of 10 and 11 QC samples were inserted during
   the lipidomics profiling for plasma samples in the discovery and
   validation stage, respectively. Ionization signals were monitored in QC
   samples based on the intensities of internal standards for individual
   lipid classes to ensure no significant drop in intensity (within 20%)
   and no drift in retention time (within 0.05 min) throughout the run.
   Lipids were identified based on structure-specific multiple reaction
   monitoring (MRMs), which comprise MRMs specific to both head groups
   distinct to individual lipid classes and fatty acyl compositions, as
   well as correct retention times by comparing to authentic lipid
   reference compounds from human lipid ID inventory constructed in-house.
   Lipid levels were expressed in moles per L (mol/L) of plasma for
   statistical analyses.

Bioinformatics and statistical analysis

   We conducted bioinformatics and statistical analyses for the lipid
   signatures associated with the risk of GC compared with the
   well-recognized mild gastric lesion group (SG/CAG) or advanced gastric
   lesion group (IM/LGIN) as references, based on the discovery and