Towards precision medicine strategies using plasma proteomic profiling for suspected gallbladder cancer: A pilot study

Abstract

Background & aims: Currently, preoperative diagnostic methods that can distinguish cancer from benign disease of the gallbladder are insufficient, and several surgical resections can be avoided if the pathology is known prior to surgery. This study aimed to assess whether preoperative plasma proteins can distinguish gallbladder cancer (GBC) from cholecystitis, with the main goal of identifying proteins for multivariate description of the postoperative diagnosis, before surgery.

Methods: Samples from 82 individuals with suspected GBC who underwent bisegmentectomy and lymphadenectomy at Karolinska University Hospital between 2009 and 2020 were included in this retrospective, observational, single-center study. Preoperative plasma samples were analyzed using a 7,500 proteomics panel from SomaScan®. High-dimensional statistical methods including machine learning regularization, were used to analyze the data.

Results: In our study, we identified and characterized a panel of 651 proteins that exhibited differential expression between GBC and cholecystitis. Through multivariate analysis, we demonstrated that circulating proteomics data provide valuable insights for diagnosing GBC before surgical intervention. Notably, we identified a subset of eight plasma proteins (PAHX, CD8A, HRG, CRIS2, Dynactin subunit 2, AT2A3, CSTN2, and DEPP) that effectively differentiated GBC from cholecystitis with a diagnostic accuracy of 94% when validated on a test set. These findings hold potential for clinical validation and could significantly aid in preoperative decision-making when GBC is suspected.

Conclusions: Our findings demonstrate that the preoperative assessment of plasma proteins can accurately differentiate cholecystitis from malignancy, supporting the potential development of a noninvasive test to assist preoperative decision-making when GBC is suspected.

Impact and implications: This study highlights the potential of plasma proteomic profiling to significantly improve the preoperative diagnostic accuracy of gallbladder cancer vs. cholecystitis. Using machine learning models, we identified biologically relevant plasma proteins associated with the diagnosis of gall bladder cancer. A noninvasive preoperative test based on selected plasma proteins could potentially enhance clinical decision-making, reduce unnecessary surgeries, and mitigate the associated risks for patients with suspected GBC, marking a step forward in precision medicine.

Keywords: Biomarkers; Cholecystitis; Machine learning; Tumor.

Graphical abstract

Fig. 1

Demonstrating the preoperative diagnostic challenge in gallbladder cancer (GBC). (A) Imaging of a patient with GBC with primary suspicion of cholecystitis on preoperative investigation. (B) Imaging of a patient with cholecystitis with primary suspicion of malignant disease on preoperative investigation.

Fig. 2

Flow chart of the study design and methods used. GBC, gallbladder cancer; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Fig. 3

A set of 651 proteins were associated with diagnosis. (A). Volcano plot of all the proteins (in blue) and those significantly (p <0.05) linked to the diagnosis (in pink), in t test. (B) Venn diagram of the overlap between results from elastic net (n = 577), and t test at p <0.05 (n = 267). (C) Principal component analysis (PCA) of the whole dataset (7,500 proteins) with no apparent data structure. (D) PCA using the 651 diagnosis-associated proteins showed separation of GBC and cholecystitis. (E) Unsupervised hierarchical clustering of diagnosis-associated proteins largely distinguished GBC and cholecystitis, yet in the central region these groups overlapped (11% of the patients). Dim, dimension; GBC, gallbladder cancer; PC, principal component.

Fig. 4

Characteristics of the most descriptive proteins of postoperative diagnosis. (A) The variance of the protein levels in patients with cholecystitis and GBC. (B) Direction and contribution of each protein to the data variance, assessed by principal component analysis (PCA). (C) Heatmap showing the difference of expression of the proteins between the two groups. (D) Data variance and group differences are well described by the proteins, visualized by PCA. (E) Low correlation between the proteins confirming non-redundant information in the ML model. (F) Thirteen proteins could separate the patients into two clusters (k-mean unsupervised clustering) that largely correspond to the cholecystitis and GBC groups (89% of the patients were rightfully assigned). All nine wrongfully assigned patients were in the middle area. PCA, principal component analysis.

Fig. 5

Enrichment analysis of the proteins linked to diagnosis. (A) Enriched KEGG pathways in the diagnosis-associated proteins (651) where cytokine receptors are the most significant. (B) Gene Ontology analysis of the molecular functions, the cellular components, and the biological processes of the diagnosis-associated proteins. (C) Enriched KEGG brites in diagnosis-associated proteins. (D) Transcription factor families enriched in the protein list. (E) Cancer pathways in KEGG database with dysregulated proteins (from this study) highlighted in red. FDR, false discovery rate; KEGG, Kyoto Encyclopedia of Genes and Genomes. N. of genes, number of genes. p values are calculated based on Fisher's exact test.