Proteomic mucin profiling for the identification of cystic precursors of pancreatic cancer

Abstract

Background: Pancreatic cystic lesions (PCLs) are increasingly frequent radiological incidentalomas, with a considerable proportion representing precursors of pancreatic cancer. Better diagnostic tools are required for patients to benefit from this development.

Methods: To evaluate whether cyst fluid mucin expression could predict malignant potential and/or transformation in PCLs, a proteomic method was devised and prospectively evaluated in consecutive patients referred to our tertiary center for endoscopic ultrasound-guided aspiration of cystic lesions from May 2007 through November 2008 (discovery cohort) and from December 2008 through October 2012 (validation cohort). Cytology and cyst fluid carcinoembryonic antigen (CEA; premalignancy > 192 ng/mL, malignancy > 1000 ng/mL) were routinely analyzed, and samples were further processed as follows: one-dimensional gel electrophoresis, excision of high-mass areas, tryptic digestion and nano-liquid chromatography-tandem mass spectrometry, with peptide identification by Mascot software and an in-house mucin database. All diagnostic evaluations were blinded to proteomics results. Histology was required to confirm the presence/absence of malignant transformation. All statistical tests were two-sided.

Results: Proteomic mucin profiling proved statistically significantly more accurate (97.5%; 95% confidence interval [CI] = 90.3% to 99.6%) than cytology (71.4%; 95% CI = 59.8% to 80.9%; P < .001) and cyst fluid CEA (78.0%; 95% CI = 65.0% to 87.3%; P < .001) in identifying the 37 (out of 79; 46.8%) lesions with malignant potential (ie, premalignant or malignant tumors). The accuracy of proteomics was nearly identical (96.6% vs 98.0%) between the discovery (n = 29) and validation (n = 50) cohorts. Furthermore, mucin profiling predicted malignant transformation, present in 16 out of 29 (discovery cohort: 9, validation cohort: 20) lesions with available histology, with 89.7% accuracy (95% CI = 71.5% to 97.3%) (for the validation cohort only: 95.0%; 95% CI = 73.1% to 99.7%). This markedly exceeded corresponding results for cytology (51.7%; 95% CI = 32.9% to 70.1%; P = .003) and CEA (57.1%; 95% CI = 34.4% to 77.4%; P = .02).

Conclusions: Proteomic cyst fluid mucin profiling robustly discriminates benign, premalignant, and malignant PCLs. Consequently, it may improve pancreatic cancer prevention and reduce the morbidity burden of unwarranted pancreatic surgery.

Figure 1.

Flow chart of patients in the study. Diagnoses provided in italics are tentative. GIST, gastrointestinal stromal tumor; IPMN, intraductal papillary mucinous neoplasm.

Figure 2.

Distribution of diagnoses for the discovery cohort (n = 29) (A) and validation cohort (n = 50) (B). IPMN, intraductal papillary mucinous neoplasm; MCN, mucinous cystic neoplasm.

Figure 3.

Identification of pancreatic cystic lesions with malignant potential: performance of proteomics, cytology, and cyst fluid CEA (192ng/mL). The vertically oriented text on the columns representing cytology and CEA refers to two-sided P values for the comparison with mucin profiling (proteomics; Fisher exact test). Error bars illustrate the 95% confidence interval (Wilson score method with continuity correction). P values < .005 (statistical significance threshold after Bonferroni correction) are in bold text. CEA, carcinoembryonic antigen; NPV, negative predictive value; PPV, positive predictive value.

Figure 4.

Step-wise approach to the assessment of pancreatic cystic lesions with regard to malignant potential. Columns represent sensitivity values for the identification of pre-/malignant lesions for MUC5AC (left), MUC5AC+MUC2 (middle), and the optimal combination of MUC5AC+MUC2+MUC1 (right). Results are from the validation cohort. Text within columns list the five most abundant unique peptides observed for each mucin.

Figure 5.

Detecting manifestly malignant pancreatic cystic lesions: performance of MUC1, cytology, cyst fluid CEA (1000ng/mL), and endoscopic ultrasound (EUS) morphology. The rightmost (light blue) columns refer to a combination of standard analyses, with at least one positive result considered indicative of malignancy. Vertical text in columns representing traditional methods show two-sided P values for the comparison with MUC1 expression (Fisher exact test). P values <.0025 (statistical significance threshold after Bonferroni correction) are in bold text. For the primary comparison of the accuracy of MUC1 expression vs conventional methods, the Holm–Bonferroni correction was separately applied (statistically significant results shown in italics). CEA, carcinoembryonic antigen; NPV, negative predictive value; PPV, positive predictive value.