In this study we identify serum peptide peaks as potential markers of BTC utilising a high-throughput C18 bead-based extraction method linked to MALDI-TOF MS profiling and LC-MS/MS for subsequent peak identification. A predictive model using an SVM consisting of nine peaks was able to distinguish BTC from healthy controls in a set of independent validation samples with a high degree of accuracy. Several of these same peptides could also distinguish BTC from benign cases of biliary disease albeit with lower accuracy, though none were discriminatory when using a smaller subset of 10 samples from patients diagnosed with PSC.
Serum profiling using proteomics has developed significantly from its early years when it was claimed that surface-enhanced laser desorption ionisation time-of-flight mass spectrometry (SELDI-TOF MS) profiling could reveal discriminatory peaks of high diagnostic accuracy for various cancers
[18–22]. Following the initial excitement, scepticism about the methodology mounted with reports of experimental bias within datasets and evidence that variations in sample handling gives rise to “differential” peptides previously reported as cancer markers
[23–27]. Our study was designed prior to sample collection, and hence all samples underwent the same process of prospective collection and were handled using a robust SOP. These aspects are crucial to avoid variability in profiling particularly as the peptides of interest are thought to be generated from products of the clotting cascade ex vivo.
Our main aim was to identify low molecular weight polypeptides in serum for detecting BTC using healthy volunteers and benign cases of biliary disease as control groups. This work was based on the hypothesis that cancer-specific exoproteases in serum act to digest products of the clotting cascade ex vivo, generating peptide fragments which may discriminate cancer from healthy samples
[13, 28, 29]. As such, serum, rather than plasma was chosen for this study. Analysis of spectral profiles from 92 serum samples gave eight significant peaks (P < 0.001) with a >2.0-fold change in peak area between BTC patients and healthy subjects. Two of these peaks (m/z 5805.0 and m/z 2903.3) had higher peak areas in BTC versus healthy, whilst the other six peaks were higher in the healthy samples. The best model generated from the discovery data using several of these peaks gave 100% sensitivity, 93.8% specificity, 92.9% PPV and 100% NPV when tested on the independent validation dataset, showing the potential of the test for accurate diagnosis of BTC.
Biliary strictures can arise from several non-malignant conditions which can mimic BTC, such as AIP/IAC, PSC and chronic pancreatitis. There were seven significant (and overlapping) peaks discriminating samples from patients with BTC versus benign biliary disease and a model was generated from the discovery data with respectable diagnostic accuracy (79.5% sensitivity, 83.9% specificity), These peaks, particularly m/z 5805.0, may therefore be of utility in differentiating BTC from benign disease. Despite this, the profiling failed to accurately differentiate patients with BTC versus the small subset of patients with PSC – a risk factor for BTC
. This may be due to the low number of samples used or the fact that the identified peaks may be markers of inflammation of the biliary epithelium. Indeed, the median level of C-reactive-protein (CRP) was significantly higher in the BTC patients than in the PSC cohort (Table
1). Whilst our preliminary findings would argue against this, further independent validation using greater numbers of samples from patients with PSC and other benign conditions will be necessary to determine the robustness of this model for the differential diagnosis of BTC. It is noteworthy that other cancer biomarker discovery studies using similar methodologies have not included benign inflammatory groups
[13, 31] and it is now becoming evident that this is absolutely critical. One recent study in pancreatobiliary disease has addressed this issue, with the study identifying a potential marker of malignancy in bile (NGAL) that was independent of markers of biliary obstruction and inflammation
. As in the present study, further validation of this marker is required.
An ion of mass [726.214]8+ was identified as the 576–628 fragment of isoform 1 of fibrinogen alpha chain by LC-MS/MS with a calculated average mass of 5805.09 Da and we matched this to the MALDI-TOF peak at m/z 5805.0. This peak may be generated by the exopeptidase-mediated loss of valine from the abundant peak at m/z 5903.9, which has been identified as a large fragment of fibrinogen alpha
. It may thus represent a surrogate marker of a tumour-derived exoprotease that is present at higher levels in the blood of BTC patients. The lower intensity peak at m/z 2903.3 behaved similarly to m/z 5805.0, but could not be identified. Its mass, isotopic pattern and expression behaviour suggest it to be a doubly-charged form of m/z 5805.0. As such, both peaks cannot be used as independent discriminating features, although we note that only the peak at m/z 2903.3 was selected by the SVM model used for validation. The other discriminatory peaks were also identified as fragments of fibrinogen alpha/fibrinopeptide A and high molecular weight kininogen, which have been reported as surrogate markers of different cancer types
Exoproteases form a heterogeneous group of enzymes that play a role in the regulation of biologically active peptides. Examples such as leucine aminopeptidase, aminopeptidase A, aminopeptidase N, carboxypeptidase N and the kininase I family of carboxypeptidases are involved in the production of angiotensin, bradykinin and vasopressin
, whilst carboxypeptidase B2 is involved in the down-regulation of fibrinolysis
. Through such activities, they may also contribute to tumour progression and invasiveness. In particular, several studies have reported the elevated expression of aminopeptidase N/CD13 in various cancers
[35–38] and it is believed to play a role in angiogenesis
. We speculate that the peptide fragments detected in our study are likely to be generated by such exopeptidase activities and thus serve as surrogate markers of the exoproteases themselves. The identity of these proteases was not the focus of this study, but their future interrogation in BTC may shed further light on its pathophysiology and lead to the identification of tumour-specific biomarkers and possible targets for therapy.
Various fibrinogen alpha chain and fibrinopeptide A fragments have been detected in hepatocellular, ovarian, urothelial and gastric cancers
[13, 40–43], although the m/z 5805.0 fragment has not been identified previously. These cleavage products may therefore be indicative of an underlying malignancy rather than be specific for BTC. However, we do note that our group could not derive a peptide signature to accurately differentiate ovarian cancer from benign ovarian disease or healthy controls using serum peptides detected on the same profiling platform
. This suggests that the signature identified herein, displays some specificity for BTC.
There are several limitations to the present study. The SOP adopted for this study used a clotting time of 60 minutes and therefore may be difficult to translate into the clinical laboratory where time and staff limitations are restrictive. In addition, the extraction and MALDI-TOF MS profiling method used here requires considerable expertise for operation and determines only relative peptide quantities. Given the nature of the peptides, the generation of fragment-specific antibodies for use in immune-based assays may be problematic and so MS-based targeted assays using synthetic, stable isotope-labelled peptide standards for accurate quantitation would be a more attractive way to validate our findings. Finally, the validation would require larger sample numbers, particularly those from cases of benign biliary disease.