Mass spectrometry protein expression profiles in colorectal cancer tissue associated with clinico-pathological features of disease

Colorectal cancer (CRC) is the second commonest malignancy and has a five-year survival rate of approximately 50% [1, 2]. The majority of patients, particularly with early stage disease (Dukes' A, Stage I), are treated with surgery [3]. For more advanced disease (Dukes' C and D, Stage III or IV) surgery combined with adjuvant chemotherapy has proven survival benefits [4–6]. However, the disease outcome is very variable and prognosis and prediction of treatment response based on conventional disease staging criteria is not reliable [6, 7]. There has therefore been considerable interest in the development of more robust prognostic and predictive disease markers for patient stratification with the ultimate aim of tailoring treatment to the individual patient [8, 9].

Markers based on circulating carcinoembryonic antigen (CEA) levels and various tumour-associated gene mutations including microsatellite instability (MSI), loss of heterozygosity of 18q, deleted in colorectal cancer (DCC), mutations in KRAS, BRAF and PIK3CA genes have all been shown to be of some prognostic or predictive value (reviewed in [8, 10]). In particular, the mutational status of KRAS, BRAF and PIK3CA genes has recently been proposed as a reliable marker for predicting responders to new targeted agents for the epidermal growth factor receptor (EGFR) [11, 12]. In addition, gene expression profiling studies of both mRNA [13] and microRNA [14] have revealed tumour-associated gene expression signatures that form the basis for a molecular classification of disease sub-types that define disease course and treatment response (reviewed in [8]). These studies on gene mutations and RNA expression have been paralleled by analysis of the tumour cell proteome, most commonly employing the technique of two-dimensional difference gel electrophoresis (2D-DIGE) to identify proteins that are differentially expressed in tumour verses normal mucosa tissue (reviewed in [15]). An expanding list of candidate prognostic markers have emerged from these studies including for example, cathepsin D, S100A4 and APAF-1 [15].

As an alternative to 2D-DIGE, studies of other tumour types have also employed the technique of direct protein expression profiling of tumour/normal tissue by surface enhanced laser desorption ionisation time-of-flight mass spectrometry (SELDI-TOF) or by matrix-assisted laser desorption ionisation time of-flight-mass spectrometry (MALDI-TOF) mass spectrometry [16, 17]. This approach, which is most commonly associated with the development of serum-based diagnostic markers, offers a number of advantages over 2D-DIGE. Although the technique yields no information on the actual identities of proteins, the reproducible spectral profiles that are relatively simple to generate in high throughput studies allow robust classification models of different proteome populations to be built. For example, studies of lung [18], breast [19], head and neck cancer [20] have all shown that the spectral profiles of tumour and normal tissue can be accurately discriminated and in some cases sub-classified by direct protein profiling using SELDI/MALDI-TOF mass spectrometry. Only one previous study has reported on the detection of differences between normal mucosa, adenoma and colorectal carcinoma by using SELDI-TOF MS [21].

In the present study, we have evaluated the potential value of protein expression profiling of CRC tissue by MALDI-TOF mass spectrometry. In addition to comparing tumour with adjacent normal mucosa, we have investigated whether spectral profiles of tumour tissue can be used to classify various clinico-pathological features of disease. Since previous 2D-DIGE studies have reported abnormalities of protein expression profiles in tumour-adjacent normal tissue [22], we have also extended this analysis to normal mucosa tissue.

Although previous studies employing 2D-DIGE analysis of CRC tissues have documented a number of proteins that are either up- or down-regulated in tumour verses normal mucosa [15], the extent to which protein expression profile differences can be detected by direct MALDI-TOF analysis in CRC was not previously known. Analysis of complex protein mixtures by MALDI-TOF MS is inherently limited by the resolution afforded by this type of instrument. Also, only a minor fraction of protein species are efficiently ionisable and therefore detectable. However, our results show that, in common with similar studies in some other solid tumour types [18–20], MALDI-TOF MS readily detects a sizable fraction of protein marker peaks whose expression level is significantly different between tumour and normal mucosa. By using an optimised kNN training model, the classification of tumour and normal tissue was correctly predicted with 100% sensitivity and specificity (95% confidence interval: 0.679-0.992) in an independent test dataset. This performance compares favourably with other studies, for example in head and neck squamous cell carcinoma, in which supervised prediction using SELDI-TOF spectral data correctly classified healthy mucosa and tumour tissue with an accuracy of 94.5% and 92.9% respectively [20].

In further evaluating the potential value of spectra generated from tumour tissue for classifying various clinic-pathological characteristics of disease, we observed low ROC errors with the kNN predictive models for differentiation (0.171) and disease recurrence (0.105). Since histological differentiation stage is a characteristic that is intrinsic to the tumour tissue (and would most closely reflect the actual tumour cell proteome), the ability of the spectra to discriminate well/moderately differentiated from poorly differentiated histologies is perhaps unsurprising. The good performance of the predictive model for disease recurrence is consistent with data from several microarray expression profiling studies that have clearly demonstrated associations between patterns of tumour-associated gene expression and prognosis/treatment response [8, 13, 14]. However, given that in our study, only six patients had succumbed to recurrent disease at the time of data analysis (median follow-up time for recurrent disease patients: 33 months; median follow-up time for disease-free patient: 27 months), our results should be interpreted with caution. It is also important to emphasise that because of the relatively small number of tumour specimens, rigorous validation of correlations with disease recurrence and histological differentiation stage in an independent 'test' datsaset was not possible in our study.

Several lines of evidence indicate that the normal mucosa from surgically resected CRC tumour specimens display abnormalities in gene and protein expression. These abnormalities have been attributed to precancerous 'field effect' changes in tumour-adjacent mucosa and have been reported to affect protein expression [22], CpG island gene methylation [31] and gene microarray expression profiles [32]. Indeed one study has reported that gene expression profiling of non-neoplastic mucosa may predict clinical outcome of CRC patients [32]. These findings are reminiscent of reports from studies of other solid tumour types, most strikingly in hepatocellular carcimoma in which gene expression patterns of non-neoplastic liver tissue were predictive of patient survival, whereas tumour tissue gene expression signatures were of no prognostic value [33]. It was therefore of interest in our study to determine whether the protein expression profiles of normal mucosa could be used to classify any clinico-patholgical characteristics. Although we found no evidence for predictive value for disease relapse (ROC error, 0.519), the kNN model of normal mucosa spectra for lymph node involvement did give a low ROC error (0.212); the corresponding kNN model for tumour spectra did not show predictive value (0.391). One plausible scenario to explain the predictive value of normal mucosa spectra for lymph node involvement is that paracrine/inflammatory mechanisms, involving proximal affected lymph nodes, may induce changes to the microenvironment of tumour-adjacent mucosa.

As an essential pre-requisite for marker validation, it would be highly desirable in future studies to determine the identities of candidate marker peaks in tumour tissue that discriminate different histological differentiation stages and predict disease recurrence. Our findings also indicate that similar studies using the alternative approach of liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) in CRC are warranted.

In summary, our study has shown that direct protein expression profiling of surgically resected CRC tissue by MALDI-TOF mass spectrometry has potential value in studies aimed at improved molecular classification of this disease. Further studies, with longer follow-up times and larger patient cohorts, that would permit independent validation of predictive models, would be required to confirm the predictive value of tumour spectra for disease recurrence/patient survival.