Kim YC Fung1*, Leanne Purins1, Ilka K Priebe1, Celine Pompeia1, Gemma V Brierley1, Bruce Tabor1, Trevor Lockett1, Peter Gibbs2, Jeannie Tie2, Paul McMurrick3, James Moore4, Andrew Ruszkiewicz5, Antony Burgess6,7, Edouard Nice8 and Leah J Cosgrove1
Received Date: April 15, 2014; Accepted Date: June 16, 2014; Published Date: June 18, 2014
Citation: Fung KYC, Purins L, Priebe IK, Pompeia C, Brierley GV, et al. (2013) Analysis of 32 Blood-Based Protein Biomarkers for their Potential to Diagnose Colorectal Cancer. J Mol Biomark Diagn S6:003. doi:10.4172/2155-9929.S6-003
Copyright: © 2014 Fung KYC, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Visit for more related articles at Journal of Molecular Biomarkers & Diagnosis
Colorectal cancer (CRC) is largely viewed as a preventable disease but the prevalence is increasing worldwide. Although many faecal and blood-based biomarkers have been proposed as potential diagnostic markers, none have been successful in large cohort studies. In this study, ELISA was used to evaluate 32 candidate protein biomarkers in a single cohort of CRC patients (n=95) and age/sex matched controls (n=50). Of these, 12 markers differed statistically between cases and controls. Receiver operating characteristic analysis identified IL8, Mac2BP, TIMP1, and OPN as the best performing markers for overall CRC diagnosis. However, further analysis determined that IL6, TGFB1, TIMP2 and IGF2 were most accurate at identifying early stage disease. We also assessed the correlation between markers and determined that the strongest correlations existed between VEGFA and TGFB1 (r=0.65, p<0.0001), M30 and M65 (r=0.59, p<0.001), and between TGFB1 and TIMP1 (r=0.55, p<0.0001). This analysis provides a consistent baseline for identifying a potential panel of diagnostic protein biomarkers in blood. Our results highlight protein biomarker combinations that reflect the disease process and which may provide the sensitivity and specificity required a reliable diagnosis of CRC.
Colorectal cancer; Diagnosis; Blood-based; Protein biomarkers
Worldwide, colorectal cancer (CRC) is the second most common cause of cancer-related death with an annual incidence over 1.2 million and an annual mortality over 600,000 . The majority of cases are sporadic with 25-30% estimated to be due to hereditary factors [2,3]. For most sporadic CRC, an accumulation of somatic genetic and epigenetic mutations underlies the transformation from normal colonic mucosa to carcinoma and this transition is believed to occur over a long period of time, i.e., 10-15 years . The high frequency of CRC, the long time frame for its development and the observation that most CRC arise from pre-malignant polyps make CRC an ideal target for population screening programs where detection and removal of premalignant (adenoma or polyp) or early stage malignant disease (Stage A) can potentially prevent the occurrence of CRC or at least significantly increase the likelihood of a complete cure. Due to the slow and multi-stage progression of this disease and the general absence of symptoms in the early stages, it is estimated that around 30-50% of patients have overt metastases at presentation .
Identification of non-invasive biomarkers for early detection of CRC, including detection of pre-malignant and clinically significant polyps and adenomas, is important for reducing both incidence and mortality. When diagnosed early, the 5 year the survival rate for CRC is 90-95% indicating a high curative rate. In comparison, when CRC is detected at later stages, the 5 year survival rate is significantly less (5-10%) . Currently, the faecal occult blood test (FOBT), faecal immunochemical test (FIT), colonoscopy and sigmoidoscopy are the only clinically accepted diagnostic tests for CRC . The FOBT and FIT are used to detect the presence of heme or blood in stool and whilst these tests have a relatively low cost, they are regarded as having poor sensitivity for early stage disease [8,9]. Because the presence of blood in stool is not specific for CRC, the FOBT and FIT also suffer from relatively high false positive rates. In contrast, while colonoscopies have high specificity for the disease, the procedure is highly invasive and expensive. Two of the most widely known serum protein biomarkers for gastrointestinal malignancies, including CRC, are the carbohydrate antigen CA19-9 and carcinoembryonic antigen (CEA) [10-12]. CEA, while useful for monitoring recurrence of CRC, exhibits specificity for cancer of 87%, but its sensitivity (35%) is too low to be useful for detection of CRC in an asymptomatic screening population . Similarly, CA19-9 has limited utility as a diagnostic marker due to its lack of specificity for malignant disease .
Currently, but still in the research phase, are several promising DNA diagnostic biomarkers for CRC, in particular methylated septin 9 (mSEPT9) measured in plasma. A stool-based DNA test consisting of a panel of four methylated genes (BMP3, NDRG4, vimentin, TFPI2) and a mutant form of KRAS is also being developed [13,14]. Initial studies evaluating mSEPT9 indicated high detection rates for CRC [14-16], and this has been confirmed in recent multicentre trials [17-20]. While these reports indicate that the stool-based DNA test is superior to plasma mSEPT for CRC detection, data from larger, longer term studies in a screening populations are required to objectively compare the performance of these two tests with other screening modalities (FOBT and FIT) under population screening conditions.
Many reviews of CRC biomarkers have been published and many factors have been suggested for the lack of success of follow-up studies and lack of consistency of results between different biomarker studies. These factors include small cohort sizes, cohort composition, differences in sample handling and processing procedures, and overrepresentation of late stage disease patients which can bias biomarker sensitivity estimates [11,21-25]. Other factors that have hampered the use of biomarkers in the clinic include assay reproducibility, biomarker stability, and biomarker variability due to comorbidities and diurnal variation. In this study we have measured the concentration of 32 candidate CRC biomarkers in serum and plasma samples from a single cohort of CRC cases (n=95) and age/sex matched controls (n=50). These protein markers were selected based on in-house proteomic and gene expression experiments on colorectal cancer cell lines (in vitro data) and colorectal cancer tissue from patients. We assessed the usefulness of these markers for detecting CRC. Our analysis was designed to minimise the effect of sample collection, processing and storage and assay variability providing an accurate comparison of the performance of these blood-based protein biomarkers for CRC diagnosis.
Patients were newly diagnosed cases of colorectal cancer (no previous history of disease) and blood was obtained prior to surgery (i.e., these are pre-surgical patients) via colorectal surgery preadmission clinics from a network of hospitals associated with the Victorian Cancer Biobank in Melbourne, Victoria, Australia, between 2005 and 2011. Patients with a previous history of CRC or who had already received chemo- and/or radio- therapy were excluded from this study. All research protocols used in this study was approved by the relevant Human Research Ethics Committees at Commonwealth Scientific Industrial Research Organisation, Adelaide, and the Royal Melbourne Hospital, Melbourne.
Serum and plasma samples from CRC patients (n=95) and healthy controls (n=50) were obtained and processed using methods previously described [26,27]. To minimise the effect of potential confounders, the normal and CRC cohorts were balanced for age, sex and disease stage. Briefly, blood was collected from each subject into serum separator tubes and EDTA plasma tubes. The blood was left at room temperature in the collection tube for 30 min and then centrifuged at (1,200 g, 10 min, room temperature). The supernatant was transferred to a new tube and centrifuged at (1,800 g, 10 min, room temperature). Aliquots of the resulting supernatant were frozen at -80ºC until analysis. The time from sampling to freezing was 2 hrs.
Protein measurements in serum and plasma by ELISA
Serum and plasma were assayed using commercially available ELISA kits or reagents according to manufacturers’ instructions unless otherwise specified. The following multiplex ELISA kits were sourced from R&D Systems (Minneapolis, MN, USA): multiplex kit for the chemokines ENA-78, MCP-1, and MIP-1β, the cytokine multiplex panel for analysis of TNF-α, IL6 and IL8, the multiplex panel for matrix metalloproteinases (MMP)-1, -3, -7, and -8. DuoSet ELISA kits for amphiregulin, DcR3, DKK3, and RegIV. ELISA kits for TGFB1, TIMP1, TIMP2, VEGFA and GRO-α were also purchased from R&D Systems (Minneapolis, MN, USA). ELISA kits for the following markers were also obtained: M30 and M65 (PEVIVA, Bromma, Sweden), IGFBP2 and IGF2 (DSL Inc., Webster, TX, USA), PKM2 (Schebo, Giessen, Germany), Mac-2BP (Bender MedSystems GmbH, Vienna, Austria), and OPN (Linco Research, St Charles, MO, USA). For the analysis of EpCAM, the DuoSet ELISA kit (R&D Systems, Minneapolis, MN, USA) was used and chemiluminescent detection was performed using the Supersignal ELISA Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific, Waltham, MA, USA).
For CEACAM6, 96 well plates were coated with polyclonal rabbit anti-human carcinoembryonic antigen (DakoCytomation, Glostrup, Denmark) (2 μg/mL in carbonate buffer, pH 9.5). The standard curve (range 0.49 – 125 ng/mL) was prepared by serial dilution of the CEACAM6 recombinant protein (R&D Systems, Minneapolis, MN, USA). Biotinylated CEACAM6 monoclonal antibody (Thermo Fisher Scientific, Waltham, MA, USA) was used for detection (1 μg/mL in PBS/1% BSA).
For SPONDIN-2, 96 well plates were coated with SPONDIN-2 monoclonal antibody (R&D Systems, Minneapolis, MN, USA) (4 μgmL carbonate buffer, pH 9.5). The standard curve (range 15.6 – 2000 ng/mL) was prepared by serial dilution of recombinant human SPONDIN-2 protein (Abnova, Taipei, Taiwan). Biotinylated antihuman SPONDIN-2 detection antibody (R&D Systems, Minneapolis, MN, USA) was prepared at a concentration of 800 ng/mL in PBS/3% BSA.
An in-house bead-based assay was used to measure P-cadherin. Monoclonal anti-human p-cadherin antibody (R&D Systems, Minneapolis, MN, USA) was coupled to carboxylated polystyrene beads (Luminex Corporation, Austin, TX, USA) according to the manufacturer’s instructions. The standard curve (range 7.8 – 2000 ng/ mL) was prepared by serial dilution of recombinant human p-cadherin protein (R&D Systems, Minneapolis, MN, USA). Streptavidin-Rphycoerythrin donkey anti-goat reporter (Thermo Fisher Scientific, Waltham, MA, USA) was used as a concentration of 0.4μg/mL (50 μL) for detection.
Two in-house quality control (QC) samples were included in each analysis. QC samples consisted of a pooled normal sample (n=41) and a pooled CRC patient sample (n=41). For commercially available ELISA kits, the intra-assay coefficients of variation (CV) were less than 10%, consistent with the manufacturers’ specifications.
Multiplex panels and p-cadherin were analysed using the Luminex bead-based system (Qiagen, Hilden, Germany) and preliminary data was analysed using the Luminex IS2.3 software. For standard ELISAs, the absorbance was determined using the Wallac Victor3V Multilabel Counter microplate reader (Perkin Elmer, Waltham, MA, USA) set to 450 nm with wavelength correction at 570 nm. Preliminary data analysis for the standard ELISA assays was performed using Workout 2.0 software (DazDaq, Brighton, UK). The Prism software package (version 6, Graphpad Software Inc., San Diego, CA, USA) was used for statistical analysis. The non-parametric Mann-Whitney U test was used to determine the statistical difference between cancer and control patients. Receiver operator characteristic (ROC) curve analysis was performed to assess the diagnostic performance for each marker and to determine the sensitivity for each marker at 95% specificity. Spearman correlation was used to determine correlations between markers. Statistical significance was defined as p<0.05.
Table 1 summarises the characteristics of the cohort used in this study. The median age was 67 yrs (range 44-93 yrs) for CRC patients and 70 yrs (range 50-85 yrs) for the control group. The CRC patient group was further stratified according to Dukes’ stage.
|Median age, yrs (range)||70 (50-85)||67 (44-93)|
Table 1: Characteristics of the study cohort.
Of the 32 proteins analysed, 23 protein markers were measured within the range of the ELISAs. Eight markers (MMP7, p-cadherin, RegIV, spondin-2, EpCam, GRO-alpha, amphiregulin, and DcR3) were undetectable in the serum or plasma of the majority of samples (CRC and control patients) and were excluded from further analysis. A summary of these 8 markers, including the standard curve range for each assay can be found in Supplementary Information 1. Of the 23 protein biomarkers that were detectable, 12 showed a significant difference (p<0.05) between the median values of the cancer and control patients (Figure 1 and Table 2). These 12 markers include Mac2BP, PKM2, IL8, IL6, IGFBP2, TGFB1, M65, IGF2, VEGFA, TIMP1 (measured in both serum and plasma), MMP1, and OPN. All of these proteins, with the exception of IGF2, were elevated in CRC patient samples in comparison to the control group. For IGF2, the median concentration was lower in CRC patients (1221 ng/mL, range 421.1-1864 ng/mL) when compared to the control group (1399 ng/mL, range 792.3-2230 ng/mL; p=0.005).
|Biomarker||Units||Serum/ plasma||median (range)||median (range)||cancer vs control|
|PKM2||U/ml||plasma||21.02 (11.26 - 73.3)||30.71 (5.81 - 98.80)||0.0006|
|IL8||pg/ml||serum||11.26 (4.36 - 49.89)||15.77 (3.71 - 103.5)||0.0006|
|IL6||pg/ml||serum||1.84 (0.95 - 55.80)||1.355 (0.93 - 4.740)||0.0006|
|IGFBP2||ng/ml||serum||490.5 (188.7 - 1583)||699.2 (163.3 - 3698)||0.0011|
|Mac2BP||ng/ml||serum||7126 (3918 - 20150)||8350 (4290 - 40870)||0.0012|
|VEGFA||pg/ml||plasma||48.69 (31.42 - 387)||78.96 (33.29 - 910.1)||0.0013|
|TGFB1||pg/ml||plasma||6034 (2070 - 21122)||8794 (1846 - 64230)||0.005|
|M65||U/L||serum||312.6 (143.3 - 783.2)||358.4 (132.8 - 1032)||0.0051|
|IGF2||ng/ml||serum||1399 (792.3 - 2230)||1221 (421.1 - 1864)||0.0052|
|TIMP1||ng/ml||plasma||74.89 (42.53 - 131.3)||84.31(49.9 - 359.6)||0.0062|
|TIMP1||ng/ml||serum||172.8 (74.3 - 314.4)||193.8 (98.6-449.4)||0.0129|
|MMP1||pg/ml||serum||2090 (258 - 14300)||3350 (249 - 56000)||0.0226|
|OPN||pg/ml||plasma||4980 (868 - 29600)||7410 (425 - 297000)||0.033|
|TNF-alpha||pg/ml||serum||7.55 (1.55 - 13.21)||7.85 (3.09 - 26.67)||0.1092|
|TIMP2||ng/ml||serum||82.5 (59.2 - 106.2)||77.6 (47.3 - 159.4)||0.1149|
|CEACAM6||ng/ml||plasma||1.62 (0.60 - 3.54)||1.63 (0.60 - 7.99)||0.1344|
|MIP-1beta||pg/ml||serum||77.00 (37.30 - 369.0)||66.95 (35.8 - 1510)||0.1355|
|Dkk3||pg/ml||plasma||58895 (37232 - 78500)||56473 (27726 - 75779)||0.3046|
|ENA-78||pg/ml||serum||1275 (302 - 5180)||1240 (264 - 4450)||0.4717|
|MMP3||pg/ml||serum||13250 (2280 - 37600)||11100 (2600 - 55500)||0.5033|
|TIMP2||ng/ml||plasma||90.71 (64.06 - 129.7)||90.86 (53.78 - 172.3)||0.6593|
|MCP1||pg/ml||serum||233.5 (80 - 526)||237.5 (20.3 - 795)||0.7915|
|MMP8||pg/ml||serum||8720 (1510 - 21600)||8190 (1490 - 50600)||0.9344|
|M30||U/L||serum||178.8 (85.7 - 500.1)||173.4 (87.17 - 901.6)||0.9791|
Table 2: Summary of ELISA results for the biomarkers analysed in colorectal cancer and control patients ranked by p value.
ROC analysis was conducted to determine the ability of each marker to distinguish between the CRC and the control groups (Figure 2 and Table 3). At 95% specificity, IL8 was the best performing biomarker (sensitivity of 38%), followed by Mac2BP (sensitivity of 35%). At 95% specificity, TIMP1 measured in plasma and in serum at sensitivity of 33% and 28% respectively, followed by OPN (sensitivity of 31%) in plasma, and IL6 (sensitivity of 27%), M65 (sensitivity of 26%) and IGFBP2 (sensitivity of 25%) in serum. PKM2 and Dkk3 in plasma and MMP1, M30, ENA-78, MMP8, MIP-1beta, MMP3 and MCP1 in serum were the poorest performing biomarkers where the sensitivity for each at 95% specificity was less than 25%. In our previous study , we determined that CEA had a sensitivity of 14% at 95% specificity indicating that it is a poor biomarker for diagnosis of CRC, consistent with other reports.
|AUC (95% CI)||p value||Sensitivity (%) at 95% specificity||Cut off value (95% specificity)|
|IL8||0.68 (0.59 - 0.77)||0.0006||38||>21.86|
|Mac2BP||0.67 (0.58 - 0.76)||0.0009||35||>9304|
|TIMP1 (plasma)||0.64 (0.55 - 0.73)||0.0061||33||>110.5|
|OPN||0.61 (0.52 - 0.70)||0.0329||31||>11800|
|TIMP1 (serum)||0.63 (0.54 - 0.72)||0.0133||28||>232.0|
|IL6||0.70 (0.61 - 0.80)||0.0002||27||>2.895|
|M65||0.66 (0.56 - 0.75)||0.0016||26||>472.4|
|IGFBP2||0.67 (0.57 - 0.76)||0.0008||25||>1225|
|TIMP2 (serum)||0.58 (0.48 - 0.67)||0.1144||25||<64.67|
|TIMP2 (plasma)||0.52 (0.43 - 0.62)||0.6578||24||<74.65|
|IGF2||0.64 (0.55 - 0.73)||0.0040||23||<1040|
|VEGFA||0.67 (0.58 - 0.77)||0.0005||23||>132.5|
|CEACAM6||0.58 (0.48 - 0.67)||0.1340||22||>2.558|
|TGFB1||0.65 (0.56 - 0.74)||0.0027||22||>16195|
|TNF-alpha||0.58 (0.49 - 0.69)||0.1088||20||>11.51|
|PKM2||0.70 (0.60 - 0.79)||0.0001||19||>60.39|
|Dkk3||0.50 (0.41 - 0.60)||0.9502||18||<42184|
|MMP1||0.62 (0.52 - 0.71)||0.0226||15||>9130|
|M30||0.51 (0.41 - 0.61)||0.8417||13||>374.2|
|MMP8||0.51 (0.41 - 0.61)||0.7808||12||>18650|
|ENA-78||0.54 (0.44 - 0.64)||0.4705||10||<488.0|
|MIP-1beta||0.56 (0.46 - 0.66)||0.2245||9||<26.35|
|MMP3||0.53 (0.44 - 0.63)||0.5744||7||<5460|
|MCP1||0.51 (0.41 - 0.61)||0.7899||6||>416.0|
AUC: Area under the receiver operator characteristic curve
Table 3: Sensitivity of biomarkers at 95% specificity.
The ability of each marker to discriminate disease stage from the control group was also evaluated (Table 4). Of the 32 markers evaluated, only IL6, TIMP2 (measured in serum), IGF2 and TGFB1 appear to identify patients with early stage disease (i.e., stage A disease, p<0.05). Although M65 was able to identify stage A disease with the highest sensitivity (38%) at 95% specificity, the area under the ROC curve was not statistically significant (p=0.057). Only TGFB1 and IL6 were able to identify patients with either Stage A or B disease (p<0.05). Although IL8 was the best performing biomarker for diagnosing CRC overall (Table 3), it was most successful at identifying stage C and D disease where its sensitivity was 53% (p<0.0001) and 75% (p=0.004), respectively. IL6 and IGF2 were the most successful markers for identifying stage A disease (sensitivities of 30% and 29%, respectively, p<0.05) however their sensitivity detecting for stage B disease dropped to 24% (p<0.05) and 19% (p=0.382), respectively.
|Stage A||Stage B||Stage C||Stage D|
|AUC (95% CI)||p value||Sensitivity (%) at 95% specificity||AUC (95% CI)||p value||Sensitivity (%) at 95% specificity||AUC (95% CI)||p value||Sensitivity (%) at 95% specificity||AUC (95% CI)||p value||Sensitivity (%) at 95% specificity|
|IL8||0.50 (0.34 - 0.67)||0.970||10||0.67 (0.55 - 0.80)||0.013||24||0.77 (0.65 - 0.88)||<0.0001||53||0.82 (0.60 - 1.05)||0.004||75|
|Mac2BP||0.64 (0.48 - 0.80)||0.073||29||0.66 (0.53 - 0.79)||0.022||38||0.71 (0.59 - 0.84)||0.001||39||0.68 (0.47 - 0.89)||0.112||25|
|TIMP1 (plasma)||0.62 (0.47 - 0.77)||0.115||23||0.66 (0.53 - 0.78)||0.018||36||0.63 (0.49 - 0.77)||0.047||39||0.65 (0.48 - 0.82)||0.132||20|
|OPN||0.51 (0.34 - 0.68)||0.900||24||0.73 (0.61 - 0.84)||0.001||42||0.52 (0.38 - 0.66)||0.761||16||0.73 (0.50 - 0.95)||0.023||60|
|TIMP1 (serum)||0.54 (0.37 - 0.70)||0.631||25||0.68 (0.56 - 0.81)||0.006||28||0.70 (0.58 - 0.83)||0.002||37||0.53 (0.30 - 0.76)||0.751||30|
|IL6||0.65 (0.50 - 0.81)||0.047||30||0.71 (0.59 - 0.83)||0.003||24||0.74 (0.63 - 0.86)||0.0003||30||0.62 (0.39 - 0.85)||0.281||13|
|M65||0.64 (0.50 - 0.79)||0.057||38||0.54 (0.42 - 0.68)||0.468||10||0.75 (0.65 - 0.86)||0.0002||25||0.77 (0.57 - 0.96)||0.008||50|
|IGFBP2||0.63 (0.48 - 0.78)||0.087||29||0.71 (0.60 - 0.82)||0.001||23||0.59 (0.47 - 0.72)||0.160||21||0.88 (0.78 - 0.97)||0.0002||40|
|TIMP2 (serum)||0.65 (0.49 - 0.81)||0.047||28||0.53 (0.37 - 0.68)||0.692||28||0.55 (0.41 - 0.68)||0.462||15||0.69 (0.48 - 0.91)||0.054||40|
|TIMP2 (plasma)||0.55 (0.40 - 0.72)||0.442||24||0.54 (0.04 - 0.68)||0.590||29||0.53 (0.40 - 0.67)||0.609||12||0.59 (0.37 - 0.81)||0.372||30|
|IGF2||0.66 (0.50 - 0.82)||0.035||29||0.56 (0.42 - 0.69)||0.382||19||0.71 (0.59 - 0.82)||0.001||27||0.68 (0.49 - 0.87)||0.074||10|
|VEGFA||0.58 (0.43 - 0.74)||0.262||14||0.76 (0.64 - 0.88)||<0.0001||40||0.67 (0.55 - 0.79)||0.010||13||0.64 (0.45 - 0.84)||0.159||20|
|CEACAM6||0.50 (0.35 - 0.65)||0.970||14||0.52 (0.39 - 0.65)||0.766||20||0.59 (0.47 - 0.72)||0.149||19||0.84 (0.70 - 0.97)||0.001||50|
|TGFB1||0.66 (0.52 - 0.80)||0.029||19||0.72 (0.60 - 0.83)||0.001||32||0.63 (0.51 - 0.76)||0.047||16||0.51 (0.29 - 0.07)||0.940||18|
|TNF-alpha||0.52 (0.37 - 0.67)||0.778||10||0.65 (0.52 - 0.78)||0.032||21||0.54 (0.41 - 0.68)||0.537||21||0.69 (0.48 - 0.91)||0.082||38|
|PKM2||0.65 (0.50 - 0.79)||0.051||19||0.67 (0.55 - 0.79)||0.010||16||0.72 (0.60 - 0.83)||0.001||15||0.79 (0.63 - 0.94)||0.004||40|
|Dkk3||0.51 (0.36 - 0.67)||0.821||0||0.52 (0.38 - 0.65)||0.789||10||0.52 (0.38 - 0.65)||0.820||21||0.57 (0.34 - 0.80)||0.475||30|
|MMP1||0.56 (0.40 - 0.73)||0.399||14||0.62 (0.49 - 0.75)||0.065||16||0.64 (0.52 - 0.77)||0.029||15||0.61 (0.42 - 0.81)||0.258||10|
|M30||0.54 (0.38 - 0.70)||0.600||14||0.62 (0.49 - 0.74)||0.080||19||0.63 (0.51 - 0.75)||0.041||12||0.60 (0.39 - 0.81)||0.321||30|
|MMP8||0.54 (0.38 - 0.69)||0.642||6||0.54 (0.41 - 0.67)||0.544||17||0.54 (0.41 - 0.66)||0.577||12||0.70 (0.49 - 0.91)||0.056||22|
|ENA-78||0.60 (0.46 - 0.75)||0.172||14||0.59 (0.37 - 0.64)||0.956||7||0.50 (0.37 - 0.63)||1.000||6||0.62 (0.43 - 0.80)||0.250||20|
|MIP-1beta||0.59 (0.44 - 0.73)||0.260||5||0.52 (0.39 - 0.65)||0.769||7||0.57 (0.45 - 0.70)||0.268||9||0.60 (0.40 - 0.80)||0.307||20|
|MMP3||0.51 (0.34 - 0.67)||0.950||14||0.55 (0.41 - 0.68)||0.466||10||0.53 (0.40 - 0.66)||0.696||12||0.69 (0.55 - 0.84)||0.054||10|
|MCP1||0.57 (0.43 - 0.72)||0.329||14||0.52 (0.39 - 0.64)||0.815||3||0.58 (0.46 - 0.71)||0.201||12||0.55 (0.34 - 0.77)||0.592||20|
Table 4: Biomarker sensitivity at 95% specificity according to disease stage.
Correlation between markers
Since the markers measured in this study represent different biological aspects of CRC (eg, inflammation, angiogenesis, metastasis, growth factor production, apoptosis), the Spearman correlation was used to determine if any relationship existed between any of the markers. Although correlations between many biomarker pairs were found to be significant, the majority of the correlations were weak (r<0.3), including correlations found between the 12 significant biomarkers (Supplementary information 2). Table 5 lists the marker pairs with Spearman r>0.3. As expected, plasma concentrations of TIMP1 and TIMP2 correlated strongly with their respective measurements in serum (r=0.63, p<0.0001 and r=0.74, p<0.0001, respectively).
|Marker 1||Marker 2||Spearman R||P value|
|TIMP2 SERUM||TIMP2 PLASMA||0.740||<0.0001|
|TIMP1 SERUM||TIMP1 PLASMA||0.626||<0.0001|
|TIMP2 PLASMA||TIMP1 PLASMA||0.393||<0.0001|
Table 5: Correlation between markers.
The strongest correlations were observed between VEGFA and TGFB1 (r=0.65, p<0.0001), M30 and M65 (r=0.59, p<0.0001), and between TGFB1 and TIMP1 measured in plasma (r=0.55, p<0.0001). For the inflammatory markers and chemokines, correlations were weak between IL6 and IL8 (r=0.274, p=0.002), between IL8 and MCP1 (r=0.218, p=0.012) and ENA-78 correlated weakly with both MCP1 (0.272, p=0.001) and MIP1B (r=0.197, p=0.024).
We have evaluated 32 protein biomarkers for their utility as diagnostic markers of CRC. Although there is an abundance of literature evaluating potential biomarkers for CRC, it is difficult to compare the performance of individual biomarkers due to the differences in cohort sizes and compositions. Differences in sample handling, storage conditions and processing for reported studies also make comparisons difficult. This report is one of the few studies evaluating a large number of proteins (> 20 proteins) in the same patient cohort [28,29], and furthermore, our cohort was balanced for age, sex and disease stage. Of the 32 biomarkers investigated, 12 were found to be significantly different between the control and CRC patient group. Of these, IL8, Mac2BP, OPN and TIMP1 (measured in both serum and plasma) were the best performing biomarkers for diagnosing CRC (sensitivities of 38%, 35%, 31%, 33% and 28%, respectively, at 95% specificity). A fixed specificity of 95% was chosen to minimise the number of false positive cases as we consider this to be an important aspect of a diagnostic assay and so that we can compare any single biomarker to the performance of the FIT assay used in screening programs.
Our data shows that the serum and/or plasma levels of proteins involved in similar pathophysiological processes did not necessarily correlate strongly. For example, proteins such as the interleukins and chemokines that are involved in the inflammatory and immune process only weakly correlated with each other, and a similar result was observed for the MMPs and TIMPs which are involved with tissue remodelling. Although the lack of correlation between markers with similar biological function is surprising, Bunger et al recently published that in a panel of 12 cytokines measured in the sera of 100 CRC patients and controls, only IL8 discriminated between controls and cancer patients and only poor to moderate correlation was found between the cytokines measured .
Strong correlations were observed between VEGFA and TGFB1 (r=0.65, p<0.0001), TGFB1 and TIMP1 (r=0.55, p<0.0001), and between VEGFA and TIMP1 (r=0.47, p<0.0001). Although cytokines have been reported to induce the expression of these markers, only relatively weak correlations were found between TIMP1 and IL8 and IL6 (r=0.36 – 0.44) and between IL6 and VEGFA (r=0.24). In a study by Biasi et al , no significant difference was found between VEGFA measured in the circulation of CRC patients and controls and a weak negative trend with disease stage was observed with TGFB1 with statistical significance occurring at stage C disease only. The authors did not report a correlation for these two markers. In contrast, our study found that the circulating level of these proteins strongly correlated, and were statistically higher in the CRC cohort in comparison to the control group. Furthermore, both markers demonstrated the highest sensitivity for predicting stage B disease. VEGFA, TGFB1 and TIMP1 reportedly encourages metastatic spread by influencing different aspects of the tumor-stromal environment including promotion of angiogenesis, by stimulating cell migration and invasion, or by promoting epithelial to mesenchymal transition [32,33].
A significant and strong correlation was found between M30 and M65 (r=0.59, p<0.001). M30 and M65 represent caspase cleaved cytokeratin 18 and total cytokeratin 18, respectively. Whereas M30 is reported to be a marker of tumour cell apoptosis, M65 is reported to be a marker of apoptosis and necrosis . Our data supports current literature which indicates circulating levels of M30 and M65 as markers of tumour burden and may be useful as diagnostic markers for epithelial cancers [35-38]. These markers have also been evaluated in pre-clinical models of cancer to assess drug and/or treatment efficacy [39-43].
The analysis that we report here is one of few studies that provide a consistent baseline for identifying a potential panel of diagnostic markers for CRC. Based on our analysis of 32 protein biomarkers in the same patient cohort, no single biomarker adequately discriminated between controls and CRC patients to be useful in a diagnostic or screening application. Further experiments are required to determine if identified protein biomarker combinations that reflect the disease process provide the sensitivity and specificity required for CRC diagnosis. Our study also highlights that a panel of markers representative of different biological processes in the carcinogenesis pathway, including inflammation, the immune response, or apoptosis may be most optimal. Simultaneous measurement of these potential biomarker combinations in a large and well-defined cohort is required to evaluate their true diagnostic ability.
The authors would like to thank the Victorian Cancer Biobank (Melbourne, Victoria) for their assistance with sample collection. This work was funded by the CSIRO Preventative Health National Research Flagship and the National Health and Medical Research Council (grant number 1017078).