Background
Colorectal cancer (CRC) is the second leading cause of death from cancer in the developed world [
1]. Randomised controlled trials (RCT) in the general population have shown that early detection by screening, such as with faecal occult blood test (FOBT) or flexible sigmoidoscopy, reduces mortality and may also reduce incidence [
2‐
6]. Reduction in mortality is dependent on treatment of curable neoplasms destined to cause death while reduction in incidence is dependent on detection and removal of pre-invasive lesions (i.e. adenomas). Given that early detection of a neoplasm is worthwhile for either a bleeding phenotype or a phenotype that enables visualisation (as detected by FOBT and flexible sigmoidoscopy, respectively), detection of a neoplasm based on other factors such as molecular characteristics may have the same benefit, but this is yet to be established.
In addition to the ability of a test to detect early curable lesions, a screening test can only be effective if the targeted individual undertakes the test. This behavioural consideration presents certain barriers for endoscopic methods and in some countries also for FOBT. Participation rates for both FOBT and endoscopic methods are highly variable and clearly sub-optimal in many settings [
7].
It has been suggested that a blood test would be more acceptable and circumvent some of the barriers with established screening methods [
8,
9]. A blood test could be deployed as an alternative frontline screening test or else as a “rescue” strategy that aims to engage those who reject the existing RCT-proven methods such as FOBT and flexible sigmoidoscopy. The appropriate manner of deployment will depend in part on the accuracy of such a blood test.
Aberrant DNA methylation is a characteristic of colorectal tumours [
10,
11].
SEPT9 is one such tumour marker methylated in colorectal neoplasia that is detectable in blood [
12,
13], but its clinical performance as a screening test is suboptimal. We have previously reported the identification and validation of a cohort of genes with hypermethylated regions that show promise for differentiating adenomas and early stage cancer from normal state and benign pathology [
14]. More recently, we have shown that cell free circulating DNA extracted from blood from CRC patients has a significantly higher fraction of methylation across two genes, namely
BCAT1 and
IKZF1, compared to normal controls [
15]. It is important to determine the accuracy of detecting methylated
BCAT1 and
IKZF1 DNA in blood across the range of neoplastic lesions encountered in the colon before proceeding to compare outcomes from screening programs using the two-marker blood test, to programs using proved screening tests. The latter step is crucial to the inclusion of tests based on blood molecular markers in screening programs since early detection alone does not guarantee program efficacy or effectiveness when the biological basis of lesion detection is different [
16,
17].
The goal of this study was to estimate true and false positive rates of the two-marker blood test for screen-relevant stages of colorectal neoplasia, namely advanced adenoma and CRC of specific stage, and across the full spectrum of non-neoplastic pathologies encountered in the colon/rectum when screening a large population.
Methods
Study overview
This was a multi-centre predominantly prospective study funded in part by the National Health and Medical Research Council (NHMRC) and Clinical Genomics Technologies Pty Ltd (CGT) to estimate the sensitivity and specificity of a test detecting methylated BCAT1 and/or IKZF1 DNA in blood from people with neoplasia or non-neoplastic pathologies likely to be encountered in the colon and rectum. Findings at colonoscopy were used as the diagnostic standard. The study was approved by the Southern Adelaide Clinical Human Research Ethics Committee (April 4, 2005) and Medical Ethical Board of Academic Medical Centre Amsterdam (July 12, 2011). Written informed consent was obtained from all recruits prior to any procedures. Clinical and research staff at the medical institutions audited clinical data and verified case classification blinded to assay results determined by CGT. The clinical data were only released subsequent to completion of testing of all collected samples. Test results were not disclosed to subjects or their physicians. The trial is registered at Australian and New Zealand Clinical Trials Registry trial registration number 12611000318987.
Population
Subjects aged 33-85 years old and either scheduled for colonoscopy for standard clinical indications (prospective element), or shown at colonoscopy within the prior ten days to have CRC that had not been treated (retrospective element), were approached about volunteering for the study. The participating centres were Repatriation General Hospital (Daw Park, South Australia), Flinders Medical Centre (Bedford Park, South Australia), Academic Medical Centre (Amsterdam, The Netherlands) and Flevo Hospital (Almere, The Netherlands). Following enrolment, cases were excluded if the scheduled colonoscopy was cancelled or if insufficient blood was available.
Clinical procedures
Venous blood was collected into two 9mL K3EDTA Vacuette tubes (Greiner Bio-One, Frickenhausen, Germany) from subjects either prior to them being sedated for colonoscopy but after consumption of bowel preparation solution, or prior to preparation for surgery but following colonoscopic diagnosis. A second sample was obtained from 26 CRC cases one month or more after surgery. Blood tubes were kept at 4 °C until commencing plasma processing. Plasma was prepared within 4 hours of blood collection by centrifugation at 1,500 g for 10 minutes at 4 °C (no braking), followed by retrieval of the plasma fraction and a repeat centrifugation. The resulting plasma was stored at -80 °C. Frozen plasma samples were shipped on dry ice to CGT and stored at -80 °C until testing.
No study-wide control of colonoscopy or pathology procedures or quality was undertaken as the study aimed to assess marker performance relative to outcomes determined in usual clinical practice. All procedures were performed by hospital-accredited specialists and so met site-specific standards for sedation, monitoring, imaging, and equipment. Histopathology and staging of neoplasia used routine procedures at each clinical site. Cases were excluded if any data crucial to clinical diagnosis was not obtainable, e.g. if colonoscopy was incomplete.
Pathological classification
An independent physician assigned diagnosis for all cases used in this study on the basis of colonoscopy, surgical and histopathological findings. CRC was staged according to AJCC 7th Edition [
18]. Advanced adenoma was defined as adenoma with any of the following characteristics: (a) ≥ 10 mm in size, (b) >20 % villous change, (c) high grade dysplasia, or (d) serrated pathology. Cases with more than two tubular adenomas or stage 0 cancer were also classified as advanced adenoma. Non-advanced adenoma refers to those not meeting the characteristics of an advanced adenoma. Hyperplastic polyps were classed as non-neoplastic pathologies. Where multiple pathologies were present, the most advanced neoplasm was used as the principal diagnosis. Location of the principal neoplasm was defined as that of the most advanced lesion in a patient with multiple neoplasms. Where multiple non-neoplastic diagnoses were present, the principal diagnosis was allocated in the following hierarchy (descending): inflammatory bowel disease (IBD), hyperplastic polyp, angiodysplasia, haemorrhoids, diverticular disease.
Test method
All plasma samples of at least 3.9mL were assayed for the presence of methylated
BCAT1 and
IKZF1 DNA at CGT’s laboratories by trained and qualified staff blinded to clinical results (see Additional file
1 for details). Samples were analysed in batches of 22 clinical samples and two process controls. Batches were loaded on a QIASymphony SP instrument (Qiagen, Hilden, Germany) and cell-free DNA was extracted using a QIASymphony Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany) according to manufacturer’s instructions (Additional file
1). The extracted DNA was bisulphite-converted using the EpiTect Fast Bisulfite Conversion kit (Qiagen) and QIACube instrument (Qiagen) as recommended by manufacturer but with minor modifications (see Additional file
1). The resulting bisulphite-converted DNA was analysed as three replicates in a triplex real-time qPCR assay (
ACTB control, methylated
BCAT1 and
IKZF1) performed on a Roche LightCycler 480 Model II instrument (see Additional file
1). A sample was deemed positive if at least one qPCR replicate was positive for either
BCAT1 or
IKZF1 DNA methylation; no cycle threshold (Ct) value cut-offs were applied. Each PCR plate included three no-template control samples and a standard curve based on 0-2ng bisulphite converted fully methylated human DNA (Merck-Millipore, MA, United States) prepared in a background of nuclease-free water (Promega, WI, United States). The mass of methylated
BCAT1 and
IKZF1 DNA in each plasma specimen was determined from the batch specific standard curve. The level of methylation was expressed as the total mass of methylated (
BCAT1 plus
IKZF1) DNA as a percentage of the total amount of recovered DNA per processed specimen.
Statistical analyses
Subjects were recruited until at least 100 cancer cases had been identified (keeping 95 % CI of sensitivity estimates to less than 20 %) with at least 25 cases at each of stages I-III (to enable determination of the relationship between positivity rate and stage). The main outcome measure was positivity rate by diagnosis. GraphPad online scientific software tool,
http://graphpad.com/scientific-software/, was used to calculate 95 % confidence intervals (binomial distribution assumed), Chi-square values (using 2x2 contingency tables without Yates’ correction) and McNemar’s test. Linear weighted Kappa statistic and odds ratios were calculated using
www.vassarstats.net and
www.medcalc.org/calc/odds_ratio.php, respectively.
Analysis of potential confounding co-variables was performed using a logistic generalised linear model fitted to a binary positivity variable (R package version 3.1.2) or by using a 2-sample z-test (two-tailed, 95 % significant level,
http://www.socscistatistics.com/tests/ztest/Default2.aspx) on sample proportions (positive results observed in a given sample size). Continuous variables included age and DNA; dichotomous variables included smoking status, gender, and family CRC history.
An ANOVA Chi-square test (R version 3.1.2) was performed on assay positivity rates corrected for stage distribution in proximal and distal cancers using a generalised linear model with a logistic regression model fitted to two covariate models including stage and lesion, or lesion only.
The log values of the percentages of methylated BCAT1 and IKZF1 DNA measured in amount of DNA retrieved per processed specimens were used to create empirical density plots for three clinical classes: non cancer (all pathologies minus CRC cases), early stage cancer (Stage I + II) and late stage cancer (Stage III + IV). A minus infinity value was assigned to all cases with no methylation signal, whereas a Gaussian distribution was assumed for all non-zero values. By fitting Gaussian distribution curves to the empirical density plots, relative risk was calculated as the ratio of the conditional probability for early or late stage cancer compared to non-cancer based on the equation \( \frac{\mathrm{P}\left(\left.\mathrm{X}=1\right|\left.\mathrm{Y}=1\right)\right.}{\mathrm{P}\left(\left.\mathrm{x}=0\right|\left.\mathrm{Y}=1\right)\right.}=\frac{{\mathrm{P}}_{11}}{{\mathrm{P}}_{01}} \), where X = 1 means cancer, X = 0 means no cancer and Y is the test result (positive (Y = 1) or negative (Y = 0)) at a given threshold value.
Reported p-values are 2-tailed and values <0.05 were considered statistically significant.
Discussion
By estimating the true- and false-positive rates of the two-marker blood test for screen-relevant stages of colorectal neoplasia, we have been able to determine that a blood test detecting methylated BCAT1 and IKZF1 DNA facilitates identification of cases with CRC relative to other clinical states encountered in the colon and rectum.
We estimated an overall sensitivity for CRC of 66 % (
n = 129, 95 % CI: 57–74), with better detection of later versus earlier stage cancers (79 % compared to 56 %). This overall sensitivity is within the upper half of the reported sensitivity range of 37–79 % for guaiac FOBT (gFOBT) in populations such as we have studied here or in true screening populations [
20]. Despite low sensitivity in the original gFOBTs, RCTs still showed effectiveness of the technology in reducing mortality from CRC [
3,
4]. In a micro-simulation model to estimate gFOBT sensitivity for CRC from the first three RCTs it was estimated that gFOBT sensitivity was 51 % for the stages of clinical diagnosis and 19 % for early stage cancer [
21]. This implies an adequate sensitivity of the two-marker blood test for reducing CRC mortality if used as a screening test but this prediction requires validation in true screening populations. The two-marker blood test has a low sensitivity for advanced adenomas and should not be expected to impact on CRC incidence as seen with certain faecal immunochemical tests (FIT) which have sensitivity for advanced adenomas in the range 29–45 % [
22,
23].
Impact of a screening test on population mortality from CRC is not dependent only on test accuracy but also on participation rates. Given the stated preference of a typical screening population for the idea of a blood test over a faecal test [
8], including a subset who had already undertaken screening with FIT [
9], one could predict that even if a lesser sensitivity were to be confirmed for the two-marker blood test when validated in true screening populations, a participatory advantage might counterbalance this.
The earlier estimates of sensitivity for cancer and advanced adenoma for methylated Septin 9 (
SEPT9) were comparable to those seen with our two-marker blood test [
12,
24‐
26], although a large-scale study in a screening population returned a cancer sensitivity of 51 % [
13]. The reported observed sensitivity for stage I cancer of 36 % was almost identical to ours (38 %), while neither study achieved a sensitivity of 10 % for advanced adenomas. Whether there is complementarity of our markers with
SEPT9 for cancer detection is unclear at present and warrants study.
To determine whether this apparent lower sensitivity for early stage cancer and adenomas was a function of the assay or a biologically-determined issue, we examined the relationship of positivity to tumour depth of invasion and modelled the biomarker mass relative to risk for different stages of neoplasia. A trend was observed between assay positivity and degree of cancer invasiveness (pT stage), which was not affected by the colonic location or other potential variables examined. By modelling the stage of neoplasia relative to marker mass, we show the potential for using the measured percentage of methylated
BCAT1 and
IKZF1 DNA in blood to estimate the relative risk of disease severity. Given that the assay is sensitive at the limits of detection to 6 DNA copies per mL of plasma (Additional file
1), some stage I cancers might escape detection due to very low amount of tumour-derived DNA reaching the blood [
27,
28]. As adenomas are non-invasive, this might account for a biological limitation in the capacity of blood tests to detect adenomas.
If methylated DNA biomarkers are fundamentally disadvantaged compared to FIT in detection of advanced adenomas, then what is their place in CRC screening? Where programs seek to detect just a proportion of cancers with high efficiency and low colonoscopy rates [
29], a blood DNA test might be acceptable as a frontline screening test if a participatory advantage can be demonstrated in practice. It seems more likely that at the present moment, blood DNA tests will be applicable to people where an FOBT is inappropriate due to bleeding benign lesions or as a second line rescue strategy for engaging those in screening who otherwise reject the faecal test.
The false-positive rate for the two-marker blood test provides insight into specificity and the factors that might influence it, and hence cost. Our observed specificity was 94–95 %, which was slightly better than the reported 91 % for
SEPT9 [
13]. Smoking, family history of CRC, gender and age were not significant predictors of assay positivity. There was no significant difference in DNA yields between non-CRC and cases with stage I-III cancers, however higher yields were observed for some stage IV cancers as reported previously [
12]. Further, we did observe an increase in assay positivity in non-neoplastic cases where recovered DNA exceeded 3ng/mL. Given the results of the technical assessment (Additional file
1), it seems likely that the false-positives (as determined by colonoscopy) reflect a true appearance of methylated
BCAT1 and
IKZF1 DNA. Longitudinal follow-up studies are required to understand whether the low false-positive rate in healthy cases reflects chance events (i.e. methylation of
BCAT1 DNA especially) of no consequence, or an early indication of colorectal neoplasia and/or other extra-colonic cancers.
The biological functions of
BCAT1 and
IKZF1 are not well understood, but both genes are involved in tumour growth and invasiveness [
30,
31]. Both genes have been demonstrated to be hypermethylated in several cancers including CRC [
10,
32]. Emerging data imply that
IKZF1 is a crucial player in proper regulation of proliferation and differentiation by controlling the activity of a small set of genes including notch [
33‐
36] which plays a crucial role in the self-renewing process of colon crypt stem cells [
37,
38].
The disappearance of circulating methylated
BCAT1 and
IKZF1 DNA after tumour resection in 10 of 12 cancer cases shows that detection of methylated
BCAT1 and
IKZF1 DNA in the blood reflects the presence of CRC rather than a risk of developing CRC. The half-life of free DNA in the blood is reportedly short at ~2 hours [
39], but 2 CRC cases remained positive for methylation even 5 months after resection. Longer follow-up is needed in the two cases with persisting methylation signal to understand the reason, as it is possible they were not cured of their cancer. Similar to observations made for other CRC methylation markers, these data suggest that the two-marker blood test may be useful to monitor tumour recurrence and adequacy of resection and/or initial therapy [
40].
There are several additional limitations with this study. The estimated sensitivities and specificities might not apply to screen-detected lesions, and comparison to other non-invasive screening tests has yet to be undertaken in this context. Actual test positivity rates in a true screening population cannot be reliably estimated from this study and so the consequences for colonoscopy follow-up rates are uncertain. As with all other DNA tests under consideration for CRC screening, how specific they are for colorectal as opposed to other organ cancers remains uncertain and long-term follow-up of false-positive cases is required.
Competing interests
Flinders Medical Centre and Academic Medical Centre received partial funding from Clinical Genomics Technologies Pty. Ltd (CGT). CGT provided salaries for LCL, SKP, RTB, AM and DHM and a consultancy fee for GPY. The specific roles of these authors are articulated in the author contribution section. LCL, RTB, AME and SKP are inventors on one or more patent applications covering the methylation DNA biomarkers described in this paper.
Authors’ contributions
SKP coordinated assay development, planned and documented the data plan, coordinated molecular testing, contributed to data analysis and manuscript preparation. ELS oversaw recruitment and collection of clinical data at the Australian hospital and contributed to data analysis and manuscript preparation. RTB, DHM and AME contributed to method development, optimisation and automation and provided qPCR experimental data. SCVD coordinated and managed recruitment at the Dutch hospitals. MWM contributed to recruitment, sample choice and provision. SRC contributed to conception of the study, sample choice and provision. GG and DM audited clinical data and verified case classifications. LCL provided ongoing input into data interpretation and project directions. ED contributed to conception of the study, clinical interpretation, sample choice and provision. GPY contributed to overall project design, clinical interpretation, sample choice and provision and manuscript preparation. All authors read and approved the final manuscript.