5hmC residues in DNA samples extracted from colon tissue and peripheral blood cells were fluorescently labelled by the two-step labelling reaction (Fig. 1). Labelled samples were then stretched on glass slides and imaged using a fluorescence microscope. DNA molecules appear as extended red lines dotted with yellow/green spots indicating 5hmC sites (Fig. 2a, b). Finally, 5hmC level in the samples was determined using image analysis (Fig. 2c), and the statistical validity of the data was assessed (Fig. 2d, e).
The main advantage of the above described assay is its extreme sensitivity, limited only by the diffraction limit of the optical imaging system which is being used. A standard fluorescence microscope equipped with an oil immersion objective will typically allow distinguishing between two adjacent fluorophores separated by 200–300 nm (300–1000 bp apart). Since close by 5hmC residues may appear as a single fluorescent spot, we quantified the statistical distribution of clustered 5hmC sites by measuring photobleaching steps (see Additional file 1: Figure S1). The majority of the analyzed 5hmC sites contained a single fluorophore, implying that isolated observable marks are comprised of mostly single 5hmC residues. Using the full photobleaching step distribution enabled us to correct for 5hmC clustering in the calibration process. The reproducibility of this assay was verified by repeating the labelling and imaging process two to four times for each of the analyzed samples, with an average standard deviation of 5% per sample (see Additional file 1: Tables S1, S2).
Colorectal cancer
Colon biopsies are commonly obtained during regular screening colonoscopy procedures. The small amount of sample needed to perform this assay allowed us to assess 5hmC levels using down to 50 ng of DNA. Significant decrease in 5hmC level is observed in CRC tissue (0.0028% 5hmC/dNTPs, n = 7) compared with healthy colon tissue (0.0059% 5hmC/dNTPs, P < 0.0005, n = 7). In addition, the 5hmC level in adjacent colon tissue taken from the same CRC patients (0.0043% 5hmC/dNTPs, n = 7) is higher than that found in tumor tissue, yet it is significantly lower than the level of colon tissue from a healthy individual (P < 0.05) indicating that staging of disease may be possible (Fig. 3a, b).
Blood cancer
Blood is the most accessible tissue for diagnostic applications. However, blood exhibits extremely low levels of 5hmC, a fact that has made it a challenging tissue for 5hmC-based studies. The sensitivity of our method allowed us to clearly differentiate between healthy and malignant blood samples. Significant decrease in 5hmC level is observed in samples of all three types of blood cancer tested, namely CLL (0.0019% 5hmC/dNTPs, P < 0.0001, n = 8), ALL (0.0031% 5hmC/dNTPs, P < 0.05, n = 3), and MM (0.0023% 5hmC/dNTPs, P < 0.005, n = 4), as compared with healthy blood samples (0.0042% 5hmC/dNTPs, n = 11, Fig. 3c, d). We note that peripheral blood samples from lung cancer and CRC patients were also analyzed. No correlation was found between 5hmC levels in these blood samples and the state of disease, indicating that, as expected, the use of 5hmC as a biomarker for cancer is tissue-specific.
Assessment of commercial kits
In order to compare our method with existing techniques, we have tested two commercial ELISA kits, which are of the most cited immuno-based methods for 5hmC quantification (Quest 5-hmC™ DNA ELISA Kit (ZYMO research) and MethylFlash™ Hydroxymethylated DNA Quantification Kit (EPIGENTEK)).
We found that the 5hmC levels measured by the Quest 5-hmC™ DNA ELISA Kit are extremely sensitive to the DNA fragment size in the analyzed sample. This is likely due to the use of anti-DNA antibody as a reporter (hence, longer DNA fragments result in stronger signal regardless of the 5hmC content). For comparison, we have measured two of the samples used in the single-molecule experiment (DNA extracted from adjacent colon tissue and healthy peripheral blood). The samples were analyzed as received (30 kbp average fragment length) and after fragmentation on a Covaris S220 focused-ultrasonicator instrument (500 bp average fragment length). As evident from the results presented in Fig. 4a, the 5hmC level for the same sample is reduced three to sixfold after fragmentation. These results imply that while different samples with similar fragment sizes may be qualitatively compared for their relative 5hmC content, an unbiased, quantitative measurement using this kit is limited.
The MethylFlash™ Hydroxymethylated DNA Quantification Kit has shown less sensitivity to DNA fragment length; however, the 5hmC levels obtained with this kit were inconsistent with previously published results as well as the results presented in our study, which were validated with HPLC-MS results (see Additional file 1: Figure S2). Specifically, we measured the 5hmC level of colon DNA and peripheral blood DNA samples (healthy peripheral blood (n = 1), CLL (n = 1), healthy colon (n = 2), adjacent tissue (n = 2), and CRC (n = 2)). The 5hmC level of the CLL sample measured with the MethylFlash kit was 20-fold higher than the 5hmC level of the healthy blood sample (Fig. 4b), most likely due to limit of detection issues. These results are inconsistent with our findings, nor are they in line with known 5hmC levels of peripheral blood tissue [27, 28]. Furthermore, the 5hmC level of colon adjacent tissue appeared higher than the 5hmC level of healthy colon tissue (Fig. 4c), in contrast to the results obtained by our assay and in disagreement with other studies [6, 10, 24].
Altogether, our findings suggest that the tested ELISA kits for 5hmC quantification have several limitations, including DNA-length dependency, low sensitivity, and poor reproducibility.