Identification of epigenetic DNA modifications with a protein nanopore

Abstract

Two DNA bases, 5-methyl cytosine (5mC) and 5-hydroxymethylcytosine (hmC), marks of epigenetic modification, are recognized in immobilized DNA strands and distinguished from G, A, T and C by nanopore current recording. Therefore, if further aspects of nanopore sequencing can be addressed, the approach will provide a means to locate epigenetic modifications in unamplified genomic DNA.

5-Methylcytosine (5mC) is the most common epigenetic modification in eukaryotic genomes. Roughly 5% of cytosines in the human genome are methylated, mainly at CpG sites,^1-3 and there is considerable variation in the pattern of methylation with cell type and state.^{4, 5} Changes in methylation patterns are required for development^{5, 6} and are associated with tumorogenesis and other disease states.² A second rarer epigenetic modification was reported recently: 5-hydroxymethylcytosine (hmC),^7-9 which arises from the oxidation of 5mC by the enzyme TET1.⁸ The role of hmC is under investigation. It has already been found to be enriched in the hippocampus and cortex of mouse brain,¹⁰ and hmC turnover has been implicated in embryonic stem cell maintenance and differentation.¹¹

Techniques have been developed to determine the sites of DNA methylation in genomic DNA, but recent developments that combine bisulfite chemistry with second generation sequencing is quickly supplanting them.⁴ Bisulfite treatment results in the conversion of C to U by deamination, while 5mC remains unchanged. In the new genome-wide approaches, genomic DNA is fragmented, treated with bisulfite, amplified and subjected to deep sequencing, revealing both the sites and extent of methylation. The complexity of the sample may be reduced by the enrichment of methylated sequences (e.g. with anti-5mC antibodies) or by the selective amplification of target sequences. The major disadvantage of this technique is incomplete conversion of C to U, which results in false positives. As yet, there are no reliable methods for the localization of hmC in a genome sequences.^12-14

Over the last 5 years, second generation sequencing has revolutionized genomics by dramatically increasing sequencing speeds and reducing costs.^15-17 Nevertheless, a third generation of sequencers based on single-molecule approaches is under development, which will have further advantages including greatly reduced cycle times for each base and the ability to read unamplified DNA.^{18, 19} Such devices based on fluorescence-detected sequencing-by-synthesis are now reaching the market.^{20, 21} 5mC and hmC have been detected indirectly by this means, by relying on changes in polymerase kinetics near a modified base, but the error rate is appreciable and multiple passes over a circular DNA template are required.²²

Nanopore sequencing is another promising third generation approach^{18, 23} with the potential to determine modified bases in unamplified genomic DNA. Two primary means for nanopore sequencing have been proposed.^{18, 24} In one of these, exo sequencing, a DNA strand is digested by a processive exonuclease to release nucleoside monophosphates that are identified by the nanopore.^{25, 26} Towards this end, nucleotides, including the 5mC monophosphate,¹⁶ have been identified with an α-hemolysin (αHL) protein nanopore equipped with a cyclodextrin adapter.^{25, 26} In a second proposed approach, strand sequencing, which is addressed in this paper, an intact DNA strand is fed through a nanopore and the bases are read off directly.^{23, 24}

At present, the speed at which DNA is translocated through protein nanopores is too fast for direct strand sequencing (around 1-3 μs per nucleobase)²⁷ and means to slow the DNA are under intense investigation.^28-30 In the interim, differences in ionic current arising from single nucleobase substitutions in immobilized ssDNA are being used to probe base identification.^31-34 Upon immobilization of the DNA within the αHL pore, the open pore current (I_O) is reduced to a new level (I_B). The residual current (I_RES) is expressed as a percentage of the open-pore currrent: I_RES = (I_B/I_O) × 100. When DNA is immobilized through a streptavidin·biotin complex, three recognition points within the β barrel of αHL, termed R₁, R₂ and R₃, are capable of distinguishing individual unmodified nucleobases in a DNA strand (Fig. 1).³³ The αHL E111N/K147N (NN) and E111N/K147N/M113Y (NNY) mutants have previously been shown to give superior discrimination between the standard bases (G, A, T and C) at R₂ and R₁, respectively.^{33, 34} The aim of the present study was to establish whether nanopore technology is able to discriminate epigenetic modifications within DNA strands. We now report conditions under which NNY can distinguish between the standard bases and 5mC and hmC, at R₁, which required the difficult and unprecedented separation of six distinct I_RES levels.

Fig. 1.

Oligonucleotides immobilized inside the αHL protein nanopore. Schematic representation of a DNA oligonucleotide (blue circles) threaded from the 5′ end and immobilized inside an αHL pore (grey, cross-section) through the use of a biotin (yellow)·streptavidin (green) linkage. The lumen of the αHL pore can be divided into two halves, each approximately 5 nm in length; an upper vestibule located between the cis entrance and the central constriction, and a fourteen-stranded, transmembrane, antiparallel β barrel, located between the central constriction and the trans entrance. The three nucleotide recognition points, R₁, R₂ and R₃ are indicated.³³ R₁ is located near the central constriction and is capable of recognizing bases at positions 8 to 11 (relative to the 3′-biotin tag, Fig. S1). R₂ is located near the center of the barrel and recognizes bases 13 to 16. The single nucleotide substitutions investigated in this study were made at position 9 (red circle) and position 14 (pink circle). Cytosine in genomic sequences can be epigenetically modified to form 5mC and hmC (right).

First, recognition point R₂ within the NN pore was probed with six poly(dC) oligonucleotides (X_R2) that contained 5mC, hmC, G, A, T and C substitutions at position 14 (relative to the 3′ biotin tag, Fig. S1a), a position known to occupy R₂. A second set of six oligonucleotides was obtained that contained in addition a G substitution at position 13 (XpG_R2), to emulate the CpG context in which 5mC, and presumably hmC, are most often found (Fig. S1b). Unfortunately, neither set of six oligonucleotides yielded the desired six distinct current levels, when elicited at +160 mV.

Similar sets of oligonucleotides were then obtained to probe recognition point R₁. The oligonucleotides in the first of these sets contained nucleobase substitutions at position 9 (Fig. 1) and were termed the 5mC_R1, hmC_R1, G_R1, A_R1, T_R1 and C_R1 (X_R1) oligonucleotides, while the oligonucleotides in the second set contained in addition a G nucleobase at position 8 and were termed the 5mCpG_R1, hmCpG_R1, GpG_R1, ApG_R1, TpG_R1 and CpG_R1 (XpG_R1) oligonucleotides (Fig. 2a). The NNY pore was probed with each set of oligonucleotides. Again, neither set produced six distinct current levels at +160 mV. However, the dispersion between the two most widely separated oligonucleotide blockade levels in the XpG_R1 set (Fig. S2d) was larger than for the X_R1 set (Fig. S2c) and only the GpG_R1 and ApG_R1 oligonucleotides gave overlapping levels (Fig. S2d). In an attempt to enhance the separation between these two sequences, applied potentials from +100 mV to +180 mV were examined and it was discovered that discrimination was improved at +110 mV. Indeed, at this potential, each XpG_R1 yielded a distinct current level (Fig. 2). Under these conditions, when immobilized within NNY, the difference in residual current levels (ΔI_RES) between the 5mCpG_R1 oligonucleotide and the CpG_R1 oligonucleotide is −8.3 ± 0.1% (n = 4) (ΔI_RES = I_RES ^5mCpGR1 – I_RES ^CpGR1). It is therefore possible to distinguish two bases in a DNA strand that differ only in a methyl group at the C5 position (Fig. 2). The hmCpG oligonucleotide also produced a distinct current block. The difference in the residual current levels between hmCpG_R1 and CpG_R1 is ΔI_RES = −6.1 ± 0.1% (n = 4). While the molecular basis for the observed differences in I_RES levels is unclear, the current levels for hmCpG and 5mCpG are distinct from all other bases.

Fig. 2.

The NNY pore discriminates between modified nucleotides at R₁. (a) Histogram (top) of residual current levels for NNY pores interrogated with six DNA strands (middle) that differ at only one position X (bold), where X represents G, A, T, C, 5mC or hmC. Gaussian fits were performed for each peak, and the mean value of the residual current for each oligonucleotide (with the standard deviation) is tabulated (bottom). The result displayed is from a typical experiment carried out at an applied potential of +110 mV. See Table S1 for full list of experimental results. (b) Typical current levels for the NNY pore when blocked at +110 mV with ApG_R1 and GpG_R1 oligonucleotides (left, filtered at 50 Hz for display). Event histogram (right) displaying the subset of residual current levels produced by ApG_R1 and GpG_R1 oligonucleotide blockades. The shaded area is the region where the fitted Gaussians overlap and comprises 7.6% of the area of the two Gaussians (see also Fig. S3).

Present techniques for mapping 5mC are tedious and usually yield a consensus methylation pattern, while the actual pattern may vary from cell to cell. By using nanopore technology, we have demonstrated that it is possible to distinguish directly all four standard bases, 5mC and hmC in individual ssDNAs. Therefore, if this technology can be integrated into a practicable nanopore device, it will benefit fundamental studies in epigenetics and medical diagnostics. In particular, the absence of bisulfite chemistry and DNA amplification will reduce errors, and there is the potential for long sequence reads that will facilitate methylation haplotyping.³⁵

Supplementary Material

01

Table S1. The open pore (I_O) and residual current levels (I_RES) produced when NNY pores are blocked with the XpG_R1 oligonucleotides, where X represents G, A, T, C, 5mC or hmC at position 9 (Fig. 2a, main text). ΔI^RES is defined as I_RES ^XpGR1 – I_RES ^CpGR1.

Fig. S1 The chemical structure of the biotin-TEG linker used to biotinylate the 3′ terminus of the DNA oligonucleotides. The structure was produced with ChemBioDraw Ultra 11.

Fig. S2 Initial investigations to determine the ability of the αHL pore to discriminate between modified bases in immobilized oligonucleotides. The histograms shown are from typical experiments (each conducted at least 3 times), where the DNA was captured at +160 mV. The left-hand column displays I_B values and the right-hand column I_RES values. (a) NN pores probed with X_R2 oligonucleotides. (b) NN pores probed with XpG_R2 oligonucleotides. (c) NNY pores probed with X_R1 oligonucleotides. (d) NNY pores probed with XpG_R1 oligonucleotides.

Supp Fig s3

Fig. S3 Gaussian fits to the ApG and GpG histogram populations from an experiment with the NNY pore and oligonucleotides XpG_R1 at +110 mV. We determined the overlap between the ApG_R1 and GpG_R1 populations from Gaussian fits to the histogram peaks. The data are binned at 0.07% I_RES, and a double Gaussian function is fitted to the data, with a constraint fixing the y-offset to zero. (a) From the Gaussian fits, we calculated that the two distributions share an overlapping area (shaded) of 7.6% of the total area of the two peaks. (b) For a more meaningful measure of the ability to accurately call nucleobases, we also calculated that the region where nucleotides cannot be called with more than 95% confidence (shaded) contains ~17% of the total events.