Detection and Characterization of Single Cisplatin Adducts on DNA by Nanopore Sequencing.

Detection and characterization of an individual cisplatin adduct on a single DNA molecule is a demanding task. We explore the characteristic features of cisplatin adducts in the nanopore sequencing signal in aspects of dwell time, genome anchored current trace, and basecalling accuracy. The offset between the motor protein and the nanopore constriction region is revealed by dwell time analysis to be about 14 bases in the nanopore device as we examined. Characteristic distortions due to cisplatin adducts are illustrated in genome anchored current trace analysis, constituting the fingerprint for identification of cisplatin adduct. The sharp increase in odds ratio at the location of adducting sites provides additional feature in the detection of the adduct. By these combined methods, single cisplatin adducts can be detected with high fidelity on a single read of the DNA sequence. The study demonstrates an effective method in the detection and characterization of single cisplatin adducts on DNA at the single-molecule level and with single nucleotide spatial resolution.

Introduction

Cisplatin is an effective, widely used drug for many cancers, which shows a high level and broad spectrum of antitumor activities.[1−6] It has been known that the binding between this drug molecule and the DNA molecule disables the physiologic activities of DNA and eventually induces cell apoptosis. Great efforts have been made to probe and to understand the interaction between the cisplatin molecules and DNA and how the cisplatin adduct affects the DNA duplication in the cell.[3,7] Many experimental methods such as nuclear magnetic resonance, optical/magnetic tweezers,[8] and atomic force microscopy[9,10] have been employed to probe the cisplatin–DNA interactions at the single-molecule level. Besides these methods, nanopore is an emerging new experimental tool that is capable of detecting single DNA molecule as well as its modifications.[11−15] For cisplatin–DNA interaction, Zhou et al. employed a solid-state nanopore for the real-time monitoring of DNA configuration and revealed three distinct stages of the interaction process.[7] Further understanding on the dynamic behavior of cisplatin–DNA interaction requires accurate information on the location and quantity of cisplatin molecules adduct bonded to a single DNA molecule. Spatial resolved detection and characterization of individual cisplatin adducts along a single DNA molecule, however, has not been reported in previous reports.

Recent development in sequence-based nanopore techniques has made it possible to detect the subtle modification along the DNA strand with single-base resolution. For example, sequencers from Oxford Nanopore Technologies (ONT) have been employed in not only DNA[16,17] and RNA sequencing[18,19] but also DNA methylation detection,[20−22] discrimination in adducts size, regiochemistry, and functional groups.[23] This unique experimental tool opens an additional door toward the characterization of cisplatin–DNA interaction at the sequence level. Here, we report the study on the detection and characterization of single cisplatin adducts on DNA molecules by using ONT’s nanopore sequencer MinION. The presence of the drug molecule shows a significant influence not only on the effective cross section of the nucleotide but also on the DNA unwinding process. As a result, the dwell time is slowed down by the presence of cisplatin adducts, and deeper current blockade occurs when cisplatin-bond nucleotide enters the nanopore constriction region. In addition, the perturbated current level reduces the accuracy in basecalling. The comprehensive analysis of these effects provides a fingerprint of cisplatin adduct on DNA with base-resolved spatial resolution.

Results and Discussion

Experimental Setup and Sample Information

The schematic diagram for nanopore[24] sequencing is shown in Figure a. Under an external drive voltage, a double-stranded DNA is unwound by a motor protein, and then, one of the two strands transit through the nanopore. As the single-stranded DNA (ssDNA) passes through the constriction region of the nanopore, variation in base size is imprinted in the current signal. The raw current signal is then basecalled to reveal the DNA sequence and other information for further process, as shown in Figure b.

Figure 1

Experimental setup and analysis methods in this study. (a) Simplified diagram for nanopore DNA sequencing. (b) Analysis pipelines as performed. The DNA sequence, as well as information on model state and move, is basecalled from raw signal. Then, dwell time analysis is performed with a homebuilt algorithm. Genome anchored current trace analysis is acquired by program Tombo. ESB and odds ratio are obtained by program Eligos2. The three analysis methods can be performed in parallel.

Four DNA samples have been used in our study. The first DNA sample is 2449 bp long, with cisplatin bonded with two adjacent guanines at the site of 2386–2387 bp. The second sample is the same DNA just without cisplatin. The third DNA sample is 2454 bp long, with a similar sequence as the first sample and with cisplatin bonded at the site of 2414–2415 bp. The fourth sample is the same DNA as the third sample, without cisplatin adduct. In the first sample, the cisplatin molecules adduct to the strand of the DNA that transits through the nanopore. As a comparison, the cisplatin-adducted ssDNA of the third sample does not transit through the nanopore, its complementary strand is sequenced instead. The second and fourth samples are used as control groups. For detailed sequence of the DNA samples and sample preparation method, please see Methods section.

Data analysis plays more and more important roles in single-molecule analysis based on nanopores.[25−27] In this study, three analysis methods are employed to characterize the effect of cisplatin adducts on the DNA sequencing results, with details explained as following.

Dwell Time

First, the dwell time for each base during sequencing is computed by summarizing the move delay in the basecalled results. Since the DNA translocation is controlled by the motor protein attached to the nanopore entrance, the dwell time for each base is associated with the DNA unwinding time. A homebuilt LabVIEW program is used to load the basecalling results and compute the dwell time for each base in the standard sequence.[28]

Figure shows the dwell time as a function of the sequence for each sample. Each curve is averaged over 4000 reads of the basecalled current signals, and the standard error is below 0.1 ms. Figure a presents the result for the first and second samples. The two curves are almost the same for most of the sequences. Evident difference, however, emerges in the last hundred bases, as shown in the zoom-in plot of Figure b. The dwell time for the cisplatin-adducted ssDNA (first sample, red line) is apparently higher than that for the control group (second sample, black line). Especially at the site of 2372 bp, the dwell time increases sharply by fourfolds, as marked by the arrow in Figure b. The site of 2372 bp is about 14 bp before the binding position of cisplatin molecule, which is at the range of 2386–2387 bp. This difference reveals the offset distance between nanopore current sensing region and the motor protein, which is consistent with the estimation from physical model structures. The influence of cisplatin on the DNA unwinding process is clearly illustrated here. The overall increase in dwell time around the binding site is due to the increased viscous force resulting from the presence of cisplatin adduct. As a contrast experiment, Figure c,d shows the dwell time for the third and fourth samples. There are still some slight variations between the two curves around the cisplatin bonding site at 2414 bp, but the difference is much smaller as compared with that in Figure a,b. This is reasonable as the cisplatin-adducted strand does not go inside the motor protein, generating much less obstruction to the unwinding process. We note that the dwell time starts to drop in the last 20 bases. This is because that the drag force becomes less as it approaches the end of translocation, resulting in an increase in speed.[29] The length of accelerated region is consistent with the Kuhn length.[29]

Figure 2

Dwell time during DNA sequencing. (a) Dwell time for each base along the cisplatin-adducted ssDNA (test group, red line) and the nonadduct ssDNA (control group, black line). The DNA sample is 2449 bp long, and the cisplatin is attached at the site of 2386–2387 bp. (b) Zoom-in of figure (a) in the range of 2200–2449 bp. There is a sharp peak at the 2372 bp site, as marked by red arrow. (c) Dwell time for each base along the ssDNA with cisplatin attached to the other chain (test group, red line) and the nonadduct ssDNA (control group, black line). The DNA sample is 2454 bp long, and the cisplatin is attached at the site of 2414–2415 bp. (d) Zoom-in of figure (c) in the range of 2200–2454 bp. The contrast is much smaller than that in (b).

Genome Anchored Current Trace Analysis

Second, a genome anchored plotting is generated to investigate the signal deviation. The calculation is performed by software Tombo, a suite of useful tools for the analysis and visualization of raw nanopore signal. It has been employed to detect base modifications in DNA sequences like DNA methylation.[20−22] The raw current signal can be normalized and assigned to each base of the sequence by a resquiggle algorithm in Tombo. Figure a shows the resquiggled signals of individual reads derived from the first and second samples, overlaid for comparison. The horizontal axis is the sequence of the ssDNA that transits through the nanopore. The signal for each base corresponds to the third base in a 5-mer model state (see k-mer signal level plot in the Supporting Information). For example, the resquiggled signals at 2380 for base “C” are associated with the 5-mer model state “CTCCT.” The test group (red line) and control group (black line) exhibit clear difference around cisplatin binding sites between base positions of 2384 and 2390. The cisplatin-adducted bases “GG” are marked by red stars. Quantitative analysis shows that the offset and standard deviation in the signal are most significant on the cisplatin-adducted sites. Figure b shows the statistics in the resquiggled current traces for the bases from 2384 to 2391 bp, for both the test group and the control group. Adduct of cisplatin results in deeper current blockade and higher standard deviation on sites of 2386–2387 bp in the test group, as compared to that in the control group. There are also some influences on the sites nearby the adducted bases. The spreading of the influence is partly due to the reason that the current blockade is indeed determined by several DNA bases within the nanopore constriction region. In addition, the cisplatin molecules adduct to two adjacent G bases, resulting in significant conformation changes in the adducted G bases, as discussed later.

Figure 3

Genome anchored current traces. (a) Genome anchored current traces for the 2449 bp DNA samples (the first and second samples) obtained by the resquiggle process. Red lines represent current signal of cisplatin-adducted samples (test group). Black lines represent samples without cisplatin (control group). Red stars indicate the cisplatin adduct sites. (b) Statistics in resquiggled current traces for specific bases of interest in (a). (c) Genome anchored current traces for the 2454 bp DNA samples (the third and fourth samples). (d) Statistics in resquiggled current traces for specific bases of interest in (c).

As a comparison, the resquiggled signal and analysis for the third and fourth samples show no sign of cisplatin adduct, as shown in Figure c,d. This is reasonable since the cisplatin does not pass through the nanopore with the DNA strand in the third sample.

Error at Specific Base and Odds Ratio

Third, the influence of cisplatin on basecalling accuracy is characterized by two terms, i.e., error at specific base (ESB) and odds ratio. The distorted ionic current signals give raise to a sequencing error following the basecalling algorithm. ESB is defined as the frequency of the sum of substitutions, insertions, and deletions of individual positions over the total mapped reads obtained from read alignment results based on the reference sequence. Odds ratios for all nucleotide positions were computed with Fisher’s exact test, comparing the error of reads derived from cisplatin-adducted DNA with that derived from the corresponding control sample. Computer program Eligos2 is used to compute the sequencing errors in individual bases and to compare the differences in error fractions, producing odds ratios for individual nucleotide positions.[23,30]

Figure a shows the ESB of the first and second samples in the range of 2384–2394 bp. The ESB in the test sample is evidently higher than that in the control sample, especially for the sites of 2388–2390 bp. The odds ratio, as shown in Figure b, presents a clear peak on the site of 2389 bp, with a peak value of 110 over a background fluctuation below 5. It shows an error in basecalling that spans three nucleotides. The center position of the peak is 2–3 bp behind the known adduct sites (2386 and 2387 bp). This lagging effect means that the presence of adduct affects the basecalling result by a few bases later, which is likely due to the basecalling algorithm. As a comparison, there is no distinct difference in ESB between the third and fourth samples, as shown in Figure c. This is consistent with the result shown in Figure c,d. As a result, there is no clear peak in the plot of odds ratio, as shown in Figure d. As described in ref (23), current signal distortion induced by DNA adducts are visualized as the π values and called as differential ionic signal (DIS) plots. Figure e,f shows DIS plots for the 2449 and 2454 bp samples, respectively. A clear peak emerges on the site of 2382 bp in Figure e, indicating that cisplatin influences mostly current signal on the site. On the contrary, there is no difference in Figure f.

Figure 4

Odds ratio, ESB, and DIS plots. (a,b) Odds ratio for the 2449 and 2454 bp samples, respectively. Insets show the zoom-in plot near the adduct sites. A clear peak emerges on the site of 2389 bp in (a). (c,d) Radar plots displaying ESB of the four samples. The test samples are plotted in red, and the control samples are plotted in black. The positions of cisplatin adduct are marked with red stars. (e,f) DIS plots for the 2449 and 2454 bp samples, respectively. A clear peak emerges on the site of 2382 bp in (e).

Discussion

In our experiment, the cisplatin adducts to the two adjacent G–C base pairs forming a 1,2-d(GG) intrastrand cisplatin cross-link, with Pt atom bonds to the N7 atom of purine bases in the adjacent guanines, as shown in Figure . The energy of activation for guanosine substitution with cisplatin is about 18 kcal/mol, and for GG adduct closure, the energy is about 21 kcal/mol.[31,32] Such strong Pt–N bonding ensures that cisplatin transits with the ssDNA through the nanopore in the first sample. There are several distorted configurations as proposed by theoretical computations, showing that the cross-link causes distortion of the adjacent G–C base pairs.[33−38]Table shows the published bond length for all six hydrogen bonds in the normal and the distorted configurations. The hydrogen bond length in a normal DNA molecule without cisplatin adduct is about 1.8 ± 0.1 Å. When a cisplatin adduct occurs, on average, the hydrogen bonds are weakened with some of the bonds elongated to beyond 2.0 Å. Figure c shows one of the 3D configurations of cisplatin-adducted G–C base pairs. The cisplatin adduct affects the DNA unwinding process if the distorted G–C base pairs enter the motor protein. The increased dwell time around the sites of cisplatin adduct may be due to two possible reasons: distorted hydrogen bonds and size effect. But considering the fact that the hydrogen bonds in the cisplatin-adducted G–C base pairs are weaker on average than that in the normal configuration, the distorted hydrogen bonds are not likely the reason to account for the increased dwell time on the corresponding sites. It is thus plausible that the increased size as well as the increased rigidity in cisplatin-adducted guanines produces strong perturbation to the unwinding process.

Figure 5

Model structure of cisplatin adduct to DNA. (a) Plan view of a cisplatin adduct to the two adjacent G–C pairs forming a 1,2-d(GG) intrastrand cisplatin cross-link. Cisplatin is colored in green. There are six hydrogen bonds between G and C pairs, as numbered from one to six. (b) 3D conformation of adjacent G–C pairs without cisplatin adduct. (c) 3D conformation of 1,2-d(GG) intrastrand cisplatin cross-link. Cisplatin is drawn in green. This model is the solution structure from ref (37).

Table 1

Published Hydrogen Bond Length in Unit of Åa

name	HB-1	HB-2	HB-3	HB-4	HB-5	HB-6
normal	1.798	1.871	1.802	1.802	1.870	1.798
2npw	2.096	1.821	1.451	1.580	1.722	1.863
2nq0	2.022	1.826	1.510	1.614	1.743	1.879
1a84	1.905	1.945	1.846	2.341	2.146	1.906
1ksb	1.916	1.851	1.911	1.943	1.702	1.793
1au5	1.718	2.038	1.873	1.969	1.702	1.793
3lpv	3.075	2.849	2.640	2.949	2.790	2.932
1aio	2.738	2.804	2.680	2.770	2.735	2.690

The names in the first column are Protein Data Bank access numbers.

Based on the results of this study, a primary scheme can be developed for the detection of single cisplatin adduct in a single DNA sequence read, especially for the 1,2-d(GG) intrastrand cisplatin cross-link. First, the location of cisplatin adduct can be statistically determined by the profile of dwell time and/or odds ratio. The actual cisplatin binding site is 14 bp after the location of the peak center in the profile of dwell time and is 2–3 bp before the peak center in the odds ratio profile. As shown in Figures and 4, the peaks in both profiles are very sharp, with peak width of a few nucleotides. The spatial resolution in determining the position of cisplatin adduct is about one nucleotide for both methods. Second, disturbances in ionic current signal can serve as fingerprint to identify the characteristics of the cisplatin adducts. Figure a shows the receiver operating characteristic (ROC) curve displaying the ability to discriminate between the adduct and control sequence based on current levels at individual sites. Figure b shows the p values derived from Eligos2 for different mixtures of various percentages of reads from adduct-containing samples in the presence of 10 000 reads from the control sample. We estimated the detection level of an individual cisplatin adduct by calculating p value using Fisher’s exact test by mixing reads for cisplatin-adducted DNA with those of the control at different percentages of cisplatin adducts. It turns out that it is possible to detect a cisplatin adduct as low as 2.5% at the p value cutoff of 0.05. Compared to previous single-molecule experimental methods, the nanopore-based method as demonstrated in this study has advantages of high detection sensitivity, base-resolved spatial resolution, and high device portability. While the accuracy in nanopore sequencing is still under improvement, the parallel acquisition with multiple channels generates massive data traces that can be used to squeeze the noise down to a very low level.

Figure 6

Precision for the detection of cisplatin adducts. (a) ROC curve displaying the ability to discriminate between the cisplatin adducts and control sequence based on different bases as indicated in the inset legend. (b) p value derived from Eligos2 for different in silico mixtures of various percentages of reads from adduct-containing samples in the presence of 10 000 reads from the control sample.

Conclusions

In summary, the cisplatin adduct on DNA has been detected and characterized by nanopore sequencing. Dwell time analysis reveals an offset of 14 bases between the motor protein and the nanopore constriction region. Genome anchored plots illustrate current signal perturbation by the presence of cisplatin adducts. Characteristic distortions, visualized in ESB profiles and DIS plots, constitute a fingerprint for the identification of cisplatin adduct. The location of adducting sites can also be measured by the sharp increase in odds ratio. Combining these paralleled analysis methods, it is possible to detect cisplatin adduct on a single read of DNA sequence with high fidelity. The results, as well as the strategy and analysis methods employed in this study, are useful for similar research tasks.

Methods

Sample Preparation

Oligonucleotides containing a unique 1,2-intrastrand cross-links were constructed as previously described with minor modifications.[39,40] Briefly, the G-residues within the 5′-TTTCTTCTCTTTGGTTCTTCCTC-3′ oligonucleotide (shown in bold for clarity) were cross-linked by incubating the oligonucleotide with activated cisplatin for 20 min at 37 °C in the dark. After purification, the site-specifically modified oligonucleotides were allowed to anneal to their complementary strands. The DNA duplexes contained 1,2-intrastrand cross-links were allowed to ligation with a 30 bp DNA duplex and 2.5 kb DNA fragment with adhesive overhangs.

1D Ligation Sequencing

Nanopore experiments were carried out with ONT’s sequencer MinION, flow cell version FLO-MIN106, and ligation kit SQK-LSK108. Single-end sequencing was implemented by designing single dA cohesive end of the sample. 1D ligation sequencing was conducted according to ONT’s protocol. Raw data were recorded by ONT’s MinKNOW software and basecalled locally by Albacore version 2.2.6. Basecalled data were resquiggled with Tombo version 1.5 as provided by ONT.

Sequence Alignment

In dwell time analysis, we utilized a normal algorithm, written in LabVIEW subVI. Detailed implementation of this alignment algorithm will be discussed elsewhere. Minimap2 is used for sequence alignment in calculating ESB and odds ratio in Eligos2.

ESB and Odds Ratio

ESB and odds ratio were acquired by Eligos2 open-source software, version 2.0.0.[23,30] Detailed process procedure can be found on the website.

Structure Visualization by VMD

CsgG nanopore in Figure a is obtained from protein data bank.[24] Cisplatin binding structures are downloaded from protein data bank.[37] 3D visualizations are generated by VMD software.[41]