Nanopore sequencing of an expanded genetic alphabet reveals high-fidelity replication of a predominantly hydrophobic unnatural base pair

Abstract

Unnatural base pairs (UBPs) have been developed and used for a variety of in vitro applications, as well as for the engineering of semi-synthetic organisms (SSOs) that store and retrieve increased information. However, these applications are limited by the availability of methods to rapidly and accurately determine the sequence of unnatural DNA. Here, we report the development and application of the MspA nanopore to sequence DNA containing the dTPT3-dNaM UBP. Analysis of two sequence contexts reveals that DNA containing the UBP is replicated with an efficiency and fidelity similar to that of natural DNA and sufficient for use as the basis of an SSO that produces proteins with non-canonical amino acids.

Graphical Abstract

We have created semi-synthetic organisms (SSOs) that store and retrieve the information made available by a family of unnatural base pairs (UBPs), typified by dTPT3-dNaM (Figure 1A). The SSOs replicate the UBPs contained within plasmid DNA^1–3 or chromosome DNA⁵ and then retrieve that genetic information via transcription and translation to produce proteins with site-specifically incorporated non-canonical amino acids (ncAAs).^3,6 UBPs are also being developed by the Hirao and Benner groups to expand the chemical space available to functional nucleic acids.^7–8 In addition, there are many other potential applications, ranging from information storage^9–10 to the development of novel materials.^11–12 However, these applications would all be facilitated by improved sequencing methods.

Figure 1.

dNaM and dTPT3 produce unique current patterns. (A) The dNaM-dTPT3 UBP. (B) The MspA/Hel308 nanopore system. The MspA porin (gold) is imbedded in a lipid bilayer (purple). The Hel308 helicase (green) draws ssDNA through the porin as a variable voltage⁴ is applied across the membrane and current is measured. (C) Consensus current patterns from 40 control measurements of the indicated sequences. Shaded areas represent ± 1 S.D. (Figure S2), demonstrating that single-molecule measurements of dNaM and dTPT3 can be identified with high confidence. Typical SNP calling accuracies using variable-voltage sequencing are >90% for the 8 hypotheses (the 6-letter alphabet plus resulting abasic site and single nucleotide deletion, Figure S5).

Methods to detect UBPs within DNA include protection from nuclease cleavage,¹³ induced strand cleavage,¹⁴ or labeling one of the unnatural nucleotides with a fluorophore or an affinity tag.^15–16 Signal termination during Sanger sequencing has been employed to sequence DNA, either when the cognate unnatural triphosphate is not present¹⁷ or when it is replaced with its dideoxynucleoside analog.¹⁸ Condition-dependent forced loss of a UBP has been used to indirectly sequence DNA containing UBPs,^7–8 but this method is labor intensive and possibly dependent on sequence context. Moreover, with all of the developed methods, determination of which unnatural nucleotide is present in a given strand is more challenging. This may be a particular concern with predominantly hydrophobic UBPs, such as those developed in the Romesberg and the Hirao groups, due to the possibility of unnatural nucleotide self-pairing (e.g. dTPT3-dTPT3 or dNaM-dNaM) and subsequent inversion of UBP strand context (i.e. 5′AXC to AYC, X=dNaM and Y=dTPT3). In fact, it has been suggested that rampant inversion of the dTPT3-dNaM UBP makes it unsuitable for expansion of the genetic alphabet.¹⁹

Next generation sequencing (NGS) has revolutionized the biological sciences. The most common NGS methods rely on sequencing-by-synthesis, but their application for the sequencing of DNA containing UBPs would require both modification of the unnatural nucleotides and significant modification of existing instrumentation. In contrast, nanopore sequencing is a potentially transformative, emerging NGS technique that obviates the need for nucleotide or instrumentation modification. The nanopore sequencing method employed in this work functions by recording an ionic current through the protein nanopore MspA as the helicase Hel308 controls the transit of DNA through the pore (Figures 1B and S1).^20–22 Hel308 moves DNA through MspA with two steps per nucleotide, such that the translocation of a single nucleotide through the pore is seen as two distinct ion-current segments²³ (Figure 1C). Because the ion-current signal is generated by the physical motion of ions past the nucleobases, nanopore sequencing is sensitive to not only the structure of the natural nucleotides but also to nucleotide modifications such as methylation^24–26 and to the unique structures of unnatural nucleotides. This has been demonstrated through direct detection of the dNaM:d5SICS pair with the MspA nanopore,²⁷ and marker-based detection of dNaM with the α-Hemolysin nanopore.²⁸ Here we use a recently developed MspA/Hel308 nanopore sequencing system that features a time-varying voltage⁴ to assess the replication fidelity of the 6-letter genetic alphabet consisting of the four natural letters, dTPT3, and dNaM.

To begin the development of a nanopore sequencing method for our expanded genetic alphabet, a reference library of consensus current patterns was generated empirically using chemically synthesized oligonucleotide standards (see Supporting Information and Table S1). 65-mer standards were synthesized and measured with the MspA/Hel308 nanopore that possessed a centrally located AXC or GXC sequence or a mutational derivative: dNaM deleted or replaced with dTPT3, a natural nucleotide, or a tetrahydrofuran abasic site analog. Each sequence’s consensus current pattern was generated by averaging the patterns of 40 unique measurements and was found to be unique from the mutationally related sequences (Figures 1C, S3, and S4). A single-nucleotide polymorphism (SNP) calling algorithm was developed from this reference library to determine the identity of the central nucleotide in the UBP site. The algorithm correctly matches single-molecule current patterns with SNP identity with >90% accuracy for most sequences (see Figure S5 and Supporting Information).

Given the importance of PCR as a tool for both constructing and analyzing DNA containing UBPs, we first sequenced DNAs produced from an unnatural PCR (uPCR, i.e. PCR supplemented with dTPT3TP and dNaMTP and with templates amplified by OneTaq, a mixture of Taq and DeepVent polymerases; see Supporting Information) templated by the same AAC, AXC, or GXC 65-mer oligonucleotides used to develop the sequence calling algorithm. Templates were amplified via uPCR. Amplification levels were estimated by PAGE analysis of control uPCRs (Supporting Information and Figure S8), and the amplicons were analyzed by streptavidin-shift PCR and nanopore sequencing after 13 doublings (Figure 2, Table S3, and Supporting Information). The mis-calling rate of the SNP calling algorithm limits our ability to detect rare mutations, generally SNPs that comprise less than 2% of a given sample. Thus, in order to improve our ability to detect rare mutation events during uPCR, uPCR products were diluted and reamplified sequentially four times for a total of 65 DNA doublings and analyzed as above.

Figure 2.

Mutation spectrum of AAC, AXC, and GXC. Left axis and dots show % Retention_A, as measured by streptavidin-shift PCR, with white dots representing measured retention for each experimental replicate. Black dots represent average retention with ± 2 S.D. error bars, n=3. Right axis and stacked bar plots show the average percent abundance of each possible single nucleotide polymorphism (% of SNPs (Table S3)), as measured by nanopore sequencing.

Fidelity was determined as the percent retention of the central A or X within the AAC, AXC, or GXC sequence, normalized by the number of DNA doublings (Supporting Information). The fidelity of the AAC template amplified via standard PCR was 99.62±0.05% per doubling (average±s.d) after 13 doublings and 99.84±0.04% after 65 doublings. When amplified by uPCR, the fidelity was slightly lower (99.20±0.13% after 13 doublings and 99.65±0.07% after 65 doublings), due to low level misinsertion of dTPT3TP. For the AXC template amplified by uPCR, the fidelities were 99.45±0.09% after 13 doublings and 99.39±0.16% after 65 doublings. Finally, for the GXC template amplified by uPCR, the fidelities were 99.30±0.53% after 13 doublings and 99.54±0.17% after 65 doublings. While these results reveal that uPCR produces DNA with a fidelity that is slightly reduced relative to natural PCR, the fidelity with which the UBP is replicated via uPCR is comparable to that of a natural base pair. Additionally, the observation that the fidelity with which the AXC and GXC templates were replicated remained constant with increased amplification demonstrates that natural DNA is not significantly more efficiently replicated than DNA containing the UBP.

For the AAC template amplified by uPCR, deletion of dA and mutation to dTPT3 (dA→dTPT3) were the most frequent mutations at 13 doublings (constituting 1.9±0.2% and 1.9±1.7% of the amplicons, respectively) and dA → dC (4.4±2.6%) and dA → dTPT3 (8.9±8.1%) at 65 doublings. For the AXC template, dNaM → dTPT3 (2.8±1.2%) and dNaM → dA (2.2±1.1%) were the most frequent mutations at 13 doublings, and dNaM → dTPT3 (17.2±3.4%) and dNaM → dC (3.9±2.4%) at 65 doublings. For the GXC template, dNaM → dTPT3 (2.7±3.2%) and dNaM → dA (3.1±0.7%) were the most frequent mutations at 13 doublings, and dNaM → dTPT3 (12.1±4.0%) and dNaM → dA (8.1±2.0%) at 65 doublings. While this data suggests that mutational introduction of dTPT3 occurs opposite both natural and unnatural nucleotides, the frequency of these events is not higher than mutational introduction of a natural nucleotide at 13 doublings. After 65 doublings, mutational introduction of a natural nucleotide remained minor, however, mutational introduction of dTPT3 increased significantly. We attribute this increase in mutational introduction of dTPT3 during later cycles of the PCR results to misinsertion and termination of synthesis during early cycles, followed by eventual reinitiation and completion of strand extension during a later cycle. However, this increase in mutational introduction of dTPT3 is only significant after more PCR amplification than is required for most practical applications.

To examine replication in the SSO, the AAC, AXC, and GXC oligonucleotides were cloned into a derivative of the pINF plasmid that we have used previously to encode and express proteins with ncAAs (see Supporting Information).³ SSOs (strain ML2⁵) were transformed with one of the three different pINF plasmids and six individual colonies for each sequence context were screened by streptavidin shift PCR to confirm that they had successfully received and initiated replication of an intact pINF plasmid. Three high retention SSO clones for each sequence context were passaged in liquid media for a total of ~75 cell doublings. Retention during the entire experiment was monitored by streptavidin shift PCR. At around 50 doublings, one of the three AXC replicates suffered pronounced loss of the UBP (see Supporting Information); contamination was surmised to be a likely cause, and this replicate is not included in the average metric reported below.

Given the difficulty of accurately measuring plasmid copy number, we estimated the in vivo amplification from the number of cell doublings that occurred in liquid media (see Supporting Information for details). This is expected to be accurate if the total pINF copy number remains constant, which is likely based on evidence that the presence of the UBP does not reduce the rate of DNA replication (as demonstrated above with uPCR, and also confirmed with replication in the SSO (see below)). In addition, as obtaining sufficient quantities of plasmid DNA directly from the SSO would have necessitated growth in prohibitively large cultures, plasmids replicated during defined amounts of growth of the SSO were recovered and subjected to 20 cycles of uPCR amplification prior to nanopore sequencing (Figure 3 and Table S3). Correspondingly, fidelity was calculated based on the total DNA amplification level, which includes the plasmid amplification during cell growth and amplicon doublings during PCR (Supporting Information).

Figure 3:

Mutation spectrum of AAC, AXC, and GXC in vivo and in uPCR. Left axis and dots show % Retention_B, as measured by streptavidin-shift PCR, with white dots representing measured retention for each experimental replicate. Red dots highlight the AXC sample that experienced catastrophic UBP loss and was excluded from reported average retention and % SNP abundance. Black dots represent average retention with ± 2 S.D. error bars, n=3 (except AXC, n=2). Right axis and stacked bar plots show the average percent abundance of each possible single nucleotide polymorphism (% of SNPs (Table S3)), as measured by nanopore sequencing.

Plasmids were isolated and analyzed after 21 SSO doublings, which corresponds to our workflow for expressing ncAA-containing proteins, and after the full 75 doublings to gauge fidelity over longer periods of growth. The fidelity of the in vivo replicated AAC sequence with uPCR analysis was 99.69±0.09% per doubling after 21 doublings and 99.88±0.03% after 75 doublings. For the unnatural sequences, AXC fidelity was 99.66±0.18% and 99.71±0.15% and GXC fidelity was 99.66±0.13% and 99.86±0.04% after 21 and 75 doublings, respectively. The observation that these fidelities are similar, or actually slightly higher than with uPCR alone indicates that the fidelity of replication in the in vivo environment of the SSO is higher than during uPCR. Moreover, because plasmid replication during colony formation was not included in the estimated amplification and because the observed mutations accumulated during both the in vivo and in vitro replication, these fidelities represent lower estimates of the true in vivo fidelity. Additionally, the fact that the observed fidelities appeared to increase as a function of growth reveals that plasmids that lost the UBP, and thus were fully natural, were not replicated more efficiently than plasmids that had retained the UBP.

Mutations detected after in vivo propagation and in vitro analysis were more variable than those observed from uPCR alone. For the AAC sequence, the most frequent mutations were dA→dT (2.6±1.0% of products) and dA→dTPT3 (2.8±1.1%) at 21 cell doublings, and dA → dC (2.0±0.8%) and dA→dTPT3 (3.7±1.4%) at 75 cell doublings. For the AXC sequence, the most frequent mutations were dNaM→dT (5.8±4.6%) and dNaM→dTPT3 (3.9±0.9%) at 21 cell doublings, and dNaM→dT (12.2±8.9%) and dNaM→dTPT3 (5.5±1.0%) at 75 cell doublings. For the GXC sequence, the most frequent mutations were dNaM→dA (2.5±2.6%) and dNaM→dTPT3 (3.5±2.4%) at 21 cell doublings; and dNaM→deletion (2.3±0.9%), dNaM→dT (4.7±4.6%), and dNaM→dTPT3 (2.3±0.8%) at 75 cell doublings. Importantly, the data demonstrate that UBP strand inversion (dNaM→dTPT3) occurs infrequently.

The developed nanopore sequencing method has enabled the first thorough analysis of the fidelity with which an expanded genetic alphabet is replicated. Empirical expansion of this method’s current pattern reference library with additional UBPs and sequence contexts would enable unrestricted sequencing of any UBP in any sequence context of interest, including both those developed by other groups as well as those formed as part of our ongoing effort to optimize dNaM-dTPT3 UBP. This method showed that, while some of this pairs high relative in vitro replication fidelity is due to a reduced fidelity of natural replication due to dTPT3 misinsertion, this only makes a significant contribution during extensive PCR amplification. Moreover, dTPT3 misinsertion does not appear to occur as frequently in the SSO, where the six-letter alphabet appears to be replicated with higher fidelity than in vitro. The increased fidelity likely results from the action of accessory proteins that are only available in vivo and highlights the importance of evaluating the UBPs in the environment where they are intended to function. Importantly, neither mutation to natural nucleotides nor strand inversion via self-pairing occur at problematic rates during in vitro or in vivo replication. Since the fidelity of unnatural protein synthesis appears to be determined by the fidelity of unnatural DNA replication,^3,6 this method will immediately facilitate improved use and further development of the SSO.

Supplementary Material

Supporting Data File

Table S3 – Summary of current pattern events from nanopore analysis of PCR and in vivo replicated UBP-containing DNA (XLSX)