Error-prone nanopore sequencing enables accurate de novo DNA sequencing of long and highly repetitive genes encoding disordered proteins

Introduction: Advanced tools for synthesis of highly repetitive genes have accelerated the engineering of “smart” biomaterials based on intrinsically disordered protein polymers (IDPPs). Unlike proteins with defined 3D structure, stimuli-responsive IDPPs dynamically sample diverse conformational ensembles. The solution and material properties of IDPPs are encoded at the repeat level. By modulating repeat composition and number, IDPPs access a wide range of solubility and self-assembling behaviors. Given the repetitiveness of IDPPs, traditional high-throughput gene synthesis methods, which are coupled with DNA sequencing, fail. Instead, IDPPs are built through slow iterative gene synthesis approaches, which allow progressive verification of up to ~800 bp via Sanger sequencing. This engineering bottleneck is exacerbated for the synthesis of IDPPs with complex domain architectures which demand multiple cycles of synthesis and sequencing. Long-read sequencing, specifically nanopore sequencing with reads exceeding 10 Kbp, is poised to circumvent the underlying sequencing bottleneck. However, nanopore sequencing suffers from high error rate and remains largely limited to genome-scale scaffolding of short-reads. Current error-correction methods rely on having a reference sequence and supplementing with short-reads, which is fundamentally incompatible with de novo sequencing of highly repetitive genes. Here, using an extensive library of sequence-verified IDPPs, we establish an experimental and bioinformatics pipeline for robust and accurate de novo DNA nanopore sequencing of highly repetitive genes encoding IDPPs. We believe that this nanopore sequencing platform will unlock the power of one-pot gene synthesis and functional screens for the discovery of IDPPs with novel functionality and biomedical utility.

Materials and

Methods: We recombinantly synthesized a library of 30 IDPPs of different syntax, repeat length (ranging from 80 to 1280 repeats; 1.2-20 Kb), and block architectures (Fig. 1A). This library was synthesized through a lengthy time-consuming process of iterative recursive directional ligation, allowing for full verification via Sanger sequencing at each gene synthesis step. Linearized, full-length plasmids were barcoded in-house and adapted per instructions for the MinION platform (Oxford Nanopore Technologies; ONT) (Fig. 1B) using the Ligation Sequencing Kit V14 from ONT. Libraries with 20 fmols of 6-12 barcoded constructs were sequenced for 16 hours using Flongle cells (R10.4.1) with 50-100 nanopores. After basecalling using Guppy, a custom-made MATLAB pipeline identified dual-barcoded reads and applied our optimized size-agnostic filters to select reads of the highest quality that encode putative full-length IDPP genes. Subsets of high-quality full-length reads were aligned using MAFFT in MegAlign Pro to build de novo consensus sequences —without using a refence gene (Fig. 1C). Exclusively for validation, we assessed performance and accuracy of our DNA nanopore sequencing approach by aligning partial (from subsets of full-length reads) or global consensus IDPP gene sequences to the corresponding reference genes in our library.

Results, Conclusions, and Discussions: Despite the error-prone nature of nanopore reads, we achieved de novo (without a reference) sequencing of all highly repetitive IDDPs in our library. The successful deployment of this long-read sequencing technology in this challenging set of long repetitive genes relied on our ability to experimentally and bioinformatically enrich for high-quality full-length IDPP reads. In our optimized pipeline, we exclusively analyzed putative full-length reads. Analysis of datasets with a mixture of full-length and truncated (up to 30% of full-length) IDPP reads demonstrated the need for >60% full-length content (Fig. 2A). Our in-house barcoding approach increased the occurrence of long-reads with dual-barcodes (not shown), simplifying the identification of full-length reads in the final dataset. Our sequencing pipeline proved length agnostic, and in any given sequencing library we consistently arrived at the known IDDP length (Fig. 2B), removing the need for prior knowledge about the IDPP gene. Error-free consensus assembly demanded filtered datasets enriched for ultra-high-quality reads (Qscore>13), higher than the typical ONT recommended Qscore of 10. Our data showed that for diverse IDPP genes, Flongle nanopore cells generated sufficient reads with a suitable distribution of Qscores (Fig. 2C). Relying exclusively on these high-quality reads, across our repeat gene library we routinely observed robust error-free consensus sequences with as few 40 full-length IDPP reads (Fig. 2D), but robust accurate assembly of challenging IDPP sequences only emerged upon applying a global consensus on 10 partial consensus sequences each from subsets of 10 full-length reads (100 total reads; as in the inset of Fig 2D). Overall, our findings demonstrate that nanopore sequencing, if properly deployed, can deliver robust and accurate reference-free sequencing of highly repetitive genes encoding IDPPs. Because functional screens for IDPPs have been largely restricted by the inability to sequence functional clones, our long-read nanopore sequencing pipeline promises to unlock new opportunities for the design and evolution of novel functional IDPPs. Beyond IDPPs, the study of repetitive DNA has lagged compared to non-repetitive DNA, and we hope that our data on sequence-verified IDPP genes will stimulate the deployment of error-prone nanopore sequencing for de novo sequencing of highly repetitive genes.

Acknowledgements (Optional):