Long-read sequencing to detect full-length protein-protein interactions

Abstract

Given the increased predictions on interactome size and demand for protein function information, methods for detecting protein-protein interactions remain a significant development area. The all-vs.-all sequencing (AVA-Seq) method utilizes a convergent fusion plasmid design to make two-hybrid technology amenable to next-generation sequencing. Here, we further innovate to take advantage of synthetic DNA technologies and Oxford Nanopore Technologies long-read sequencing improvements to allow us to determine full-length protein-protein interactions. We tested 3,115 human protein-protein pairs using this approach and recovered 159 protein-protein interactions from a set of 57 full-length human proteins. Fifteen of the 159 full-length protein-protein interactions matched known human interactions. When referencing a human gold standard set of interactions, eight full-length protein-protein interactions were recovered from an expected 28 interaction pairs (28.6%), a typical recovery rate for two-hybrid technologies. The AVA-Seq method, in combination with the ease of synthetic DNA production and the MinION platform, offers a low-cost, high-throughput alternative for determining protein-protein interactions, which can be utilized in research labs at all stages.

Keywords: Long-read sequencing; Oxford nanopore technologies; Protein–protein interaction; Single-molecule sequencing; Two-hybrid.

Fig. 1

Comparison of a two-hybrid method for determining protein-protein interactions and the all-vs.-all sequencing (AVA-Seq) method. Top panel: represents a traditional two-hybrid experiment where the two proteins being tested are each on different plasmids (bait and prey). Each plasmid contains half of an essential transcription factor that reconstitutes when they are within proximity of each other. When there is an interaction between the bait and prey, the transcription factor is reconstituted, and the system can recruit RNA polymerase which facilitates the readout of the HIS3 gene. The selection conditions for this example include a competitive inhibitor of HIS3 called 3-AT. As the concentration of 3-AT increases the interaction strength must also increase in order for the interaction to overcome the inhibition. The readout in yeast and bacterial two-hybrid is colony growth. Each individual colonies harbors information about which bait and prey plasmid successfully recruited RNAp. Each colony is selected, amplified, plasmid DNA is extracted, and Sanger sequenced. Bottom panel: represents the all-vs.-all sequencing (AVA-Seq) method where the readout for an interaction is growth in liquid culture. This growth (and change in growth) can be later quantified (log2FC after sequencing). Because the information for which bait and prey are on a single plasmid DNA can be extracted in bulk and next generation sequencing can be utilized.

Fig. 2

Growth comparison of full-length known human interacting pairs in both orientations. Data normalized to 0 mM 3-AT 9-hours growth time. The 2 mM growth is shown in green and 5 mM growth is shown in black with error bars (SD). The gene listed first is the DBD fusion.

Fig. 3

AVA-Seq method experimental design using full-length proteins. (A) Full-length proteins that are > 1300 bp paired and screened against themselves. (B) Full-length proteins from human proteins > 1300 bp paired with full-length proteins < 1299 bp. (C) Full-length proteins < 1299 bp paired and screened against themselves. All convergent insert sizes reflect the final DNA length (including adapters) amplified for NGS. Individual protein sequences were synthesized by TWIST Biosciences and included custom adapters added to the 5’ and 3’ ends of the design process. All pathways utilize two sets of primers. Each set has a complementary stop insert, allowing the two fragments to be ‘stitched’ together (represented as orange on the convergent insert).

Fig. 4

Heatmap of coverage and recovered interactions. The gene pairs tested but did not interact in 3 experiments are colored dark grey and 2 experiments medium grey. The gene pairs tested and determined to interact are shaded red with a darker color indicating higher number of tested experiments. Protein pairs shown in white were not abundant enough to be considered ‘tested’.

Fig. 5

Correlation of low-throughput (individual growth in liquid culture) and high-throughput all-vs.-all sequencing (AVA-Seq). Log₂FC max is shown for 2 mM and 5 mM 3-AT growth conditions. FDR < 0.1. Interactions were considered if the log₂FC > 0.8 and in two experiments. Known human interacting pairs from this study are listed in both orientations in the left column. Not all orientations were successfully cloned resulting in missing data for the individual growth (Supplemental Table 1). Selected individual non-interacting pairs (as reported in HsRRS-V2 ) ERBB3|C3orf38, SLC22A15|PSMD5, ZCCHC9|GCDH, ZNF350|PPP6C, PPP6C|ZNF350 were tested in the individual liquid growth (in those specific orientations only) and AVA-Seq high throughput. ERBB3|C3orf38 and SLC22A15|PSMD5 showed AVA-Seq growth in only 1 experiment out of three tested so they do not meet the criteria for being considered an interaction. SLC22A15 is annotated as a ‘sticky’ protein which may explain its growth.