Solanum tuberosum (potato) is among the most important food crops and a model tuber species. The crop is highly useful as an organism for studying plant responses to environmental stress factors (Lukan et al., 2022; Bleker et al., 2024), such as herbivory (Petek et al., 2020a), viral diseases (Baebler et al., 2020), transcriptional (Tomaž et al., 2023) and small RNA regulation (Križnik et al., 2020), single and combined abiotic stress responses (Demirel et al., 2020), and growth-defense trade-offs (Huot et al., 2014). Potato also serves as an excellent platform for transferring and testing vast amounts of knowledge garnered in Arabidopsis with an agriculturally relevant crop (Ramšak et al., 2018; Zagorščak et al., 2018; Schwacke et al., 2019), aiding toward solving present-day food security issues (Cole et al., 2018).

Apart from novel wild potato (Tang et al., 2022) and pan-genome assemblies (Hoopes et al., 2022; Bozan et al., 2023), the double monoploid (DM) clone of group Phureja DM1–3 516 R44 was until recently the standard variety on which gene models were defined. DM genome assemblies and gene models have been introduced by multiple consortia, including the Potato Genome Sequencing Consortium (PGSC), International Tomato Genome Consortium (ITAG), and Buell Lab (University of Georgia). These sequenced and assembled (Yandell and Ence, 2012) up to 88% of the potato genome that included between 35,004 (ITAG) (Tomato Genome Consortium, 2012) and 39,428 (PGSCv4.04) (Potato Genome Sequencing Consortium et al., 2011) gene models, with the recent nanopore long read assembly DMv6.1 (Pham et al., 2020) annotating 40,652 genes (in the working version). Moreover, we previously unified the PGSC v4 and ITAG gene models into a merged DMv4n version containing 49,322 genes (Petek et al., 2020b).

In contrast to Arabidopsis gene models, where version control and gene model tracking have been utilized for over a decade and gene annotations are optimized (Rhee et al., 2003; Lamesch et al., 2012), this is not the case with potato and similar less matured crop assemblies. Here, sequencing assemblies and gene models are only slowly advancing, and while each subsequent version improves sequencing depth, coverage, and assembly statistics, the gene models are reformulated. Frequently, previous gene model versions are not accounted for, and mapping and translation tables are not provided (e.g., Pham et al., 2020; Hoopes et al., 2022). This unfortunately limits potato research in multiple ways: i) hindering comparison of gene models across versions and experiments, ii) limiting the reuse and integration of experimental data based on older model versions (e.g., v4) with the latest version (v6) without extensive reprocessing of the RNA-Seq data, and iii) impeding the use of certain popular comparative genomics resources, such as Plaza and Ensembl plants (Valentin et al., 2021; Van Bel et al., 2022; Yates et al., 2022), which, as of writing this paper, have not been updated to the latest gene model versions (v6). Plaza is a platform for comparative, evolutionary, and functional plant genomics, which even in its latest version (5.0) uses v4 potato gene models (Van Bel et al., 2022). On the other hand, Ensembl plants (Yates et al., 2022), a plant genome analysis platform, is based on even older PGSC v3 potato gene models (Visser et al., 2009), yet is a source for other derived ontology (Schwacke et al., 2019) and transcription factor databases (Tian et al., 2020).

In addition to the previously mentioned issues, inadequate consideration of previous gene model information has resulted in the omission of a number of known genes (Visser et al., 2009; Potato Genome Sequencing Consortium et al., 2011; Tomato Genome Consortium, 2012; Petek et al., 2020b). In the case of the recent v6 gene models, we observed that they do not include certain well-known genes with important molecular functions, as they do not account for previous gene model information (Pham et al., 2020). An example is the transcription factor TGA2, an essential regulator of hormonal signaling (Tomaž et al., 2023). Aside from such missing genes, some are also moved, merged, or split (as presented in the sections below). These deviations can lead to differences in interpretation in downstream analyses (e.g., gene family expansion, differential expression, marker selection, gene set enrichment analysis) (Yandell and Ence, 2012). In addition to these imperfect annotations negatively affecting future experiments, existing published results using previous gene models, including, e.g., AlphaFold structure predictions (Tomaž et al., 2023), have become outdated, making it essential to update and consolidate gene predictions.

To help resolve these issues, here, we expand the ITAG and PGSC v4 annotations with v6 annotations (Pham et al., 2020), unifying the different gene models. In addition, we include experiment-based evidence from our pan-transcriptome (Petek et al., 2020b), short- and long-read sequencing data (Lukan et al., 2020; Hoopes et al., 2022), and Solanaceae proteomes (Hosmani et al., 2019; Wang et al., 2024), thereby creating an improved and more accurate potato gene annotation model for downstream analyses. To ensure the transparency and accuracy of future gene models, we present the Unified poTato genome annotation resource (UniTato). UniTato is provided through an Apollo web interface (Dunn et al., 2019), enabling a community-driven effort for real-time revision and enhancement of gene models by experts. This will increase the interpretational power of experimental datasets and facilitate the reuse of experimental analyses conducted on v4, thus expediting progress in potato research.

The advancement and maturation of high-throughput and long-read sequencing has led to several different potato genome assemblies, gene annotations, and transcriptomic datasets. Sequencing the group Phureja DM (Potato Genome Sequencing Consortium et al., 2011) still enables functional studies of polyploid potato cultivars using RNA-Seq technologies, although with the limitation of not covering cultivar-specific gene expression (Petek et al., 2020b; Hoopes et al., 2022). For practical reasons, most potato researchers use only one genome annotation, either PGSC (Potato Genome Sequencing Consortium et al., 2011) or ITAG (Tomato Genome Consortium, 2012), especially when conducting high-throughput analyses. However, using an incomplete gene set can lead to false outcomes regarding gene presence or gene family diversity, severely affecting downstream results (Yandell and Ence, 2012; Petek et al., 2020b). It is well known that incorrect or incomplete annotations corrupt all subsequent experiments that rely on them, making it essential to have the ability to share accurate and up-to-date annotations (Yandell and Ence, 2012; Bolger et al., 2018).

Our motivation here was thus two-fold: first, to transfer both gene model sets from the older PGSC assembly (Potato Genome Sequencing Consortium et al., 2011; Tomato Genome Consortium, 2012) to the new DMv6.1 assembly (Pham et al., 2020) and, at the same time, to merge the gene models ( Figure 2 ), allowing for data interoperability of previous experimental results (e.g., from RNA-Seq) (Petek et al., 2020b) with the unified gene model set, UniTato. Annotation merging was performed using an in-house-developed bioinformatics pipeline that utilizes open-source software and complementing it with evidence from published tetraploid transcriptomes (Petek et al., 2020b) ( Figure 1 ). The resulting annotation files were incorporated into an Apollo web server (Dunn et al., 2019), which enables the potato community to curate and refine potato gene models collaboratively and in real time, facilitating the establishment of a single standardized potato genome annotation ( Figure 3 ). Moreover, by comparing v4 and v6 annotations with UniTato, we observed multiple complex cases of gene models that cannot be straightforwardly resolved and will need to be manually curated (see “overlaps.xlsx” on the UniTato GitHub for a list of gene identifiers for these complicated cases). We thus show how UniTato can be used to identify gene models that are either missing or were moved, merged, or split (see Figures 3 , 4 ; Supplementary Figure S2 ).

This showcases the usefulness of the established resource for resolving genome assembly and annotation issues. Bioinformatics users can thus i) compare gene models visually across versions and tracks, pinpointing and resolving errors and ensuring that the most accurate gene models are constructed and applied; ii) compare experimental results obtained on v4 to those obtained on the new v6 assembly or higher, such as for instance with RNA-Seq, where results with old identifiers can be incorporated with new results using v6 identifiers (via the translation table), without requiring repeated read mapping and computations; iii) curate potato gene models in problematic regions, such as determining gene structures in tandemly repeated gene regions, which cause problems with most annotation pipelines (multiple long-read and short-read tracks available in UniTato, see Table 3 ); and iv) use current data with popular genome analysis resources that still rely on older annotations (Van Bel et al., 2022; Yates et al., 2022), facilitating, e.g., translation of gene descriptions and ontologies via orthology from model plants. Furthermore, with the provided v4–v6 mapping and available evidence tracks ( Table 3 ), UniTato also aids wet lab research. This includes i) guiding experiment design and interpretation, enabling users to check for off-target effects across different gene models; ii) defining and cloning functional orthologs based on experimental results and not merely partial sequence similarity, by revealing if orthologs from another plant map to the v6 assembly (see RNA-Seq and proteome tracks, Table 3 ); and iii) primer design, since the unified gene models are an improvement over the initial v4 and v6 models, enhancing gene coverage and accuracy (e.g., the user can visually determine if the amplicon is covering variations in the RNA-Seq tracks, Table 3 ).

In conclusion, we believe that building upon existing gene models to improve and unify them in a community-wise manner is a reasonable and transparent way to improve potato gene model annotations. The repeated creation of new genome model versions, without interlinking, is not contributing to the FAIR data paradigm (Wilkinson et al., 2016; Petek et al., 2022) and thus hinders agricultural research, including precision agriculture and food safety (Cole et al., 2018). The requirements of periodic annotation curation and incorporating experimental data and novel findings into the annotation process are inherent also to other plant species (Yandell and Ence, 2012; Kersey, 2019). Even in model plants, up to 40% of protein-coding genes can still be of unknown function, suggesting that much work is still required to fully resolve, annotate, and understand most plant genomes (Horan et al., 2008; Wang et al., 2023). We propose that a similar approach for evidence- and community-based revision as the one presented here can be utilized for any other insufficiently annotated species, for which genome models of closely related species are available. Apart from updating our database with new assemblies as they become available (Yang et al., 2023), future developments include the addition of novel experimental omics datasets and expansion to related genomes.

Data and code to reproduce the analysis are available at the GitHub repository https://github.com/NIB-SI/unitato/. The GFF and GTF files and the identifier translation table are also available from https://unitato.nib.si/downloads/. Publicly available RNA-Seq datasets were used in the study. The data can be found under the following SRA accession numbers: SRR8281993-SRR8282008 (Rywal IsoSeq reads; SRA study SRP172523), SRR14298411-SRR14298459 (Altus, Avenger, Colomba and Spunta IsoSeq reads; SRA study SRP315827), SRR11431596-SRR11431617 (PRJNA612026 ONT reads; SRA study SRP254248), SRR10690850, SRR10690852, SRR10690854, SRR10690856, SRR10690857, SRR10690858 (Rywal Illumina reads; SRA study SRP237525; GEO accession GSE142002), SRR122108, SRR122109, SRR122113, SRR122122, SRR122124, SRR122129, SRR122139 (Phureja Illumina reads; SRA study SRP005965), SRR7047512 (Phureja Illumina reads; SRA study SRP141363), SRR8457030-SRR8457059 (Phureja Illumina reads; SRA study SRP180310), SRR14627804-SRR14627805 (Phureja Illumina reads; SRA study SRP321011), SRR17202512-SRR17202515 (Phureja Illumina reads; SRA study SRP350333), SRR17244262-SRR17244298 (Phureja Illumina reads; SRA study SRP350981), and SRR10153126 (Phureja Illumina reads; SRA study SRP222783).

MZ: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing. JZ: Formal analysis, Funding acquisition, Investigation, Methodology, Writing – original draft, Writing – review & editing. CB: Formal analysis, Funding acquisition, Investigation, Methodology, Software, Visualization, Writing – review & editing. NN: Formal analysis, Investigation, Validation, Writing – review & editing. MJ: Formal analysis, Investigation, Validation, Writing - review and editing. ŽR: Investigation, Writing – review & editing. KG: Conceptualization, Investigation, Project administration, Supervision, Writing – review & editing, Funding acquisition, Resources. MP: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing.