CD-CODE: crowdsourcing condensate database and encyclopedia (2023)

Biomolecular condensates are membraneless organelles that selectively concentrate biomolecules (for example, proteins and nucleic acids) in the cell, with spatial and temporal precision¹. In recent years, their role was implicated in several biochemical processes, in physiology and disease². Consequently, biomolecular condensates are now leveraged as a new class of therapeutic targets^3,4.

Basic science and drug discovery advances build upon published reports and the rate of new discoveries depends on timely accessibility to relevant data. However, as with every novel paradigm, new terms and concepts emerge and evolve as the field develops. Accordingly, currently available databases which catalog proteins involved in condensate formation use various definitions and criteria to define condensates and their constituent proteins and RNAs^5,6,7,8. These are excellent databases curating proteins that phase separate. Specifically, LLPSDB⁷ and PhaSePro⁶ collect proteins that are thought to drive liquid–liquid phase separation, with the former curating exclusively in vitro data.

However, these databases do not answer the following questions regarding biomolecular condensates: What are the biomolecular condensates discovered and verified to date? What are their known protein components? Which condensates is a given protein known to belong to? What are the experimental evidences supporting the existence of a particular condensate? Our goal is to generate answers for these and other important questions, and to create a community-editable database to facilitate the dynamic data updates. Therefore, we designed a condensate-centric database, which is based on the scientific literature, and provides experimental evidences, scores and references for each condensate–protein relationship (Extended Data Figs. 1 and 2). This database is updated dynamically by contributors to keep up with the growing knowledge in the field. We call our platform CD-CODE, which consists of three main parts: (1) a database of biomolecular condensates and their protein constituents; (2) an encyclopedia for the scientific terms used in condensate biology; and (3) a crowdsourcing web application (Extended Data Fig. 3).

CD-CODE is a ‘living database’ designed for dynamic and rapid addition and review of information about condensates and proteins by users (Fig. 1) and is open to any expert researcher who wishes to contribute. Our user management system supports three types of users: viewers, contributors and maintainers. Viewers can read and download the curated information. Contributors can suggest edits and propose new condensate and protein entries (Extended Data Figs. 4 and 5). Maintainers are part of the development team, who curate the changes and accept or reject suggestions by contributors, who are then notified about the status of their suggestions and can engage in further discussion. To keep up with the rapidly evolving definitions, nomenclature and growing scientific evidence, the crowdsourcing platform allows the community to aggregate scientific findings in condensate biology.

At the time of this report, CD-CODE (cd-code.org) contains 9,861 proteins linked to 244 unique biomolecular condensates (and 375 in vitro synthetic condensates) across 49 different organisms. Notably, these numbers are continuously changing as contributors add and review more data. CD-CODE, as a semi-manually curated and annotated resource, aggregates information from the primary literature (to date, PubMed references published until 1 June 2022 were manually curated) and other databases^5,6,7,8 (Extended Data Fig. 6 and Extended Data Tables 1 and 2). To promote easy integration with other resources, protein entries are cross-referenced with UniProt⁹, Ensembl¹⁰ and the Human Protein Atlas (proteinatlas.org)¹¹. Common sequence properties of condensate proteins are also displayed graphically, such as disorder score¹² and amino acid composition (Extended Data Fig. 7), facilitating the identification of regions that may drive condensate partitioning.

We standardized the names of condensates by creating an ontology from the literature and grouped condensates by functional categories (Supplementary Table 1) to reveal the evolutionary history of condensates. Most known condensates are found in mammals and many are clade-specific (Fig. 2a). Since our current knowledge is sparse and likely biased, the evolutionary origin of condensates remains an open future research direction that CD-CODE can facilitate.

While many proteins undergo liquid–liquid phase separation in vitro, it is unclear which proteins form condensates in cells and which condensates they partition into. To facilitate our understanding of condensate-specificity of proteins, we collected all known condensates a given protein was found in, and we curated the experimental evidence for association of each protein with a given condensate (confidence score, corresponding to zero to five stars: 1 star: literature evidence, PubMed identifier (ID); 2 stars, high-throughput; 3 stars, in vitro; 4 stars, in cellulo; and 5 stars, in vivo evidence). Condensates and proteins that have zero or one star rating have not been manually curated yet.

As expected, for dynamic cellular compartments, many proteins partition into different condensates and the overlap between condensate proteomes is substantial (Fig. 2b). While proteins may localize to multiple condensates (members), a few are obligate and essential components (drivers). We annotated 205 driver proteins in specific condensates, providing the corresponding experimental evidences. Our database revealed that several proteins that are drivers in one condensate are nonessential members of another (for example, G3BP1, a driver of stress granules, is also present in processing bodies (P-bodies) and neuronal ribonucleoprotein particle granules). CD-CODE will aid our understanding of the determinants of condensate-specific driver behavior, and whether a driver protein can be used as a ‘marker’ of a condensate in experiments.

Marker proteins are used to define the identity of the condensates and inform designing of condensate-targeting drug screening pipelines³. They are thought to be uniquely associated with a given condensate, and are commonly used to visualize condensates using microscopy, for example, in colocalization experiments to prove the localization of proteins into condensates. Our database revealed that several known marker proteins are not specific to a condensate. For example, whilst DCP1A is used as a marker for P-bodies, it also localizes to stress granules and nucleoli. Knowing specific protein components will facilitate the experimental design for accurate, specific identification of condensates.

CD-CODE enables us to answer the questions posed at the beginning: (1) there are currently 136 unique biomolecular condensates documented in the literature; (2) as an example, P-granules, which are the germ granules of Caenorhabditis elegans, have 190 documented protein components: one of them, pgl-3 (PGL3_CAEEL), is a driver for P-granule formation, and its presence within P-granules is supported by in vivo experimental evidence (5 star). Pgl-3 is exclusively reported to be associated with P-granules; thus, it is a P-granule-specific marker protein.

Databases that curate proteins undergoing liquid–liquid phase separation have facilitated the development of machine learning algorithms to predict phase separation^13,14,15 and the discovery of what protein properties drive phase separation¹⁵. The next open question is which biomolecular condensate does a specific protein belong to. Our database contains a curated list of condensate proteomes (Fig. 2c), which can facilitate investigations of protein recruitment into specific condensates. Our resource can provide high-quality benchmarking data for machine learning algorithms aimed at predicting the protein components of condensates.

Furthermore, our comprehensive curation of condensate types and their respective composition in multiple species, and the level of experimental support, provides a valuable resource for drug hunters, which can inform the design of assays and screening pipelines. For example, in high-content imaging phenotypic screens, it is desired that the protein or protein combination chosen to be monitored is/are selective for the target condensate³. Additionally, through regular updating of the database by the community and via curation of new publications, CD-CODE supports and accelerates nomination of new condensate-associated drug targets.

The field of biomolecular condensates is highly transdisciplinary and ever-developing, where definitions and terms keep changing, creating a need for constant updates that require consensus within the community. The encyclopedia, as a standalone wiki, serves as a platform to aggregate knowledge about condensate research. In the future, we are planning weekly updates to integrate new data from the users, and yearly updates with new features and data points that become relevant to store, as the research field develops.

The main feature of CD-CODE is that it contains experimentally validated entries. However, caution should be exercised by users when interpreting lack of data on a particular protein, condensate or species, as this may simply reflect the biased interest of the community towards particular model systems and biological pathways. Any missing information could mean that (1) the protein or condensate has not been studied yet; (2) there is a research paper but the information has not been added to the database yet; (3) the condensate truly does not exist; or (4) the protein truly does not belong to a given condensate. As such, CD-CODE aims to highlight the unknowns in the field to guide future research questions to fill the gaps. These gaps in experimental evidence can be bridged by computational predictions^16,17, which are beyond the scope of CD-CODE. Evolution of the CD-CODE database through ongoing curation of new experimental evidence will lead to a progressive increase in high-scoring condensate entries.

In summary, we present CD-CODE, a semi-manually curated condensate database, and a community-editable web application. The crowdsourcing platform allows the community to further scrutinize definitions and evidence as the field evolves. This will ensure that the ever-growing knowledge on condensate research is integrated into the database and into the encyclopedia in a timely manner.