Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Pfam: The protein families database in 2021

Pfam: The protein families database in 2021 Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 D412–D419 Nucleic Acids Research, 2021, Vol. 49, Database issue Published online 30 October 2020 doi: 10.1093/nar/gkaa913 1,* 1 1 1 Jaina Mistry , Sara Chuguransky , Lowri Williams , Matloob Qureshi , 1 2 3 3 Gustavo A. Salazar , Erik L.L. Sonnhammer , Silvio C.E. Tosatto , Lisanna Paladin , 1 1 1 1 Shriya Raj , Lorna J. Richardson , Robert D. Finn and Alex Bateman European Molecular Biology Laboratory, European Bioinformatics Institute (EMBL-EBI), Wellcome Genome Campus, Hinxton CB10 1SD, UK, Department of Biochemistry and Biophysics, Science for Life Laboratory, Stockholm University, Box 1031, 17121 Solna, Sweden and Department of Biomedical Sciences, University of Padua, 35131 Padova, Italy Received September 11, 2020; Revised October 01, 2020; Editorial Decision October 02, 2020; Accepted October 06, 2020 ABSTRACT per-family gathering thresholds. Pfam entries are manually annotated with functional information from the literature The Pfam database is a widely used resource for clas- where available. sifying protein sequences into families and domains. Since Pfam release 29.0, pfamseq is based on UniPro- Since Pfam was last described in this journal, over tKB reference proteomes, whilst prior to that, it was based 350 new families have been added in Pfam 33.1 and on the whole of UniProtKB (3,4). Although the underly- numerous improvements have been made to existing ing sequence database is based on reference proteomes, all of the profile HMMs are also searched against UniProtKB entries. To facilitate research on COVID-19, we have and the resulting matches are made available on the Pfam revised the Pfam entries that cover the SARS-CoV-2 website and in a flatfile format. Similarly, the match data proteome, and built new entries for regions that were for representative proteomes (5) are provided in the same not covered by Pfam. We have reintroduced Pfam-B way. Pfam can be accessed via the website at https://pfam. which provides an automatically generated supple- xfam.org. The flatfiles and exports of the Pfam MySQL ment to Pfam and contains 136 730 novel clusters of database are provided under a CC0 license for each re- sequences that are not yet matched by a Pfam fam- lease of the database, and can be found on the FTP site ily. The new Pfam-B is based on a clustering by the ftp://ftp.ebi.ac.uk/pub/databases/Pfam/releases. MMseqs2 software. We have compared all of the re- When a Pfam entry is built, we typically search it itera- gions in the RepeatsDB to those in Pfam and have tively against pfamseq in order to find more distant homo- started to use the results to build and refine Pfam logues. Pfam entries are built such that there are no over- repeat families. Pfam is freely available for browsing laps between them; this means that the same region of a se- quence should not match more than one family. This non- and download at http://pfam.xfam.org/. overlap rule proves to be an excellent quality control cri- teria which helps to avoid including false positive matches INTRODUCTION into a family. From Pfam 28.0, we relaxed this rule to allow small overlaps between families as it had become increas- Pfam is a database of protein families and domains that is ingly time-consuming to resolve all such overlaps each time widely used to analyse novel genomes, metagenomes and we updated pfamseq (see (3) for more details). to guide experimental work on particular proteins and sys- Sets of Pfam entries that we believe to be evolutionarily tems (1,2). Each Pfam family has a seed alignment that related are grouped together into clans (6). Relationships contains a representative set of sequences for the entry. between entries are identified through sequence similarity, A profile hidden Markov model (HMM) is automatically structural similarity, functional similarity and/or profile- built from the seed alignment and searched against a se- profile comparisons using software such as HHsearch ( 7). quence database called pfamseq using the HMMER soft- Where possible, we build a single comprehensive profile ware (http://hmmer.org/). All sequence regions that satisfy HMM to detect all members of a family. For some of the a family-specific curated threshold, also known as the gath- larger superfamilies where this is not possible, we build ering threshold, are aligned to the profile HMM to create multiple profile HMMs and put them in the same clan. the full alignment. It is worth noting that a common mis- As families in a clan are evolutionarily related, we allow use of Pfam is to use a single E-value threshold across all them to overlap with other members of the same clan. The Pfam HMMs, which results in lower sensitivity and an in- clans are competed such that if there are multiple over- crease in false positive matches when compared to using the To whom correspondence should be addressed. Tel: +44 1223 494100; Fax: +44 1223 494468; Email: jaina@ebi.ac.uk C The Author(s) 2020. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 Nucleic Acids Research, 2021, Vol. 49, Database issue D413 lapping profile HMM matches to families from a single FAMILY IMPROVEMENTS clan, only the match with the lowest E-value is displayed Despite almost 25 years of active curation, there remain on the website, or included in the full alignment of the many protein domains and families that are as yet unclassi- entry. fied by Pfam. New families added to Pfam are created from Here we describe Pfam 33.1 and some of the family up- a range of sources, including Pfam-B families and protein dates we have made since the last release. We also detail the structure. Pfam-B alignments in particular have been a very Pfam coverage of the SARS-CoV-2 proteome, and the new fruitful substrate for families, with historically nearly a third method to create Pfam-B which we re-introduced in Pfam of all Pfam entries being built from them. The new Pfam- 33.1. Finally we describe the results of the analysis we car- B (described below) was used to build 18 entries in Pfam ried out on comparing repeats in the Pfam database to Re- 33.1. We expect that Pfam-B will again become a very use- peatsDB. ful source of additional families in the coming years. Protein structures from databases such as the Protein PFAM 33.1 RELEASE Data Bank (PDB) (9,10) are particularly amenable as a source of new Pfam entries because the Pfam domain Pfam 33.0 was due to be released in March 2020, however boundaries can be defined precisely from the structure. due to the global COVID-19 pandemic, we delayed its re- There is often an associated paper to the structure which lease so that we could focus on improving our SARS-CoV- we use to help annotate the Pfam entry. Building families 2 models (see ‘COVID-19 updates’ section for more de- using protein structures is an ongoing activity, and some of tail). The updated SARS-CoV-2 models, and a handful of the new SARS-CoV-2 families were built in this way. Ad- other new Pfam entries were added to Pfam 33.0 to create ditionally, we have made 37 new families based on clus- Pfam 33.1, which was released in May 2020. Release 33.0 ters of sequences from the MGnify metagenomic protein was never officially released on the Pfam website, although database (11). The MGnify clusters may be another large the database files and flat files are made available on the source from which we could build families from in the fu- ftp site. ture. Using metagenomics clusters can help to cover regions Pfam 33.1 contains 18 259 families and 635 clans. Since of sequence space that are not well covered by existing ge- Pfam 32.0, we have built 355 new families, deleted 25 fami- nomic sequencing projects. lies and built 8 new clans. Just over 39% of all Pfam families Pfam has created numerous entries for domains of un- are placed within a clan. Of the sequences in UniProtKB, known function (DUF) and uncharacterized protein fami- 77.0% have at least one match to a Pfam entry, and 53.2% lies (UPF). Over time the functions of some of these are dis- residues in UniProtKB fall within a Pfam entry. These fig- covered. We continue to scan the literature to identify newly ures, termed the sequence and residue coverage respectively, identified functions as well as receiving updates from the have remained fairly constant over the past vfi e Pfam re- community via our helpdesk. Many functions are also iden- leases despite a 240% increase in UniProtKB size during tified by the InterPro database ( 12) update process which that time (see Figure 1). Since 2015, highly redundant bac- checks whether the UniProtKB/Swiss-Prot descriptions of terial proteomes have been identified and removed from proteins in each InterPro entry have changed between re- UniProtKB (8). This process ensures a level of diversity in leases. When these changes identify a function, Pfam is no- the sequences added to UniProtKB, and prevents, for ex- tified. To date we have changed the identifiers of 1132 DUF ample, multiple strains of a particular species being added. or UPF families indicating that a function has been iden- Pfam is able to maintain a sequence coverage of ∼77%, and tified for these families. As of release 33.1, Pfam contains a residue coverage of ∼53% largely due to the new sequences 4244 DUF or UPF families, which is 23% of all Pfam fami- being added to UniProtKB matching existing Pfam models. lies. This suggests that there are still a lot of uncharacterized As UniProtKB grows, it is progressively harder to increase families and domains for molecular biologists to study. the Pfam coverage since newer Pfam entries tend to cover a We regularly receive feedback from users about families small taxonomic range. Although new models add little in or domains that are missing in Pfam, and typically add terms of coverage, they may represent medically important many user submitted families at each release. We include the proteins. For example, during our SARS-CoV-2 model im- submitters name and ORCID identifier ( https://orcid.org/) provements, we built the new entry Pfam:PF19213, which where available as an author of such Pfam entries. This corresponds to non-structural protein 6 (NSP6), a protein helps people to get credit for community activities that im- involved in autophagosome generation. prove molecular biology databases such as Pfam. One such The Pfam 33.1 sequence and residue coverage of UniPro- user submission was from Heli Monttinen ¨ (University of tKB reference proteomes is 75.1 and 49.4%, respectively Helsinki) who submitted a large scale clustering of virus (slightly lower than the figure for all of UniProtKB men- families. Based on this clustering we added 88 new families tioned above). This is a 0.6% increase in sequence coverage, to Pfam. We currently have 144 non-Pfam authors listed by and 0.7% decrease in residue coverage compared to Pfam their ORCID and we encourage our users to continue to 32.0. Although there was only a 3% increase in the num- submit interesting potential new domains and families. ber of sequences in reference proteomes between Pfam re- leases 32.0 and 33.1, around 2000 bacterial proteomes were removed from the reference proteomes (due to quality is- COVID-19 UPDATES sues) during this time. The changes in coverage since Pfam 32.0 are likely to be due to the change in sequences within The SARS-CoV-2 pandemic has mobilized a worldwide the reference proteomes set. research effort to understand the pathogen itself and the Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 D414 Nucleic Acids Research, 2021, Vol. 49, Database issue Figure 1. Growth of UniProtKB, and its coverage by Pfam over the last vfi e Pfam releases. As UniProtKB grows in size, the Pfam sequence and residue coverage is maintained at ∼77 and ∼53%, respectively. The UniProtKB size in the figure corresponds to the version of UniProtKB we used for each Pfam release. mechanism of COVID-19 disease, as well as to identify Structural and accessory proteins treatment options. Pfam already provided useful annotation One of the most important proteins of this virus is the for SARS-CoV-2, but we sought to update our models, fam- Spike protein (S), which aids viral entry into the host ily names and annotations for this virus in an effort to help cell, and is key for its pathogenicity. The models that the research community. were already present in the database have been improved We assessed all the protein sequences provided by and we added a new domain entry. Following this, as the UniProt via its new COVID-19 portal (https://covid-19. S protein is translated into a large polypeptide which is uniprot.org/), identified those which lacked an existing cleaved by host proteases to produce S1 and S2 peptides, Pfam model, and built new models as required. We now we have now three domains corresponding to S1, the N- cover almost all gene products encoded by SARS-CoV-2 terminal domain (Pfam:PF16451), the receptor binding (Figure 2). Orf10, a small putative protein encoded at the 3 - domain (RBD) (Pfam:PF09408) and the new C-terminal end of the SARS-CoV-2 genome is the only protein which domain (Pfam:PF19209). S2 is described in the family remains unannotated by Pfam. It was not possible to build Pfam:PF01601, which contains an additional S2 cleavage a Pfam entry for it since it lacked any detectable homo- site, a fusion peptide (FP), internal fusion peptide (IFP), logues in UniProtKB. The mean percentage identity for heptad repeat 1/2 (HR1/2), and the transmembrane do- the full alignments (based on reference proteomes) of Pfam main (TM). entries that match SARS-CoV-2 proteins is 49% (range Entities for the other structural proteins in Pfam 15–96%). were updated, such as the Nucleocapsid protein (N) in We have standardized our identifier nomenclature and Pfam:PF00937, the matrix protein (M) Pfam:PF01635 and descriptions of the families to ensure they are both cor- the envelope E protein in Pfam:PF02723. Additional acces- rect and consistent. The majority of the family identifiers sory proteins encoded by coronaviruses, usually called non- now begin with either CoV, for coronavirus specific fami- structural accessory proteins (NS), although some of them lies, or bCoV for the families which are specific to the beta- constitute structural parts of the virion, were updated. For coronavirus clade, which SARS-CoV-2 belongs to. We have example, NS3a corresponds to the betacoronavirus viro- also fixed inconsistencies in the naming and descriptions porin in the Pfam entry Pfam:PF11289, and Pfam:PF09399 of the various non-structural proteins (NSPs), using NSPx describes the protein 9b encoded within the nucleocapsid for those proteins encoded by the replicase polyprotein and gene, which contains a lipid binding domain. The acces- NSx for those encoded by other ORFs. sory proteins NS7a and NS7b in the entries Pfam:PF08779 Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 Nucleic Acids Research, 2021, Vol. 49, Database issue D415 Figure 2. Schematic representation of Pfam coverage of the SARS-CoV-2 proteome. The top row of boxes represent the individual virus proteins processed from the precursor polyproteins. These boxes are coloured when they contain more than one Pfam domain and the individual Pfam entries are expanded below. and Pfam:PF11395, respectively, are important during the new family Pfam:PF19217 whilst its C-terminal domain replication cycle. NS8 in Pfam:PF12093 family may mod- is described in Pfam:PF16348 and the DUF5881 was ulate viral pathogenicity or replication in favour of human identified as NSP6 in Pfam:PF19213. The other protease adaptation and it has been to be one of the most important activity in these viruses is described in Pfam:PF05409 usu- genes in the study of the human adaptation of the virus. The ally known as Main protease (M-pro domain) or 3C-like accessory protein NS6 (Pfam:PF12133) is highly conserved proteinase (3CL-pro), corresponding to NSP5, a member amongst SARS-related coronaviruses and it can modulate of MEROPS (15) peptidase family C30. Other NSPs were host immune responses by inhibiting synthesis and sig- in the Pfam database and most of these families have been nalling of interferon-beta. The family Pfam:PF17635 de- extended to include SARS-Cov-2 sequences. For example, scribes the protein 14 encoded in Orf14 and its function is Pfam:PF08716 and Pfam:PF08717 corresponds to NSP7 currently unknown. and NSP8, respectively, which both form a hexadecameric supercomplex that adopts a hollow cylinder-like structure that play a role in the stabilization of NSP12 regions in- Non-structural proteins volved in RNA binding and are essential for a highly active NSP12 polymerase complex. NSP12 is an RNA-directed Regarding NSPs from coronaviruses, encoded by RNA polymerase described in two entries, Pfam:PF06478 ORF1a/1ab (replicase 1a/1ab), we updated the exist- and Pfam:PF00680 describing the N- and C-terminal ing entries and added new ones where appropriate. NSPs domains, respectively (16). form the replication–transcription complexes (RTCs) that Following annotation updates, NSP9 belongs to the are essential for the synthesis of viral RNA necessary family Pfam:PF08710, which is a single-stranded RNA- for the infection process or to avoid the host immune binding viral protein involved in RNA synthesis, essen- response. Amongst these proteins, NSP3 is the largest tial for the coronavirus replication. NSP10 included in replicase product including several conserved domains Pfam:PF09401 is amongst the more conserved coronavirus whose organization differ amongst coronavirus genera proteins. It interacts with NSP14 (Pfam:PF06471) and (13). It has a crucial function as it cleaves the proteins NSP16 (Pfam:PF06460) and regulates their respective ex- encoded by the replicase, including NSP3 itself. It is an onuclease (ExoN) and ribose-2 -O-MTase (2 -O-MTase) ac- essential component of the RTC and serves as a scaffold tivities. It has a Tyr-96 specific for SARS-CoV and it is of protein to interact with itself and other NSPs. We describe particular interest as it plays a crucial role in the NSP10– it in separate entries: the existing N-terminal domain NSP16 interaction and in the activation of the NSP16 2 -O- Pfam:PF12379 was improved and updated, and includes MTase activity as well as in the NSP10–NSP14 interaction the NSP3a domain which acts as a scaffold through its (17). interaction with numerous proteins involved in replication Based on recently solved structures of SARS-CoV-2 pro- and transcription processes and contains the ubiquitin-like teins, we were able to build new families, as is the case of 1 (UB1) and glutamic acid-rich acidic (AC) hypervariable NSP15. We built three new entries representing the three domains. NSP3a is essential for localizing the RNA in structural domains of the NSP15 protein (Figure 3)based the replicase–transcriptase complex at the first steps of on the structure by Kim et al.(18). infection, as it interacts with the nucleocapsid protein N in the nascent replicase/transcriptase complex (14). The other domains encoded in this protein are the Macro PFAM-B domain included in Pfam:PF01661 and SUD-M domain in Pfam:PF11633 that binds single-stranded poly(A). In addition to our HMM-based Pfam entries (Pfam-A), The domain Pfam:PF12124 was updated and is now we used to provide a set of unannotated, computation- described as the SUD-C or DPUP domain which binds ally generated multiple sequence alignments called Pfam- to single-stranded RNA and recognizes purine bases. B. The method used for Pfam-B construction changed sig- The papain-like protease (PLPro) crucial for polypeptide nificantly over the years. The final prior version of Pfam-B processing is described in Pfam:PF08715; the nucleic alignments were created from clusters generated by apply- acid-binding domain (NAR) belongs to Pfam:PF16251 ing the ADDA algorithm (19) to an all-against-all BLAST family and, lastly, the C-terminal domain of NSP3 was (20) search of a non-redundant version of UniProtKB, and added as the new entry Pfam:PF19218. Additionally, we removing any regions covered by Pfam-A. The version of built the N-terminal domain of NSP4 represented in the UniProtKB used in the all-against-all clustering became in- Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 D416 Nucleic Acids Research, 2021, Vol. 49, Database issue Table 1. The number of each Pfam type for the last two Pfam releases 32.0 and 33.1 Type Pfam release 32.0 Pfam release 33.1 Family 11 177 11 242 Domain 6248 6406 Repeat 260 268 Coiled-coil 95 171 Motif 82 91 Disordered 67 81 der and Repeat types within Pfam and report these results below. DISORDER IN PFAM We investigated the level of predicted disorder across all Figure 3. The structure of NSP15 (PDB ID: 6VWW) from Kim et al. Pfam families. The mean percentage identity for full align- shows the three new Pfam domains. (i) CoV NSP15 N (Pfam:PF19219) Coronavirus replicase NSP15, N-terminal oligomerization domain in red, ments (based on reference proteomes) of Pfam entries with (ii) CoV NSP15 M (Pfam:PF19216) Coronavirus replicase NSP15, mid- type Disordered is 55% (range 18–94%). For each family dle domain in blue and (iii) CoV NSP15 C (Pfam:PF19215) Coronavirus we determined the fraction of residues in the seed align- replicase NSP15, uridylate-specific endoribonuclease in green. ment that were predicted to be low complexity by segmasker (24) or disordered by MobiDB-lite (25). A histogram of scores is shown in Figure 4. There are 303 Pfam families creasingly out of sync with that used by Pfam. This, com- that have over 80% of seed alignment residues predicted as bined with the time it took to produce it meant that as of disordered/low complexity, and yet only 81 are currently Pfam 28.0 (released in 2015), it was no longer feasible to classified as disordered. We have begun to reclassify these make Pfam-B (see (3) for a longer discussion on why we families for release 34.0. stopped producing Pfam-B). We have now devised an alter- native method of making Pfam-B that uses the MMSeqs2 software (21) which has a much lower computational cost. REPEATS IN PFAM The new pipeline was also designed to produce alignments that are more likely to represent new domain families. To investigate the Repeat type entries in Pfam, we compared Pfam-B was created by clustering pfamseq sequence seg- them to the gold-standard repeat database RepeatsDB (26). ments longer than 50 residues not covered by Pfam-A do- This resource contains a rigorous structural classification mains, using MMSeqs2 run with the cluster option and of repeat proteins from the PDB and places each into a re- bidirectional coverage mode. Multiple sequence alignments peat ontology. Repeat regions in RepeatsDB are detected by were generated for each cluster with more than 20 sequences structure and annotated with the position of each repeated using FAMSA (22). This resulted in 136,730 Pfam-B fami- unit. The comparison to Pfam highlights the differences be- lies that on average contain 99 sequences (maximum 40 912) tween sequence- and structure-based repeat identification, and are 310 positions wide (maximum 29 216). domain annotation as well as classification, and is relevant The Pfam-B alignments are presently only released as a in the context of our ongoing effort of improving repeat def- tar archive on the Pfam FTP site (file Pfam-B.tgz). We have initions. not built profile HMMs for the entries or integrated them Pfam covers 64.2% of the repeat regions found in Re- into the Pfam website. The entries are sorted such that the peatsDB (Figure 5) and shows a bimodal distribution, with first entries have an optimal combination of size and con- peaks at 0 and 100%. Therefore, most RepeatsDB entries servation and would therefore have the highest chance of (1415 UniProtKB proteins) are either fully characterized representing novel and useful domain families. or uncharacterized by Pfam families. This analysis allowed us to identify candidates for future Pfam curation, that present 0% coverage, such as the STU2 protein (UniPro- PFAM TYPE DEFINITIONS tKB:P46675) containing HEAT repeats (Figure 5), the - Pfam type definitions divide entries into one of six types propeller domain in DDB1- and CUL4-associated factor and they can help users in selecting which Pfam families to 1 (UniProtKB:Q9Y4B6) and the LRR domain in Ran use in their analyses. Over the past year we have made nu- GTPase-activating protein 1 (UniProtKB:P41391). These merous changes and updates to the Pfam type definitions sequences will be aligned with structurally similar entries for families (Table 1). In particular, we have carried out a in RepeatsDB to derive the input alignments for Pfam, and large scale screen of Pfam families using the ncoils software the eventual overlap with existing sequence models (e.g. the (23) to identify families with a high proportion of predicted existing Pfam LRR and HEAT repeat domains) will be in- coiled-coil, and after inspection of such families, we were vestigated and resolved. able to change their type. We investigated the number and types of Pfam domains We were interested to see if we can improve the definitions detected in RepeatsDB regions. A total of 573 Pfam do- of the other types in Pfam. We have investigated the Disor- mains in 106 Pfam clans are mapped to repeats, of which Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 Nucleic Acids Research, 2021, Vol. 49, Database issue D417 Figure 4. Percent of residues in the seed alignment of Pfam entries that are low complexity or disordered as predicted by segmasker and MobiDB-lite, respectively. only 91 (15.9%) are of type Repeat. A total of 176 (30.7%) of the following one (Figure 5). This type of unit is called are of type Family, 303 (52.9%) Domain, 1 Coiled-coil ‘Velcro’ and has stabilizing functions (27). The two phases (mapping to entries in RepeatsDB fibrous structures) and are therefore related to two different concepts, and these 2 Motif. The large majority of Family and Domain types types of examples highlight cases where the structural re- may be justified by the fact that RepeatsDB includes a class, peat patterns may be used to revise the sequence one or Beads-on-a-string, whose repeat units are large enough to vice versa. (iii) Finally, different Pfam models may map to fold independently (26), likely corresponding to Pfam type the same repeat region, and correspond to different num- domain rather than type repeat. Indeed, class V entries ber of units, such as in the Ankyrin region (Figure 5)of mostly (86.5%) map to Domains in Pfam. However, Re- the Neurogenic locus notch homologue protein 1 (UniPro- peatsDB elongated structures, i.e. the most canonical re- tKB:P46531), where the three types of Pfam models de- peats, including solenoids (26), are still mostly mapped to tected (Pfam:PF00023, Pfam:PF13637 and Pfam:PF12796) non-Repeat Pfam types (22.2% are Repeat), including the map respectively to one, two and three repeat units. With two Motif entries. this arrangement, one unit and a fragment are actually These data will support the revision of Pfam types as- skipped by Pfam models; these cases will require an exten- signment in repeat entries, also in light of the RepeatsDB sive investigation of the optimal number of units included units to Pfam domains correspondence. Pfam models may in Pfam models and the potential overlaps. be defined in different ways: (i) they may map to the entire repeat region, such as the Ricin-type beta-trefoil lectin do- DISCUSSION main (Pfam:PF00652) in the rRNA N-glycosidase (UniPro- tKB:B7 × 8M2). The model for this domain was initially de- The Pfam sequence and residue coverage of UniProtKB has signed to correspond to single units, but it was later revised remained fairly constant over the past few years at ∼77 and and updated to the current version to increase its sensitiv- ∼53%, respectively. This means that although Pfam aims ity, supported by the observation that this type of repeat to be comprehensive, there remains a significant area of se- region usually includes a tandem repeat of three units. In quence space that has no annotation by Pfam. As UniPro- this case, the type assignment to Domain should be kept. tKB grows, it becomes progressively harder to increase the (ii) Pfam models may map to each repeat unit, with the coverage as the larger ubiquitous families have already been same phase, as the RCC1 repeats (Pfam:PF00415) in the built, and newer families tend to have a smaller taxonomic human Regulator of chromosome condensation (UniPro- range. Nevertheless there are still many important families tKB:P18754), or with a different phase, as the WD do- left to build, and we plan to concentrate our efforts on fam- main (Pfam:PF00400) in the WD repeat-containing pro- ily building for the next release. New families will be built tein 5 (UniProtKB:P61964). In the latter, whilst the Re- from a range of sources such that we represent the diver- peatsDB phase is defined as corresponding to a single en- sity in the tree of life. Sources will include Pfam-B clus- tire blade of the -propeller, the Pfam model maps to a ters, which we re-introduced into Pfam in the current release segment including most of each blade and the first -sheet (version 33.1). We will also continue to build families from Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 D418 Nucleic Acids Research, 2021, Vol. 49, Database issue Figure 5. Pfam coverage of repeat regions in UniProtKB entries from RepeatsDB. Three examples are shown represented by their PDB structures. On the top left PDB ID: 4ffb, chain C, mapping to a region of the HEAT repeats in UniProtKB:P46675 (residues 1–272), with Pfam coverage 0%. In the centre, the Ankyrin region of UniProtKB:P46531, PDB ID: 6py8, chain K (residues 1759–2127), with Pfam coverage 89.4%. Three Pfam domains are detected: Pfam:PF12796 in yellow, Pfam:PF13637 in orange and Pfam:PF00023 in red. On the top right, PDB ID: 3ur4, chain A, mapping to the -propeller UniProtKB:P61964 (residues 24–334), with Pfam coverage 93%. Only one type of Pfam domain is detected (Pfam:PF00400), shown in alternating shades of blue to facilitate the visualization of the Pfam model phase. metagenomic sequence clusters, PDB structures and com- Pfam and to the numerous scientists who contributed sug- munity submissions. gestions and families via our helpdesk. In response to the COVID-19 pandemic, we have revised all existing families that match SARS-CoV-2 proteins, and FUNDING built new profile HMMs to cover regions previously unan- notated by Pfam. All proteins in the SARS-CoV-2 pro- European Union’s Horizon 2020 MSCA-RISE ac- teome are covered by Pfam apart from the putative pro- tion [823886]; Wellcome [108433/Z/15/Z]; BBSRC tein encoded by Orf10, which lacked any detectable homo- [BB/S020381/1]; Open Targets; European Molecular logues in UniProtKB. We hope our models help the research Biology Laboratory Core Funds. Funding for open access charge: Wellcome; Research Councils UK (RCUK). community to identify and annotate coronavirus sequences. Conflict of interest statement. None declared. Additionally, multiple sequence alignments generated using Pfam HMMs may prove useful in tracking the evolution of coronaviruses. REFERENCES Since the last release, we have updated some of our type 1. Sonnhammer,E.L., Eddy,S.R. and Durbin,R. (1997) Pfam: a definitions, particularly those that are predicted to have comprehensive database of protein domain families based on seed coiled-coil regions. We have also identified a set of families alignments. Proteins, 28, 405–420. that have a high proportion of disordered residues, and plan 2. El-Gebali,S., Mistry,J., Bateman,A., Eddy,S.R., Luciani,A., to reclassify these families as type disorder for the next Pfam Potter,S.C., Qureshi,M., Richardson,L.J., Salazar,G.A., Smart,A. et al. (2019) The Pfam protein families database in 2019. Nucleic release. We have also compared Pfam to the RepeatsDB Acids Res., 47, D427–D432. database. Using the comparison, we will look to change the 3. Finn,R.D., Coggill,P., Eberhardt,R.Y., Eddy,S.R., Mistry,J., type of some of the Pfam entries to repeat, incorporate addi- Mitchell,A.L., Potter,S.C., Punta,M., Qureshi,M., tional RepeatsDB regions into Pfam and revise the domain Sangrador-Vegas,A. et al. (2016) The Pfam protein families database: towards a more sustainable future. Nucleic Acids Res., 44, boundaries on some of our repeat entries. D279–D285. 4. Consortium,U.P.(2019) UniProt: a worldwide hub of protein knowledge. Nucleic Acids Res., 47, D506–D515. ACKNOWLEDGEMENTS 5. Chen,C., Natale,D.A., Finn,R.D., Huang,H., Zhang,J., Wu,C.H. and Mazumder,R. (2011) Representative proteomes: a stable, scalable and We are grateful to Philippe Le Mercier from the Swiss In- unbiased proteome set for sequence analysis and functional stitute of Bioinformatics who gave us valuable guidance for annotation. PLoS One, 6, e18910. our SARS-CoV-2 nomenclature. We are grateful to Layla 6. Finn,R.D., Mistry,J., Schuster-Bockler,B ¨ ., Griffiths-Jones,S., Hirsh Martinez and Aleix Lafita for adding families to Hollich,V., Lassmann,T., Moxon,S., Marshall,M., Khanna,A., Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 Nucleic Acids Research, 2021, Vol. 49, Database issue D419 Durbin,R. et al. (2006) Pfam: clans, web tools and services. Nucleic 17. Ma,Y., Wu,L., Shaw,N., Gao,Y., Wang,J., Sun,Y., Lou,Z., Yan,L., Acids Res., 34, D247–D251. Zhang,R. and Rao,Z. (2015) Structural basis and functional analysis 7. Soding,J ¨ . (2005) Protein homology detection by HMM-HMM of the SARS coronavirus nsp14-nsp10 complex. Proc. Natl. Acad. comparison. Bioinformatics, 21, 951–960. Sci. U.S.A., 112, 9436–9441. 8. Bursteinas,B., Britto,R., Bely,B., Auchincloss,A., Rivoire,C., 18. Kim,Y., Jedrzejczak,R., Maltseva,N.I., Wilamowski,M., Endres,M., Redaschi,N., O’Donovan,C. and Martin,M.J. (2016) Minimizing Godzik,A., Michalska,K. and Joachimiak,A. (2020) Crystal structure proteome redundancy in the UniProt Knowledgebase. Database, of Nsp15 endoribonuclease NendoU from SARS-CoV-2. Protein Sci., 2016, baw139. 29, 1596–1605. 9. Berman,H., Henrick,K. and Nakamura,H. (2003) Announcing the 19. Heger,A., Wilton,C.A., Sivakumar,A. and Holm,L. (2005) ADDA: a worldwide Protein Data Bank. Nat. Struct. Biol., 10, 980–980. domain database with global coverage of the protein universe. Nucleic 10. Berman,H., Henrick,K., Nakamura,H. and Markley,J.L. (2007) The Acids Res., 33, D188–D191. worldwide Protein Data Bank (wwPDB): ensuring a single, uniform 20. Altschul,S.F., Madden,T.L., Schaf ¨ fer,A.A., Zhang,J., Zhang,Z., archive of PDB data. Nucleic Acids Res., 35, D301–D303. Miller,W. and Lipman,D.J. (1997) Gapped BLAST and PSI-BLAST: 11. Mitchell,A.L., Almeida,A., Beracochea,M., Boland,M., Burgin,J., a new generation of protein database search programs. Nucleic Acids Cochrane,G., Crusoe,M.R., Kale,V., Potter,S.C., Richardson,L.J. Res., 25, 3389–3402. et al. (2020) MGnify: the microbiome analysis resource in 2020. 21. Hauser,M., Steinegger,M. and Soding,J ¨ . (2016) MMseqs software Nucleic Acids Res., 48, D570–D578. suite for fast and deep clustering and searching of large protein 12. Mitchell,A.L., Attwood,T.K., Babbitt,P.C., Blum,M., Bork,P., sequence sets. Bioinformatics, 32, 1323–1330. Bridge,A., Brown,S.D., Chang,H.-Y., El-Gebali,S., Fraser,M.I. et al. 22. Deorowicz,S., Debudaj-Grabysz,A. and Gudys,A. ´ (2016) FAMSA: (2019) InterPro in 2019: improving coverage, classification and access fast and accurate multiple sequence alignment of huge protein to protein sequence annotations. Nucleic Acids Res., 47, D351–D360. families. Sci. Rep., 6, 33964. 13. Lei,J., Kusov,Y. and Hilgenfeld,R. (2018) Nsp3 of coronaviruses: 23. Lupas,A., Van Dyke,M. and Stock,J. (1991) Predicting coiled coils Structures and functions of a large multi-domain protein. Antiviral from protein sequences. Science, 252, 1162–1164. Res., 149, 58–74. 24. Wootton,J.C. and Federhen,S. (1993) Statistics of local complexity in 14. Cong,Y., Ulasli,M., Schepers,H., Mauthe,M., V’kovski,P., amino acid sequences and sequence databases. Comput. Chem., 17, Kriegenburg,F., Thiel,V., de Haan,C.A.M. and Reggiori,F. (2020) 149–163. Nucleocapsid protein recruitment to replication–transcription 25. Piovesan,D., Tabaro,F., Paladin,L., Necci,M., Micetic,I., complexes plays a crucial role in coronaviral life cycle. J. Virol., 94, Camilloni,C., Davey,N., Dosztan ´ yi,Z., Mesz ´ ar ´ os,B., Monzon,A.M. e01925-19. et al. (2018) MobiDB 3.0: more annotations for intrinsic disorder, 15. Rawlings,N.D., Barrett,A.J., Thomas,P.D., Huang,X., Bateman,A. conformational diversity and interactions in proteins. Nucleic Acids and Finn,R.D. (2018) The MEROPS database of proteolytic Res., 46, D471–D476. enzymes, their substrates and inhibitors in 2017 and a comparison 26. Paladin,L., Hirsh,L., Piovesan,D., Andrade-Navarro,M.A., with peptidases in the PANTHER database. Nucleic Acids Res., 46, Kajava,A.V. and Tosatto,S.C.E. (2017) RepeatsDB 2.0: improved D624–D632. annotation, classification, search and visualization of repeat protein 16. Kirchdoerfer,R.N. and Ward,A.B. (2019) Structure of the structures. Nucleic Acids Res., 45, D308–D312. SARS-CoV nsp12 polymerase bound to nsp7 and nsp8 co-factors. 27. Kopec,K.O. and Lupas,A.N. (2013) -Propeller blades as ancestral Nat. Commun., 10, 2342. peptides in protein evolution. PLoS One, 8, e77074. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nucleic Acids Research Oxford University Press

Loading next page...
 
/lp/oxford-university-press/pfam-the-protein-families-database-in-2021-a0El280DWX

References (55)

Publisher
Oxford University Press
Copyright
© The Author(s) 2020. Published by Oxford University Press on behalf of Nucleic Acids Research.
ISSN
0305-1048
eISSN
1362-4962
DOI
10.1093/nar/gkaa913
Publisher site
See Article on Publisher Site

Abstract

Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 D412–D419 Nucleic Acids Research, 2021, Vol. 49, Database issue Published online 30 October 2020 doi: 10.1093/nar/gkaa913 1,* 1 1 1 Jaina Mistry , Sara Chuguransky , Lowri Williams , Matloob Qureshi , 1 2 3 3 Gustavo A. Salazar , Erik L.L. Sonnhammer , Silvio C.E. Tosatto , Lisanna Paladin , 1 1 1 1 Shriya Raj , Lorna J. Richardson , Robert D. Finn and Alex Bateman European Molecular Biology Laboratory, European Bioinformatics Institute (EMBL-EBI), Wellcome Genome Campus, Hinxton CB10 1SD, UK, Department of Biochemistry and Biophysics, Science for Life Laboratory, Stockholm University, Box 1031, 17121 Solna, Sweden and Department of Biomedical Sciences, University of Padua, 35131 Padova, Italy Received September 11, 2020; Revised October 01, 2020; Editorial Decision October 02, 2020; Accepted October 06, 2020 ABSTRACT per-family gathering thresholds. Pfam entries are manually annotated with functional information from the literature The Pfam database is a widely used resource for clas- where available. sifying protein sequences into families and domains. Since Pfam release 29.0, pfamseq is based on UniPro- Since Pfam was last described in this journal, over tKB reference proteomes, whilst prior to that, it was based 350 new families have been added in Pfam 33.1 and on the whole of UniProtKB (3,4). Although the underly- numerous improvements have been made to existing ing sequence database is based on reference proteomes, all of the profile HMMs are also searched against UniProtKB entries. To facilitate research on COVID-19, we have and the resulting matches are made available on the Pfam revised the Pfam entries that cover the SARS-CoV-2 website and in a flatfile format. Similarly, the match data proteome, and built new entries for regions that were for representative proteomes (5) are provided in the same not covered by Pfam. We have reintroduced Pfam-B way. Pfam can be accessed via the website at https://pfam. which provides an automatically generated supple- xfam.org. The flatfiles and exports of the Pfam MySQL ment to Pfam and contains 136 730 novel clusters of database are provided under a CC0 license for each re- sequences that are not yet matched by a Pfam fam- lease of the database, and can be found on the FTP site ily. The new Pfam-B is based on a clustering by the ftp://ftp.ebi.ac.uk/pub/databases/Pfam/releases. MMseqs2 software. We have compared all of the re- When a Pfam entry is built, we typically search it itera- gions in the RepeatsDB to those in Pfam and have tively against pfamseq in order to find more distant homo- started to use the results to build and refine Pfam logues. Pfam entries are built such that there are no over- repeat families. Pfam is freely available for browsing laps between them; this means that the same region of a se- quence should not match more than one family. This non- and download at http://pfam.xfam.org/. overlap rule proves to be an excellent quality control cri- teria which helps to avoid including false positive matches INTRODUCTION into a family. From Pfam 28.0, we relaxed this rule to allow small overlaps between families as it had become increas- Pfam is a database of protein families and domains that is ingly time-consuming to resolve all such overlaps each time widely used to analyse novel genomes, metagenomes and we updated pfamseq (see (3) for more details). to guide experimental work on particular proteins and sys- Sets of Pfam entries that we believe to be evolutionarily tems (1,2). Each Pfam family has a seed alignment that related are grouped together into clans (6). Relationships contains a representative set of sequences for the entry. between entries are identified through sequence similarity, A profile hidden Markov model (HMM) is automatically structural similarity, functional similarity and/or profile- built from the seed alignment and searched against a se- profile comparisons using software such as HHsearch ( 7). quence database called pfamseq using the HMMER soft- Where possible, we build a single comprehensive profile ware (http://hmmer.org/). All sequence regions that satisfy HMM to detect all members of a family. For some of the a family-specific curated threshold, also known as the gath- larger superfamilies where this is not possible, we build ering threshold, are aligned to the profile HMM to create multiple profile HMMs and put them in the same clan. the full alignment. It is worth noting that a common mis- As families in a clan are evolutionarily related, we allow use of Pfam is to use a single E-value threshold across all them to overlap with other members of the same clan. The Pfam HMMs, which results in lower sensitivity and an in- clans are competed such that if there are multiple over- crease in false positive matches when compared to using the To whom correspondence should be addressed. Tel: +44 1223 494100; Fax: +44 1223 494468; Email: jaina@ebi.ac.uk C The Author(s) 2020. Published by Oxford University Press on behalf of Nucleic Acids Research. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 Nucleic Acids Research, 2021, Vol. 49, Database issue D413 lapping profile HMM matches to families from a single FAMILY IMPROVEMENTS clan, only the match with the lowest E-value is displayed Despite almost 25 years of active curation, there remain on the website, or included in the full alignment of the many protein domains and families that are as yet unclassi- entry. fied by Pfam. New families added to Pfam are created from Here we describe Pfam 33.1 and some of the family up- a range of sources, including Pfam-B families and protein dates we have made since the last release. We also detail the structure. Pfam-B alignments in particular have been a very Pfam coverage of the SARS-CoV-2 proteome, and the new fruitful substrate for families, with historically nearly a third method to create Pfam-B which we re-introduced in Pfam of all Pfam entries being built from them. The new Pfam- 33.1. Finally we describe the results of the analysis we car- B (described below) was used to build 18 entries in Pfam ried out on comparing repeats in the Pfam database to Re- 33.1. We expect that Pfam-B will again become a very use- peatsDB. ful source of additional families in the coming years. Protein structures from databases such as the Protein PFAM 33.1 RELEASE Data Bank (PDB) (9,10) are particularly amenable as a source of new Pfam entries because the Pfam domain Pfam 33.0 was due to be released in March 2020, however boundaries can be defined precisely from the structure. due to the global COVID-19 pandemic, we delayed its re- There is often an associated paper to the structure which lease so that we could focus on improving our SARS-CoV- we use to help annotate the Pfam entry. Building families 2 models (see ‘COVID-19 updates’ section for more de- using protein structures is an ongoing activity, and some of tail). The updated SARS-CoV-2 models, and a handful of the new SARS-CoV-2 families were built in this way. Ad- other new Pfam entries were added to Pfam 33.0 to create ditionally, we have made 37 new families based on clus- Pfam 33.1, which was released in May 2020. Release 33.0 ters of sequences from the MGnify metagenomic protein was never officially released on the Pfam website, although database (11). The MGnify clusters may be another large the database files and flat files are made available on the source from which we could build families from in the fu- ftp site. ture. Using metagenomics clusters can help to cover regions Pfam 33.1 contains 18 259 families and 635 clans. Since of sequence space that are not well covered by existing ge- Pfam 32.0, we have built 355 new families, deleted 25 fami- nomic sequencing projects. lies and built 8 new clans. Just over 39% of all Pfam families Pfam has created numerous entries for domains of un- are placed within a clan. Of the sequences in UniProtKB, known function (DUF) and uncharacterized protein fami- 77.0% have at least one match to a Pfam entry, and 53.2% lies (UPF). Over time the functions of some of these are dis- residues in UniProtKB fall within a Pfam entry. These fig- covered. We continue to scan the literature to identify newly ures, termed the sequence and residue coverage respectively, identified functions as well as receiving updates from the have remained fairly constant over the past vfi e Pfam re- community via our helpdesk. Many functions are also iden- leases despite a 240% increase in UniProtKB size during tified by the InterPro database ( 12) update process which that time (see Figure 1). Since 2015, highly redundant bac- checks whether the UniProtKB/Swiss-Prot descriptions of terial proteomes have been identified and removed from proteins in each InterPro entry have changed between re- UniProtKB (8). This process ensures a level of diversity in leases. When these changes identify a function, Pfam is no- the sequences added to UniProtKB, and prevents, for ex- tified. To date we have changed the identifiers of 1132 DUF ample, multiple strains of a particular species being added. or UPF families indicating that a function has been iden- Pfam is able to maintain a sequence coverage of ∼77%, and tified for these families. As of release 33.1, Pfam contains a residue coverage of ∼53% largely due to the new sequences 4244 DUF or UPF families, which is 23% of all Pfam fami- being added to UniProtKB matching existing Pfam models. lies. This suggests that there are still a lot of uncharacterized As UniProtKB grows, it is progressively harder to increase families and domains for molecular biologists to study. the Pfam coverage since newer Pfam entries tend to cover a We regularly receive feedback from users about families small taxonomic range. Although new models add little in or domains that are missing in Pfam, and typically add terms of coverage, they may represent medically important many user submitted families at each release. We include the proteins. For example, during our SARS-CoV-2 model im- submitters name and ORCID identifier ( https://orcid.org/) provements, we built the new entry Pfam:PF19213, which where available as an author of such Pfam entries. This corresponds to non-structural protein 6 (NSP6), a protein helps people to get credit for community activities that im- involved in autophagosome generation. prove molecular biology databases such as Pfam. One such The Pfam 33.1 sequence and residue coverage of UniPro- user submission was from Heli Monttinen ¨ (University of tKB reference proteomes is 75.1 and 49.4%, respectively Helsinki) who submitted a large scale clustering of virus (slightly lower than the figure for all of UniProtKB men- families. Based on this clustering we added 88 new families tioned above). This is a 0.6% increase in sequence coverage, to Pfam. We currently have 144 non-Pfam authors listed by and 0.7% decrease in residue coverage compared to Pfam their ORCID and we encourage our users to continue to 32.0. Although there was only a 3% increase in the num- submit interesting potential new domains and families. ber of sequences in reference proteomes between Pfam re- leases 32.0 and 33.1, around 2000 bacterial proteomes were removed from the reference proteomes (due to quality is- COVID-19 UPDATES sues) during this time. The changes in coverage since Pfam 32.0 are likely to be due to the change in sequences within The SARS-CoV-2 pandemic has mobilized a worldwide the reference proteomes set. research effort to understand the pathogen itself and the Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 D414 Nucleic Acids Research, 2021, Vol. 49, Database issue Figure 1. Growth of UniProtKB, and its coverage by Pfam over the last vfi e Pfam releases. As UniProtKB grows in size, the Pfam sequence and residue coverage is maintained at ∼77 and ∼53%, respectively. The UniProtKB size in the figure corresponds to the version of UniProtKB we used for each Pfam release. mechanism of COVID-19 disease, as well as to identify Structural and accessory proteins treatment options. Pfam already provided useful annotation One of the most important proteins of this virus is the for SARS-CoV-2, but we sought to update our models, fam- Spike protein (S), which aids viral entry into the host ily names and annotations for this virus in an effort to help cell, and is key for its pathogenicity. The models that the research community. were already present in the database have been improved We assessed all the protein sequences provided by and we added a new domain entry. Following this, as the UniProt via its new COVID-19 portal (https://covid-19. S protein is translated into a large polypeptide which is uniprot.org/), identified those which lacked an existing cleaved by host proteases to produce S1 and S2 peptides, Pfam model, and built new models as required. We now we have now three domains corresponding to S1, the N- cover almost all gene products encoded by SARS-CoV-2 terminal domain (Pfam:PF16451), the receptor binding (Figure 2). Orf10, a small putative protein encoded at the 3 - domain (RBD) (Pfam:PF09408) and the new C-terminal end of the SARS-CoV-2 genome is the only protein which domain (Pfam:PF19209). S2 is described in the family remains unannotated by Pfam. It was not possible to build Pfam:PF01601, which contains an additional S2 cleavage a Pfam entry for it since it lacked any detectable homo- site, a fusion peptide (FP), internal fusion peptide (IFP), logues in UniProtKB. The mean percentage identity for heptad repeat 1/2 (HR1/2), and the transmembrane do- the full alignments (based on reference proteomes) of Pfam main (TM). entries that match SARS-CoV-2 proteins is 49% (range Entities for the other structural proteins in Pfam 15–96%). were updated, such as the Nucleocapsid protein (N) in We have standardized our identifier nomenclature and Pfam:PF00937, the matrix protein (M) Pfam:PF01635 and descriptions of the families to ensure they are both cor- the envelope E protein in Pfam:PF02723. Additional acces- rect and consistent. The majority of the family identifiers sory proteins encoded by coronaviruses, usually called non- now begin with either CoV, for coronavirus specific fami- structural accessory proteins (NS), although some of them lies, or bCoV for the families which are specific to the beta- constitute structural parts of the virion, were updated. For coronavirus clade, which SARS-CoV-2 belongs to. We have example, NS3a corresponds to the betacoronavirus viro- also fixed inconsistencies in the naming and descriptions porin in the Pfam entry Pfam:PF11289, and Pfam:PF09399 of the various non-structural proteins (NSPs), using NSPx describes the protein 9b encoded within the nucleocapsid for those proteins encoded by the replicase polyprotein and gene, which contains a lipid binding domain. The acces- NSx for those encoded by other ORFs. sory proteins NS7a and NS7b in the entries Pfam:PF08779 Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 Nucleic Acids Research, 2021, Vol. 49, Database issue D415 Figure 2. Schematic representation of Pfam coverage of the SARS-CoV-2 proteome. The top row of boxes represent the individual virus proteins processed from the precursor polyproteins. These boxes are coloured when they contain more than one Pfam domain and the individual Pfam entries are expanded below. and Pfam:PF11395, respectively, are important during the new family Pfam:PF19217 whilst its C-terminal domain replication cycle. NS8 in Pfam:PF12093 family may mod- is described in Pfam:PF16348 and the DUF5881 was ulate viral pathogenicity or replication in favour of human identified as NSP6 in Pfam:PF19213. The other protease adaptation and it has been to be one of the most important activity in these viruses is described in Pfam:PF05409 usu- genes in the study of the human adaptation of the virus. The ally known as Main protease (M-pro domain) or 3C-like accessory protein NS6 (Pfam:PF12133) is highly conserved proteinase (3CL-pro), corresponding to NSP5, a member amongst SARS-related coronaviruses and it can modulate of MEROPS (15) peptidase family C30. Other NSPs were host immune responses by inhibiting synthesis and sig- in the Pfam database and most of these families have been nalling of interferon-beta. The family Pfam:PF17635 de- extended to include SARS-Cov-2 sequences. For example, scribes the protein 14 encoded in Orf14 and its function is Pfam:PF08716 and Pfam:PF08717 corresponds to NSP7 currently unknown. and NSP8, respectively, which both form a hexadecameric supercomplex that adopts a hollow cylinder-like structure that play a role in the stabilization of NSP12 regions in- Non-structural proteins volved in RNA binding and are essential for a highly active NSP12 polymerase complex. NSP12 is an RNA-directed Regarding NSPs from coronaviruses, encoded by RNA polymerase described in two entries, Pfam:PF06478 ORF1a/1ab (replicase 1a/1ab), we updated the exist- and Pfam:PF00680 describing the N- and C-terminal ing entries and added new ones where appropriate. NSPs domains, respectively (16). form the replication–transcription complexes (RTCs) that Following annotation updates, NSP9 belongs to the are essential for the synthesis of viral RNA necessary family Pfam:PF08710, which is a single-stranded RNA- for the infection process or to avoid the host immune binding viral protein involved in RNA synthesis, essen- response. Amongst these proteins, NSP3 is the largest tial for the coronavirus replication. NSP10 included in replicase product including several conserved domains Pfam:PF09401 is amongst the more conserved coronavirus whose organization differ amongst coronavirus genera proteins. It interacts with NSP14 (Pfam:PF06471) and (13). It has a crucial function as it cleaves the proteins NSP16 (Pfam:PF06460) and regulates their respective ex- encoded by the replicase, including NSP3 itself. It is an onuclease (ExoN) and ribose-2 -O-MTase (2 -O-MTase) ac- essential component of the RTC and serves as a scaffold tivities. It has a Tyr-96 specific for SARS-CoV and it is of protein to interact with itself and other NSPs. We describe particular interest as it plays a crucial role in the NSP10– it in separate entries: the existing N-terminal domain NSP16 interaction and in the activation of the NSP16 2 -O- Pfam:PF12379 was improved and updated, and includes MTase activity as well as in the NSP10–NSP14 interaction the NSP3a domain which acts as a scaffold through its (17). interaction with numerous proteins involved in replication Based on recently solved structures of SARS-CoV-2 pro- and transcription processes and contains the ubiquitin-like teins, we were able to build new families, as is the case of 1 (UB1) and glutamic acid-rich acidic (AC) hypervariable NSP15. We built three new entries representing the three domains. NSP3a is essential for localizing the RNA in structural domains of the NSP15 protein (Figure 3)based the replicase–transcriptase complex at the first steps of on the structure by Kim et al.(18). infection, as it interacts with the nucleocapsid protein N in the nascent replicase/transcriptase complex (14). The other domains encoded in this protein are the Macro PFAM-B domain included in Pfam:PF01661 and SUD-M domain in Pfam:PF11633 that binds single-stranded poly(A). In addition to our HMM-based Pfam entries (Pfam-A), The domain Pfam:PF12124 was updated and is now we used to provide a set of unannotated, computation- described as the SUD-C or DPUP domain which binds ally generated multiple sequence alignments called Pfam- to single-stranded RNA and recognizes purine bases. B. The method used for Pfam-B construction changed sig- The papain-like protease (PLPro) crucial for polypeptide nificantly over the years. The final prior version of Pfam-B processing is described in Pfam:PF08715; the nucleic alignments were created from clusters generated by apply- acid-binding domain (NAR) belongs to Pfam:PF16251 ing the ADDA algorithm (19) to an all-against-all BLAST family and, lastly, the C-terminal domain of NSP3 was (20) search of a non-redundant version of UniProtKB, and added as the new entry Pfam:PF19218. Additionally, we removing any regions covered by Pfam-A. The version of built the N-terminal domain of NSP4 represented in the UniProtKB used in the all-against-all clustering became in- Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 D416 Nucleic Acids Research, 2021, Vol. 49, Database issue Table 1. The number of each Pfam type for the last two Pfam releases 32.0 and 33.1 Type Pfam release 32.0 Pfam release 33.1 Family 11 177 11 242 Domain 6248 6406 Repeat 260 268 Coiled-coil 95 171 Motif 82 91 Disordered 67 81 der and Repeat types within Pfam and report these results below. DISORDER IN PFAM We investigated the level of predicted disorder across all Figure 3. The structure of NSP15 (PDB ID: 6VWW) from Kim et al. Pfam families. The mean percentage identity for full align- shows the three new Pfam domains. (i) CoV NSP15 N (Pfam:PF19219) Coronavirus replicase NSP15, N-terminal oligomerization domain in red, ments (based on reference proteomes) of Pfam entries with (ii) CoV NSP15 M (Pfam:PF19216) Coronavirus replicase NSP15, mid- type Disordered is 55% (range 18–94%). For each family dle domain in blue and (iii) CoV NSP15 C (Pfam:PF19215) Coronavirus we determined the fraction of residues in the seed align- replicase NSP15, uridylate-specific endoribonuclease in green. ment that were predicted to be low complexity by segmasker (24) or disordered by MobiDB-lite (25). A histogram of scores is shown in Figure 4. There are 303 Pfam families creasingly out of sync with that used by Pfam. This, com- that have over 80% of seed alignment residues predicted as bined with the time it took to produce it meant that as of disordered/low complexity, and yet only 81 are currently Pfam 28.0 (released in 2015), it was no longer feasible to classified as disordered. We have begun to reclassify these make Pfam-B (see (3) for a longer discussion on why we families for release 34.0. stopped producing Pfam-B). We have now devised an alter- native method of making Pfam-B that uses the MMSeqs2 software (21) which has a much lower computational cost. REPEATS IN PFAM The new pipeline was also designed to produce alignments that are more likely to represent new domain families. To investigate the Repeat type entries in Pfam, we compared Pfam-B was created by clustering pfamseq sequence seg- them to the gold-standard repeat database RepeatsDB (26). ments longer than 50 residues not covered by Pfam-A do- This resource contains a rigorous structural classification mains, using MMSeqs2 run with the cluster option and of repeat proteins from the PDB and places each into a re- bidirectional coverage mode. Multiple sequence alignments peat ontology. Repeat regions in RepeatsDB are detected by were generated for each cluster with more than 20 sequences structure and annotated with the position of each repeated using FAMSA (22). This resulted in 136,730 Pfam-B fami- unit. The comparison to Pfam highlights the differences be- lies that on average contain 99 sequences (maximum 40 912) tween sequence- and structure-based repeat identification, and are 310 positions wide (maximum 29 216). domain annotation as well as classification, and is relevant The Pfam-B alignments are presently only released as a in the context of our ongoing effort of improving repeat def- tar archive on the Pfam FTP site (file Pfam-B.tgz). We have initions. not built profile HMMs for the entries or integrated them Pfam covers 64.2% of the repeat regions found in Re- into the Pfam website. The entries are sorted such that the peatsDB (Figure 5) and shows a bimodal distribution, with first entries have an optimal combination of size and con- peaks at 0 and 100%. Therefore, most RepeatsDB entries servation and would therefore have the highest chance of (1415 UniProtKB proteins) are either fully characterized representing novel and useful domain families. or uncharacterized by Pfam families. This analysis allowed us to identify candidates for future Pfam curation, that present 0% coverage, such as the STU2 protein (UniPro- PFAM TYPE DEFINITIONS tKB:P46675) containing HEAT repeats (Figure 5), the - Pfam type definitions divide entries into one of six types propeller domain in DDB1- and CUL4-associated factor and they can help users in selecting which Pfam families to 1 (UniProtKB:Q9Y4B6) and the LRR domain in Ran use in their analyses. Over the past year we have made nu- GTPase-activating protein 1 (UniProtKB:P41391). These merous changes and updates to the Pfam type definitions sequences will be aligned with structurally similar entries for families (Table 1). In particular, we have carried out a in RepeatsDB to derive the input alignments for Pfam, and large scale screen of Pfam families using the ncoils software the eventual overlap with existing sequence models (e.g. the (23) to identify families with a high proportion of predicted existing Pfam LRR and HEAT repeat domains) will be in- coiled-coil, and after inspection of such families, we were vestigated and resolved. able to change their type. We investigated the number and types of Pfam domains We were interested to see if we can improve the definitions detected in RepeatsDB regions. A total of 573 Pfam do- of the other types in Pfam. We have investigated the Disor- mains in 106 Pfam clans are mapped to repeats, of which Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 Nucleic Acids Research, 2021, Vol. 49, Database issue D417 Figure 4. Percent of residues in the seed alignment of Pfam entries that are low complexity or disordered as predicted by segmasker and MobiDB-lite, respectively. only 91 (15.9%) are of type Repeat. A total of 176 (30.7%) of the following one (Figure 5). This type of unit is called are of type Family, 303 (52.9%) Domain, 1 Coiled-coil ‘Velcro’ and has stabilizing functions (27). The two phases (mapping to entries in RepeatsDB fibrous structures) and are therefore related to two different concepts, and these 2 Motif. The large majority of Family and Domain types types of examples highlight cases where the structural re- may be justified by the fact that RepeatsDB includes a class, peat patterns may be used to revise the sequence one or Beads-on-a-string, whose repeat units are large enough to vice versa. (iii) Finally, different Pfam models may map to fold independently (26), likely corresponding to Pfam type the same repeat region, and correspond to different num- domain rather than type repeat. Indeed, class V entries ber of units, such as in the Ankyrin region (Figure 5)of mostly (86.5%) map to Domains in Pfam. However, Re- the Neurogenic locus notch homologue protein 1 (UniPro- peatsDB elongated structures, i.e. the most canonical re- tKB:P46531), where the three types of Pfam models de- peats, including solenoids (26), are still mostly mapped to tected (Pfam:PF00023, Pfam:PF13637 and Pfam:PF12796) non-Repeat Pfam types (22.2% are Repeat), including the map respectively to one, two and three repeat units. With two Motif entries. this arrangement, one unit and a fragment are actually These data will support the revision of Pfam types as- skipped by Pfam models; these cases will require an exten- signment in repeat entries, also in light of the RepeatsDB sive investigation of the optimal number of units included units to Pfam domains correspondence. Pfam models may in Pfam models and the potential overlaps. be defined in different ways: (i) they may map to the entire repeat region, such as the Ricin-type beta-trefoil lectin do- DISCUSSION main (Pfam:PF00652) in the rRNA N-glycosidase (UniPro- tKB:B7 × 8M2). The model for this domain was initially de- The Pfam sequence and residue coverage of UniProtKB has signed to correspond to single units, but it was later revised remained fairly constant over the past few years at ∼77 and and updated to the current version to increase its sensitiv- ∼53%, respectively. This means that although Pfam aims ity, supported by the observation that this type of repeat to be comprehensive, there remains a significant area of se- region usually includes a tandem repeat of three units. In quence space that has no annotation by Pfam. As UniPro- this case, the type assignment to Domain should be kept. tKB grows, it becomes progressively harder to increase the (ii) Pfam models may map to each repeat unit, with the coverage as the larger ubiquitous families have already been same phase, as the RCC1 repeats (Pfam:PF00415) in the built, and newer families tend to have a smaller taxonomic human Regulator of chromosome condensation (UniPro- range. Nevertheless there are still many important families tKB:P18754), or with a different phase, as the WD do- left to build, and we plan to concentrate our efforts on fam- main (Pfam:PF00400) in the WD repeat-containing pro- ily building for the next release. New families will be built tein 5 (UniProtKB:P61964). In the latter, whilst the Re- from a range of sources such that we represent the diver- peatsDB phase is defined as corresponding to a single en- sity in the tree of life. Sources will include Pfam-B clus- tire blade of the -propeller, the Pfam model maps to a ters, which we re-introduced into Pfam in the current release segment including most of each blade and the first -sheet (version 33.1). We will also continue to build families from Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 D418 Nucleic Acids Research, 2021, Vol. 49, Database issue Figure 5. Pfam coverage of repeat regions in UniProtKB entries from RepeatsDB. Three examples are shown represented by their PDB structures. On the top left PDB ID: 4ffb, chain C, mapping to a region of the HEAT repeats in UniProtKB:P46675 (residues 1–272), with Pfam coverage 0%. In the centre, the Ankyrin region of UniProtKB:P46531, PDB ID: 6py8, chain K (residues 1759–2127), with Pfam coverage 89.4%. Three Pfam domains are detected: Pfam:PF12796 in yellow, Pfam:PF13637 in orange and Pfam:PF00023 in red. On the top right, PDB ID: 3ur4, chain A, mapping to the -propeller UniProtKB:P61964 (residues 24–334), with Pfam coverage 93%. Only one type of Pfam domain is detected (Pfam:PF00400), shown in alternating shades of blue to facilitate the visualization of the Pfam model phase. metagenomic sequence clusters, PDB structures and com- Pfam and to the numerous scientists who contributed sug- munity submissions. gestions and families via our helpdesk. In response to the COVID-19 pandemic, we have revised all existing families that match SARS-CoV-2 proteins, and FUNDING built new profile HMMs to cover regions previously unan- notated by Pfam. All proteins in the SARS-CoV-2 pro- European Union’s Horizon 2020 MSCA-RISE ac- teome are covered by Pfam apart from the putative pro- tion [823886]; Wellcome [108433/Z/15/Z]; BBSRC tein encoded by Orf10, which lacked any detectable homo- [BB/S020381/1]; Open Targets; European Molecular logues in UniProtKB. We hope our models help the research Biology Laboratory Core Funds. Funding for open access charge: Wellcome; Research Councils UK (RCUK). community to identify and annotate coronavirus sequences. Conflict of interest statement. None declared. Additionally, multiple sequence alignments generated using Pfam HMMs may prove useful in tracking the evolution of coronaviruses. REFERENCES Since the last release, we have updated some of our type 1. Sonnhammer,E.L., Eddy,S.R. and Durbin,R. (1997) Pfam: a definitions, particularly those that are predicted to have comprehensive database of protein domain families based on seed coiled-coil regions. We have also identified a set of families alignments. Proteins, 28, 405–420. that have a high proportion of disordered residues, and plan 2. El-Gebali,S., Mistry,J., Bateman,A., Eddy,S.R., Luciani,A., to reclassify these families as type disorder for the next Pfam Potter,S.C., Qureshi,M., Richardson,L.J., Salazar,G.A., Smart,A. et al. (2019) The Pfam protein families database in 2019. Nucleic release. We have also compared Pfam to the RepeatsDB Acids Res., 47, D427–D432. database. Using the comparison, we will look to change the 3. Finn,R.D., Coggill,P., Eberhardt,R.Y., Eddy,S.R., Mistry,J., type of some of the Pfam entries to repeat, incorporate addi- Mitchell,A.L., Potter,S.C., Punta,M., Qureshi,M., tional RepeatsDB regions into Pfam and revise the domain Sangrador-Vegas,A. et al. (2016) The Pfam protein families database: towards a more sustainable future. Nucleic Acids Res., 44, boundaries on some of our repeat entries. D279–D285. 4. Consortium,U.P.(2019) UniProt: a worldwide hub of protein knowledge. Nucleic Acids Res., 47, D506–D515. ACKNOWLEDGEMENTS 5. Chen,C., Natale,D.A., Finn,R.D., Huang,H., Zhang,J., Wu,C.H. and Mazumder,R. (2011) Representative proteomes: a stable, scalable and We are grateful to Philippe Le Mercier from the Swiss In- unbiased proteome set for sequence analysis and functional stitute of Bioinformatics who gave us valuable guidance for annotation. PLoS One, 6, e18910. our SARS-CoV-2 nomenclature. We are grateful to Layla 6. Finn,R.D., Mistry,J., Schuster-Bockler,B ¨ ., Griffiths-Jones,S., Hirsh Martinez and Aleix Lafita for adding families to Hollich,V., Lassmann,T., Moxon,S., Marshall,M., Khanna,A., Downloaded from https://academic.oup.com/nar/article/49/D1/D412/5943818 by DeepDyve user on 05 January 2021 Nucleic Acids Research, 2021, Vol. 49, Database issue D419 Durbin,R. et al. (2006) Pfam: clans, web tools and services. Nucleic 17. Ma,Y., Wu,L., Shaw,N., Gao,Y., Wang,J., Sun,Y., Lou,Z., Yan,L., Acids Res., 34, D247–D251. Zhang,R. and Rao,Z. (2015) Structural basis and functional analysis 7. Soding,J ¨ . (2005) Protein homology detection by HMM-HMM of the SARS coronavirus nsp14-nsp10 complex. Proc. Natl. Acad. comparison. Bioinformatics, 21, 951–960. Sci. U.S.A., 112, 9436–9441. 8. Bursteinas,B., Britto,R., Bely,B., Auchincloss,A., Rivoire,C., 18. Kim,Y., Jedrzejczak,R., Maltseva,N.I., Wilamowski,M., Endres,M., Redaschi,N., O’Donovan,C. and Martin,M.J. (2016) Minimizing Godzik,A., Michalska,K. and Joachimiak,A. (2020) Crystal structure proteome redundancy in the UniProt Knowledgebase. Database, of Nsp15 endoribonuclease NendoU from SARS-CoV-2. Protein Sci., 2016, baw139. 29, 1596–1605. 9. Berman,H., Henrick,K. and Nakamura,H. (2003) Announcing the 19. Heger,A., Wilton,C.A., Sivakumar,A. and Holm,L. (2005) ADDA: a worldwide Protein Data Bank. Nat. Struct. Biol., 10, 980–980. domain database with global coverage of the protein universe. Nucleic 10. Berman,H., Henrick,K., Nakamura,H. and Markley,J.L. (2007) The Acids Res., 33, D188–D191. worldwide Protein Data Bank (wwPDB): ensuring a single, uniform 20. Altschul,S.F., Madden,T.L., Schaf ¨ fer,A.A., Zhang,J., Zhang,Z., archive of PDB data. Nucleic Acids Res., 35, D301–D303. Miller,W. and Lipman,D.J. (1997) Gapped BLAST and PSI-BLAST: 11. Mitchell,A.L., Almeida,A., Beracochea,M., Boland,M., Burgin,J., a new generation of protein database search programs. Nucleic Acids Cochrane,G., Crusoe,M.R., Kale,V., Potter,S.C., Richardson,L.J. Res., 25, 3389–3402. et al. (2020) MGnify: the microbiome analysis resource in 2020. 21. Hauser,M., Steinegger,M. and Soding,J ¨ . (2016) MMseqs software Nucleic Acids Res., 48, D570–D578. suite for fast and deep clustering and searching of large protein 12. Mitchell,A.L., Attwood,T.K., Babbitt,P.C., Blum,M., Bork,P., sequence sets. Bioinformatics, 32, 1323–1330. Bridge,A., Brown,S.D., Chang,H.-Y., El-Gebali,S., Fraser,M.I. et al. 22. Deorowicz,S., Debudaj-Grabysz,A. and Gudys,A. ´ (2016) FAMSA: (2019) InterPro in 2019: improving coverage, classification and access fast and accurate multiple sequence alignment of huge protein to protein sequence annotations. Nucleic Acids Res., 47, D351–D360. families. Sci. Rep., 6, 33964. 13. Lei,J., Kusov,Y. and Hilgenfeld,R. (2018) Nsp3 of coronaviruses: 23. Lupas,A., Van Dyke,M. and Stock,J. (1991) Predicting coiled coils Structures and functions of a large multi-domain protein. Antiviral from protein sequences. Science, 252, 1162–1164. Res., 149, 58–74. 24. Wootton,J.C. and Federhen,S. (1993) Statistics of local complexity in 14. Cong,Y., Ulasli,M., Schepers,H., Mauthe,M., V’kovski,P., amino acid sequences and sequence databases. Comput. Chem., 17, Kriegenburg,F., Thiel,V., de Haan,C.A.M. and Reggiori,F. (2020) 149–163. Nucleocapsid protein recruitment to replication–transcription 25. Piovesan,D., Tabaro,F., Paladin,L., Necci,M., Micetic,I., complexes plays a crucial role in coronaviral life cycle. J. Virol., 94, Camilloni,C., Davey,N., Dosztan ´ yi,Z., Mesz ´ ar ´ os,B., Monzon,A.M. e01925-19. et al. (2018) MobiDB 3.0: more annotations for intrinsic disorder, 15. Rawlings,N.D., Barrett,A.J., Thomas,P.D., Huang,X., Bateman,A. conformational diversity and interactions in proteins. Nucleic Acids and Finn,R.D. (2018) The MEROPS database of proteolytic Res., 46, D471–D476. enzymes, their substrates and inhibitors in 2017 and a comparison 26. Paladin,L., Hirsh,L., Piovesan,D., Andrade-Navarro,M.A., with peptidases in the PANTHER database. Nucleic Acids Res., 46, Kajava,A.V. and Tosatto,S.C.E. (2017) RepeatsDB 2.0: improved D624–D632. annotation, classification, search and visualization of repeat protein 16. Kirchdoerfer,R.N. and Ward,A.B. (2019) Structure of the structures. Nucleic Acids Res., 45, D308–D312. SARS-CoV nsp12 polymerase bound to nsp7 and nsp8 co-factors. 27. Kopec,K.O. and Lupas,A.N. (2013) -Propeller blades as ancestral Nat. Commun., 10, 2342. peptides in protein evolution. PLoS One, 8, e77074.

Journal

Nucleic Acids ResearchOxford University Press

Published: Jan 8, 2021

There are no references for this article.