Donovan H. ParkspersName, Maria ChuvochinapersName, Christian RinkepersName, Aaron J. MussigpersName, Pierre-Alain ChaumeilpersName, Philip HugenholtzpersName (2021date). “GTDB: an ongoing census of bacterial and archaeal diversity through a phylogenetically consistent, rank normalized and complete genome-based taxonomy.” Nucleic Acids Researchtitle·venue.
Object of study — A SPECIESterm·def BOUNDARY WHOSE JUSTIFICATION WAS TESTED, DISPROVED AND KEPT ON PURPOSE - the 95-96% average nucleotide identity threshold, defended in the literature by a claimed genetic discontinuity that this paper looks for in its own 254,090 genomes and does not find
Research question — whether prokaryotic speciesterm·def are discrete enough for a similarity threshold to be a natural boundary, and how to keep a genome-based taxonomy stable while it grows by 270% and its ranks are set by two continuous statistics
Contribution — a rank-normalised genome taxonomy in which higher ranks are delineated by relative evolutionary divergence and speciesterm·def by ANI, with a published policy for names, a quantitative rule for replacing type genomes, and a measurement showing the speciesterm·def threshold has no natural discontinuity behind it
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | prokaryotic diversity as a continuum, most of it never cultured and most of it unnamed, being cut into named classes so that it can be talked about at all — “The use of ANI to delineate species despite the lack of clear evidence for discrete species boundaries in the GTDB dataset is a pragmatic approach for organizing the rapidly growing biodiversity being discovered with metagenomic approaches ( 2 , 3 , 15 ). Leaving this diversity unclassified would severely limit the ability to effectively communicate scientific results as exemplified by 28.5% of genomes at NCBI lacking a species classification and 77.0% of GTDB R06-RS202 species clusters having an alphanumeric placeholder name.” | |
| claim-type | method — “We conclude with a discussion on the use of average nucleotide identities as a pragmatic approach for delineating prokaryotic species.” | |
| method — evidence | — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives as previously described ( 5 ).” | |
| object of study — evidence | — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | |
| research question — evidence | — “An ANI threshold of 95–96% is widely considered the ‘gold standard’ for quantitatively delineating species ( 27–30 ) and it has recently been proposed that a genetic discontinuity exists between 83% and 95% ANI which would further support the use of ANI for defining species boundaries ( 31 , 32 ). This genetic discontinuity has also been observed between genomes recovered from specific environments such as the human gut and soil ( 33 ).” | |
| contribution — evidence | — “In order to taxonomically organise this large and growing genome dataset, the GTDB uses relative evolutionary divergence (RED) to delineate higher-rank taxa and average nucleotide identity (ANI) to delineate species clusters ( 4 , 5 ). Use of these quantitative criteria for circumscribing taxa allows for automated classification of new genomes ( 6 ), ensures all genomes are classified from species to domain, and normalizes the definition of taxonomic ranks across the bacterial and archaeal domains.” | |
| mechanism | THE JUSTIFICATION FOR THE SPECIES THRESHOLD WAS TESTED AND FAILED: `An ANI threshold of 95-96% is widely considered the `GOLD STANDARD` ... and it has recently been proposed that a genetic discontinuity exists between 83% and 95% ANI which would further support the use of ANI for defining species boundaries.` Re-tested here: `If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% ... In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%.` The gap the line was supposed to sit inside is not there — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | mechanisms[0] |
| mechanism | AND THE THRESHOLD IS KEPT, WITH THE REASON WRITTEN DOWN: `The use of ANI to delineate species despite the lack of clear evidence for discrete species boundaries in the GTDB dataset is a pragmatic approach for organizing the rapidly growing biodiversity ... Leaving this diversity unclassified would severely limit the ability to effectively communicate scientific results.` Read against simonneau2018 in the same corpus - `THIS DEFINITION IS, THEREFORE, NO LONGER ARBITRARY, BUT BASED ON A SCIENTIFIC APPROACH`. One field says it has removed the arbitrariness; this one measures for a natural boundary, does not find it, and keeps the rule anyway — “The use of ANI to delineate species despite the lack of clear evidence for discrete species boundaries in the GTDB dataset is a pragmatic approach for organizing the rapidly growing biodiversity being discovered with metagenomic approaches ( 2 , 3 , 15 ). Leaving this diversity unclassified would severely limit the ability to effectively communicate scientific results as exemplified by 28.5% of genomes at NCBI lacking a species classification and 77.0% of GTDB R06-RS202 species clusters having an alphanumeric placeholder name.” | mechanisms[1] |
| mechanism | AND THE ARGUMENT FOR DOING SO IS QUOTED FROM SOMEBODY ELSE, IN FULL: `What we want from a species definition is a set of easily applied and stable rules by which to decide when two organisms are similar enough ... to be given the same name. The needs for such a guide to taxonomic practice in medicine, biotechnology and defense are obvious, and even arbitrary rules to satisfy them would be better than no rules at all.` The clearest statement in the corpus that a definition can be justified by its usefulness rather than by nature - and it is a quotation, not this paper's own sentence — “Doolittle and Papke ( 37 ) best articulate the rationale for a pragmatic species definition: ‘What we want from a species definition is a set of easily applied and stable rules by which to decide when two organisms are similar enough in their genomic and/or phenotypic properties to be given the same name. The needs for such a guide to taxonomic practice in medicine, biotechnology and defense are obvious, and even arbitrary rules to satisfy them would be better than no rules at all.’” | mechanisms[2] |
| mechanism | AND THE RANKS ARE NORMALISED BY A CONTINUOUS STATISTIC: `the GTDB uses RELATIVE EVOLUTIONARY DIVERGENCE (RED) to delineate higher-rank taxa and average nucleotide identity (ANI) to delineate species clusters. Use of these quantitative criteria for circumscribing taxa allows for automated classification ..., ensures all genomes are classified from species to domain, and normalizes the definition of taxonomic ranks across the bacterial and archaeal domains.` A genus in one phylum and a genus in another are made comparable by a number - which is the thing traditional taxonomy could never guarantee — “In order to taxonomically organise this large and growing genome dataset, the GTDB uses relative evolutionary divergence (RED) to delineate higher-rank taxa and average nucleotide identity (ANI) to delineate species clusters ( 4 , 5 ). Use of these quantitative criteria for circumscribing taxa allows for automated classification of new genomes ( 6 ), ensures all genomes are classified from species to domain, and normalizes the definition of taxonomic ranks across the bacterial and archaeal domains.” | mechanisms[3] |
| mechanism | AND STABILITY IS ENFORCED BY A SCORE WITH AN EXPLICIT ASYMMETRY: `representatives should only be replaced by equivalent assemblies of demonstrably higher quality ... quantitatively defined by the BALANCED ANI SCORE (BAS) which is 0.5 x (ANI score) + 0.5 x (quality score) ... An existing representative is only replaced ... IF IT HAS A BAS >= 10 ABOVE the BAS of the current representative. This achieves the goal of taxonomic stability by requiring A NEW REPRESENTATIVE TO BE OF INCREASINGLY HIGHER QUALITY THE MORE DISSIMILAR IT IS from the current representative.` The cost of changing what a name points at rises with how much the pointing would move — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” | mechanisms[4] |
| mechanism | AND A NAME CAN TURN OUT TO HAVE POINTED AT THE WRONG ORGANISM: `GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus WHILE GCF_004359515.1 is assembled from the type strain of M. carouselicus, indicating that prior to the ... assembly being updated, an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB.` For at least one release a species cluster carried one name and contained the other organism, and the correction came from an assembly update rather than from biology — “This discrepancy can be explained by the fact that GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. carouselicus indicating that prior to the GCF_004359525.1 assembly being updated, an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB. While the exception, such large-scale reassignments and assembly errors do occur and we make a best effort to identify and handle all such cases.” | mechanisms[5] |
| mechanism | AND MOST OF THE TAXONOMY HAS NO NAME AND NO ORGANISM: `28.5% of genomes at NCBI lacking a species classification and 77.0% of GTDB R06-RS202 species clusters having an alphanumeric placeholder name`; and `over 70% of archaeal species, genera, families, and orders lacking an available cultured representative`. Three quarters of the species in a species database are named by a string like CG2-30-70-394, and half the taxa exist only as assembled sequence — “Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGs with over 70% of archaeal species, genera, families, and orders lacking an available cultured representative (Figure 1F ).” | mechanisms[6] |
| mechanism | AND THE NAMING POLICY IS FIVE RULES ABOUT WHO MAY NAME WHAT: Latin names are `no longer proposed by GTDB curators unless there is an associated publication with taxon descriptions`; names above genus `without designated type material ... will no longer be incorporated`; `Latin names in preprints are also no longer incorporated ... as names can change between preprint and peer-reviewed publication`; and placeholder names are never swapped for other placeholder names, `EVEN WHEN [THEY] ARE UNWIELDY, e.g. CG2-30-70-394 which is used as A genus, family, order, class and phylum name`. Continuity of reference beats readability, explicitly — “Finally, to preserve name continuity, we do not replace alphanumeric placeholder names with other placeholder names, even when names (mostly introduced in early releases) are unwieldy, e.g. CG2-30-70-394 which is used as a genus, family, order, class and phylum name.” | mechanisms[7] |
| mechanism | AND A DEFINITIONAL PARAMETER IS ABOUT TO MOVE BECAUSE OF DATA QUALITY: `we expect to change the AF criteria used for assigning genomes to GTDB species clusters from 0.65 to 0.5 starting in the next GTDB release in order to better accommodate the growing number of incomplete MAGs contained in the GTDB which can cause the AF between MAGs to be artificially low.` The membership rule is loosened because the evidence got worse, not because the organisms changed - and the sentence says so — “Related to MAG quality, we expect to change the AF criteria used for assigning genomes to GTDB species clusters from 0.65 to 0.5 starting in the next GTDB release in order to better accommodate the growing number of incomplete MAGs contained in the GTDB which can cause the AF between MAGs to be artificially low.” | mechanisms[8] |
| mechanism | AND DISCRETENESS IS MEASURED RATHER THAN ASSUMED: `2.2% can be described as fuzzy as they have one or more genomes with an interspecific ANI value >=95% AND 89.5% can be described as discrete as they contain no genomes with an interspecific ANI value >94% ... but this result should be interpreted with care as most bacterial and archaeal species lack genomic representation.` The clusters are mostly clean; the boundary between them is not a gap in nature - both are true, and the paper keeps both — “In terms of species, 2.2% can be described as fuzzy as they have one or more genomes with an interspecific ANI value ≥95% and 89.5% can be described as discrete as they contain no genomes with an interspecific ANI value >94% (Figure 4F ). Encouragingly, this indicates that GTDB species clusters are largely discrete, but this result should be interpreted with care as most bacterial and archaeal species lack genomic representation based on even conservative estimates of the number of prokaryotic species in nature ( 36 ).” | mechanisms[9] |
| extends | the earlier GTDB papers that introduced relative evolutionary divergence for higher ranks and ANI-based species clusters, whose criteria this release applies and re-tests — “Importantly, we implemented an ANI-based method for delineating species starting with R04-RS89 (June 2019) which allows all genomes to be classified at the rank of species and provides a uniform estimate of species-level diversity within the GTDB ( 5 ).” | claim_relations[0] |
| critiques | the claim that a genetic discontinuity exists between 83% and 95% ANI, which is the main argument for treating the species threshold as a natural boundary — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | claim_relations[1] |
| extends | LPSN, the NCBI Taxonomy and the Living Tree Project, which supply type material, nomenclatural status and 16S classification to this taxonomy — “The LPSN (List of Prokaryotic names with Standing in Nomenclature) database ( 12 ) is used to establish co-identical strain identifiers for the type strains of species and subspecies, the types of higher-rank taxa, and the nomenclatural status of newly proposed or reclassified taxa. The Living Tree Project ( 13 ) is used to classify 16S rRNA sequences and help resolve ambiguity in regard to the correct classification of genomes.” | claim_relations[2] |
| extends | Doolittle and Papke's argument for a pragmatic species definition, quoted in full as the rationale for keeping a threshold with no natural boundary behind it — “Doolittle and Papke ( 37 ) best articulate the rationale for a pragmatic species definition: ‘What we want from a species definition is a set of easily applied and stable rules by which to decide when two organisms are similar enough in their genomic and/or phenotypic properties to be given the same name. The needs for such a guide to taxonomic practice in medicine, biotechnology and defense are obvious, and even arbitrary rules to satisfy them would be better than no rules at all.’” | claim_relations[3] |
| sense | `species` (GTDB sense) - a cluster around a single representative genome, entered by ANI and alignment fraction to that representative; NOT a claim about a discontinuity in nature — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives as previously described ( 5 ).” | senses[0] |
| sense | `rank normalized` - higher ranks set by relative evolutionary divergence so that a genus in one phylum is comparable with a genus in another — “In order to taxonomically organise this large and growing genome dataset, the GTDB uses relative evolutionary divergence (RED) to delineate higher-rank taxa and average nucleotide identity (ANI) to delineate species clusters ( 4 , 5 ). Use of these quantitative criteria for circumscribing taxa allows for automated classification of new genomes ( 6 ), ensures all genomes are classified from species to domain, and normalizes the definition of taxonomic ranks across the bacterial and archaeal domains.” | senses[1] |
| sense | `fuzzy species` - a species with one or more genomes at >=95% ANI to a genome in another species; 2.2% of those examined. The complement, `discrete`, is 89.5% — “In terms of species, 2.2% can be described as fuzzy as they have one or more genomes with an interspecific ANI value ≥95% and 89.5% can be described as discrete as they contain no genomes with an interspecific ANI value >94% (Figure 4F ). Encouragingly, this indicates that GTDB species clusters are largely discrete, but this result should be interpreted with care as most bacterial and archaeal species lack genomic representation based on even conservative estimates of the number of prokaryotic species in nature ( 36 ).” | senses[2] |
| sense | `placeholder name` - an alphanumeric string standing in for a Latin name, carried by 77.0% of GTDB species clusters and never replaced by another placeholder — “Finally, to preserve name continuity, we do not replace alphanumeric placeholder names with other placeholder names, even when names (mostly introduced in early releases) are unwieldy, e.g. CG2-30-70-394 which is used as a genus, family, order, class and phylum name.” | senses[3] |
| sense | `Balanced ANI Score` - `0.5 x (ANI score) + 0.5 x (quality score)`, the rule deciding when the genome a species name points at may be swapped — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” | senses[4] |
| control | repeating the discontinuity analysis using only genomes assembled from type strains — held — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” — rules out the objection that the missing discontinuity is an artefact of how GTDB picks species representatives — “This lack of a genetic discontinuity also persists when considering only genomes assembled from the type strain of a species, indicating that this result is independent of the methodology used to select GTDB species representatives (Figure 4B ).” | controls[0] |
| control | requiring a Balanced ANI Score at least 10 above the incumbent before a species representative may be replaced — held — “GTDB species representatives have been largely stable since the introduction of ANI-based species clusters in RS04-RS89 with 96.7% of representatives being unchanged on average between releases (Table 2 ). This stability is particularly encouraging given the rapid growth in the number of GTDB species clusters (Figure 1C and D ; Table 2 ). Numerous factors account for the changes that have occurred illustrating the benefits and need to regularly update taxonomic frameworks.” — rules out a name drifting to a different genome every release as new assemblies arrive — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” | controls[1] |
| control | refusing Latin names without an associated publication, without designated type material, or published only as a preprint — held — “Latin names in preprints are also no longer incorporated into the GTDB even if their type genomes are already in the database as names can change between preprint and peer-reviewed publication ( 25 , 26 ).” — rules out names that cannot be anchored to a genome, and names that change between preprint and publication — “Effectively published Latin names above the rank of genus without designated type material, either a sequenced type strain or MAG, will no longer be incorporated into GTDB, and those that do will only be introduced when the associated type genome is present in GTDB. This change is necessary as establishing the correct interior node in the reference tree for taxa without type material can be ambiguous, particularly when the addition of new genomes or alternative inference methods results in the named taxon becoming polyphyletic in later releases.” | controls[2] |
| control | never replacing an alphanumeric placeholder with another placeholder, however unwieldy — held — “As of R06-RS202, Latin names are no longer proposed by GTDB curators unless there is an associated publication with taxon descriptions ( 22 , 23 ). This change reflects our efforts to follow the International Code of Nomenclature of Prokaryotes ( 24 ) for determining correct names, including the proposal of Candidatus names.” — rules out breaking the link between a published result and the taxon it named — “Finally, to preserve name continuity, we do not replace alphanumeric placeholder names with other placeholder names, even when names (mostly introduced in early releases) are unwieldy, e.g. CG2-30-70-394 which is used as a genus, family, order, class and phylum name.” | controls[3] |
| weight | the earlier GTDB papers that introduced relative evolutionary divergence for higher ranks and ANI-based species clusters, whose criteria this release applies and re-tests — 0.95 — “Importantly, we implemented an ANI-based method for delineating species starting with R04-RS89 (June 2019) which allows all genomes to be classified at the rank of species and provides a uniform estimate of species-level diversity within the GTDB ( 5 ).” | claim_relations[0] |
| weight | the claim that a genetic discontinuity exists between 83% and 95% ANI, which is the main argument for treating the species threshold as a natural boundary — 0.95 — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | claim_relations[1] |
| weight | LPSN, the NCBI Taxonomy and the Living Tree Project, which supply type material, nomenclatural status and 16S classification to this taxonomy — 0.9 — “The LPSN (List of Prokaryotic names with Standing in Nomenclature) database ( 12 ) is used to establish co-identical strain identifiers for the type strains of species and subspecies, the types of higher-rank taxa, and the nomenclatural status of newly proposed or reclassified taxa. The Living Tree Project ( 13 ) is used to classify 16S rRNA sequences and help resolve ambiguity in regard to the correct classification of genomes.” | claim_relations[2] |
| weight | Doolittle and Papke's argument for a pragmatic species definition, quoted in full as the rationale for keeping a threshold with no natural boundary behind it — 0.85 — “Doolittle and Papke ( 37 ) best articulate the rationale for a pragmatic species definition: ‘What we want from a species definition is a set of easily applied and stable rules by which to decide when two organisms are similar enough in their genomic and/or phenotypic properties to be given the same name. The needs for such a guide to taxonomic practice in medicine, biotechnology and defense are obvious, and even arbitrary rules to satisfy them would be better than no rules at all.’” | claim_relations[3] |
| note | Each GTDB species cluster —operationalised-as→ a single representative genome — 'Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives.' A definition by exemplar - and it is why the paper needs a rule for replacing representatives at all: change the exemplar and the class moves. Elsewhere in the paper, only '0.26% of genomes on average are assigned to a different species cluster between releases' — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives as previously described ( 5 ).” | concept_edges[0] |
| relation | Each GTDB species cluster —operationalised-as→ a single representative genome — a species is whatever falls within an ANI and alignment-fraction radius of one chosen genome, so the class is defined by an exemplar rather than by a property — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives as previously described ( 5 ).” — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives” | concept_edges[0] |
| weight | Each GTDB species cluster —operationalised-as→ a single representative genome — 0.95 — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives as previously described ( 5 ).” | concept_edges[0] |
| note | two primary principles —operationalised-as→ the Balanced ANI Score (BAS) — 'GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain ... and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality. These two principles are quantitatively defined by the Balanced ANI Score (BAS) WHICH IS 0.5 x (ANI SCORE) + 0.5 x (QUALITY SCORE).' A qualitative preference turned into an arithmetic one - and the quality score behind it encodes an ordinal ranking as powers of ten, so the sum is only meaningful because one term dominates — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” | concept_edges[1] |
| relation | two primary principles —operationalised-as→ the Balanced ANI Score (BAS) — prefer the type strain, and replace only for demonstrably higher quality - expressed as a weighted sum with a fixed replacement margin — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score)” | concept_edges[1] |
| weight | two primary principles —operationalised-as→ the Balanced ANI Score (BAS) — 0.95 — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” | concept_edges[1] |
| note | a meaningful discontinuity —contrasted-with→ nearly equal numbers of pairs with ANI values between 78% and 95% — 'If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%.' A prediction stated before the test and a negative result published against the authors' own threshold - and the result 'persists when considering only genomes assembled from the type strain', so it is not an artefact of how representatives were chosen — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concept_edges[2] |
| relation | a meaningful discontinuity —contrasted-with→ nearly equal numbers of pairs with ANI values between 78% and 95% — the discontinuity that would justify the species threshold predicts few pairs in the 83-95% band; the data show a flat distribution across 78-95% — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” — “If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%.” | concept_edges[2] |
| weight | a meaningful discontinuity —contrasted-with→ nearly equal numbers of pairs with ANI values between 78% and 95% — 0.95 — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concept_edges[2] |
| audit | The addition of non-representative genomes —causes→ a genetic continuum between species — Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span says the addition can only increase the LIKELIHOOD of such genomes; causes is loose. — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concept_edges[3] |
| effect direction | The addition of non-representative genomes —causes→ a genetic continuum between species — increases — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” — “The addition of non-representative genomes can only increase the likelihood” | concept_edges[3] |
| note | The addition of non-representative genomes —causes→ a genetic continuum between species — 'The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.' A boundary claim that gets weaker with every new observation, by construction - the opposite of how thresholds usually behave as evidence accumulates — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concept_edges[3] |
| relation | The addition of non-representative genomes —causes→ a genetic continuum between species — every additional strain can only fill in the gap the threshold was supposed to sit in, so the evidence for a natural boundary weakens as the database grows — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” — “The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concept_edges[3] |
| weight | The addition of non-representative genomes —causes→ a genetic continuum between species — 0.9 — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concept_edges[3] |
| note | Enterocloster bolteae —contrasted-with→ Bradyrhizobium elkanii — 'the pairwise ANI values between ENTEROCLOSTER BOLTEAE and E. clostridioformis genomes FORM DISCRETE CLUSTERS and have a BIMODAL DISTRIBUTION with a discontinuity between 92% and 96% ANI. IN CONTRAST, pairwise ANI values between BRADYRHIZOBIUM ELKANII and B. pachyrhizi genomes DISPLAY A GENETIC CONTINUUM and can be described as FUZZY SPECIES.' One threshold, two kinds of object - and the paper measures how often each occurs rather than choosing between them — “For example, the pairwise ANI values between Enterocloster bolteae and E. clostridioformis genomes form discrete clusters and have a bimodal distribution with a discontinuity between 92% and 96% ANI (Figure 4C ). In contrast, pairwise ANI values between Bradyrhizobium elkanii and B. pachyrhizi genomes display a genetic continuum (Figure 4D ) and can be described as fuzzy species ( 35 ).” | concept_edges[4] |
| relation | Enterocloster bolteae —contrasted-with→ Bradyrhizobium elkanii — the same ANI rule produces a clean bimodal separation in one genus and an unbroken continuum in another - the species concept behaves differently by clade — “For example, the pairwise ANI values between Enterocloster bolteae and E. clostridioformis genomes form discrete clusters and have a bimodal distribution with a discontinuity between 92% and 96% ANI (Figure 4C ). In contrast, pairwise ANI values between Bradyrhizobium elkanii and B. pachyrhizi genomes display a genetic continuum (Figure 4D ) and can be described as fuzzy species ( 35 ).” | concept_edges[4] |
| weight | Enterocloster bolteae —contrasted-with→ Bradyrhizobium elkanii — 0.95 — “For example, the pairwise ANI values between Enterocloster bolteae and E. clostridioformis genomes form discrete clusters and have a bimodal distribution with a discontinuity between 92% and 96% ANI (Figure 4C ). In contrast, pairwise ANI values between Bradyrhizobium elkanii and B. pachyrhizi genomes display a genetic continuum (Figure 4D ) and can be described as fuzzy species ( 35 ).” | concept_edges[4] |
| note | Macrococcus equipercicus —distinct-from→ M. carouselicus — 'GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. CAROUSELICUS, indicating that ... an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB.' The distinction was always real; the pointer was wrong - which is a different failure from a definition being wrong, and only visible because the assembly was updated — “This discrepancy can be explained by the fact that GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. carouselicus indicating that prior to the GCF_004359525.1 assembly being updated, an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB. While the exception, such large-scale reassignments and assembly errors do occur and we make a best effort to identify and handle all such cases.” | concept_edges[5] |
| relation | Macrococcus equipercicus —distinct-from→ M. carouselicus — two species whose type strains were confused in the source database, so a cluster carried one name and contained the other until an assembly was corrected — “This discrepancy can be explained by the fact that GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. carouselicus indicating that prior to the GCF_004359525.1 assembly being updated, an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB. While the exception, such large-scale reassignments and assembly errors do occur and we make a best effort to identify and handle all such cases.” — “GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. carouselicus” | concept_edges[5] |
| weight | Macrococcus equipercicus —distinct-from→ M. carouselicus — 0.95 — “This discrepancy can be explained by the fact that GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. carouselicus indicating that prior to the GCF_004359525.1 assembly being updated, an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB. While the exception, such large-scale reassignments and assembly errors do occur and we make a best effort to identify and handle all such cases.” | concept_edges[5] |
| note | MAGs and/or SAGs —part-of→ bacterial taxa — 'Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes.' The taxa and the genomes are differently distributed: uncultured sequence is a minority of the data and a majority of the classes, because isolate genomes pile up inside a few well-sampled pathogens — “Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGs with over 70% of archaeal species, genera, families, and orders lacking an available cultured representative (Figure 1F ).” | concept_edges[6] |
| relation | MAGs and/or SAGs —part-of→ bacterial taxa — over half of bacterial taxa at every rank contain nothing but assembled sequence, although such genomes are only a fifth of the database — “Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGs with over 70% of archaeal species, genera, families, and orders lacking an available cultured representative (Figure 1F ).” — “Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite” | concept_edges[6] |
| weight | MAGs and/or SAGs —part-of→ bacterial taxa — 0.9 — “Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGs with over 70% of archaeal species, genera, families, and orders lacking an available cultured representative (Figure 1F ).” | concept_edges[6] |
| outcome | control — held — “This lack of a genetic discontinuity also persists when considering only genomes assembled from the type strain of a species, indicating that this result is independent of the methodology used to select GTDB species representatives (Figure 4B ).” — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | controls[0] |
| outcome | control — held — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” — “GTDB species representatives have been largely stable since the introduction of ANI-based species clusters in RS04-RS89 with 96.7% of representatives being unchanged on average between releases (Table 2 ). This stability is particularly encouraging given the rapid growth in the number of GTDB species clusters (Figure 1C and D ; Table 2 ). Numerous factors account for the changes that have occurred illustrating the benefits and need to regularly update taxonomic frameworks.” | controls[1] |
| outcome | control — held — “Effectively published Latin names above the rank of genus without designated type material, either a sequenced type strain or MAG, will no longer be incorporated into GTDB, and those that do will only be introduced when the associated type genome is present in GTDB. This change is necessary as establishing the correct interior node in the reference tree for taxa without type material can be ambiguous, particularly when the addition of new genomes or alternative inference methods results in the named taxon becoming polyphyletic in later releases.” — “Latin names in preprints are also no longer incorporated into the GTDB even if their type genomes are already in the database as names can change between preprint and peer-reviewed publication ( 25 , 26 ).” | controls[2] |
| outcome | control — held — “Finally, to preserve name continuity, we do not replace alphanumeric placeholder names with other placeholder names, even when names (mostly introduced in early releases) are unwieldy, e.g. CG2-30-70-394 which is used as a genus, family, order, class and phylum name.” — “As of R06-RS202, Latin names are no longer proposed by GTDB curators unless there is an associated publication with taxon descriptions ( 22 , 23 ). This change reflects our efforts to follow the International Code of Nomenclature of Prokaryotes ( 24 ) for determining correct names, including the proposal of Candidatus names.” | controls[3] |
| limitation | THE SPECIES DEFINITION IS ADMITTED TO HAVE NO NATURAL BOUNDARY BEHIND IT: `The use of ANI to delineate species despite the lack of clear evidence for discrete species boundaries in the GTDB dataset is a pragmatic approach.` This is a stronger disclosure than most definitional papers make - the authors ran the test that would have vindicated their own threshold and published the negative result — “The use of ANI to delineate species despite the lack of clear evidence for discrete species boundaries in the GTDB dataset is a pragmatic approach for organizing the rapidly growing biodiversity being discovered with metagenomic approaches ( 2 , 3 , 15 ). Leaving this diversity unclassified would severely limit the ability to effectively communicate scientific results as exemplified by 28.5% of genomes at NCBI lacking a species classification and 77.0% of GTDB R06-RS202 species clusters having an alphanumeric placeholder name.” | limitations[0] |
| limitation | AND THE DISCRETENESS RESULT IS BOUNDED BY WHAT HAS BEEN SEQUENCED: `this result should be interpreted with care as most bacterial and archaeal species lack genomic representation based on even conservative estimates of the number of prokaryotic species in nature.` The clusters look discrete partly because the space between them has not been sampled — “In terms of species, 2.2% can be described as fuzzy as they have one or more genomes with an interspecific ANI value ≥95% and 89.5% can be described as discrete as they contain no genomes with an interspecific ANI value >94% (Figure 4F ). Encouragingly, this indicates that GTDB species clusters are largely discrete, but this result should be interpreted with care as most bacterial and archaeal species lack genomic representation based on even conservative estimates of the number of prokaryotic species in nature ( 36 ).” | limitations[1] |
| limitation | AND HALF THE TAXA HAVE NEVER BEEN GROWN: `over 70% of archaeal species, genera, families, and orders lacking an available cultured representative`, and over 50% of bacterial taxa `consist exclusively of MAGs and/or SAGs` although those are only 20.2% of the genomes. A taxonomy of organisms nobody has in a flask — “Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGs with over 70% of archaeal species, genera, families, and orders lacking an available cultured representative (Figure 1F ).” | limitations[2] |
| limitation | AND THE DISCRETENESS ANALYSIS TAKES TWO SHORTCUTS, BOTH NAMED: `it was assumed that if the closest GTDB representative genome for species A was found to be species B, that the closest genome for any genome in species A will be in species B. Furthermore, for the 87 species containing >150 genomes only a randomly selected subset of 150 genomes was considered.` Declared in the Methods, and both would move the fuzzy/discrete split in the same direction - toward discreteness — “In order to reduce computational requirements, it was assumed that if the closest GTDB representative genome for species A was found to be species B, that the closest genome for any genome in species A will be in species B. Furthermore, for the 87 species containing >150 genomes only a randomly selected subset of 150 genomes was considered though these were compared to all genomes in species B regardless of the number of genomes in this species.” | limitations[3] |
| limitation | AND THE TYPE MATERIAL ITSELF IS NOT ALWAYS KNOWABLE: `Establishing the provenance of type material remains an ongoing challenge and this decision is made by considering a number of factors including the classification of 16S rRNA genes ..., the type status of genomes at NCBI, previous GTDB classifications, consideration of all pairwise ANI values and literature review.` One of the inputs to the decision is the database's own previous answer — “Establishing the provenance of type material remains an ongoing challenge and this decision is made by considering a number of factors including the classification of 16S rRNA genes against the Living Tree Project database ( 13 ), the type status of genomes at NCBI, previous GTDB classifications, consideration of all pairwise ANI values and literature review.” | limitations[4] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | content_tags.findings[0] |
| finding | — “The use of ANI to delineate species despite the lack of clear evidence for discrete species boundaries in the GTDB dataset is a pragmatic approach for organizing the rapidly growing biodiversity being discovered with metagenomic approaches ( 2 , 3 , 15 ). Leaving this diversity unclassified would severely limit the ability to effectively communicate scientific results as exemplified by 28.5% of genomes at NCBI lacking a species classification and 77.0% of GTDB R06-RS202 species clusters having an alphanumeric placeholder name.” | content_tags.findings[1] |
| finding | — “Doolittle and Papke ( 37 ) best articulate the rationale for a pragmatic species definition: ‘What we want from a species definition is a set of easily applied and stable rules by which to decide when two organisms are similar enough in their genomic and/or phenotypic properties to be given the same name. The needs for such a guide to taxonomic practice in medicine, biotechnology and defense are obvious, and even arbitrary rules to satisfy them would be better than no rules at all.’” | content_tags.findings[2] |
| finding | — “This lack of a genetic discontinuity also persists when considering only genomes assembled from the type strain of a species, indicating that this result is independent of the methodology used to select GTDB species representatives (Figure 4B ).” | content_tags.findings[3] |
| finding | — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | content_tags.findings[4] |
| finding | — “Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGs with over 70% of archaeal species, genera, families, and orders lacking an available cultured representative (Figure 1F ).” | content_tags.findings[5] |
| finding | — “This discrepancy can be explained by the fact that GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. carouselicus indicating that prior to the GCF_004359525.1 assembly being updated, an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB. While the exception, such large-scale reassignments and assembly errors do occur and we make a best effort to identify and handle all such cases.” | content_tags.findings[6] |
| finding | — “Related to MAG quality, we expect to change the AF criteria used for assigning genomes to GTDB species clusters from 0.65 to 0.5 starting in the next GTDB release in order to better accommodate the growing number of incomplete MAGs contained in the GTDB which can cause the AF between MAGs to be artificially low.” | content_tags.findings[7] |
| finding | — “Finally, to preserve name continuity, we do not replace alphanumeric placeholder names with other placeholder names, even when names (mostly introduced in early releases) are unwieldy, e.g. CG2-30-70-394 which is used as a genus, family, order, class and phylum name.” | content_tags.findings[8] |
| quantity | ANI 95-96% as the `gold standard`; the discontinuity said to lie between 83% and 95% is not found — — “An ANI threshold of 95–96% is widely considered the ‘gold standard’ for quantitatively delineating species ( 27–30 ) and it has recently been proposed that a genetic discontinuity exists between 83% and 95% ANI which would further support the use of ANI for defining species boundaries ( 31 , 32 ). This genetic discontinuity has also been observed between genomes recovered from specific environments such as the human gut and soil ( 33 ).” | content_tags.quantities[0] |
| quantity | 77.0% placeholder names; 28.5% of NCBI genomes with no species — — “The use of ANI to delineate species despite the lack of clear evidence for discrete species boundaries in the GTDB dataset is a pragmatic approach for organizing the rapidly growing biodiversity being discovered with metagenomic approaches ( 2 , 3 , 15 ). Leaving this diversity unclassified would severely limit the ability to effectively communicate scientific results as exemplified by 28.5% of genomes at NCBI lacking a species classification and 77.0% of GTDB R06-RS202 species clusters having an alphanumeric placeholder name.” | content_tags.quantities[1] |
| quantity | 2.2% fuzzy, 89.5% discrete — — “In terms of species, 2.2% can be described as fuzzy as they have one or more genomes with an interspecific ANI value ≥95% and 89.5% can be described as discrete as they contain no genomes with an interspecific ANI value >94% (Figure 4F ). Encouragingly, this indicates that GTDB species clusters are largely discrete, but this result should be interpreted with care as most bacterial and archaeal species lack genomic representation based on even conservative estimates of the number of prokaryotic species in nature ( 36 ).” | content_tags.quantities[2] |
| quantity | 96.7% of representatives unchanged between releases — — “GTDB species representatives have been largely stable since the introduction of ANI-based species clusters in RS04-RS89 with 96.7% of representatives being unchanged on average between releases (Table 2 ). This stability is particularly encouraging given the rapid growth in the number of GTDB species clusters (Figure 1C and D ; Table 2 ). Numerous factors account for the changes that have occurred illustrating the benefits and need to regularly update taxonomic frameworks.” | content_tags.quantities[3] |
| quantity | 0.65, moving to 0.5 next release — — “Related to MAG quality, we expect to change the AF criteria used for assigning genomes to GTDB species clusters from 0.65 to 0.5 starting in the next GTDB release in order to better accommodate the growing number of incomplete MAGs contained in the GTDB which can cause the AF between MAGs to be artificially low.” | content_tags.quantities[4] |
| variable | average nucleotide identity (ANI) — “An ANI threshold of 95–96% is widely considered the ‘gold standard’ for quantitatively delineating species ( 27–30 ) and it has recently been proposed that a genetic discontinuity exists between 83% and 95% ANI which would further support the use of ANI for defining species boundaries ( 31 , 32 ). This genetic discontinuity has also been observed between genomes recovered from specific environments such as the human gut and soil ( 33 ).” | content_tags.variables[0] |
| variable | relative evolutionary divergence (RED) — “In order to taxonomically organise this large and growing genome dataset, the GTDB uses relative evolutionary divergence (RED) to delineate higher-rank taxa and average nucleotide identity (ANI) to delineate species clusters ( 4 , 5 ). Use of these quantitative criteria for circumscribing taxa allows for automated classification of new genomes ( 6 ), ensures all genomes are classified from species to domain, and normalizes the definition of taxonomic ranks across the bacterial and archaeal domains.” | content_tags.variables[1] |
| variable | AF — “Related to MAG quality, we expect to change the AF criteria used for assigning genomes to GTDB species clusters from 0.65 to 0.5 starting in the next GTDB release in order to better accommodate the growing number of incomplete MAGs contained in the GTDB which can cause the AF between MAGs to be artificially low.” | content_tags.variables[2] |
| definition | species — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives as previously described ( 5 ).” | content_tags.definitions[0] |
| definition | fuzzy species — “In terms of species, 2.2% can be described as fuzzy as they have one or more genomes with an interspecific ANI value ≥95% and 89.5% can be described as discrete as they contain no genomes with an interspecific ANI value >94% (Figure 4F ). Encouragingly, this indicates that GTDB species clusters are largely discrete, but this result should be interpreted with care as most bacterial and archaeal species lack genomic representation based on even conservative estimates of the number of prokaryotic species in nature ( 36 ).” | content_tags.definitions[2] |
| definition | the Balanced ANI Score — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” | content_tags.definitions[3] |
| definition | a pragmatic species definition — “Doolittle and Papke ( 37 ) best articulate the rationale for a pragmatic species definition: ‘What we want from a species definition is a set of easily applied and stable rules by which to decide when two organisms are similar enough in their genomic and/or phenotypic properties to be given the same name. The needs for such a guide to taxonomic practice in medicine, biotechnology and defense are obvious, and even arbitrary rules to satisfy them would be better than no rules at all.’” | content_tags.definitions[4] |
| concept | Each GTDB species cluster — not defined in this paper's own words; the unit the taxonomy counts — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives as previously described ( 5 ).” | concepts[0] |
| concept | a single representative genome — not defined in this paper's own words; the one sequence a species name actually points at — “Each GTDB species cluster is defined by a single representative genome and species assignments are established by considering the ANI and AF (alignment fraction) of genomes to these representatives as previously described ( 5 ).” | concepts[1] |
| concept | two primary principles — not defined in this paper's own words; type-strain preference and quality-only replacement — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” | concepts[2] |
| concept | the Balanced ANI Score (BAS) — not defined in this paper's own words; the quantity those two principles are turned into — “GTDB representatives are updated according to two primary principles: (i) representatives should be assembled from the type strain of a species whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS) which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).” | concepts[3] |
| concept | Enterocloster bolteae — not defined in this paper's own words; a species pair that separates cleanly on ANI — “For example, the pairwise ANI values between Enterocloster bolteae and E. clostridioformis genomes form discrete clusters and have a bimodal distribution with a discontinuity between 92% and 96% ANI (Figure 4C ). In contrast, pairwise ANI values between Bradyrhizobium elkanii and B. pachyrhizi genomes display a genetic continuum (Figure 4D ) and can be described as fuzzy species ( 35 ).” | concepts[4] |
| concept | Bradyrhizobium elkanii — not defined in this paper's own words; a species pair that does not - 'a genetic continuum ... fuzzy species' — “For example, the pairwise ANI values between Enterocloster bolteae and E. clostridioformis genomes form discrete clusters and have a bimodal distribution with a discontinuity between 92% and 96% ANI (Figure 4C ). In contrast, pairwise ANI values between Bradyrhizobium elkanii and B. pachyrhizi genomes display a genetic continuum (Figure 4D ) and can be described as fuzzy species ( 35 ).” | concepts[5] |
| concept | a meaningful discontinuity — not defined in this paper's own words; the gap in ANI space that would make the species threshold a natural boundary — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concepts[6] |
| concept | nearly equal numbers of pairs with ANI values between 78% and 95% — not defined in this paper's own words; what was found instead - a continuum where the gap should have been — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concepts[7] |
| concept | Macrococcus equipercicus — not defined in this paper's own words; the name a cluster carried — “This discrepancy can be explained by the fact that GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. carouselicus indicating that prior to the GCF_004359525.1 assembly being updated, an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB. While the exception, such large-scale reassignments and assembly errors do occur and we make a best effort to identify and handle all such cases.” | concepts[8] |
| concept | M. carouselicus — not defined in this paper's own words; the organism the cluster actually contained — “This discrepancy can be explained by the fact that GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicus while GCF_004359515.1 is assembled from the type strain of M. carouselicus indicating that prior to the GCF_004359525.1 assembly being updated, an M. carouselicus strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB. While the exception, such large-scale reassignments and assembly errors do occur and we make a best effort to identify and handle all such cases.” | concepts[9] |
| concept | bacterial taxa — not defined in this paper's own words; the classes at every rank — “Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGs with over 70% of archaeal species, genera, families, and orders lacking an available cultured representative (Figure 1F ).” | concepts[10] |
| concept | MAGs and/or SAGs — not defined in this paper's own words; metagenome-assembled and single-amplified genomes - sequence with no organism behind it — “Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxa, regardless of rank, consist exclusively of MAGs and/or SAGs despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGs with over 70% of archaeal species, genera, families, and orders lacking an available cultured representative (Figure 1F ).” | concepts[11] |
| concept | The addition of non-representative genomes — not defined in this paper's own words; what happens to the picture as more strains are sequenced — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concepts[12] |
| concept | a genetic continuum between species — not defined in this paper's own words; the unbroken distribution the threshold has to cut — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | concepts[13] |
| a negative result published against the authors' own threshold | — “We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB species within the same genus (Figure 4A ). If a meaningful discontinuity existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated species boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%. The addition of non-representative genomes can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a species representative and thus a genetic continuum between species.” | content_tags.statistics[0] |
| two Methods assumptions that both push toward discreteness | — “In order to reduce computational requirements, it was assumed that if the closest GTDB representative genome for species A was found to be species B, that the closest genome for any genome in species A will be in species B. Furthermore, for the 87 species containing >150 genomes only a randomly selected subset of 150 genomes was considered though these were compared to all genomes in species B regardless of the number of genomes in this species.” | content_tags.statistics[1] |
| a percentage that is a property of sampling, not of nature | — “In terms of species, 2.2% can be described as fuzzy as they have one or more genomes with an interspecific ANI value ≥95% and 89.5% can be described as discrete as they contain no genomes with an interspecific ANI value >94% (Figure 4F ). Encouragingly, this indicates that GTDB species clusters are largely discrete, but this result should be interpreted with care as most bacterial and archaeal species lack genomic representation based on even conservative estimates of the number of prokaryotic species in nature ( 36 ).” | content_tags.statistics[2] |
Full text — tagged
The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.
4 nar Nucleic Acids Research Nucleic Acids Res Oxford University Press PMC8728215 8728215 8728215 34520557 10.1093/nar/gkab776 GTDB: an ongoing census of bacterial and archaeal diversity through a phylogenetically consistent, rank normalized and complete genome-based taxonomy ParkspersName Donovan H 1 ✉ ChuvochinapersName Maria 2 RinkepersName Christian 3 MussigpersName Aaron J 4 ChaumeilpersName Pierre-Alain 5 HugenholtzpersName Philip 6 ✉ 1 The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia 2 The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia 3 The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia 4 The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia 5 The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia 6 The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia ✉ To whom correspondence should be addressed. Email: donovan.parkspersName@gmail.com ✉ Correspondence may also be addressed to Philip HugenholtzpersName. Email: p.hugenholtzpersName@uq.edu.au 14 9 2021 50 D1 D785 D785–D794 5 1 2022 © The Author(s) 2021. 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 ( https://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Abstract The Genome Taxonomy Database (GTDB; https://gtdb.ecogenomic.org ) provides a phylogenetically consistent and rank normalized genome-based taxonomy for prokaryotic genomes sourced from the NCBI Assembly database. GTDB R06-RS202 spans 254 090 bacterial and 4316 archaeal genomes, a 270% increase since the introduction of the GTDB in November, 2017. These genomes are organized into 45 555 bacterial and 2339 archaeal speciesterm·def clusters which is a 200% increase since the integration of speciesterm·def clusters into the GTDB in June, 2019. Here, we explore prokaryotic diversity from the perspective of the GTDB and highlight the importance of metagenome-assembled genomes in expanding available genomic representation. We also discuss improvements to the GTDB website which allow tracking of taxonomic changes, easy assessment of genome assembly quality, and identification of genomes assembled from type material or used as speciesterm·def representatives. Methodological updates and policy changes made since the inception of the GTDB are then described along with the procedure used to update speciesterm·def clusters in the GTDB. We conclude with a discussion on the use of average nucleotide identities as a pragmatic approach for delineating prokaryotic speciesterm·def.
status released display-pdf yes is-olf no is-manuscript no is-preprint no is-journal-matter no is-scanned no is-retracted no Accepted 2021 Aug 28; Revised 2021 Aug 18; Received 2021 Jul 27; Collection date 2022 Jan 7.
INTRODUCTION The Genome Taxonomy Database (GTDB) was developed to provide a phylogenetically consistent bacterial and archaeal taxonomy which can accommodate isolate genomes and the tens of thousands of metagenome-assembled genomes (MAGs) now being obtained from environmental and clinical samples ( 1–3 ). In order to taxonomically organise this large and growing genome dataset, the GTDB uses relative evolutionary divergence (RED)term·var to delineate higher-rank taxa and average nucleotide identity (ANI)term·var to delineate speciesterm·def clusters ( 4 , 5 ). Use of these quantitative criteria for circumscribing taxa allows for automated classification of new genomes ( 6 ), ensures all genomes are classified from speciesterm·def to domain, and normalizes the definition of taxonomic ranks across the bacterial and archaeal domains.
GTDB builds upon a number of existing public resources in order to provide a taxonomic resource that reflects recently proposed taxa, changes in taxonomic opinion, and the wealth of publicly available genomes. The NCBI Taxonomy database ( 7 ) aids in the discovery of newly proposed taxa, provides initial speciesterm·def assignments and specifies co-identical strain identifiers for genomes in the NCBI Assembly database ( 8 ). We currently use the NCBI Assembly database as the sole genome repository for the GTDB as it is a member of the International Nucleotide Sequence Database Collaboration (INSDC; 9 ) which ensures it also contains genomes submitted to DDBJ ( 10 ) and EMBL-EBI ( 11 ). The LPSN (List of Prokaryotic names with Standing in Nomenclature) database ( 12 ) is used to establish co-identical strain identifiers for the type strains of speciesterm·def and subspecies, the types of higher-rank taxa, and the nomenclatural status of newly proposed or reclassified taxa. The Living Tree Project ( 13 ) is used to classify 16S rRNA sequences and help resolve ambiguity in regard to the correct classification of genomes.
The latest version of the GTDB is R06-RS202 which was released in April, 2021. This is the sixth release (i.e. R06) of the GTDB since its inception in November 2017 and comprises genomes in the NCBI Assembly database as of September 2020, i.e. the release date of RefSeq 202 ( 14 ).
RESOURCE CONTENT Growth of GTDB The number of genomes in the GTDB has grown by over 270% since its inception in November 2017 (Figure 1A and B ; Supplementary Table S1 ). Perhaps surprisingly, bacterial isolates account for the majority of this growth rather than MAGs (or single amplified genomes; SAGs) despite recent metagenomic studies recovering tens of thousands of MAGs ( 2 , 3 , 15 ). This is a result of both the large numbers of genomes from human pathogens deposited in INSDC repositories ( Supplementary Table S2 ) and many metagenomic studies providing only a representative MAG per operational speciesterm·def cluster or no MAGs to an INSDC repository. Unfortunately, this means INSDC repositories are missing much of the strain diversity which has recently been discovered. Interestingly, this pattern is not observed for archaea where MAGs are responsible for the majority of additional genomes (Figure 1B ), perhaps reflecting both challenges in cultivating archaeal strains and the lack of archaeal human pathogens.
Figure 1. Growth of the GTDB since its inception in November 2017. ( A , B ) Number of bacterial and archaeal isolates, MAGs, and SAGs in the GTDB along with the total number of genomes. Archaea were introduced into the GTDB starting with R03-RS86 in August, 2018. ( C , D ) Percent growth in the number of bacterial and archaeal taxa in the GTDB. ( E , F ) Proportion of bacterial and archaeal taxa at each taxonomic rank in GTDB R06-RS202 comprised exclusively of environmental genomes (MAGs and/or SAGsterm), exclusively of isolates, or both isolate and environmental genomes. For comparison, the proportion of isolate and environmental genomes is shown in the right bar plot.
Growth in the number of genomes has coincided with a large increase in the number of bacterial (513%) and archaeal (459%) speciesterm·def in GTDB (Figure 1C and D ; Supplementary Table S1 ). Importantly, we implemented an ANI-based method for delineating speciesterm·def starting with R04-RS89 (June 2019) which allows all genomes to be classified at the rank of speciesterm·def and provides a uniform estimate of species-level diversity within the GTDB ( 5 ). As expected, growth in the number of higher-rank taxa has been more modest (Figure 1C and D ) though still appreciable, with the number of bacterial and archaeal genera increasing by 222% and 193%, respectively.
Unlike the number of bacterial genomes, MAGs account for the majority of taxonomic diversity covered by GTDB R06-RS202 (Figure 1E and F ; Supplementary Table S1 ). Over 50% of bacterial taxaterm, regardless of rank, consist exclusively of MAGs and/or SAGsterm despite MAGs (50 669 genomes) and SAGs (745 genomes) representing only 20.2% of the 254 090 bacterial genomes. Similarly, nearly 50% of all archaeal taxa, regardless of rank, consist exclusively of MAGs and/or SAGsterm with over 70% of archaeal speciesterm·def, genera, families, and orders lacking an available cultured representative (Figure 1F ).
Changes to the GTDB Website There have been a number of additional features and improvements to the GTDB website since its inception in November 2017, which we illustrate here using the Enterocloster bolteaeterm genome GCF_002234575.2 (Figure 2 ). Each genome has an associated GTDB Genome page (Figure 2A – D ) indicating taxonomic, nomenclatural, and assembly quality information. A badge system is used to provide a quick assessment of overall genome quality and indicate that GCF_002234575.2 is a high-quality (HQ) isolate genome assembly which contains the 5S, 16S, and 23S rRNA genes along with genes for all 20 tRNAs (Figure 2A ). Additional assembly statistics such as CheckM completeness and contamination estimates ( 16 ), number of contigs, N50, genome size, protein count, and GC content are also provided on the Genome page (data not shown). A link to the NCBI Assembly page for GCF_002234575.2 is provided along with a link to the LPSN page for E. bolteae as this genome is recognized as being assembled from the type strain of the speciesterm·def according to LPSN (Figure 2B ). Both GTDB and NCBI classifications are provided to allow for easy comparison between these taxonomic resources which have identical classifications for GCF_002234575.2 (Figure 2C ). GCF_002234575.2 is annotated as being assembled from the strain ATCC BAA-613 at NCBI and considered to be assembled from the type strain of the speciesterm·def within the GTDB framework based on LPSN nomenclatural information for this speciesterm·def (Figure 2C ). This genome has also been selected as the GTDB representative of the E. bolteae speciesterm·def cluster which comprises 24 genomes in GTDB R06-RS202 (Figure 2C ).
Figure 2. Taxonomic, nomenclatural, and assembly quality information provided for individual genomes. ( A ) NCBI genome assembly accession and GTDB quality badges associated with this genome. Hovering over a tag provides information about the criteria used to establish the tag. ( B ) An external link is always provided to the NCBI Assembly page of the genome as all GTDB genomes are sourced from NCBI. A link is also provided to an LPSN speciesterm·def page when a genome is established to be assembled from the type strain of a speciesterm·def based on LPSN nomenclatural information. ( C ) GTDB and NCBI classifications for this genome along with its strain identifiers, nomenclatural status at GTDB (i.e. type strain of the speciesterm·def), and GTDB speciesterm·def representative status. GTDB taxa link to their corresponding position in the GTDB Taxonomy Tree (i.e. Figure 2F ) while NCBI taxa link to NCBI Taxonomy Browser pages. Each genome also links to a table indicating all genomes in the same GTDB speciesterm·def cluster. ( D ) GTDB classification of the genome in each GTDB release. GTDB taxa link to their corresponding Taxon History page (i.e. Figure 2E ). ( E ) GTDB Taxon History view for genomes classified as Enterocloster bolteaeterm indicating this speciesterm·def was reclassified from Clostridium_M to Enterocloster in GTDB R95. Numbers in parenthesis indicate the number of genomes assigned to a taxon. The Not Present label indicates genomes that were not available at the time of a GTDB release or failed the GTDB quality-control criteria used for the release, and thus had no GTDB classification. ( F ) GTDB Taxonomy Tree which provides a hierarchical exploration of the GTDB taxonomy and indicates nomenclatural type information, genomes selected as GTDB representatives, and Latin names in the GTDB which remain to be validated. Genomes link to their corresponding GTDB Genome page (i.e. A–D).
We provide two avenues for exploring how GTDB classifications have changed between releases. At the bottom of each Genome page, the classification of the genome is provided for each GTDB release (Figure 2D ). For example, GCF_002234575.2 was reclassified from Clostridium_M bolteae to E. bolteae in GTDB R95 in accordance with a new taxonomic opinion ( 17 ). Changes in GTDB classifications and their relationship to the NCBI taxonomy can also be explored at the taxon level using the GTDB Taxon History tool (Figure 2E ). This visualization illustrates that E. bolteae was reclassified from Clostridium_M to Enterocloster in GTDB R05-RS95, that the number of genomes assigned to this speciesterm·def has steadily increased between releases, and that there are a number of incongruent classifications at NCBI including one genome assigned as Clostridium bolteae and another as Hungatella hathewayi .
The GTDB taxonomy can be explored hierarchically using the Taxonomy Tree tool (Figure 2F ) or as an alphabetically sortable table (not shown). The tree view provides an approximation of the reference phylogenetic tree where taxa within a group are listed alphabetically with Latin names listed ahead of alphanumeric placeholder names. Type speciesterm·def of genera and type strains of speciesterm·def and subspecies are highlighted by color-coding. For example, there are three genomes assembled from the type strain of E. bolteae with the genome selected to represent this speciesterm·def in GTDB marked with an asterisk (Figure 2F ) ( 5 ). GTDB includes over 400 Latin names that have been assigned by the GTDB curators but are still to be formally proposed (see Policy Changes ). These are identified in the Taxonomy Tree with hashes and can also be marked and downloaded from the table view.
Searching functionality on the GTDB website has been expanded to allow for searches that are restricted to the GTDB taxonomy, NCBI taxonomy, NCBI organism name, or NCBI genome ID. This is useful for identifying genomes assigned to a specific GTDB or NCBI taxon. For example, this can be used to identify the 35 Enterocloster clostridioformis genomes in GTDB R06-RS202 without retrieving the two additional genomes with this classification at NCBI which are considered misclassified E. bolteae genomes in the GTDB (Figure 2E ). An advanced search feature has been added to the GTDB website to allow specific subsets of genomes to be identified across all GTDB metadata fields. For example, this can be used to identify the 94 genomes in GTDB R06-RS202 which are classified as Enterocloster and have an estimated completeness >90%, estimated contamination <10%, and contain at least one 16S rRNA gene.
A primary goal of taxonomy is to aid in scientific communication. With this in mind, all GTDB web tools produce URLs which allow specific results to be communicated between researchers. The following URLs produce the Genome page, Taxonomy Tree view, Taxon History view, and described Advanced search results for E. bolteae (Figure 2 ): Genome page: https://gtdb.ecogenomic.org/genomes?gid=GCF_002234575.2 Taxonomy Tree : https://gtdb.ecogenomic.org/tree?r=s__Enterocloster%20bolteae Taxon History : https://gtdb.ecogenomic.org/taxon_history/?from=R89&to=NCBI&query=s__Enterocloster%20bolteae Advanced Search : https://gtdb.ecogenomic.org/advanced?1=MX4yfmdfX0VudGVyb2Nsb3N0ZXI∼&3=Nn4xMn45MA∼∼&4=N34xMH4xMA∼∼&5=MTB.MTJ.MQ∼∼&exp=KDEmMyY0JjUp RESOURCE METHODOLOGY AND POLICIES Methodological updates A number of methodological changes have occurred since the original description of the GTDB ( 4 ) including the addition of Archaea starting with GTDB R03-RS86 ( 18 ) and adoption of ANI-based genomic similarity criteria for delineating speciesterm·def in GTDB R04-RS89 ( 5 ), which is discussed in more detail below. The 120 bacterial and 122 archaeal marker genes used to infer the domain-specific GTDB reference trees have not changed but identification and alignment of these genes is performed using Pfam v33.1 instead of v27 ( 19 ) starting with R06-RS202. The archaeal tree was originally inferred with FastTree v2.1.10 ( 20 ) under the WAG + GAMMA model but has been inferred using IQ-TREE v1.6.9 ( 21 ) under the C10 + PMSF model as of GTDB R04-RS89 ( 18 ). The initial GTDB releases included in-house MAGs that at the time had not been deposited in an INSDC repository ( 1 , 4 ). However, starting with GTDB R05-RS95, the GTDB only incorporates genomes from the NCBI Assembly database ( 8 ) and explicitly excludes assemblies annotated as being from ‘large multi-isolate projects’ at NCBI as these comprise genomically well-represented speciesterm·def and would require substantial additional computational resources if considered. The methodology and reference data used for each release have been provided in the METHODS file starting with GTDB R04-RS89 ( https://data.gtdb.ecogenomic.org/ ).
Updating GTDB speciesterm·def clusters Each GTDB species clusterterm is defined by a single representative genometerm and speciesterm·def assignments are established by considering the ANI and AFterm·var (alignment fraction) of genomes to these representatives as previously described ( 5 ). Here, we explain the methodology used to update these speciesterm·def clusters with each GTDB release. Speciesterm·def representatives are re-evaluated each release with an emphasis placed on retaining the same representative genomes for previously named speciesterm·def to preserve consistency between releases. However, the goal of stable representatives must be balanced with the utility of high-quality genomes as representatives, the need to incorporate changes in taxonomic opinion, and the need to correct genome classification or assembly errors.
Updating GTDB speciesterm·def clusters consists of four steps (Figure 3A ). Genomes in the current and previous releases are compared to identify new and updated genome assemblies, along with genomes that have been suppressed at NCBI indicating that they are no longer considered reliable (e.g. GCA_005039905.1). This is done by directly comparing the NCBI accession numbers (e.g. GCF_000267585.2) of the genomes in each release. Genomes of insufficient quality, as previously defined ( 5 ), are then identified and removed from further consideration. Next, speciesterm·def which contain multiple genomes identified as being assembled from the type strain of the speciesterm·def are examined. If any pair of these genomes has an ANI <99%, the genomes are manually inspected in order to establish which genome or genomes are most likely to represent the type strain of the speciesterm·def. Establishing the provenance of type material remains an ongoing challenge and this decision is made by considering a number of factors including the classification of 16S rRNA genes against the Living Tree Project database ( 13 ), the type status of genomes at NCBI, previous GTDB classifications, consideration of all pairwise ANI values and literature review.
Figure 3. Updating speciesterm·def clusters with each GTDB release. ( A ) Workflow for updating GTDB speciesterm·def clusters with results for the most recent GTDB release, R06-RS202, given below each step. There were 90 368 new genomes in this release, 987 genomes where the assembly at NCBI was updated, and 158 genomes where the assembly was suppressed at NCBI and thus not used in this release. All genomes were subjected to quality control which resulted in 26 407 (9.3%) genomes being removed from consideration. There were 2,458 speciesterm·def where multiple genomes were identified as being assembled from the type strain of the speciesterm·def. Of these, 130 speciesterm·def had genomes that were sufficiently divergent to warrant manual inspection to establish the genome most likely to be from the type strain. The 31 910 representatives from the previous GTDB release, R05-RS95, were examined and 1131 (3.5%) updated to a new genome. In addition, 6 speciesterm·def defined in R05-RS95 were retired as the sole genome representing the speciesterm·def was suppressed at NCBI. ( B ) Illustrative example of a GTDB speciesterm·def cluster with previous and new genomes. Genomes are depicted by shapes and the distance between genomes scales with their ANI divergence. The large red circle indicates the ANI circumscription radii for assigning genomes to the current speciesterm·def clusters. The new/updated genome (blue triangle) will only replace the existing GTDB speciesterm·def representative (red circle) if the ANI between these genomes is sufficiently high and the new/updated genome is of sufficient quality (see Table 1 ). This decision is determined quantitatively using the balanced ANI scoreterm·def (see main text). ( C ) Updating the Macrococcus equipercicusterm speciesterm·def cluster from GTDB R05-RS95 to R06-RS202. The M. equipercicus genome assembly, GCF_004359525.1, was updated and found to be distinct from the previous assembly (ANI = 80.6%). Consequently, this genome formed a new speciesterm·def cluster and the genome GCF_004359515.1 was promoted to a speciesterm·def representative. GCF_004359525.2 is assembled from the type strain of M. equipercicus and GCF_004359515.1 assembled from the type strain of M. carouselicusterm indicating the M. equipercicus cluster in GTDB R05-RS95 actually represented the speciesterm·def M. carouselicusterm and was incorrectly classified as a result of the GCF_004359525.1 assembly being incorrect.
GTDB representatives are updated according to two primary principlesterm: (i) representatives should be assembled from the type strain of a speciesterm·def whenever possible, and (ii) representatives should only be replaced by equivalent assemblies of demonstrably higher quality (Figure 3B ). These two principles are quantitatively defined by the Balanced ANI Score (BAS)term which is 0.5 × (ANI score) + 0.5 × (quality score), where the ANI score is 100 – 20 × (100 – ANI to current representative) and the quality score is defined by the criteria given in Table 1 . An existing representative is only replaced by a new or updated genome if it has a BAS ≥ 10 above the BAS of the current representative. This achieves the goal of taxonomic stability by requiring a new representative to be of increasingly higher quality (as defined by the quality score) the more dissimilar it is from the current representative (as defined by the ANI score).
Table 1. Criteria used to establish the quality score of genome assemblies Criteria Score Assembled from type strain of speciesterm·def 1 000 000 Effective type strain of speciesterm·def according to NCBI 100 000 NCBI representative of speciesterm·def 10 000 Assembled from type strain of subspecies 1000 Complete genome 100 CheckM quality estimate Completeness – 5 × contamination MAG or SAG –100 Contig count –5 × (no. contigs/100) Undetermined bases –5 × (no. undetermined bases/10 000) Full length 16S rRNA gene 10 There are two special cases that must be handled when updating speciesterm·def representatives. The first is when the assembly of an existing representative is updated. Although atypical, it is possible the new assembly will be sufficiently divergent from the previous assembly (e.g. by removal of a substantial amount of contamination) that a new representative should be selected. This is determined by calculating the BAS between all genomes in the speciesterm·def cluster, including the new assembly, to the previous assembly of the representative genome. The selected representative is the genome with the highest BAS which is typically the new assembly. The second special case occurs when a GTDB representative becomes suppressed at NCBI indicating that the assembly is no longer reliable. In such cases, the representative is replaced with the genome in the speciesterm·def cluster with the highest BAS to the suppressed representative in order to provide stability to the set of GTDB representatives to the extent possible. If the speciesterm·def cluster does not contain any genomes other than the suppressed representative, the speciesterm·def cluster is retired and will no longer appear in the GTDB.
GTDB speciesterm·def representatives have been largely stable since the introduction of ANI-based speciesterm·def clusters in RS04-RS89 with 96.7% of representatives being unchanged on average between releases (Table 2 ). This stability is particularly encouraging given the rapid growth in the number of GTDB speciesterm·def clusters (Figure 1C and D ; Table 2 ). Numerous factors account for the changes that have occurred illustrating the benefits and need to regularly update taxonomic frameworks. Between R05-RS95 and R06-RS202 there were 111 representatives with an updated genome assembly at NCBI of which two changed sufficiently that a new GTDB representative was selected for the speciesterm·def cluster. For example, this occurred for GCF_004359525.1 as it has an ANI of only 80.6% to its updated assembly GCF_004359525.2, but the original speciesterm·def cluster was still deemed valid as the representative could be replaced with GCF_004359515.1 which is 99.99% similar to GCF_004359525.1 (Figure 3C ). This discrepancy can be explained by the fact that GCF_004359525.2 is assembled from the type strain of Macrococcus equipercicusterm while GCF_004359515.1 is assembled from the type strain of M. carouselicusterm indicating that prior to the GCF_004359525.1 assembly being updated, an M. carouselicusterm strain was erroneously considered as the type strain for M. equipercicus at NCBI and in the GTDB. While the exception, such large-scale reassignments and assembly errors do occur and we make a best effort to identify and handle all such cases. There were 158 genomes in R06-RS202 which were removed from the GTDB as they were suppressed at NCBI. Among these 158 genomes, were 20 GTDB representatives which resulted in 14 speciesterm·def having new representatives selected and six speciesterm·def being retired as they were singleton clusters. A relatively larger number of changes were the result of replacing a non-type representative with a genome assembled from the type strain of the speciesterm·def (312 cases) or subspecies (7 cases). However, the majority of changes (595 cases) were the result of replacing a non-type representative with a higher-quality genome assembly with sufficiently high ANI to the existing representative (i.e. they satisfied the BAS replacement criterion).
Table 2. Updated GTDB speciesterm·def representatives and cluster assignments between releases R04-RS89 to R05-RS95 R05-RS95 to R06-RS202 Speciesterm·def representative 24 706 to 31 910 31 910 to 47 894 Unchanged 23 957 (96.97%) 30 773 (96.44%) Changed 743 (3.01%) 1131 (3.54%) New speciesterm·def 7218 (29.2% increase) 16 004 (50.2% increase) Genome cluster assignment 145 896 to 194 600 194 600 to 258 406 Same cluster 145 566 (99.77%) 193 668 (99.52%) Different cluster 172 (0.12%) 774 (0.40%) Suppressed 158 (0.11%) 158 (0.08%) After updating previous GTDB representatives, representatives for speciesterm·def with new effectively or validly published names must be selected and all remaining genomes are assigned to these speciesterm·def clusters or de novo speciesterm·def clusters using ANI to delineate speciesterm·def as previously described ( 5 ). Only 0.26% of genomes on average are assigned to a different speciesterm·def cluster between releases (Table 2 ). This is despite the assignment of genomes to speciesterm·def clusters needing to be determined de novo each release in order to account for new and modified GTDB speciesterm·def representatives. Such reassignments are the direct result of allowing representatives to change between releases but as illustrated above there are situations where new representatives must be selected (e.g. replacing GCF_004359525.1). We consider this small amount of reassignment acceptable as it allows genomes assembled from type strains and high-quality genome assemblies to become speciesterm·def representatives.
Policy changes The GTDB is an evolving resource that aims to balance continuity between releases with changes in classification methodology and taxonomic opinion to best serve the user community, which has necessitated a number of policy changes over time. As of R06-RS202, Latin names are no longer proposed by GTDB curators unless there is an associated publication with taxon descriptions ( 22 , 23 ). This change reflects our efforts to follow the International Code of Nomenclature of Prokaryotes ( 24 ) for determining correct names, including the proposal of Candidatus names. We currently discover new effectively published names by consulting the NCBI taxonomy and LPSN, receiving feedback from the community, and our own best efforts to read the literature. Effectively published Latin names above the rank of genus without designated type material, either a sequenced type strain or MAG, will no longer be incorporated into GTDB, and those that do will only be introduced when the associated type genome is present in GTDB. This change is necessary as establishing the correct interior node in the reference tree for taxa without type material can be ambiguous, particularly when the addition of new genomes or alternative inference methods results in the named taxon becoming polyphyletic in later releases. Latin names in preprints are also no longer incorporated into the GTDB even if their type genomes are already in the database as names can change between preprint and peer-reviewed publication ( 25 , 26 ). Finally, to preserve name continuity, we do not replace alphanumeric placeholder names with other placeholder names, even when names (mostly introduced in early releases) are unwieldy, e.g. CG2-30-70-394 which is used as a genus, family, order, class and phylum name. Placeholder names for new taxa in each release are now manually selected from a longer list of unique identifiers to avoid unwieldy names: NCBI organism name, NCBI infraspecific identifier, NCBI WGS identifier, or NCBI assembly identifier. We had previously aimed to release an update of the GTDB every six months and have averaged a new release every seven months. However, starting with R06-RS202 we plan to release updates every nine months in order to accommodate development of the GTDB between releases.
AVERAGE NUCLEOTIDE IDENTITY FOR DELINEATING SPECIESterm·def Observations on the discreteness of ANI-based speciesterm·def in GTDB An ANI threshold of 95–96% is widely considered the ‘gold standard’ for quantitatively delineating speciesterm·def ( 27–30 ) and it has recently been proposed that a genetic discontinuity exists between 83% and 95% ANI which would further support the use of ANI for defining speciesterm·def boundaries ( 31 , 32 ). This genetic discontinuity has also been observed between genomes recovered from specific environments such as the human gut and soil ( 33 ). However, when genome data are compared globally, i.e. not taking strain cohabitation into account, it is not uncommon for pairs of strains to have ANI values between 83% and 95% ( 5 , 34 ).
This issue is of interest as the GTDB uses a speciesterm·def definition based on the ANI to genomes selected as speciesterm·def representatives ( 5 ). Here we re-evaluated the evidence for a genetic discontinuity as the number of speciesterm·def and genomes within the GTDB has increased substantially since our original study. We demonstrate the lack of such a discontinuity by considering the ANI between closest pairs of representative genomes from GTDB speciesterm·def within the same genus (Figure 4A ). If a meaningful discontinuityterm existed, we would expect to find few such pairs with an ANI between 83% and 95% supporting the idea of discrete, well-separated speciesterm·def boundaries. In contrast, we find nearly equal numbers of pairs with ANI values between 78% and 95%term. The addition of non-representative genomesterm can only increase the likelihood of there being genomes with ANI values between 83% and 95% to a speciesterm·def representative and thus a genetic continuum between speciesterm. This lack of a genetic discontinuity also persists when considering only genomes assembled from the type strain of a speciesterm·def, indicating that this result is independent of the methodology used to select GTDB speciesterm·def representatives (Figure 4B ). The above analyses do not speak to the possibility of discrete groups of strains defined by a large genetic discontinuity existing within specific environments as suggested by Olm et al. ( 33 ). However, further work is required to establish this discontinuity and specify how an environment-specific speciesterm·def definition might be used in practice.
Figure 4. Use of genomic similarity for delineating speciesterm·def. ( A ) ANI values between 36 781 GTDB speciesterm·def representatives and their closest representative within the same genus. ( B ) Same plot as in A but restricted to the 9,687 speciesterm·def where the GTDB representative genome is assembled from the type strain. ( C ) Pairwise ANI between the 24 Enterocloster bolteaeterm and 34 E. clostridioformis genomes in GTDB RS06-RS202. ( D ) Pairwise ANI between the 14 Bradyrhizobium elkaniiterm and 8 B. pachyrhizi genomes in GTDB RS06-RS202. ( E ) ANI between genomes and their closest genome in a different, intrageneric speciesterm·def cluster (108 503 total pairs). ( F ) ANI between the closest pairs in plot E for each of the 35 147 speciesterm·def considered.
While the above analysis indicates that there is no global genetic discontinuity between 83% and 95% ANI in the GTDB, it remains unclear as to whether strains generally form discrete, ANI-based speciesterm·def clusters. For example, the pairwise ANI values between Enterocloster bolteaeterm and E. clostridioformis genomes form discrete clusters and have a bimodal distribution with a discontinuity between 92% and 96% ANI (Figure 4C ). In contrast, pairwise ANI values between Bradyrhizobium elkaniiterm and B. pachyrhizi genomes display a genetic continuum (Figure 4D ) and can be described as fuzzy speciesterm·def ( 35 ). The discreteness of GTDB speciesterm·def clusters was assessed by considering the ANI between each genome and its closest genome in a different, intrageneric speciesterm·def cluster (see Methods). Discrete speciesterm·def will contain few or ideally no genomes with a high ANI to genomes from a different speciesterm·def. Disregarding GTDB speciesterm·def clusters delineated at an ANI >95% indicated that only 6.0% of the closest interspecific genome pairs have an ANI ≥ 95% and 80.8% of pairs are well-separated as indicated by having an ANI ≤94% (Figure 4E ). In terms of speciesterm·def, 2.2% can be described as fuzzy as they have one or more genomes with an interspecific ANI value ≥95% and 89.5% can be described as discrete as they contain no genomes with an interspecific ANI value >94% (Figure 4F ). Encouragingly, this indicates that GTDB speciesterm·def clusters are largely discrete, but this result should be interpreted with care as most bacterial and archaeal speciesterm·def lack genomic representation based on even conservative estimates of the number of prokaryotic speciesterm·def in nature ( 36 ).
The use of ANI to delineate speciesterm·def despite the lack of clear evidence for discrete speciesterm·def boundaries in the GTDB dataset is a pragmatic approach for organizing the rapidly growing biodiversity being discovered with metagenomic approaches ( 2 , 3 , 15 ). Leaving this diversity unclassified would severely limit the ability to effectively communicate scientific results as exemplified by 28.5% of genomes at NCBI lacking a speciesterm·def classification and 77.0% of GTDB R06-RS202 speciesterm·def clusters having an alphanumeric placeholder name. Doolittle and Papke ( 37 ) best articulate the rationale for a pragmatic species definitionterm·def: ‘What we want from a speciesterm·def definition is a set of easily applied and stable rules by which to decide when two organisms are similar enough in their genomic and/or phenotypic properties to be given the same name. The needs for such a guide to taxonomic practice in medicine, biotechnology and defense are obvious, and even arbitrary rules to satisfy them would be better than no rules at all.’
CONCLUDING REMARKS AND FUTURE PLANS GTDB is in active development and there are a number of initiatives underway to make this resource more comprehensive and accurate. As MAGs are not always submitted to INSDC repositories, we are exploring incorporation of additional repositories into the GTDB such as MGnify ( 38 ). The quality of MAGs is an ongoing concern and we are evaluating methods for decontaminating MAGs ( 39 ) and identifying and removing MAGs that may contribute to phylogenetic instability ( 40 ). Related to MAG quality, we expect to change the AFterm·var criteria used for assigning genomes to GTDB speciesterm·def clusters from 0.65 to 0.5 starting in the next GTDB release in order to better accommodate the growing number of incomplete MAGs contained in the GTDB which can cause the AFterm·var between MAGs to be artificially low. There is also an ongoing need to evaluate best practices in terms of tree inference methods, which has been initially explored for the archaeal tree and resulted in the adoption of IQ-TREE for this domain. We envisage that further evaluations will result in modifications to the bacterial and archaeal marker sets in future GTDB releases. Feedback is essential for the GTDB to best serve the community and we encourage suggestions and comments on either the GTDB Forum ( https://forum.gtdb.ecogenomic.org/ ) or directly to the GTDB development team ( https://gtdb.ecogenomic.org/about ).
METHODS ANI and AFterm·var values were calculated with FastANI v1.3 ( 31 ). The discreteness of GTDB speciesterm·def clusters was assessed by identifying the closest genome to a genome within a different, intrageneric speciesterm·def cluster. The 35 147 speciesterm·def in genera with multiple speciesterm·def and with an ANI circumscription radius of 95% were considered in this analysis (speciesterm·def with an ANI radius >95% are exceptional cases in the GTDB and result from another speciesterm·def being within 95% ANI that was retained in the GTDB to preserve historical classifications). In order to reduce computational requirements, it was assumed that if the closest GTDB representative genome for speciesterm·def A was found to be speciesterm·def B, that the closest genome for any genome in speciesterm·def A will be in speciesterm·def B. Furthermore, for the 87 speciesterm·def containing >150 genomes only a randomly selected subset of 150 genomes was considered though these were compared to all genomes in speciesterm·def B regardless of the number of genomes in this speciesterm·def.
DATA AVAILABILITY The GTDB can be accessed at https://gtdb.ecogenomic.org/ and data files for each GTDB release are available from https://data.gtdb.ecogenomic.org/ . Genomes comprising the GTDB are obtained from the NCBI Assembly database ( www.ncbi.nlm.nih.gov/assembly ). GTDB speciesterm·def clusters are updated using the Python code at https://github.com/Ecogenomics/gtdb-species-clusters which is made available under the GNU GPL v3.0.
Supplementary Material gkab776_Supplemental_File Click here for additional data file.
ACKNOWLEDGEMENTS We thank Brian Kemish for systems engineering support, and members of the Australian Centre for Ecogenomics for beta-testing new features. We also thank the user community for feedback and highlighting errors in the taxonomy.
Contributor Information Donovan H ParkspersName, The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia.
Maria ChuvochinapersName, The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia.
Christian RinkepersName, The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia.
Aaron J MussigpersName, The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia.
Pierre-Alain ChaumeilpersName, The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia.
Philip HugenholtzpersName, The University of Queensland, School of Chemistry and Molecular Biosciences, Australian Centre for Ecogenomics, QLD 4072, Australia.
SUPPLEMENTARY DATA Supplementary Data are available at NAR Online.
FUNDING GTDB is supported by Australian Research Council Laureate Fellowship [FL150100038 to P.H.]; Australian Research Council Future Fellowship [FT170100213 to C.R.]; strategic funding from the University of Queensland. Funding for open access charge: Australian Research Council Laureate Fellowship [FL150100038 to P.H.].
Conflict of interest statement . None declared.