Benjamin J. CallahanpersName, Paul J. McMurdiepersName, Susan HolmespersName (2017date). “Exact sequence variants should replace operational taxonomic units in marker-gene data analysis.” The ISME Journaltitle·venue.
Object of study — the unit of analysis in marker-gene sequencing - what a row of the feature table denotes - and whether it should be a cluster defined by a dissimilarity threshold or an exactly resolved sequence
Research question — should amplicon sequence variantsterm·def replace operational taxonomic unitsterm as the standard unit of marker-gene analysis and reporting
Contribution — reframes the case for ASVsterm away from RESOLUTION - the advantage everyone had argued - and onto CONSISTENT LABELLING: an ASVterm names a DNA sequence that exists outside any data set, so it can be compared, merged and predicted from across studies, which neither kind of OTU can do at once
phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | the same construct serving two orthogonal purposes - translating taxonomy and absorbing sequencing error - and doing neither well — “Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUs are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence ( Eren et al. , 2016 ).” | |
| claim-type | critique — “A consequence of this often unacknowledged dual mandate is that OTUs struggle to serve both purposes well; the connection between OTUs and species is largely unfounded ( Stackebrandt and Ebers, 2006 ), and the most common methods often output numbers of OTUs an order of magnitude higher than the number of strains present in mock communities ( Kopylova et al. , 2016 ).” | |
| method — evidence | — “As a demonstration of reusability and computational tractability on large data sets, the DADA2 method for ASV inference was used to process the 18S rRNA gene amplicon data from the TARA Oceans project, which contains ~766 million reads ( De Vargas et al. , 2015 ), on a 2016 Macbook Pro.” | |
| object of study — evidence | — “The analysis of marker-gene data customarily begins with the construction of molecular operational taxonomic units (OTUs): clusters of reads that differ by less than a fixed sequence dissimilarity threshold, most commonly 3% ( Westcott and Schloss, 2015 ; Kopylova et al. , 2016 ).” | |
| research question — evidence | — “For analysis to be reproducible the fundamental units must be reproducible, and de novo OTUs are not. For analysis to be comprehensive the fundamental units must be comprehensive, and closed-reference OTUs are not.” | |
| contribution — evidence | — “However, unlike de novo OTUs, ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism.” | |
| mechanism | A de novo OTU is a property of its data set, not of the biology. `de novo OTUs are emergent features of a data set, with boundaries and membership that depend on the data set in which they are defined` — “Many methods for constructing these clusters have been developed, but in all cases de novo OTUs are emergent features of a data set, with boundaries and membership that depend on the data set in which they are defined.” | mechanisms[0] |
| mechanism | and the dependence is not an artefact of finite sampling - `the delineation of de novo OTUs depends on the relative abundances of the sampled community even in the limit of infinite sequencing depth and zero errors`. Better data does not fix it — “This data set dependence is not just a practical concern: the delineation of de novo OTUs depends on the relative abundances of the sampled community even in the limit of infinite sequencing depth and zero errors.” | mechanisms[1] |
| mechanism | so the consequence is flat: `de novo OTUs defined in two different data sets cannot be compared` — “As a result, de novo OTUs defined in two different data sets cannot be compared.” | mechanisms[2] |
| mechanism | A closed-reference OTU is a property of the database. `each reference sequence in the database defines and labels an associated closed-reference OTU` - so two studies using the same database can be compared, and that is the property the paper names `consistent labeling` — “If the same reference database is used, closed-reference OTU assignments from independently processed data sets can be validly compared, a property we refer to as consistent labeling.” | mechanisms[3] |
| mechanism | but the price is everything the database does not contain: `biological variation that is not represented in the reference database is necessarily lost during assignment` — “However, biological variation that is not represented in the reference database is necessarily lost during assignment to closed-reference OTUs.” | mechanisms[4] |
| mechanism | and the database is not a fixed point either - closed-reference OTUs `even depend on the order of the reference sequences`, so `reference databases must remain static or closed-reference OTU tables must be regenerated` to stay comparable — “In order to maintain comparability over time, reference databases must remain static or closed-reference OTU tables must be regenerated whenever a new or upgraded reference database is adopted.” | mechanisms[5] |
| mechanism | An ASV is a property of the organism. `ASVs represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism` - which is why ASVs inferred in different studies can be compared without a shared database — “However, unlike de novo OTUs, ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism.” | mechanisms[6] |
| mechanism | the inference is still not read-by-read - `the smallest unit of data from which ASVs can be inferred is a sample`, because a sequence is distinguished from an error partly by being seen more than once — “As a result, ASV inference cannot be performed independently on each read—the smallest unit of data from which ASVs can be inferred is a sample.” | mechanisms[7] |
| mechanism | Without consistent labels a result cannot be tested. `The absence of consistent labels makes replicating or falsifying previous results problematic, even impossible` - an association with a de novo OTU can only be retested by first coarsening it to taxonomy or to a diversity metric — “The absence of consistent labels makes replicating or falsifying previous results problematic, even impossible. Consider a significant association reported between a particular de novo OTU and a condition of interest. This association cannot be tested in a new data set, because that de novo OTU only exists within the data set in which it was defined. Therefore, results from the analysis of de novo OTUs can only be tested indirectly, by mapping de novo OTUs onto consistent labels such as taxonomy or reducing community composition to summary measures such as a diversity metric, and then testing the results at that coarser level.” | mechanisms[8] |
| mechanism | THE DIAGNOSIS. `Molecular OTUs serve two different and orthogonal purposes` - translating taxonomic concepts into sequencing, and absorbing amplicon error - and `A consequence of this often unacknowledged dual mandate is that OTUs struggle to serve both purposes well` — “A consequence of this often unacknowledged dual mandate is that OTUs struggle to serve both purposes well; the connection between OTUs and species is largely unfounded ( Stackebrandt and Ebers, 2006 ), and the most common methods often output numbers of OTUs an order of magnitude higher than the number of strains present in mock communities ( Kopylova et al. , 2016 ).” | mechanisms[9] |
| mechanism | and the taxonomic half is the one that fails: `the connection between OTUs and species is largely unfounded`, with common methods outputting OTU counts an order of magnitude higher than the strains actually present in mock communities — “A consequence of this often unacknowledged dual mandate is that OTUs struggle to serve both purposes well; the connection between OTUs and species is largely unfounded ( Stackebrandt and Ebers, 2006 ), and the most common methods often output numbers of OTUs an order of magnitude higher than the number of strains present in mock communities ( Kopylova et al. , 2016 ).” | mechanisms[10] |
| critiques | the customary construction of molecular OTUs at a fixed dissimilarity threshold, most commonly 3% - the standard first step of marker-gene analysis — “The analysis of marker-gene data customarily begins with the construction of molecular operational taxonomic units (OTUs): clusters of reads that differ by less than a fixed sequence dissimilarity threshold, most commonly 3% ( Westcott and Schloss, 2015 ; Kopylova et al. , 2016 ).” | claim_relations[0] |
| critiques | the equation of an OTU with a species, which the paper calls largely unfounded while noting the field routinely treats observing an OTU as observing a species — “The sample-by-OTU feature table serves as the basis for further analysis, with the observation of an OTU often treated as akin to the observation of a ‘species’ in the taxonomic profiling application.” | claim_relations[1] |
| extends | the existing case for ASV methods, which had rested on improved RESOLUTION; this paper argues resolution is the lesser benefit and consistent labelling the greater — “ASV methods infer the biological sequences in the sample prior to the introduction of amplification and sequencing errors, and distinguish sequence variants differing by as little as one nucleotide.” | claim_relations[2] |
| concedes | the partial remedies already available for OTUs - open-reference methods, linear-time clustering, and aggressive filtering with complete read overlap — “Appealing terminology such as ‘resolution of exact sequence variants’ does not eliminate the limitations inherent to representing a complex biological organism by a short genetic barcode.” | claim_relations[3] |
| sense | `operational taxonomic unit` - a cluster of reads below a fixed dissimilarity threshold, most commonly 3%; the customary unit of marker-gene analysis — “The analysis of marker-gene data customarily begins with the construction of molecular operational taxonomic units (OTUs): clusters of reads that differ by less than a fixed sequence dissimilarity threshold, most commonly 3% ( Westcott and Schloss, 2015 ; Kopylova et al. , 2016 ).” | senses[0] |
| sense | `de novo OTU` - a cluster emergent from one data set, whose boundaries and membership depend on that data set and are undefined outside it — “Many methods for constructing these clusters have been developed, but in all cases de novo OTUs are emergent features of a data set, with boundaries and membership that depend on the data set in which they are defined.” | senses[1] |
| sense | `closed-reference OTU` - a property of a reference database: each reference sequence defines and labels one — “Closed-reference OTUs are properties of a reference database; each reference sequence in the database defines and labels an associated closed-reference OTU.” | senses[2] |
| sense | `consistent labeling` - the property that assignments made independently in different data sets can be validly compared — “If the same reference database is used, closed-reference OTU assignments from independently processed data sets can be validly compared, a property we refer to as consistent labeling.” | senses[3] |
| sense | `amplicon sequence variant` - a biological sequence inferred prior to amplification and sequencing error, distinguished to a single nucleotide — “ASV methods infer the biological sequences in the sample prior to the introduction of amplification and sequencing errors, and distinguish sequence variants differing by as little as one nucleotide.” | senses[4] |
| control | measuring OTU output against MOCK COMMUNITIES - samples whose true strain composition is known by construction — failed — “The sample-by-OTU feature table serves as the basis for further analysis, with the observation of an OTU often treated as akin to the observation of a ‘species’ in the taxonomic profiling application.” — rules out the circularity of judging a clustering method by the clusters it produces; with a known answer, an order-of-magnitude excess of OTUs is a fact about the method rather than a discovery about the community — “A consequence of this often unacknowledged dual mandate is that OTUs struggle to serve both purposes well; the connection between OTUs and species is largely unfounded ( Stackebrandt and Ebers, 2006 ), and the most common methods often output numbers of OTUs an order of magnitude higher than the number of strains present in mock communities ( Kopylova et al. , 2016 ).” | controls[0] |
| control | asking what de novo OTUs do in the limit of INFINITE SEQUENCING DEPTH AND ZERO ERRORS - a thought-experiment control that removes sampling and error entirely — held — “As a result, de novo OTUs defined in two different data sets cannot be compared.” — rules out the reading that data-set dependence is a practical nuisance that better sequencing would cure; the delineation still depends on relative abundances — “This data set dependence is not just a practical concern: the delineation of de novo OTUs depends on the relative abundances of the sampled community even in the limit of infinite sequencing depth and zero errors.” | controls[1] |
| control | running ASV inference on TARA Oceans - ~766 million 18S reads - on a 2016 MacBook Pro, and depositing the resulting tables — held — “As a result, ASV inference cannot be performed independently on each read—the smallest unit of data from which ASVs can be inferred is a sample.” — rules out the objection that per-sample inference cannot scale; the demonstration is chosen at the size where de novo OTU methods are said to become prohibitive — “As a demonstration of reusability and computational tractability on large data sets, the DADA2 method for ASV inference was used to process the 18S rRNA gene amplicon data from the TARA Oceans project, which contains ~766 million reads ( De Vargas et al. , 2015 ), on a 2016 Macbook Pro.” | controls[2] |
| weight | the customary construction of molecular OTUs at a fixed dissimilarity threshold, most commonly 3% - the standard first step of marker-gene analysis — 0.95 — “The analysis of marker-gene data customarily begins with the construction of molecular operational taxonomic units (OTUs): clusters of reads that differ by less than a fixed sequence dissimilarity threshold, most commonly 3% ( Westcott and Schloss, 2015 ; Kopylova et al. , 2016 ).” | claim_relations[0] |
| weight | the equation of an OTU with a species, which the paper calls largely unfounded while noting the field routinely treats observing an OTU as observing a species — 0.95 — “The sample-by-OTU feature table serves as the basis for further analysis, with the observation of an OTU often treated as akin to the observation of a ‘species’ in the taxonomic profiling application.” | claim_relations[1] |
| weight | the existing case for ASV methods, which had rested on improved RESOLUTION; this paper argues resolution is the lesser benefit and consistent labelling the greater — 0.9 — “ASV methods infer the biological sequences in the sample prior to the introduction of amplification and sequencing errors, and distinguish sequence variants differing by as little as one nucleotide.” | claim_relations[2] |
| weight | the partial remedies already available for OTUs - open-reference methods, linear-time clustering, and aggressive filtering with complete read overlap — 0.85 — “Appealing terminology such as ‘resolution of exact sequence variants’ does not eliminate the limitations inherent to representing a complex biological organism by a short genetic barcode.” | claim_relations[3] |
| note | closed-reference methods —distinct-from→ de novo methods — 'the most substantive distinction is between closed-reference methods ... and de novo methods' - the paper says the many published OTU methods differ in ways that do not matter, and in exactly one way that does. — “Many methods for defining molecular OTUs have been proposed, but the most substantive distinction is between closed-reference methods—in which reads sufficiently similar to a sequence in a reference database are recruited into a corresponding OTU—and de novo methods—in which reads are grouped into OTUs as a function of their pairwise sequence similarities.” | concept_edges[0] |
| relation | closed-reference methods —distinct-from→ de novo methods — the substantive division among OTU methods: one recruits reads to a database entry, the other clusters reads against each other — “Many methods for defining molecular OTUs have been proposed, but the most substantive distinction is between closed-reference methods—in which reads sufficiently similar to a sequence in a reference database are recruited into a corresponding OTU—and de novo methods—in which reads are grouped into OTUs as a function of their pairwise sequence similarities.” — “but the most substantive distinction is between closed-reference methods—in which reads sufficiently similar to a sequence in a reference database are recruited into a corresponding OTU—and de novo methods—in which reads are grouped into OTUs as a function of their pairwise sequence similarities.” | concept_edges[0] |
| weight | closed-reference methods —distinct-from→ de novo methods — 0.95 — “Many methods for defining molecular OTUs have been proposed, but the most substantive distinction is between closed-reference methods—in which reads sufficiently similar to a sequence in a reference database are recruited into a corresponding OTU—and de novo methods—in which reads are grouped into OTUs as a function of their pairwise sequence similarities.” | concept_edges[0] |
| note | ASVs —distinct-from→ de novo OTUs — THE WHOLE ARGUMENT IS IN THIS ONE CONTRAST. 'ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed' - not that ASVs are finer, which is what everyone else had argued, but that they REFER to something the data does not constitute. — “However, unlike de novo OTUs, ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism.” | concept_edges[1] |
| relation | ASVs —distinct-from→ de novo OTUs — an ASV denotes a DNA sequence that exists whether or not it was sampled; a de novo OTU denotes a cluster that exists only inside the data set that made it — “However, unlike de novo OTUs, ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism.” — “However, unlike de novo OTUs, ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed” | concept_edges[1] |
| weight | ASVs —distinct-from→ de novo OTUs — 0.95 — “However, unlike de novo OTUs, ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism.” | concept_edges[1] |
| note | ribosomal sequence variant —contrasted-with→ 3% ribosomal OTUs — 'while necessarily better than the customary 3% ribosomal OTUs, there is still no guarantee of ecological coherence or even monophyly'. The authors are the ASV method's own builders and they place the limit in the paper arguing for it. — “For example, while necessarily better than the customary 3% ribosomal OTUs, there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variant ( Berry et al. , 2017 ).” | concept_edges[2] |
| relation | ribosomal sequence variant —contrasted-with→ 3% ribosomal OTUs — finer than the customary unit and still not guaranteed to correspond to a coherent or even monophyletic group of genomes — “For example, while necessarily better than the customary 3% ribosomal OTUs, there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variant ( Berry et al. , 2017 ).” — “while necessarily better than the customary 3% ribosomal OTUs, there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variant” | concept_edges[2] |
| weight | ribosomal sequence variant —contrasted-with→ 3% ribosomal OTUs — 0.9 — “For example, while necessarily better than the customary 3% ribosomal OTUs, there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variant ( Berry et al. , 2017 ).” | concept_edges[2] |
| effect direction | grouping errors together with the error-free sequence —causes→ amplicon sequencing error — reduces — “Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUs are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence ( Eren et al. , 2016 ).” — “to reduce the impact of amplicon sequencing error” | concept_edges[3] |
| note | grouping errors together with the error-free sequence —causes→ amplicon sequencing error — 'to reduce the impact of amplicon sequencing error ... by grouping errors together with the error-free sequence'. Direction is 'reduces'. This is the job OTUs do well, and naming it is what makes the dual-mandate diagnosis land. — “Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUs are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence ( Eren et al. , 2016 ).” | concept_edges[3] |
| relation | grouping errors together with the error-free sequence —causes→ amplicon sequencing error — clustering absorbs erroneous reads into the same unit as the correct one, which is what reduces the error's effect on diversity and composition measures — “Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUs are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence ( Eren et al. , 2016 ).” — “The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence” | concept_edges[3] |
| weight | grouping errors together with the error-free sequence —causes→ amplicon sequencing error — 0.9 — “Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUs are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence ( Eren et al. , 2016 ).” | concept_edges[3] |
| note | closed-reference OTU —distinct-from→ reference sequence — 'it does not guarantee that the reference sequence itself was observed - to the contrary, that is often not the case.' A unit whose representative is usually absent, used downstream to index into other data as though it were present. — “The observation of a closed-reference OTU indicates that at least one sequencing read was sufficiently similar to the associated reference sequence to be mapped to its OTU. However, it does not guarantee that the reference sequence itself was observed—to the contrary, that is often not the case.” | concept_edges[4] |
| relation | closed-reference OTU —distinct-from→ reference sequence — the OTU can be observed while the sequence that names and represents it was never observed at all — “The observation of a closed-reference OTU indicates that at least one sequencing read was sufficiently similar to the associated reference sequence to be mapped to its OTU. However, it does not guarantee that the reference sequence itself was observed—to the contrary, that is often not the case.” | concept_edges[4] |
| weight | closed-reference OTU —distinct-from→ reference sequence — 0.9 — “The observation of a closed-reference OTU indicates that at least one sequencing read was sufficiently similar to the associated reference sequence to be mapped to its OTU. However, it does not guarantee that the reference sequence itself was observed—to the contrary, that is often not the case.” | concept_edges[4] |
| outcome | control — failed — “A consequence of this often unacknowledged dual mandate is that OTUs struggle to serve both purposes well; the connection between OTUs and species is largely unfounded ( Stackebrandt and Ebers, 2006 ), and the most common methods often output numbers of OTUs an order of magnitude higher than the number of strains present in mock communities ( Kopylova et al. , 2016 ).” — “The sample-by-OTU feature table serves as the basis for further analysis, with the observation of an OTU often treated as akin to the observation of a ‘species’ in the taxonomic profiling application.” | controls[0] |
| outcome | control — held — “This data set dependence is not just a practical concern: the delineation of de novo OTUs depends on the relative abundances of the sampled community even in the limit of infinite sequencing depth and zero errors.” — “As a result, de novo OTUs defined in two different data sets cannot be compared.” | controls[1] |
| outcome | control — held — “As a demonstration of reusability and computational tractability on large data sets, the DADA2 method for ASV inference was used to process the 18S rRNA gene amplicon data from the TARA Oceans project, which contains ~766 million reads ( De Vargas et al. , 2015 ), on a 2016 Macbook Pro.” — “As a result, ASV inference cannot be performed independently on each read—the smallest unit of data from which ASVs can be inferred is a sample.” | controls[2] |
| limitation | The replacement inherits a limit the paper states itself: `The same genome can contain multiple ASVs if there are multiple copies of the targeted genetic locus` - one organism, several units — “The same genome can contain multiple ASVs if there are multiple copies of the targeted genetic locus.” | limitations[0] |
| limitation | and exact resolution does not buy biological coherence: `there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variant` — “For example, while necessarily better than the customary 3% ribosomal OTUs, there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variant ( Berry et al. , 2017 ).” | limitations[1] |
| limitation | and the authors say the naming does not do the work - `Appealing terminology such as 'resolution of exact sequence variants' does not eliminate the limitations inherent to representing a complex biological organism by a short genetic barcode.` A paper proposing a new unit, warning against being persuaded by its name — “Appealing terminology such as ‘resolution of exact sequence variants’ does not eliminate the limitations inherent to representing a complex biological organism by a short genetic barcode.” | limitations[2] |
| limitation | the promised merging is conditional: ASV tables `can only be simply merged if the underlying sequence data was derived from the same genetic locus`, in practice the same primer set, or after trimming to the overlapping region — “ASV tables can only be simply merged if the underlying sequence data was derived from the same genetic locus. In practice, this means that only studies using the same primer set can be simply merged, although ASVs from different-but-overlapping primer sets can be merged if the ASVs are first trimmed to the overlapping gene region.” | limitations[3] |
| limitation | This is a Perspective, and the case is argued rather than measured. The only quantitative demonstration is a tractability run - TARA Oceans 18S, ~766 million reads, on a laptop - which shows ASV inference SCALES, not that ASVs recover communities better than OTUs do. The accuracy claims are carried by citation — “As a demonstration of reusability and computational tractability on large data sets, the DADA2 method for ASV inference was used to process the 18S rRNA gene amplicon data from the TARA Oceans project, which contains ~766 million reads ( De Vargas et al. , 2015 ), on a 2016 Macbook Pro.” | limitations[4] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | — “Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUs are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence ( Eren et al. , 2016 ).” | content_tags.findings[0] |
| finding | — “A consequence of this often unacknowledged dual mandate is that OTUs struggle to serve both purposes well; the connection between OTUs and species is largely unfounded ( Stackebrandt and Ebers, 2006 ), and the most common methods often output numbers of OTUs an order of magnitude higher than the number of strains present in mock communities ( Kopylova et al. , 2016 ).” | content_tags.findings[1] |
| finding | — “This data set dependence is not just a practical concern: the delineation of de novo OTUs depends on the relative abundances of the sampled community even in the limit of infinite sequencing depth and zero errors.” | content_tags.findings[2] |
| finding | — “The absence of consistent labels makes replicating or falsifying previous results problematic, even impossible. Consider a significant association reported between a particular de novo OTU and a condition of interest. This association cannot be tested in a new data set, because that de novo OTU only exists within the data set in which it was defined. Therefore, results from the analysis of de novo OTUs can only be tested indirectly, by mapping de novo OTUs onto consistent labels such as taxonomy or reducing community composition to summary measures such as a diversity metric, and then testing the results at that coarser level.” | content_tags.findings[3] |
| finding | — “The observation of a closed-reference OTU indicates that at least one sequencing read was sufficiently similar to the associated reference sequence to be mapped to its OTU. However, it does not guarantee that the reference sequence itself was observed—to the contrary, that is often not the case.” | content_tags.findings[4] |
| finding | — “The same genome can contain multiple ASVs if there are multiple copies of the targeted genetic locus.” | content_tags.findings[5] |
| finding | — “Appealing terminology such as ‘resolution of exact sequence variants’ does not eliminate the limitations inherent to representing a complex biological organism by a short genetic barcode.” | content_tags.findings[6] |
| finding | — “For analysis to be reproducible the fundamental units must be reproducible, and de novo OTUs are not. For analysis to be comprehensive the fundamental units must be comprehensive, and closed-reference OTUs are not.” | content_tags.findings[7] |
| quantity | 3%, `most commonly` — — “The analysis of marker-gene data customarily begins with the construction of molecular operational taxonomic units (OTUs): clusters of reads that differ by less than a fixed sequence dissimilarity threshold, most commonly 3% ( Westcott and Schloss, 2015 ; Kopylova et al. , 2016 ).” | content_tags.quantities[0] |
| quantity | one nucleotide — — “nucleotide” | content_tags.quantities[1] |
| quantity | upwards of 90%, in the best-studied environments and loci - and `far fewer` elsewhere — — “In the best studied environments and for the most common genetic loci, such as 16S rRNA gene sequencing of the human gut, upwards of 90% of sequencing reads can typically be assigned to closed-reference OTUs. However, far fewer sequencing reads might be assigned to closed-reference OTUs in less-studied environments and when sequencing other genetic loci, because of limited representation in reference databases.” | content_tags.quantities[2] |
| quantity | an order of magnitude higher than the strains present — — “magnitude” | content_tags.quantities[3] |
| quantity | ~766 million reads, one 2016 MacBook Pro — — “As a demonstration of reusability and computational tractability on large data sets, the DADA2 method for ASV inference was used to process the 18S rRNA gene amplicon data from the TARA Oceans project, which contains ~766 million reads ( De Vargas et al. , 2015 ), on a 2016 Macbook Pro.” | content_tags.quantities[4] |
| variable | relative abundances — “abundances” | content_tags.variables[1] |
| variable | order of the reference sequences — “order” | content_tags.variables[2] |
| definition | operational taxonomic unit — “OTUs” | content_tags.definitions[0] |
| definition | consistent labeling — “If the same reference database is used, closed-reference OTU assignments from independently processed data sets can be validly compared, a property we refer to as consistent labeling” | content_tags.definitions[1] |
| definition | amplicon sequence variant — “ASVs” | content_tags.definitions[2] |
| definition | de novo OTU — “membership” | content_tags.definitions[3] |
| concept | operational taxonomic units — not defined in this paper's own words; a cluster of reads differing by less than a fixed dissimilarity threshold, most commonly 3% — “The analysis of marker-gene data customarily begins with the construction of molecular operational taxonomic units (OTUs): clusters of reads that differ by less than a fixed sequence dissimilarity threshold, most commonly 3% ( Westcott and Schloss, 2015 ; Kopylova et al. , 2016 ).” | concepts[0] |
| concept | de novo methods — not defined in this paper's own words; reads grouped into OTUs by their pairwise similarity to each other, so the clusters are properties of the data set — “Many methods for defining molecular OTUs have been proposed, but the most substantive distinction is between closed-reference methods—in which reads sufficiently similar to a sequence in a reference database are recruited into a corresponding OTU—and de novo methods—in which reads are grouped into OTUs as a function of their pairwise sequence similarities.” | concepts[1] |
| concept | closed-reference methods — not defined in this paper's own words; reads recruited to an OTU defined by a sequence in a reference database, so the clusters are properties of the database — “Many methods for defining molecular OTUs have been proposed, but the most substantive distinction is between closed-reference methods—in which reads sufficiently similar to a sequence in a reference database are recruited into a corresponding OTU—and de novo methods—in which reads are grouped into OTUs as a function of their pairwise sequence similarities.” | concepts[2] |
| concept | de novo OTUs — not defined in this paper's own words; emergent features of one data set; undefined outside it — “However, unlike de novo OTUs, ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism.” | concepts[3] |
| concept | closed-reference OTU — not defined in this paper's own words; a label owned by a reference database, observable without its own representative sequence ever being observed — “The observation of a closed-reference OTU indicates that at least one sequencing read was sufficiently similar to the associated reference sequence to be mapped to its OTU. However, it does not guarantee that the reference sequence itself was observed—to the contrary, that is often not the case.” | concepts[4] |
| concept | reference sequence — not defined in this paper's own words; the database entry that names and represents a closed-reference OTU — “The observation of a closed-reference OTU indicates that at least one sequencing read was sufficiently similar to the associated reference sequence to be mapped to its OTU. However, it does not guarantee that the reference sequence itself was observed—to the contrary, that is often not the case.” | concepts[5] |
| concept | ASVs — not defined in this paper's own words; amplicon sequence variants - biological sequences inferred exactly, to a single nucleotide, prior to amplification and sequencing error — “However, unlike de novo OTUs, ASVs are consistent labels because ASVs represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism.” | concepts[6] |
| concept | consistent labeling — not defined in this paper's own words; the property that assignments made independently in different data sets can be validly compared — “If the same reference database is used, closed-reference OTU assignments from independently processed data sets can be validly compared, a property we refer to as consistent labeling.” | concepts[7] |
| concept | ribosomal sequence variant — not defined in this paper's own words; an exactly resolved ribosomal sequence - finer than a 3% OTU and still not a guarantee of a coherent organism-level group — “For example, while necessarily better than the customary 3% ribosomal OTUs, there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variant ( Berry et al. , 2017 ).” | concepts[8] |
| concept | 3% ribosomal OTUs — not defined in this paper's own words; the customary threshold-defined unit the field has treated as 'like species' — “For example, while necessarily better than the customary 3% ribosomal OTUs, there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variant ( Berry et al. , 2017 ).” | concepts[9] |
| concept | amplicon sequencing error — not defined in this paper's own words; the PCR and sequencing artefacts that inflate apparent diversity — “Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUs are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence ( Eren et al. , 2016 ).” | concepts[10] |
| concept | grouping errors together with the error-free sequence — not defined in this paper's own words; OTU clustering used as de facto error correction - the construct's second job — “Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUs are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing error on measures of diversity and community composition by grouping errors together with the error-free sequence ( Eren et al. , 2016 ).” | concepts[11] |
| an order of magnitude, against a known answer | — “communities” | content_tags.statistics[0] |
| 90% is a best case reported as a norm | — “90%” | content_tags.statistics[1] |
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.
ISME J ISME J 1478 ismej The ISME Journal 1751-7362 1751-7370 Oxford University Press PMC5702726 PMC5702726.1 5702726 5702726 28731476 10.1038/ismej.2017.119 ismej2017119 1 Perspective Exact sequence variants should replace operational taxonomic unitsterm in marker-gene data analysis Exact sequence variants should replace OTUs CallahanpersName Benjamin J 1 * McMurdiepersName Paul J 2 HolmespersName Susan P 3 1 Department of Population Health and Pathobiology, NC State University , Raleigh NC, USA 2 Whole Biome Inc , San Francisco CA, USA 3 Department of Statistics, Stanford University , Stanford CA, USA * Department of Population Health and Pathobiology, NC State University , 1051 William Moore Drive, Raleigh, NC 27606, USA . E-mail: benjamin.j.callahanpersName@gmail.com 12 2017 21 07 2017 11 12 302112 2639 2643 21 03 2017 20 05 2017 07 06 2017 01 12 2017 01 12 2017 30 11 2017 01 12 2017 Copyright © 2017 The Author(s) 2017 The Author(s) https://creativecommons.org/licenses/by/4.0/ This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/ Recent advances have made it possible to analyze high-throughput marker-gene sequencing data without resorting to the customary construction of molecular operational taxonomic unitsterm (OTUs): clusters of sequencing reads that differ by less than a fixed dissimilarity threshold. New methods control errors sufficiently such that amplicon sequence variantsterm·def (ASVsterm) can be resolved exactly, down to the level of single-nucleotide differences over the sequenced gene region. The benefits of finer resolution are immediately apparent, and arguments for ASVterm methods have focused on their improved resolution. Less obvious, but we believe more important, are the broad benefits that derive from the status of ASVsterm as consistent labels with intrinsic biological meaning identified independently from a reference database. Here we discuss how these features grant ASVsterm the combined advantages of closed-reference OTUsterm—including computational costs that scale linearly with study size, simple merging between independently processed data sets, and forward prediction—and of de novo OTUsterm—including accurate measurement of diversity and applicability to communities lacking deep coverage in reference databases. We argue that the improvements in reusability, reproducibility and comprehensiveness are sufficiently great that ASVsterm should replace OTUs as the standard unit of marker-gene analysis and reporting.
pmc-status-qastatus 0 pmc-status-live yes pmc-status-embargo no pmc-status-released yes pmc-prop-open-access yes pmc-prop-olf no pmc-prop-manuscript no pmc-prop-legally-suppressed no pmc-prop-has-pdf yes pmc-prop-has-supplement no pmc-prop-pdf-only no pmc-prop-suppress-copyright no pmc-prop-is-real-version no pmc-prop-is-scanned-article no pmc-prop-preprint no pmc-prop-in-epmc yes pmc-license-ref CC BY Introduction High-throughput sequencing of PCR-amplified marker genes has grown explosively over the past decade, especially as a means of taxonomically profiling microbial communities. Increasing use of marker-gene sequencing has been accompanied by increasing data set sizes; this year, we expect thousands of marker-gene studies to generate millions to billions of sequencing reads each.
The analysis of marker-gene data customarily begins with the construction of molecular operational taxonomic unitsterm (OTUs): clusters of reads that differ by less than a fixed sequence dissimilarity threshold, most commonly 3% ( Westcott and Schloss, 2015 ; Kopylova et al. , 2016 ). The sample-by-OTU feature table serves as the basis for further analysis, with the observation of an OTU often treated as akin to the observation of a ‘species’ in the taxonomic profiling application. Many methods for defining molecular OTUs have been proposed, but the most substantive distinction is between closed-reference methodsterm—in which reads sufficiently similar to a sequence in a reference database are recruited into a corresponding OTU—and de novo methodsterm—in which reads are grouped into OTUs as a function of their pairwise sequence similarities.
Recently, new methods have been developed that resolve amplicon sequence variantsterm·def (ASVsterm) from Illumina-scale amplicon data without imposing the arbitrary dissimilarity thresholds that define molecular OTUs ( Eren et al. , 2013 ; Tikhonov et al. , 2015 ; Eren et al. , 2015 ; CallahanpersName et al. , 2016a ; Edgar, 2016 ; Amir et al. , 2017 ). ASVterm methods infer the biological sequences in the sample prior to the introduction of amplification and sequencing errors, and distinguish sequence variants differing by as little as one nucleotide. A similar class of methods developed for 454-scale data was typically used to ‘denoise’ sequencing data prior to constructing OTUs ( Quince et al. , 2011 ), while new ASVterm methods are explicitly intended to replace OTUs as the atomic unit of analysis. ASVterm methods have demonstrated sensitivity and specificity as good or better than OTU methods and better discriminate ecological patterns ( Eren et al. , 2013 ; Eren et al. , 2015 ; CallahanpersName et al. , 2016a ; Needham et al. , 2017 ). The higher resolution afforded by ASVterm methods has self-evident benefits—for example, it is clearly useful to distinguish Neisseria gonorrhoeae from the many other Neisseria species commonly found in the human microbiota—and initial evaluation has focused on that improved resolution. However, we argue here that the more important, and overlooked, advantage of ASVsterm is that they combine the benefits for subsequent analysis of closed-reference and de novo OTUsterm: ASVsterm are reusable across studies, reproducible in future data sets and are not limited by incomplete reference databases.
Description of OTU and ASVterm methods De novo OTUsterm are constructed by clustering sequencing reads that are sufficiently similar to one another. Many methods for constructing these clusters have been developed, but in all cases de novo OTUsterm are emergent features of a data set, with boundaries and membership that depend on the data set in which they are defined. This data set dependence is not just a practical concern: the delineation of de novo OTUsterm depends on the relative abundancesterm·var of the sampled community even in the limit of infinite sequencing depth and zero errors. As a result, de novo OTUsterm defined in two different data sets cannot be compared.
Closed-reference OTUsterm are properties of a reference database; each reference sequenceterm in the database defines and labels an associated closed-reference OTUterm. Sequencing reads are assigned to a closed-reference OTUsterm if they are sufficiently similar to the associated reference sequenceterm. If the same reference database is used, closed-reference OTUterm assignments from independently processed data sets can be validly compared, a property we refer to as consistent labelingterm. However, biological variation that is not represented in the reference database is necessarily lost during assignment to closed-reference OTUsterm.
ASVsterm are inferred by a de novo process in which biological sequences are discriminated from errors on the basis of, in part, the expectation that biological sequences are more likely to be repeatedly observed than are error-containing sequences. As a result, ASVterm inference cannot be performed independently on each read—the smallest unit of data from which ASVsterm can be inferred is a sample. However, unlike de novo OTUsterm, ASVsterm are consistent labels because ASVsterm represent a biological reality that exists outside of the data being analyzed: the DNA sequence of the assayed organism. Thus, ASVsterm inferred independently from different studies or different samples can be validly compared.
We schematically represent the validity of de novo OTUsterm, closed-reference OTUsterm and ASVsterm assigned from a common focal data set in Figure 1 . The x axis represents all biological variation that exists at the sequenced genetic locus. The y axis represents all amplicon data generated from that locus and all future data that may be generated. The region of validity for feature type is shaded.
This schematic emphasizes the limitations inherent to both classes of OTUs. De novo OTUsterm are invalid outside of the data set in which they were defined. Closed-reference OTUsterm cannot capture biological variation outside of the reference database used in their construction. ASVsterm transcend those limitations: ASVsterm capture all biological variation present in the data, and ASVsterm inferred from a given data set can be reproduced in future data sets and validly compared between data sets.
Practical consequences of consistent labels Computational tractability Consistent labels allow the assignment of closed-reference OTUsterm to be split into the independent assignment of subsets of the data that are then merged together. De novo OTUsterm lack consistent labels so all data must be pooled for assignment, resulting in a number of potential sequence comparisons that scales quadratically with total sequencing effort and rendering common de novo methodsterm prohibitively costly as study sizes increase ( Rideout et al. , 2014 ; Mahé et al. , 2015 ).
ASVterm inference cannot be performed on each read independently, but can be performed on each sample independently. As long as the sequencing effort devoted to individual samples remains tractable, independent inference by sample is trivially parallelizable and enables total computation time to scale linearly and memory requirements to remain flat with increasing sample number, allowing ASVsterm to be inferred from arbitrarily large data sets.
Meta-analysis The growing number of marker-gene studies in similar environments creates opportunities for new analyses that combine studies for more power and generality. The consistent labelingterm of closed-reference OTUsterm and ASVsterm allows per-study tables to be merged into a cross-study table. Meta-analysis is much more difficult with de novo OTUsterm, as the raw sequence data from each study must be compiled, pooled and reprocessed into new cross-study OTUs.
Replication The absence of consistent labels makes replicating or falsifying previous results problematic, even impossible. Consider a significant association reported between a particular de novo OTUterm·def and a condition of interest. This association cannot be tested in a new data set, because that de novo OTUterm·def only exists within the data set in which it was defined. Therefore, results from the analysis of de novo OTUsterm can only be tested indirectly, by mapping de novo OTUsterm onto consistent labels such as taxonomy or reducing community composition to summary measures such as a diversity metric, and then testing the results at that coarser level.
Forward prediction A major area of translational research is the use of microbial community composition as a predictive biomarker ( DiGiulio et al. , 2015 ; Baxter et al. , 2016 ). For example, the relative abundancesterm·var of a set of OTUs or ASVsterm might be used to predict a health condition. Predictive biomarkers can be constructed through statistical methods such as regression or by various machine learning methods, and their accuracy evaluated within the study by splitting the data into training and validation subsets ( CallahanpersName et al. , 2016b ). However, predictive biomarkers are only useful if they can be applied to new data. De novo OTUsterm exist only in the data set in which the the predictor was trained and evaluated, so predictive biomarkers based on de novo OTUsterm can’t predict from new data. Biomarkers based on consistently labeled features such as ASVsterm and closed-reference OTUsterm can be applied to new data, although closed-reference OTUsterm may limit predictive power by omitting predictive features absent from the reference database.
Practical consequences of reference independence Diversity measurement Reference databases are incomplete. As a result, the assignment of reads to closed-reference OTUsterm removes that portion of the data that is unrepresented in the reference database. This limitation is problematic if community diversity is of interest. The absence of unrepresented members of the community in closed-reference OTUterm tables can systematically skew diversity measures, potentially in a condition-dependent manner if some conditions are associated with a higher proportion of unrepresented taxa.
Application across environments and genetic loci The extent to which microbes inhabiting different environments are represented in reference databases varies greatly. In the best studied environments and for the most common genetic loci, such as 16S rRNA gene sequencing of the human gut, upwards of 90% of sequencing reads can typically be assigned to closed-reference OTUsterm. However, far fewer sequencing reads might be assigned to closed-reference OTUsterm in less-studied environments and when sequencing other genetic loci, because of limited representation in reference databases. ASVsterm and de novo OTUsterm more accurately represent the extent of biological variation in environments and loci without comprehensive coverage in reference databases.
Guaranteed observation The observation of a closed-reference OTUterm indicates that at least one sequencing read was sufficiently similar to the associated reference sequenceterm to be mapped to its OTU. However, it does not guarantee that the reference sequenceterm itself was observed—to the contrary, that is often not the case. Nevertheless, the often-absent reference sequenceterm serves as the representative of the closed-reference OTUterm, which can give misleading results when downstream analyses use the unobserved representative sequences to index into other sources of data.
Changing references Closed-reference OTUsterm are independent of the data, but are wholly dependent on the set of reference sequencesterm used in their assignment (closed-reference OTUsterm assigned by current methods even depend on the order of the reference sequencesterm·var ( Westcott and Schloss, 2015 )). Therefore, closed-reference OTUsterm assigned against different reference databases are not comparable. In order to maintain comparability over time, reference databases must remain static or closed-reference OTUterm tables must be regenerated whenever a new or upgraded reference database is adopted. Because ASVsterm are consistent labels derived from the data alone, ASVsterm remain consistent into the indefinite future. Discussion Molecular OTUs serve two different and orthogonal purposes. The first purpose is to translate taxonomic concepts developed in other systems into the context of high-throughput marker-gene sequencing of microbial communities by the ostensible equation of OTUs defined at certain thresholds with particular taxonomic levels (for example, 3% ribosomal OTUsterm are ‘like species’). The second purpose is to reduce the impact of amplicon sequencing errorterm on measures of diversity and community composition by grouping errors together with the error-free sequenceterm ( Eren et al. , 2016 ). A consequence of this often unacknowledged dual mandate is that OTUs struggle to serve both purposes well; the connection between OTUs and species is largely unfounded ( Stackebrandt and Ebers, 2006 ), and the most common methods often output numbers of OTUs an order of magnitude higher than the number of strains present in mock communities ( Kopylova et al. , 2016 ). Some of the shortcomings of molecular OTUs can be ameliorated. Open-reference OTU methods combine closed-reference OTUterm assignment with subsequent construction of de novo OTUsterm from the unassigned sequencing reads, gleaning the benefits of closed-reference OTUsterm without entirely sacrificing the unassignable portion of the data ( Rideout et al. , 2014 ). A clever algorithm has been developed that linearizes the computational time of a special case of de novo OTUterm·def assignment (single-linkage clustering with a linkage threshold of 1) allowing computational tractability on extremely large data sets ( Mahé et al. , 2015 ). Aggressive filtering and complete overlap between paired-end reads can reduce the rate at which OTU methods misinterpret sequencing artifacts as biological variation ( Bokulich et al. , 2013 ; Kozich et al. , 2013 ; Edgar and Flyvbjerg, 2015 ). Furthermore, the inference of exact sequence variants does not solve all problems. The same genome can contain multiple ASVsterm if there are multiple copies of the targeted genetic locus. Appealing terminology such as ‘resolution of exact sequence variants’ does not eliminate the limitations inherent to representing a complex biological organism by a short genetic barcode. For example, while necessarily better than the customary 3% ribosomal OTUsterm, there is still no guarantee of ecological coherence or even monophyly among genomes with the same ribosomal sequence variantterm ( Berry et al. , 2017 ). The merging of ASVsterm from different studies is attractive, but can be limited in practice. ASVterm tables can only be simply merged if the underlying sequence data was derived from the same genetic locus. In practice, this means that only studies using the same primer set can be simply merged, although ASVsterm from different-but-overlapping primer sets can be merged if the ASVsterm are first trimmed to the overlapping gene region. ASVsterm generated by different methods can be validly merged, but downstream analyses should account for the possibility of ‘batch effects’ due to methodological differences. These caveats stated, the breadth of issues that ASVsterm cleanly solve, and the more powerful and reproducible analyses that ASVsterm enable, makes a dispositive case in our opinion for replacing OTUs with ASVsterm. ASVsterm have an intrinsic biological meaning as a DNA sequence. ASVterm inference from large marker-gene data sets is both tractable and comprehensive. ASVsterm improve the return-on-investment of marker-gene sequencing by better leveraging the corpus of such data sets for further discovery, especially in oft-investigated communities like those inhabiting the human body. And the ASVterm methods that are now available provide better resolution and accuracy than OTU methods ( Eren et al. , 2015 ; CallahanpersName et al. , 2016a ). For analysis to be reproducible the fundamental units must be reproducible, and de novo OTUsterm are not. For analysis to be comprehensive the fundamental units must be comprehensive, and closed-reference OTUsterm are not. Replacing OTUs with ASVsterm makes marker-gene sequencing more precise, reusable, reproducible and comprehensive. We believe that ASVsterm should be the standard way that marker-gene data is processed and reported going forward. Data Supplement As a demonstration of reusability and computational tractability on large data sets, the DADA2 method for ASVterm inference was used to process the 18S rRNA gene amplicon data from the TARA Oceans project, which contains ~766 million reads ( De Vargas et al. , 2015 ), on a 2016 Macbook Pro. The resulting ASVterm tables have been deposited at Zenodo (DOI: 10.5281/zenodo.581694) alongside the workflow scripts that generated them.