Taxonomic Unit and OTUs Explained

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-09-20

A taxonomic unit is any named group in a biological classification system, whether that group is a rank such as species or genus or a specific named taxon such as Escherichia coli. An operational taxonomic unit, usually shortened to OTU, is something different: a cluster of DNA sequences grouped together because they share a defined level of similarity, classically 97 percent identity for the 16S ribosomal RNA gene.

The distinction matters because the two terms get used interchangeably in lab meetings and papers, and that confusion causes real errors. A taxonomic unit is a hypothesis about evolutionary relatedness that a taxonomist names and defends. An OTU is a computational convenience that a clustering algorithm produces from a fasta file. Treating the second as if it were the first is one of the most common mistakes in microbiome science.

What a Taxonomic Unit Actually Is

A taxonomic unit is the basic building block of biological classification. Every level of the Linnaean hierarchy, domain, kingdom, phylum, class, order, family, genus, and species, is a rank, and the actual named group sitting at that rank is a taxonomic unit. Homo sapiens is a taxonomic unit at the species rank. Mammalia is a taxonomic unit at the class rank. Firmicutes is a taxonomic unit at the phylum rank in bacterial taxonomy.

Two properties define a taxonomic unit. First, it has a name that follows a formal code, such as the International Code of Nomenclature for algae, fungi, and plants or the International Code of Nomenclature of Prokaryotes. Second, it has a type specimen or type strain that anchors the name to a physical reference. When a microbiologist says E. coli K-12, they are pointing at a specific strain that has been cultured, characterized, and deposited in a culture collection.

That anchoring is what makes a taxonomic unit falsifiable. If new genomic data show that a named species is actually two distinct lineages, taxonomists can revise the classification and split the unit. The name changes. The type strain stays put.

Ranks Versus Clades

Modern taxonomy increasingly recognizes that ranks are arbitrary. A clade is a group consisting of an ancestor and all of its descendants, and clades exist in nature whether or not anyone assigns them a rank. Ranks are human labels layered on top. This is why some bacteriologists now write about "phylotypes" or "ribotypes" instead of forcing every lineage into a genus or species box.

For uncultured microorganisms, which make up the vast majority of environmental bacteria, there is often no type strain and no formal name. The organism exists, its DNA is in your sample, and it has never been grown on a plate. Taxonomy has no clean slot for it. This gap is exactly where the operational taxonomic unit was born.

Why OTUs Exist

The operational taxonomic unit concept came from numerical taxonomy in the 1960s, when researchers needed an objective way to group organisms by measured characters without relying on expert judgment. The term migrated into molecular ecology in the 1990s as Sanger sequencing of cloned 16S rRNA genes became routine.

The problem was scale and error. Early sequencing produced reads with base-calling mistakes. Two reads from the same organism could differ by a few bases purely because of technical noise. If you treated every unique sequence as its own unit, you would count one bacterium as five or ten "species." Clustering sequences at a similarity threshold collapsed that noise into manageable groups.

The 97 percent threshold became the field standard. The reasoning was that 16S rRNA genes from organisms generally recognized as the same species typically share more than 97 percent sequence identity, while different species usually fall below that line. Stackebrandt and Goebel formalized this in 1994, and the number stuck for three decades.

A 97 percent OTU is therefore a cluster of sequences that are at least 97 percent identical across the aligned region. Everything inside the cluster is treated as one unit for downstream analysis. The centroid sequence, or a representative sequence chosen from the cluster, gets compared against a reference database to assign a taxonomy.

The Clustering Workflow

Building OTUs from raw amplicon data follows a predictable path.

1. Quality filtering removes reads with low base quality scores or unexpected lengths.
2. Primer and adapter trimming strips the synthetic sequences used to initiate amplification.
3. Dereplication collapses identical reads into unique sequences with abundance counts.
4. Chimera removal discards hybrid sequences formed when two templates partially co-amplify.
5. Clustering groups the remaining unique sequences at the chosen threshold.
6. Representative picking selects one sequence per cluster.
7. Taxonomic assignment compares representatives against a reference database such as SILVA or EzBioCloud.

Reference-based clustering, sometimes called closed-reference OTU picking, matches reads against a pre-built database and discards reads that do not hit anything. De novo clustering builds clusters from the data itself with no reference. Open-reference approaches do both, clustering against the database first and then clustering the leftovers de novo. Each choice changes the final OTU table, which is the core problem with the method.

A Worked Example

Suppose you sequence the V4 region of 16S rRNA from a soil sample and get 50,000 quality-filtered reads. After dereplication you have 8,000 unique sequences. Clustering at 97 percent yields 1,200 OTUs. One of those OTUs contains 4,300 reads spread across 12 unique sequences that differ by one to three bases. You pick a representative, BLAST it against SILVA, and the best hit is Bacillus subtilis at 99.2 percent identity. You label the OTU "Bacillus" at genus level and move on.

Now sequence a second soil sample from a nearby site. The same organism is present, but this time its reads cluster with a slightly different set of neighbors, and the representative sequence is a different variant. You get a different OTU identifier. The two samples now appear to share no Bacillus OTU even though they clearly share the same organism. This is the reproducibility problem in miniature.

OTUs Are Not Species

This is the single most important point in the entire topic. An OTU is a cluster defined by a similarity cutoff and a clustering algorithm. A species is a taxonomic hypothesis defined by taxonomists using multiple lines of evidence. They are not the same thing, and an OTU should never be reported as a species without qualification.

Three specific reasons make the equivalence fail.

First, the 97 percent threshold is a heuristic, not a law. Some valid species share more than 97 percent 16S identity, so a single OTU can contain multiple species. Other species have 16S genes that diverge below 97 percent, so one species can be split across multiple OTUs. The threshold cuts through biological reality rather than tracking it.

Second, the result depends on the clustering algorithm. Different tools make different decisions about where to draw cluster boundaries, especially for sequences sitting near the threshold. A study that benchmarked eight clustering and denoising methods on a 227-strain mock community found that OTU algorithms led by UPARSE produced clusters with lower error rates but more over-merging, while ASV algorithms led by DADA2 gave consistent output but suffered from over-splitting [1]. The same input data yields different OTU counts depending on which tool you run.

Third, the result depends on the reference database. Assigning taxonomy to a representative sequence requires a curated database, and different databases contain different sequences with different naming conventions. A study of 16S-ITS-23S operon sequencing compared four reference databases and two classification methods to see which combinations gave consistent species-level calls [2]. The database you choose shapes the answer you get.

The practical consequence is that OTU identifiers are not portable. An OTU labeled OTU_47 in one paper means nothing in another paper unless both used identical primers, identical clustering parameters, and identical databases. Reviewers increasingly ask for representative sequences to be deposited so the claim can at least be checked.

The Rise of Amplicon Sequence Variants

Amplicon sequence variants, or ASVs, are exact DNA sequences inferred after correcting sequencing errors at single-nucleotide resolution. Instead of clustering reads into fuzzy groups, denoising algorithms such as DADA2 and Deblur model the error process and separate true biological variation from technical noise.

The result is that each ASV corresponds to one exact sequence. Two studies that sequence the same organism with the same primers will recover the same ASV, regardless of which samples were included or which other organisms were present. That property, reproducibility across studies, is the main reason ASVs have largely displaced OTUs in new work [3][4].

The trade-off is real. ASV methods are more aggressive about quality control and can discard a large fraction of reads. A streamlined pipeline built on HmmUFOtu retained 93.2 percent of over 104 million read pairs in one stool dataset while DADA2 retained only 44.6 percent, and in a second dataset HmmUFOtu retained 89.2 percent against DADA2's 18.4 percent [5]. Both approaches still produced qualitatively similar beta-diversity plots, which is reassuring for ecological conclusions even when the read counts differ dramatically.

ASVs also tend to over-split. A single organism can carry multiple copies of the 16S rRNA gene that differ slightly from each other, and a denoiser will call each variant a separate ASV. The same benchmarking study that found DADA2 consistent also found it prone to over-splitting [1]. So ASVs are not magically closer to species than OTUs. They are more precise about sequences and less precise about organisms.

OTU Versus ASV at a Glance

Both columns describe computational objects, not organisms. Neither column is a species.

How This Plays Out in Real Studies

Microbiome research uses these units across an enormous range of hosts and environments, and the choice of unit shapes what the study can claim.

In a colorectal cancer screening cohort of 358 clinical stool samples, researchers compared de novo OTU clustering against DADA2 and Deblur. All three approaches produced similar taxonomic profiles, with no significant difference in PERMANOVA and strong correlation in the Mantel test, but the number of OTUs or ASVs and the alpha-diversity indices varied considerably [3]. The ecological conclusion held. The diversity numbers did not.

A study of dental plaque in caries-active and caries-free children used OTU clustering to compare groups and found no significant differences in alpha or beta diversity. Notably, Streptococcus mutans, long treated as the primary cariogenic pathogen, was absent in 65 percent of caries-active subjects, while 17 genera appeared exclusively in plaque from caries-affected surfaces [6]. The OTU framework was good enough to surface that finding.

In captive coconut lorikeets, full-length 16S rRNA sequencing on the Oxford Nanopore platform was used to compare oral and cloacal microbiomes across adults and juveniles, with observed OTU counts feeding into alpha-diversity metrics [7]. In yaks, V4 region sequencing with 97 percent OTU clustering and SILVA annotation was used to compare captive and free-range animals, with functional prediction layered on top [8]. In silkworms, OTU data from EzBioCloud were used to estimate bacterial richness in guts of larvae fed on two mulberry species [9].

The pattern is consistent. OTUs remain perfectly serviceable for comparing groups within a single study where all samples were processed identically. They become unreliable when you try to compare across studies or accumulate data over years.

When Closed-Reference OTUs Still Win

Closed-reference OTU clustering has one genuine advantage that ASVs do not match: it makes datasets mergeable. Because every read is matched against a fixed reference database, two datasets processed months apart end up with the same OTU identifiers for the same reference sequences. In one large-scale comparison, shared OTUs between two independently processed Korean stool datasets showed a correlation coefficient of 0.92 in total abundance on a log scale [5]. That kind of cross-dataset consistency is hard to achieve with de novo methods.

For long-read platforms, OTU assembly is also still an active area. The Pike tool was built specifically to assemble OTU sequences de novo from Oxford Nanopore amplicon data without restrictions on amplicon size [10]. Full-length 16S and 16S-ITS-23S operon sequencing create new clustering challenges that the short-read ASV toolkit does not fully address.

The Main Workflow, Step by Step

The decision path from raw reads to a final unit table can be drawn as a simple flow.

flowchart TD
    A[Raw amplicon reads] --> B[Quality filtering]
    B --> C[Primer trimming]
    C --> D[Dereplication]
    D --> E[Chimera removal]
    E --> F{Choose unit type}
    F --> G[Cluster at threshold]
    F --> H[Denoise to exact variants]
    G --> I[Pick representative]
    H --> J[Keep each variant]
    I --> K[Assign taxonomy]
    J --> K
    K --> L[Build unit table]
    L --> M[Diversity analysis]

The branch at the center is the decision that defines everything downstream. Choosing the left path gives you operational taxonomic units. Choosing the right path gives you amplicon sequence variants. Both paths converge on the same taxonomic assignment step, which is where the reference database enters.

Common Mistakes and Limitations

Treating OTUs as species is the first and worst mistake. An OTU at 97 percent identity is a sequence cluster, and calling it Lactobacillus acidophilus overstates what the data show. Report the unit as an OTU and report the best taxonomic assignment separately, with the identity percentage attached.

Reporting OTU counts as absolute richness is the second mistake. OTU counts depend on sequencing depth, clustering threshold, and the tool used. A sample with more reads will usually yield more OTUs even if its true diversity is identical. Rarefaction, coverage-based normalization, or diversity estimators that account for unseen taxa are needed before comparing richness across samples.

Assuming OTU identifiers transfer between studies is the third mistake. OTU_47 in your dataset and OTU_47 in someone else's dataset have no relationship unless both used the same primers, the same clustering algorithm, the same threshold, and the same reference database. They almost never do.

Ignoring the threshold's arbitrariness is the fourth. Clustering at 99 percent and at 97 percent on the same data produces different OTU counts and different diversity values [11]. There is no single correct threshold. There is only the threshold you chose and the reason you chose it.

Assuming ASVs solve everything is the fifth. ASVs are reproducible and exact, but they over-split within-taxon variation and they discard reads aggressively in some datasets [1][5]. An ASV is a sequence variant, not an organism, and a single genome can contribute several ASVs.

Finally, remember that all of these methods operate on a marker gene. 16S rRNA copy number varies between taxa, primer bias favors some groups over others, and the region you amplify determines what you can resolve. No clustering choice fixes a biased amplicon.

Quick Review

• A taxonomic unit is any named group at any rank, anchored by a type specimen and governed by a nomenclature code.
• An operational taxonomic unit is a cluster of sequences sharing a similarity threshold, classically 97 percent for 16S rRNA.
• OTUs are not species. They are computational objects whose boundaries depend on the clustering algorithm, the threshold, and the reference database.
• Amplicon sequence variants are exact sequences recovered by denoising, and they are reproducible across studies in a way OTUs are not.
• ASVs over-split within-taxon variation while OTUs over-merge distinct taxa near the threshold.
• Closed-reference OTUs remain useful for merging datasets processed at different times.
• Every result depends on the marker gene, the primers, and the reference database, not just on the clustering step.

Frequently Asked Questions

What is an operational taxonomic unit in simple terms?

An operational taxonomic unit is a group of DNA sequences that are similar enough to be treated as one unit in an analysis. The classic definition clusters 16S rRNA sequences at 97 percent identity, meaning sequences that match at 97 percent or more get placed in the same OTU.

Is an OTU the same as a species?

No. An OTU is a sequence cluster defined by a similarity cutoff and a clustering algorithm. A species is a taxonomic hypothesis defined by taxonomists using multiple lines of evidence and anchored to a type strain. A single OTU can contain multiple species, and a single species can be split across multiple OTUs.

Why is 97 percent the standard threshold for OTUs?

The 97 percent figure came from observations that 16S rRNA genes from organisms generally recognized as the same species usually share more than 97 percent identity. It became a field convention rather than a biological law, and it is now understood as a heuristic that does not perfectly track species boundaries.

What is the difference between an OTU and an ASV?

An OTU is a cluster of similar sequences, while an ASV is an exact sequence recovered after error correction at single-nucleotide resolution. ASVs are reproducible across studies because the same sequence always produces the same variant, whereas OTU identifiers depend on the dataset and parameters used to build them.

Can I compare OTUs between two different studies?

Usually not. OTU identifiers are only comparable when both studies used the same primers, the same clustering algorithm, the same similarity threshold, and the same reference database. Closed-reference clustering against a shared database improves comparability, but most published OTU tables cannot be merged directly.

Do ASVs replace OTUs completely now?

ASVs have become the default for new short-read 16S and ITS studies, but OTUs have not disappeared. Closed-reference OTU clustering is still used to merge datasets processed at different times, and OTU assembly remains relevant for long-read platforms where amplicon sizes vary widely.

Related Articles

• Experimental Unit
• Operational Definitions in Quantitative Research
• Metagenomic Taxonomic Classification
• Nucleosome Structure: The Fundamental Unit of Chromatin
• pal genetic research unit tower
• Experimental Units: Definition, Examples, and Common Mistakes

Sources

1. The unresolved struggle of 16S rRNA amplicon sequencing: a benchmarking analysis of clustering and denoising methods.
2. Evaluating the efficiency of 16S-ITS-23S operon sequencing for species level resolution in microbial communities.
3. An independent evaluation in a CRC patient cohort of microbiome 16S rRNA sequence analysis methods: OTU clustering, DADA2, and Deblur.
4. RAPID: an interactive R/Shiny platform for end-to-end 16S rRNA and ITS amplicon sequence analysis using DADA2.
5. A streamlined pipeline based on HmmUFOtu for microbial community profiling using 16S rRNA amplicon sequencing.
6. Microbial profiling of dental plaque in caries-active and caries-free children: A cross-sectional 16S rRNA gene sequencing study.
7. Comparative characterization of oral and cloacal microbiomes in captive adult and juvenile coconut lorikeets (Trichoglossus haematodus) using full-length 16S rRNA sequencing.
8. Comparative gut microbiome composition and predicted microbial functions in captive and free-range yaks (Bos grunniens).
9. Comparative analysis of gut microbiota in Bombyx mori fed on M. alba and M. nigra using 16S rRNA amplicon sequencing.
10. Pike: OTU-Level Analysis for Oxford Nanopore Amplicon Metagenomics.
11. ASV vs OTUs clustering: Effects on alpha, beta, and gamma diversities in microbiome metabarcoding studies.