Technical Note

Genus, species, subspecies: what each rank on a label actually guarantees

The three ranks printed on a probiotic specification are not equally solid. Species rests on a numerical threshold that was calibrated to reproduce existing names, and whose empirical justification is itself disputed. Subspecies has no field-wide threshold at all, and a 2022 paper argues the rank should be discontinued. And the unit actually being bought and sold — the strain — is not a taxonomic rank in the first place. This note sets out what each level does and does not guarantee, and why two organisms sharing a species name can be nearly identical in one species and profoundly different in another.

Published 23 October 2026 · 11 min read

Why bacteria get thresholds instead of a definition

For animals and plants, a species can be defined by whether members interbreed to produce fertile offspring. Prokaryotes reproduce asexually and exchange genes across enormous phylogenetic distances by horizontal transfer, so that criterion is simply unavailable.

Prokaryotic taxonomy therefore fell back on something frankly pragmatic: measure similarity, then draw a line. Everything that follows — including the parts that cause trouble in a registration dossier — comes from the fact that the line is a convention rather than a discovery.

The species threshold, and how the number was chosen

Criterion	Threshold	Standing today
DNA–DNA hybridisation (DDH)	70%	The 1987 committee recommendation. Wet-lab, laborious, hard to reproduce between labs
16S rRNA gene identity	~98.7%	A screening step only. Cannot resolve close relatives — see E. faecium versus E. lactis
Average Nucleotide Identity (ANI)	95–96%	The current working standard; roughly equivalent to the 70% DDH line
Digital DDH (dDDH)	70%	In-silico substitute for the hybridisation assay

The 95% figure was not derived from biology. It was reverse-calibrated: a value was sought that best reproduced the species taxonomists had already named on phenotypic grounds. The names came first; the threshold was fitted to them, then applied going forward.

The empirical argument, and the objection to it

The strongest support for the line is a reported discontinuity in sequence space. Analysing roughly 90,000 genomes and 8 billion genome pairs, 99.8% of pairs fell either above 95% ANI (within species) or below 83% (between species) — the middle band was nearly empty, and the pattern held after removing the most heavily sequenced species.

That interpretation has been challenged. A 2021 analysis argues the sparse middle may reflect the hierarchical structure of phylogeny plus redundancy in the reference databases rather than a genuine boundary, noting that both benchmarks were calibrated on a highly redundant set of named genomes; a separate survey of about 150,000 genomes reported no discontinuity across the 78–95% range.

So the honest description is: a convention, with empirical support that is still under argument. Nothing in a dossier depends on resolving that argument — but it explains why the next section is not an anomaly.

The asymmetry that explains everything: a ceiling, not a floor

Here is the single most useful thing to take from this note. The rule says members of a species are at least ~95% identical to one another. It caps the maximum divergence. It says nothing about the minimum.

That leaves the whole span from 95% to 100% available, and different species occupy wildly different amounts of it:

Lactiplantibacillus plantarum — fills the space

A 2026 analysis of 1,374 quality-filtered genomes found a pan-genome of 32,500 gene families of which only 1,175 were core — about 3.6% — and the pan-genome was still not saturating. This is a broad-niche organism found on plants, in fermented foods, in guts and in insects, with a correspondingly large accessory genome.

Bifidobacterium animalis subsp. lactis — sits at the top of it

Commercially circulating strains differ very little; the subspecies was long described as monomorphic or clonal. But the authors who proposed that noted the low diversity probably reflects sampling bias — the same handful of widely distributed commercial strains being sequenced repeatedly. A later comparison of 71 strains drawn from human and animal faeces, breast milk, fermented foods and supplements reported substantial diversity with an open pan-genome and a core of around 35%.

The precise statement is therefore “the strains in commerce are very similar to each other”, not “this subspecies lacks diversity”.

Lacticaseibacillus rhamnosus — differs where it matters

The spaCBA pilus operon is reported as a genomic rarity across this species. LGG carries it on a genomic island absent from other strains, and SpaC is the key factor in that strain’s adhesion to intestinal mucus. One accessory island, one headline functional property.

What determines how much room a species occupies is not its rank but its biology: niche breadth, rate of horizontal transfer and mobile-element activity, population history (a recent bottleneck or clonal expansion flattens diversity), and where anyone has bothered to sample.

Subspecies: the rank with no threshold

Species has a number. Subspecies does not have an agreed one. Most contemporary subspecies are defined by phenotypic coherence and monophyly rather than by a cutoff, and the quantitative proposals that exist are scattered and group-specific:

dDDH 70–80% has been proposed as a quantitative guide for subspecies delineation.
98% ANI was used to separate subspecies within Salmonella enterica in one study; a Mycobacterium abscessus analysis used 95% for species and 98% for subspecies within that group, with the authors noting the threshold’s general applicability was unverified.
For the Bacillus cereus group, a framework uses roughly 92.5% ANI for genomospecies while retaining formal subspecies names for medically important lineages.

And the rank itself is under challenge. A 2022 paper in New Microbes and New Infections argues that the taxonomic category of subspecies “has become irrelevant” and that its use “should be discontinued”, on the reasoning that well-defined lineages carry all the hallmarks of species — so most present-day subspecies are probably species in their own right.

This is not an academic curiosity for this industry. Two of the most commercially significant labels in probiotics are subspecies labels: Bifidobacterium animalis subsp. lactis and Bifidobacterium longum subsp. infantis. Whatever a claim rests on at that level rests on the softest-defined rank in the hierarchy.

Below species there is a second gap — and “strain” is not a rank

Work on 18,123 complete isolate genomes across 330 bacterial species reports a further discontinuity at around 99.5% ANI, proposed as the basis for defining intra-species units such as genomovars.

That is roughly the resolution at which commercially meaningful differences live. Two points follow:

“Strain” is not a taxonomic rank. It is a culture with a provenance and a designation. No nomenclatural code governs it, which is exactly why a strain number is a commercial and documentary instrument rather than a taxonomic one.
ANI cannot answer a strain question. At 95–96% it decides species membership; it is far too coarse to distinguish two strains of the same species. That requires MLST, cgMLST or whole-genome SNP comparison — a different assay answering a different question.

What anchors a name: type material

A species name means something only because it is tied to a designated reference. Under the International Code of Nomenclature of Prokaryotes, the nomenclatural type must be a viable pure culture deposited in public culture collections — which is why uncultivated organisms could not be validly named for decades, however well sequenced they were.

SeqCode, published in 2022, changes that by allowing genome sequences to serve as nomenclatural types — isolate genomes, metagenome-assembled genomes or single-amplified genomes — registered through its own registry, while otherwise following the existing code on name formation and priority.

The practical consequence for anyone reading an identification report: ask what the comparison was made against. A species assignment is a statement about distance from a type, so the type’s identity and the method are part of the result, not footnotes to it.

Names change; the organism does not

Three patterns worth recognising, because each one shows up differently in paperwork:

Reclassification. In 2020 the genus Lactobacillus was split into 25 genera, which is why specification sheets now carry Lactiplantibacillus, Lacticaseibacillus, Limosilactobacillus and the rest. Nothing about any organism changed; the labels, dossiers and literature searches all did.
Convention overriding the threshold. Shigella falls inside Escherichia coli by ANI, yet is maintained as separate for medical and nomenclatural reasons. Thresholds are applied by people, with reasons.
Normalisation. The Genome Taxonomy Database applies ANI-based criteria uniformly and reassigns names accordingly, which produces a more consistent system at the cost of breaking familiar ones. Two databases can legitimately give one organism two names.

A name change is a documentation event, not a biological one. Unfortunately, documentation is the regulated part.

What this means for labels, dossiers and claims

Level on the label	What it guarantees	Confirmed by
Genus	Very little functionally; and the name may be reassigned	16S, MALDI-TOF
Species	A ceiling on genetic distance (~95% ANI), never a floor	ANI / dDDH against type material
Subspecies	No agreed threshold; the rank itself is contested	Group-specific criteria; state which were used
Strain	The identity the evidence and the dossier actually attach to	MLST / cgMLST / WGS SNP, plus provenance records

1. “Same species” carries different information depending on the species

In a near-clonal species it nearly amounts to “genomically comparable”. In an open pan-genome species it says almost nothing. You cannot infer how similar two strains are from the taxonomic rank; you have to know that species’ population structure.

2. Where genomics will not carry the argument, two other legs still do

For a near-clonal species, arguing strain specificity from genomic difference does not work — the strains really are nearly identical. Strain identity still matters for two reasons that are independent of genetic distance: the clinical evidence is attached to a named strain (one strain’s trial data does not transfer to another of the same species), and traceability and registration require a strain identity. Neither collapses just because two genomes look alike.

3. Admissibility is decided by name, so a renaming is a real event

Permitted-strain lists, catalogues and product categories are all keyed to names. When taxonomy moves, the organism in the fermenter does not — but the entry it corresponds to may. Keep the synonym chain in the file (old name, current name, the reclassification that connects them), and verify category admissibility against the current official announcement rather than against a remembered name.

4. Match the level of the claim to the level of the identification

An ANI report supports a statement about species. It does not support a statement about a strain, and it was never designed to. If a claim is written at strain level, the identification behind it has to be at strain level too — which is the same discipline we have described for matching cited evidence to the organism in hand.

If you are checking a name

We can provide species confirmation by whole-genome ANI against type material, with the method and the comparison recorded in the accompanying documentation, and we can supply the synonym history for names affected by reclassification. Strain-level discrimination is a separate exercise and we will say so rather than present an ANI figure in its place.

This note is about taxonomy and documentation. It makes no claim about the efficacy or safety of any product, ours or anyone else’s.

References

Jain C, Rodriguez-R LM, Phillippy AM, Konstantinidis KT, Aluru S. High throughput ANI analysis of 90K prokaryotic genomes reveals clear species boundaries. Nature Communications, 2018;9:5114. PubMed record
Murray CS, Gao Y, Wu M. Re-evaluating the evidence for a universal genetic boundary among microbial species. Nature Communications, 2021;12:4059. Full text (open access)
Venter SN, Palmer M, Steenkamp ET. Relevance of prokaryotic subspecies in the age of genomics. New Microbes and New Infections, 2022;48:101024. Full text (open access)
Rodriguez-R LM, et al. An ANI gap within bacterial species that advances the definitions of intra-species units. Record
Hedlund BP, et al. SeqCode: a nomenclatural code for prokaryotes described from sequence data. Nature Microbiology, 2022;7:1702–1708. Full text (open access)
Large-Scale Comparative Genomics of 1,374 Lactiplantibacillus plantarum Genomes Reveals an Open Pan-genome and Strain-Level Variation in Probiotic-Associated and Safety-Related Genomic Features. Current Microbiology, 2026. PubMed record
Zheng J, et al. A taxonomic note on the genus Lactobacillus: Description of 23 novel genera, emended description of the genus Lactobacillus Beijerinck 1901, and union of Lactobacillaceae and Leuconostocaceae. International Journal of Systematic and Evolutionary Microbiology, 2020;70:2782–2858.
Wayne LG, et al. Report of the ad hoc committee on reconciliation of approaches to bacterial systematics. International Journal of Systematic Bacteriology, 1987;37:463–464.

This note summarises published work on prokaryotic taxonomy in order to explain what each taxonomic rank does and does not establish. It is not a claim about the efficacy or safety of any product, is not directed at any company or brand, and is not medical, legal or regulatory advice. Threshold values cited for subspecies are proposals from individual studies or specific taxonomic groups and are not field-wide standards. Statements about permitted-strain lists and product categories are general; the current official announcement governs in every case.