16S profiling and its taxonomic limits

16S profiling amplifies part of the bacterial and archaeal small-subunit ribosomal gene and classifies those amplicons. It is a marker method with a ceiling. The troubleshooting problem this page is for is a species list, a stacked bar, or a pathway claim that the marker cannot support, even when the sequencing itself succeeded. Shotgun sequencing of the whole community is a different experiment, described in metagenomic sequencing overview. The amplicon logic, primers and all, is amplicon sequencing for a defined region. The reads still come from a library and a run, next-generation sequencing from library to reads.

If you need the exact base sequence of one cultured colony's ribosomal gene, a Sanger read of that single amplicon can be the cleaner record. It has the same taxonomic ceiling. A prettier chromatogram does not split Escherichia from Shigella.

Variable regions are a choice of window

The 16S gene alternates conserved stretches and variable regions, conventionally numbered V1 through V9. Primers sit in the conserved stretches so many taxa amplify, and the variable stretch between them supplies the differences a classifier can see. No window sees every difference the full gene contains. Taxa that happen to vary outside your window look identical.

V3-V4 is commonly discussed because the amplicon is a few hundred base pairs, short enough that paired short reads can cover it and often overlap. That practical fit is why so many protocols use it. It is not a universal taxonomic optimum. Some groups resolve better in other regions, and some groups amplify poorly with the primer pair a V3-V4 protocol treats as standard. V4 alone is another common short window, with its own blind spots. Full-length 16S, read by a long-read method, sees more of the gene and still fails to separate species that are identical along the whole marker. Length removes the window problem. It does not remove the biology of a slow-evolving gene.

Write the primer sequences, not only "V3-V4". Two pairs that both claim that region are not the same assay. Check the pair against the group you care about with a tool such as NCBI Primer-BLAST, and accept that in silico coverage is a prediction. Empirical dropout still happens.

What the read fraction is made of

Copy number is the bias people forget because the bar chart looks like abundance. Ribosomal operon counts differ among bacteria, from a single copy to many. A taxon with ten operons can contribute ten times the reads of a taxon with one, at the same cell count, before primer bias is even considered. Comparing read fractions and calling them community structure is already a distortion. It can be a useful distortion if every sample was measured the same way. It is not a cell count, and it is not comparable to a shotgun relative abundance without a model you state.

Primer bias sits on top. A mismatch near the 3-prime end drops a lineage out. Entire clades can be silent with a given pair, including some archaea when the primers were optimised on bacteria. The lineage is absent from the table. It may not be absent from the sample. A mock community with a known member that vanishes is the control that reveals this. A mock community that lacks that member will not.

PCR adds chimeras, as incomplete products jump templates. Mixed 16S reactions are among the worst environments for that jump, because so many similar templates share the tube. A chimera looks like a novel organism that matches no database entry, or matches badly. Screen for chimeras before you celebrate a new lineage. The screen is a filter with its own false negatives. Say that you ran one.

The sequencing depth of a 16S run is depth of a PCR. Coverage depth is not the same as accuracy applies with extra force: you can have immense depth of a biased amplification. Rarefaction and similar curves tell you whether you keep finding new sequence types as you add reads. They do not correct primer bias, and they do not turn a genus into a species.

Denoising into exact sequence variants, or clustering into operational taxonomic units at a similarity cutoff, are analysis classes. Exact variants are sharper objects. They are still 16S sequences. Clustering at a historical similarity cutoff does not create a species concept. Either output is then placed on a reference taxonomy. The reference has a version. GenBank is one home for the sequences; curated ribosomal databases are another. Name the one you used. A taxonomy that forces every variant to a species will produce species names the marker did not earn. Prefer the rank at which the placement is stable.

Where a species name fails, and what to do instead

Take a species-level call and ask three questions. Do multiple species share this sequence? Did the classifier break a tie by a rule you would defend? Does the primer pair even amplify the neighbours that could have tied? If you cannot answer, report the genus, or report the sequence variant and the list of species it matches. Escherichia and Shigella are the teaching case because the conflict is famous and unresolved by this gene. They are not the only case. Many Bacillus, Staphylococcus, and Mycobacterium species will be over-split by a confident-looking table.

A genus call has failure modes too. Primer dropout can delete a genus. Copy number can inflate one. A contaminant from a reagent kit can introduce a genus that is real DNA and unreal ecology, especially in low-biomass samples. The extraction blank is the control. Taxa in the blank are not part of the sample's story until their abundance is overwhelmingly higher in the sample than in the blank. Low-biomass studies that skip the blank discover the kit.

Compositional data add a further trap. Fractions sum to one. If one genus blooms in the PCR, every other genus's fraction falls, including genera whose absolute amount did not change. A decrease in a bar chart is not a decrease in cells. If the claim needs absolute direction, you need a measurement the marker's composition does not contain, such as a spike-in or a separate quantification. State the limit when you plot relative abundance.

Do not attach functions. Antibiotic resistance, virulence, and metabolism are gene contents that vary inside a genus and inside a species. 16S does not carry them. Shotgun metagenomics can, when the genes are actually seen. A citation that a genus "is associated with" a pathway belongs in the discussion as literature, not in the results as a measurement.

ITS and 18S deserve one clear separation and then no further treatment here. ITS is the fungal internal transcribed spacer, a different marker with different primers and, for some fungi, finer resolution that still has its own biases. 18S targets eukaryotes more broadly. Running bacterial 16S primers and reporting fungi, or mixing the tables because all three are "amplicon sequencing", produces a community that never existed. Commission them as separate assays if you need them.

Failures that survive a deep, pretty run

A deep run of a contaminated PCR characterises the contaminant. The no-template control and the extraction blank have to be sequenced, not merely run on a gel, because a faint band becomes a large fraction after indexing. If the blank's profile resembles the samples, the batch is not interpretable. Repeating the analysis with a rarer threshold will not remove a reagent genus.

Index hopping between 16S libraries is particularly misleading. Every sample shares the same primer insert, so a hopped read is a perfect amplicon of the wrong sample. It creates shared rare variants across the plate. Adapters indexes and barcode hopping is the mechanism. Unique dual indexes and a blank lane or well are the check. A variant that appears at low level in every sample, including the blank, is a method variant until proven otherwise.

Over-classification is a software failure mode. Forcing a species label, then running a differential-abundance test on those labels, produces significant species that are threshold artefacts. Test at the rank you trust, or test the sequence variants and annotate them cautiously. Changing the database version halfway through a project splits the cohort into two taxonomies. Freeze the version.

Sanger of a picked colony tells you the dominant 16S sequence in that colony's amplicon. It does not survey the community, and mixed peaks mean the colony was not pure or the primers sat down more than once. Use it to check an isolate. Do not use it to "confirm" a species the short-read profile should never have named.

Not a clinical identification

A 16S profile from a wound swab, a sputum, or a stool is not a diagnostic identification of an infection. Research taxonomy sits under an ethics approval if the sample is human, and under an institutional biosafety decision because the sample may be infectious. The Laboratory biosafety manual is a reference point for that decision, not a substitute for it. This page does not authorise clinical use. Do not treat a genus call as a pathogen report, and do not treat the absence of a genus as clearance. Primer dropout makes absence a weak sentence.

Holding temperature, and the region written down

A swab that sits warm grows some organisms and loses others before extraction. The profile then describes the delay. Write the holding condition, the time, and the preservative into the specification, and record excursions when a freezer warms after a power cut. Do not analyse those samples as if they shared a cold history with the rest of the batch. In the same specification, write the variable region, the primer sequences, the chemistry of the reads, and the database version. A collaborator who is handed "16S data" and a species table cannot see the ceiling you already hit. Humidity and an unsealed amplicon plate matter for the same reason they matter in any PCR: the reaction changes volume, and the chimera rate is not the one you validated.

What to put in the enquiry

Ask for the primer pair, the variable region, the read layout, the denoising or clustering choice, the taxonomy version, and whether blanks are sequenced. Ask which rank will be reported, and ask that species calls be limited to placements the marker can bear. Say explicitly if the biological question needs functions or species resolution. In that case the enquiry is for shotgun community sequencing, not a deeper 16S run.

16S as a defined amplicon can be discussed against the Sanger DNA sequencing enquiry reference when the object is one colony, and a community shotgun alternative can be discussed against the whole-genome sequencing enquiry reference. Reagent classes are in the genomics and sequencing catalogue. The design context is the genomics research pathway. Put primers, rank, blanks, and database in the quote request. What to check before a sequencing run covers the blank and the library. Commissioning a sequencing or proteomics study is where the rank you will defend is written before the bar chart exists.