II. 2. 16S rRNA-Based Amplicon Sequencing – Powerful but Limited

16S rRNA sequencing is cheap and well-validated, reliably revealing the major bacterial groups — yet it sees only part of the gut ecosystem.

The 16S ribosomal RNA gene is present in every prokaryote, but its sequence differs between species – this is what makes bacterial identification possible. In amplicon sequencing, DNA is extracted from a stool sample, a defined segment of the 16S gene (a so-called variable region, V1–V9) is amplified and then sequenced. The result is a composition map: which bacterial taxon is present in what proportion in the sample [15].

The method's advantages are real. It can be performed on large sample numbers at relatively low cost. It works with well-validated databases (SILVA, Greengenes). It is capable of identifying the most characteristic phyla of the human gut microbiome. For research purposes – particularly in population studies – it is a proven tool [16].

The method has clinically significant, methodologically documented limitations. These are discussed in detail in the "Five Fundamental Limits" subsection below. One limitation is worth highlighting separately:

Limitations in species-level identification: Most 16S analyses can only reliably identify down to genus level; species- or strain-level differentiation is often not possible [15]. This can be clinically critical, since pathogenic and harmless strains can occur within the same genus.

Imagine thousands of different LEGO sets with all their pieces thrown into the air simultaneously, then from the pieces scattered across the floor, we must identify exactly which sets were there and how many pieces each had. The difficulty is compounded by not being able to pick up every piece – some rolled under the bed, some got stuck together, and some look so similar to others that they cannot be distinguished by eye. This is the essence of 16S rRNA analysis: we read from gut DNA which bacterial "set" was there – but we only examine one characteristic element from each (a variable region of the 16S gene), not the whole set. The result is a real but incomplete picture: it shows the main groups, but similar elements can be confused, pieces under the bed (non-culturable strains) remain invisible, and we cannot tell exactly how many pieces were present from each set.

Analytical algorithms: interpolation and hallucination

When software processes the raw data of 16S sequencing – millions of short DNA fragments – it goes through two critical steps: clustering (OTU or ASV formation) and database matching [15], [18]. In OTU formation, similar sequences are grouped into a single unit; the ASV method by contrast considers differences down to single nucleotide level. The two approaches identify different "species" from the same data.

In database matching, the program assigns an unknown sequence to a known bacterium – but only if there is a sufficiently close reference in the database. If there is none, the software interpolates: it infers what the unknown organism "might be" based on its nearest known relative. This is a useful approximation – but not a measurement. Interpolation means that what we see in the report is partly genuine observation and partly algorithmic estimation [15], [18]. The two are generally indistinguishable in the final result.

A more serious case is when the program hallucinates: it "identifies" a bacterium that is not actually present in the sample, having incorrectly matched it to a database entry based on sequence similarity [49].. This occurs particularly with low-abundance taxa, where a single "false read" can appear as a percentage presence in the report. The phenomenon is not theoretical: in reproducibility studies, "detected" taxa regularly appear in negative controls – which should theoretically contain no bacteria [17].

What does this mean for the patient?

When someone has a 16S rRNA-based microbiome test done, the resulting report typically contains colourful pie charts, percentages and reference bands [15], [16]. Interpretation seems temptingly simple: this value is low, we are above or below the "optimal" range. The reality, however, is considerably more complex.

First, defining a "reference range" is not trivial [18], [16]. The healthy human microbiome shows extremely high individual variability – the gut microbiota composition of two perfectly healthy people can differ by as much as 90%, while both are functionally stable and healthy. Comparing to an "average healthy" microbiome can therefore be misleading: every person's individual "optimum" is different.

Second, the 16S analysis only shows who is present – not what they are doing [18]. The presence of a bacterium does not mean it is active, that it produces metabolites in functionally relevant quantities, or that it is currently involved in a pathological process. There are multiple steps between a genetic map and functional activity – and 16S measures none of them.

None of this means the test has no value. Compared to itself – using the same method, the same laboratory, the same sampling protocol, repeated over time – it can genuinely show how the microbiome composition changes in response to a treatment (such as FMT). This longitudinal, follow-up application is where 16S diagnostics has its most convincing clinical utility today [15], [18].

The Five Fundamental Limits of 16S rRNA Technology

Alongside its technological elegance, 16S rRNA-based amplicon sequencing carries five methodologically documented limits that affect clinical reliability:

1. Reagent and laboratory contamination ("kit-ome"). DNA from laboratory kits, reagents, and equipment can produce background noise greater than that of the sample itself — especially for low-biomass samples [17], [471]. Stool samples are usually rich enough to dilute contamination, but skin or mucosal samples can be substantially distorted.

2. Low biomass — high error. The lower the bacterial content, the higher the relative share of contaminating DNA, and the result can become entirely misleading [471]. Clinically, sample-collection conditions (timing, method, storage) directly influence outcome.

3. V-region choice affects the profile. The 16S gene's hypervariable regions (V1–V2, V3–V4, V4, V6–V8, etc.) are targeted by different "universal" primer pairs. The same sample analysed with different V-region targeting produces different community profiles; certain bacterial groups are well-detected by one region and poorly by another [472]. Two laboratories using different primers can report different "microbiotas" for the same patient.

4. Inter-lab variability. The MBQC (Microbiome Quality Control) consortium demonstrated that identical samples processed in different labs with different kits and bioinformatic pipelines yield substantially different results [16]. Clinical implication: tests from different providers are not directly comparable.

5. Relative proportions, not absolute counts. 16S (and shotgun) yield compositional data: the relative proportions of taxa, not their absolute cell counts. If a taxon "increases" in the report, it may actually reflect a decrease in other taxa [23]. Absolute quantification (qPCR or flow cytometry calibration) can yield a different picture — addressed in detail in Chapter II.4.

These limits do not invalidate 16S as a research tool, but they justify interpretive caution in clinical decision-making.

References

[15] Jovel J, Patterson J, Wang W et al. Characterization of the human gut microbiome using 16S or WGS: a comparative study on the outcome of different analysis pipelines. Front Microbiol. 2016. Link

Jovel and colleagues compare 16S rRNA gene sequencing and whole-genome shotgun (WGS) metagenomics on the same human gut samples to assess analytical pipeline impact. They demonstrate that taxonomic resolution, observed diversity and the reproducibility of differential abundance calls depend strongly on the choice of variable region, reference database, OTU/ASV clustering approach and bioinformatic pipeline. WGS recovered more species-level and functional information than 16S, but at higher cost. The authors benchmark QIIME, mothur, MG-RAST and assembly-based pipelines, showing meaningful disagreement between them. They recommend transparent reporting of every analytical step and caution against over-interpreting single-pipeline microbiome results.

[16] Sinha R, Abu-Ali G, Vogtmann E et al. Assessment of Variation in Microbial Community Amplicon Sequencing by the Microbiome Quality Control (MBQC) Project Consortium. Nature Biotechnology. 2017. Link

The Microbiome Quality Control (MBQC) baseline study assessed taxonomic profiling variability across 15 laboratories and 9 bioinformatics protocols using blinded stool, chemostat and artificial community specimens. Variability depended most on biospecimen type and origin, followed by DNA extraction, sample handling environment and bioinformatics pipeline. Artificial community analyses revealed quantitative differences in extraction efficiency and bioinformatic classification. The findings highlight the need for standardisation to enable meta-analysis of population-scale microbiome studies.

[17] Salter SJ, Cox MJ, Turek EM et al. Reagent and Laboratory Contamination Can Critically Impact Sequence-Based Microbiome Analyses. BMC Biology. 2014. Link

Contaminating DNA is shown to be ubiquitous in commonly used DNA extraction kits and laboratory reagents, varying greatly in composition between different kits and batches. This contamination critically distorts results from low-microbial-biomass samples in both 16S rRNA gene surveys and shotgun metagenomics. The authors provide an extensive list of potential contaminating genera and mitigation guidelines, and recommend caution when applying sequence-based techniques to low-biomass microbial environments.

[18] Gloor GB, Macklaim JM, Pawlowsky-Glahn V, Egozcue JJ. Microbiome datasets are compositional: and this is not optional. Front Microbiol. 2017. Link

Microbiome datasets generated by high-throughput sequencing of 16S rRNA amplicons, metagenomes or metatranscriptomes are inherently compositional because the instrument imposes an arbitrary total. The review explains the pathologies that arise when compositional data are analysed with non-compositional methods and provides guidance for applying compositional data analysis throughout microbiome study workflows. The compositional framework is presented as essential, not optional, for valid inference.

[23] Vandeputte D, Kathagen G, D'hoe K et al. Quantitative Microbiome Profiling Links Gut Community Variation to Microbial Load. Nature. 2017. Link

Conventional sequencing-based faecal microbiota analyses provide only relative abundances, hampering the link between microbiome features and quantitative host parameters when microbial load varies between samples. The authors argue that relative profiling can mask altered total microbiota abundance as a key disease-associated signal and call for quantitative microbiome profiling that pairs relative composition with cell-density counts to enable genuine characterisation of host-microbiota interactions.

[49] DeFilipp Z, Bloom PP, Torres Soto M et al. Drug-Resistant E. coli Bacteremia (the presence of bacteria in the bloodstream) Transmitted by Fecal Microbiota Transplant. N Engl J Med. 2019. Link

Case report of two patients in independent FMT clinical trials who developed ESBL-producing Escherichia coli bacteremia after the procedure; both cases were linked to the same stool donor by genomic sequencing, and one patient died. Highlights the risk of multidrug-resistant organism transmission via FMT and supports enhanced donor screening protocols. The report underpins regulatory updates requiring multidrug-resistant pathogen screening of all FMT donor material.

[471] Eisenhofer R, Minich JJ, Marotz C et al. Contamination in Low Microbial Biomass Microbiome Studies: Issues and Recommendations. mSystems. 2019. Link

To evaluate the impact of decreasing microbial biomass on 16S rRNA gene sequencing, the authors generated a mock community dilution series and tested four computational decontamination approaches: filtering by negative-control sequences, by relative abundance, by inverse correlation with DNA concentration (Decontam), and by contaminant-source modeling (SourceTracker). As expected, the proportion of contaminant bacterial DNA rose as starting biomass fell, reaching 80,1% in the most diluted sample. The benchmark provides practical guidance for choosing decontamination methods in low-biomass microbiome studies and highlights the limits of in silico approaches.

[472] Yang B, Wang Y, Qian PY. Sensitivity and Correlation of Hypervariable Regions in 16S rRNA Genes in Phylogenetic Analysis. BMC Bioinformatics. 2017. Link

In the Normative Aging Study, a binomial model was used to investigate the association between a metabolic-syndrome index and DNA methylation. Iterative Sure Independence Screening (ISIS) with elastic-net penalty was applied to methylation levels at 484 548 CpG markers from 659 human subjects. The screening step significantly improved elastic-net performance. The method identified four CpGs mapping to two biologically relevant, functional genes. These markers may have practical implications for prevention and treatment of metabolic syndrome.