Bacterial Taxonomy and Phylogenetics

Traditional bacterial taxonomy classified organisms by observable traits: cell shape, Gram stain, spore formation, oxygen requirements, and biochemical tests such as sugar fermentation. This approach, sometimes called numerical taxonomy or phenetics, groups organisms by overall similarity without any explicit claim about ancestry. The problem is that bacteria evolve their traits through convergence, loss, and horizontal gene transfer so easily that two unrelated species can look and behave almost identically, while close relatives can look wildly different. A classification built on appearance is therefore not guaranteed to reflect real evolutionary history. Molecular phylogenetics solves this by comparing informational molecules — DNA or RNA sequences — that accumulate changes over time in a comparatively clock-like way. The chosen molecule for bacteria (and, more broadly, for all cellular life) is the gene encoding the small ribosomal subunit RNA, called 16S rRNA in prokaryotes (18S rRNA in eukaryotes). This gene has three properties that make it an ideal "universal chronometer": it is present in every free-living organism because ribosomes are essential; it is functionally constrained, so it changes slowly enough to preserve deep evolutionary signal over billions of years; and it contains both highly conserved stretches (useful for designing PCR primers that work across nearly all bacteria) and variable stretches (useful for telling species apart). A phylogenetic tree built from such sequences is a testable hypothesis of common descent, not merely a similarity chart — branching patterns are meant to reflect the actual historical order in which lineages split. Taxonomy and phylogenetics are related but distinct: taxonomy is the practice of naming and classifying organisms into a nested hierarchy (domain, phylum, class, order, family, genus, species), while phylogenetics is the science of reconstructing evolutionary trees from data. Modern bacterial taxonomy is explicitly phylogenetic — taxonomic ranks are meant to correspond to clades (monophyletic groups: an ancestor and all its descendants) recovered from molecular trees, rather than to arbitrary trait combinations.

Step 1 — Choosing and amplifying the marker gene. The 16S rRNA gene is roughly 1,500 base pairs long and consists of about nine variable regions (V1–V9) interspersed among highly conserved regions. Because the conserved regions are nearly identical across all bacteria, researchers design "universal" PCR primers that bind there and amplify the variable regions in between, from essentially any bacterial DNA sample, without needing to grow the organism in culture. This is what makes 16S sequencing so powerful for environmental and clinical samples containing organisms that will not grow on any known medium — estimates suggest well over 99% of environmental bacterial species have never been cultured.

Step 2 — Sequencing and alignment. The amplified fragments are sequenced (historically by Sanger sequencing of cloned fragments, now typically by high-throughput sequencing of many samples in parallel). Sequences from different organisms are then aligned — conserved regions are lined up as anchors, and the variable regions in between are compared position by position. Alignment is not trivial: insertions and deletions (indels) must be accounted for using gap penalties, since evolution adds and removes nucleotides as well as substituting them.

Step 3 — Converting sequence difference into evolutionary distance. The simplest measure is p-distance: the fraction of aligned sites that differ between two sequences. But raw percent difference underestimates true evolutionary change, because multiple substitutions can occur at the same site over long timescales (a site that mutated twice and reverted looks unchanged), a phenomenon called multiple hits or saturation. Correction models — the Jukes-Cantor model being the simplest, more elaborate models like Kimura-2-parameter or general time-reversible models being more realistic — mathematically estimate the true number of substitutions per site, correcting for this hidden multiple-hit problem.

Step 4 — Tree building. Once pairwise distances (or the full aligned character matrix) are in hand, one of several algorithms builds a tree:

• Distance methods (e.g., neighbor-joining) cluster sequences by overall similarity, iteratively joining the closest pairs while correcting for distances to all other sequences. Fast, but discards some information by reducing sequences to single distance numbers.
• Maximum parsimony seeks the tree requiring the fewest total evolutionary changes to explain the observed sequences — an application of Occam's razor to evolution.
• Maximum likelihood asks, for a given model of sequence evolution, which tree topology makes the observed data most probable; it is more computationally demanding but generally more accurate.
• Bayesian inference treats tree topology and branch lengths as unknowns with prior probability distributions, updated using the sequence data via Markov chain Monte Carlo sampling, yielding trees with explicit posterior probability support values.

Step 5 — Assessing confidence. Because a single dataset gives only one tree estimate, biologists test robustness using bootstrapping: resampling alignment columns with replacement thousands of times, rebuilding the tree each time, and reporting what percentage of replicate trees recover each branching point (a bootstrap value). Branches with less than roughly 70% support are generally treated with caution.

Step 6 — Reading the tree. A phylogenetic tree is composed of tips (extant or sampled taxa), internal nodes (inferred common ancestors), and branches whose length (in most representations) is proportional to the amount of evolutionary change — not necessarily to elapsed time. A "root" designates the direction of evolutionary time, usually placed using an outgroup — a taxon known independently to have diverged before the rest of the group. Sister taxa share a more recent common ancestor with each other than either does with any other group on the tree.

The molecular clock. If substitutions accumulate at an approximately constant rate over time, then the number of differences between two sequences is roughly proportional to the time since they diverged from a common ancestor — this is the molecular clock hypothesis, first proposed by Emile Zuckerkandl and Linus Pauling in the 1960s from studies of hemoglobin. Calibrating the clock requires an independent estimate of divergence time, usually from the fossil record or from a well-dated geological event, to convert "substitutions per site" into "years." For bacteria this is especially fraught: there are almost no bacterial fossils suitable for calibration, rates vary between lineages and even between genes within a lineage (rate heterogeneity), and horizontal gene transfer can make different genes in the "same" genome tell different evolutionary stories. Molecular clock estimates for bacterial divergences are therefore treated as rough, model-dependent hypotheses rather than precise dates.

Carl Woese and George Fox's 16S rRNA survey (1977). Working at the University of Illinois, Woese and Fox catalogued short oligonucleotide sequences from the 16S (and 18S) rRNA of dozens of organisms — bacteria, eukaryotic cells, and a group of unusual microbes called methanogens that produce methane gas and live in anaerobic environments such as swamp mud and animal guts. Methanogens had always been classified as unusual bacteria on morphological grounds. Woese and Fox instead compared their rRNA "signature sequences" using pairwise association coefficients (SAB values). The methanogens' rRNA signatures were roughly as different from ordinary bacteria as they were from eukaryotes — a result nobody expected. Rather than clustering with bacteria, methanogens formed a third, deeply separate group. Woese and Fox published this in the Proceedings of the National Academy of Sciences in 1977, initially calling the group "archaebacteria."

What this proved. Cell structure (methanogens are prokaryotic, lacking a nucleus, just like bacteria) does not track evolutionary history reliably. Molecular sequence, in contrast, revealed a genuine third branch of life with its own ancient origin, independent of both bacteria and eukaryotes. This directly overturned the long-standing two-empire classification of life into prokaryotes and eukaryotes, replacing it with the three-domain system: Bacteria, Archaea, and Eukarya — the framework foreshadowed in The Three Domains of Life and elaborated further in Archaea and Extremophiles. Later biochemical work reinforced the molecular result: archaea have distinctive ether-linked membrane lipids (rather than bacteria's and eukaryotes' ester-linked lipids), different RNA polymerase structures more closely resembling eukaryotic enzymes, and cell walls lacking the peptidoglycan found in nearly all bacteria — independent lines of evidence converging on the same conclusion the rRNA tree had already revealed.

Rooting the universal tree — the second key experiment. A tree built purely from 16S/18S rRNA comparisons is unrooted; it shows relationships but not which branch point came first. In 1989, two independent groups (led by J. Peter Gogarten and by Naoyuki Iwabe, working with Takashi Miyata) solved this using ancient gene duplications. Certain gene families — such as the genes encoding elongation factors (EF-Tu/EF-1α and EF-G/EF-2) and the catalytic subunits of ATP synthase — duplicated before the three domains diverged. Because both paralogous copies exist in all three domains, each copy can be used to root the tree using the other copy as an outgroup for itself. Both studies independently placed the root of the universal tree between Bacteria on one side and an Archaea/Eukarya clade on the other, meaning Archaea and Eukarya share a more recent common ancestor than either does with Bacteria — a result now well supported, though the exact placement of Eukarya (nested within or sister to Archaea) remains an active research question given growing genomic evidence, such as the 2015 discovery of Lokiarchaeota, for eukaryotes having emerged from within archaeal diversity.

Sanger sequencing made it possible. None of this would have worked without Frederick Sanger's dideoxy chain-termination sequencing method (1977), which allowed Woese's group (and everyone since) to read out RNA/DNA sequences directly rather than relying on indirect fingerprinting. Modern high-throughput ("next-generation") sequencing has since scaled the same logic from single-gene surveys of a handful of organisms to metagenomic surveys of entire microbial communities containing thousands of species simultaneously.

Example 1 — Estimating divergence time with a molecular clock. Suppose two bacterial lineages' 16S rRNA genes differ at 3% of aligned sites (p-distance = 0.03) after correcting for multiple hits with a Jukes-Cantor model, giving a corrected distance of d = 0.031 substitutions per site. Suppose a calibration point (from a well-dated symbiotic association, since bacterial fossils are essentially unusable) estimates that 16S rRNA in this lineage evolves at a rate of roughly 1% sequence divergence per 50 million years (a commonly cited but approximate figure for some bacterial groups). Then: Time since divergence ≈ (0.031 / 0.01) × 50 million years ≈ 155 million years. Caveats to report alongside this number: 16S rRNA evolves slowly and fairly uniformly compared to protein-coding genes, which makes it useful for deep phylogeny but gives it low resolution for very recently diverged species (two species can share 99.9% identical 16S rRNA yet have diverged millions of years ago and be metabolically very different) — this is why species-level and strain-level identification increasingly relies on whole-genome comparisons rather than 16S rRNA alone.

Example 2 — Reading a tree and diagnosing an infection. A hospital microbiology lab sequences the 16S rRNA gene (V3–V4 region) from a blood culture that failed to grow on standard media after 5 days, a scenario suggesting a fastidious or slow-growing organism. The resulting sequence is compared by BLAST against a curated reference database (such as SILVA or the Ribosomal Database Project) and by neighbor-joining tree placement. The unknown sequence groups, with 98% bootstrap support, as sister to Tropheryma whipplei with 99.6% sequence identity — well above the roughly 98.7% identity threshold conventionally used as a rough (and imperfect) guideline for same-species assignment in 16S taxonomy. Combined with the clinical picture (chronic diarrhea, joint pain, weight loss), this molecular identification supports a diagnosis of Whipple's disease, caused by an organism notoriously difficult to culture by standard methods. This illustrates the central clinical payoff of molecular taxonomy: a pathogen never grown on a plate can still be identified with confidence, because identity is defined by sequence, not by culture behavior.

Carl Woese (1928–2012) trained originally in biophysics, earning a PhD from Yale in 1953 studying bacteriophage before pivoting toward the origin of the genetic code and, eventually, evolutionary microbiology at the University of Illinois at Urbana-Champaign. Colleagues describe him as intensely focused and somewhat reclusive, uninterested in the conventional academic reward system, and willing to spend a decade on a painstaking cataloguing project — sequencing short oligonucleotide fragments of ribosomal RNA one at a time using two-dimensional gel fingerprinting, long before automated sequencing existed — because he believed it was the only rigorous way to ask how life's major lineages related to one another. When his 1977 paper with George Fox proposed a third domain of life, the reaction from much of the microbiology community was skeptical dismissal; a widely reported New York Times article on the discovery drew scorn from some biologists who considered the claim implausible on its face. Woese was, in time, thoroughly vindicated: the three-domain system is now the standard framework taught in every introductory biology course, and Woese received a MacArthur "genius" grant (1984) and the Crafoord Prize (2003, an honor the Royal Swedish Academy created partly because no Nobel Prize exists for the field Woese effectively founded).

Clinical microbiology laboratories now routinely use 16S rRNA gene sequencing (and increasingly whole-genome sequencing) to identify pathogens directly from patient samples, bypassing culture entirely for organisms that grow slowly, require unusual conditions, or have already been killed by antibiotics before the sample was taken — critical in cases of culture-negative endocarditis, meningitis, and sepsis. Public health agencies use bacterial phylogenomics to trace outbreaks in near-real time: during outbreaks of Escherichia coli O104:H4, Listeria monocytogenes, and multidrug-resistant Klebsiella pneumoniae, whole-genome phylogenies distinguish a true point-source outbreak (a tight cluster of nearly identical genomes) from unrelated background cases, and can identify the specific hospital ward or food-processing facility responsible. In biotechnology and industrial microbiology, 16S and whole-genome phylogenetics guide bioprospecting for novel enzymes and antibiotics by flagging environmental samples containing lineages phylogenetically distant from well-studied groups — distant relatives are more likely to harbor biochemically novel pathways. In agriculture, phylogenetic surveillance tracks the evolution and spread of plant pathogens such as Xylella fastidiosa and monitors the diversity of nitrogen-fixing rhizobia in soil, informing crop management. In Microbial Ecology and the Microbiome, 16S amplicon sequencing (metabarcoding) remains the standard low-cost method for surveying "who is there" in environments ranging from the human gut to deep-sea hydrothermal vents to Antarctic ice cores, and is the entry point for the deeper study of Archaea and Extremophiles living in the planet's most inhospitable niches. Conservation biology uses environmental DNA phylogenetics to detect invasive or endangered microbial and eukaryotic species from water samples without ever observing the organism directly. In every one of these applications, the underlying logic traces back to the same insight Woese formalized: sequence comparison of a conserved, essential molecule is a far more reliable guide to relatedness than anything visible under a microscope.

Ask the tutor: How do scientists actually root a phylogenetic tree when there's no outside reference to compare to?