The unculturable majority: reading microbes we cannot grow

Take pond water. Put a drop under a microscope and count the cells. Then spread a measured volume on nutrient agar and count the colonies. The two numbers should roughly agree. They do not.

Staley and Konopka named this the great plate count anomaly in 1985: the gap between cells countable by direct microscopy and cells that form colonies on a plate. In low-nutrient aquatic habitats the plate count commonly runs one to four orders of magnitude below the microscopic count.

Key idea: For roughly a century, culture-based microbiology was studying whichever organisms happened to grow on the media people had invented. The catalogue was not a sample of microbial life. It was a sample of the cooperative fraction.

Nothing is wrong with the cells. The plate is simply the wrong offer, and there are several distinct reasons why.

• Missing partners. Many organisms live syntrophically, consuming a product another species makes. Isolated on a plate, that supply is gone.
• Missing signals. Some cells will not resume division without chemical cues from neighbours.
• Too rich a medium. Organisms adapted to scarcity are damaged by standard laboratory broth, which is a feast by their standards.
• Wrong physical conditions. Pressure, temperature, oxygen tension or redox state that no incubator was set to.
• Too slow. Cells with generation times of days or weeks never form a visible colony before the plate is discarded.
• Viable but nonculturable. The stress-induced state from the previous lesson, intact and active, but not dividing.

The escape was to abandon culture as a prerequisite. If you cannot grow an organism, extract its DNA from the sample and identify it by sequence.

That needs a marker gene present in every organism, and the 16S ribosomal RNA gene is close to ideal. Carl Woese used it to define the domains for exactly these properties:

The conserved regions give you a handle that fits every organism; the variable regions between them carry the identity.

No step in this pipeline requires the organism to be alive, let alone willing to grow. That is the entire point of it.

An amplicon run does not report cells. It reports reads, and reads are converted to relative abundance.

reads = {"Bacteroides": 41230, "Prevotella": 8115, "Faecalibacterium": 12960}
total = sum(reads.values())            # 62,305 reads
{name: round(100 * n / total, 1) for name, n in reads.items()}
# {'Bacteroides': 66.2, 'Prevotella': 13.0, 'Faecalibacterium': 20.8}

Those percentages must sum to 100 by construction. If one taxon genuinely triples in a sample, every other taxon's percentage falls, even where their absolute numbers never moved. Data that behaves this way is called compositional, and comparing such percentages between samples as though they were counts is one of the most common errors in the field.

Compositionality is not the only caveat, and each of these has a direction you can reason about.

Gotcha: Genome copy number varies between species, commonly from one to around fifteen copies of the 16S gene. An organism with ten copies contributes ten times the reads of an equally abundant organism with one, so an uncorrected profile systematically over-represents high-copy taxa.

Beyond that: universal primers are not truly universal and miss taxa whose binding sites differ; most studies sequence a single variable region rather than the full gene, which usually caps resolution near genus level; and DNA persists after death, so a profile includes organisms that are no longer alive. A 16S profile tells you which lineages left DNA in the sample, in roughly what proportion. That is genuinely useful and it is not a census.

Shotgun metagenomics drops the marker gene and sequences all the DNA in the sample, then assembles reads into draft genomes, called metagenome-assembled genomes.

The decisive gain is functional. An amplicon profile tells you who is present; a metagenome shows the genes they carry, so you can ask whether a community holds a particular resistance gene or metabolic pathway without knowing which organism to look in.

Lloyd, Steen and colleagues put numbers on it in mSystems in 2018 by combining cell counts with sequence surveys across Earth's major environments.

About 81 percent of microbial cells, some $7.3 \times 10^{29}$, belong to genera with no cultured representative. Around 25 percent, some $2.2 \times 10^{29}$, belong to phyla with none. Not unnamed species: entire phyla, the rank that separates you from a mushroom, known only as sequence.

That same study found one environment where the uncultured fraction is small: the human body, where most resident microbes have been cultured. Decades of clinical effort went into growing exactly these organisms. It is the best-characterised microbial habitat on the planet and it is thoroughly unrepresentative of the rest.

The human microbiome also carries microbiology's most-repeated wrong number.

Sequencing made the invisible majority countable. It did not make it understood.

Gotcha: Nearly all microbiome findings are associations. Sequencing a hundred patients and a hundred controls and finding a taxon enriched in the patients tells you nothing about direction. The organism may drive the disease, the disease may alter the gut and favour the organism, or a medication, a diet or a transit time may drive both.

Breaking the tie takes intervention rather than more sequencing: colonising germ-free animals with a defined community, or transferring a community and watching what follows. That is also where the inability to culture bites hardest, because you cannot add an organism to a gnotobiotic mouse if nobody can grow it.

The resolution is not to choose between the plate and the sequencer. It is to let one instruct the other.

A metagenome-assembled genome lists the transporters an organism has, the amino acids it cannot make, and the electron acceptors it can use. That is a recipe. Media designed from genomes, diffusion chambers that let a cell sit in its own habitat while being isolated, and very long incubations have all brought organisms into culture that plates had refused for decades.

The arc across this cursus ends where it started. Lesson one described a cell built on a different plan; lesson two counted it; lesson three attacked it. This lesson is the reminder that the catalogue those three describe covers a minority of what is out there, and that the boundary is a limit of method rather than of biology.