Four methods are in routine use, they give related but not identical answers, and knowing which produced a number changes how much weight it deserves. Here is what happens on the bench.
Broth microdilution, the reference method
A 96 well plate, a doubling dilution series across the rows, a standardised inoculum in every well, and a read after overnight incubation. The steps that matter:
- Medium. Cation adjusted Mueller Hinton broth. The cation adjustment is not a detail: divalent cation concentration materially changes aminoglycoside and polymyxin results.
- Inoculum. Approximately 5 x 105 colony forming units per millilitre in the final well. Prepared from a fresh culture adjusted to a 0.5 McFarland standard and then diluted.
- Compound. Dissolved, usually in DMSO, then diluted so that the solvent concentration is low enough not to inhibit growth by itself. Solvent controls exist for exactly this reason.
- Incubation. 35 to 37 degrees Celsius, ambient air, 16 to 20 hours.
- Reading. By eye against a dark background, or by optical density, or with a metabolic indicator such as resazurin where visual reads are unreliable.
- Controls. Growth control, sterility control and a quality control strain with a published expected range. If the QC strain is out of range, the plate is void, not interesting.
Agar dilution
Compound is incorporated into molten agar at each concentration, plates are poured, and organisms are spotted on top with a replicator. Laborious to set up and excellent when you want many organisms against a few concentrations, which makes it the traditional choice for surveillance work and for organisms that grow poorly in broth.
Gradient diffusion strips
A plastic strip carrying an exponential concentration gradient is laid on an inoculated agar plate. An elliptical zone of inhibition forms and the MIC is read where the ellipse meets the strip. Fast, convenient, and able to report values between doubling steps, which reads as extra precision but is really a different scale. Agreement with broth microdilution is usually within one doubling dilution, which is the same tolerance applied everywhere else.
Disk diffusion is not a MIC
A paper disk with a fixed compound load, a zone diameter in millimetres, and a species specific interpretive table. It is a susceptibility test, and a good one, but the output is a category rather than a concentration. Converting a zone diameter into a MIC is not valid without a validated regression for that agent and organism.
Where the error comes from
| Source | Effect | Control |
|---|---|---|
| Inoculum size | Too heavy shifts MICs up, the inoculum effect, pronounced for beta-lactams | Colony count the prepared inoculum periodically |
| Medium composition | Cations, thymidine, serum content all shift specific classes | Use the standard medium, state deviations |
| Compound solubility | Precipitation in the top wells reads as activity loss | Inspect wells, check solvent limits |
| Plate binding | Cationic peptides adsorb to polystyrene | Treated plates, or state the plate type |
| Reading time | Early reads flatter the compound | Fixed window, same reader |
| Subjectivity | Trailing endpoints are read differently by different people | Defined endpoint criteria, or an indicator dye |
The last row is the one that surprises people new to the assay. A trailing endpoint, where growth fades gradually across several wells rather than stopping cleanly, is common with bacteriostatic agents, and the convention for reading it has to be agreed within the group or the numbers will not be comparable between two people at the same bench.
What this means when you read a published number
Keep the method with the number. A MIC from a gradient strip, a MIC from broth microdilution in standard medium and a MIC from a screening assay in a non standard buffer are three different quantities that share a unit. This is the main reason assembling a picture from public bioactivity data takes an afternoon per compound.
And when you read a predicted number
A prediction is standing on the same heterogeneous literature, so it inherits the same uncertainty and should say so. That is why every predicted row carries a confidence level, and why a compound with no published neighbourhood gets no band rather than a plausible one. The MIC reference page covers the definition and interpretation side, and the method page explains how the confidence level is assigned.