Ask five people in a discovery group what a MIC is and you will get the same one line definition and five different working assumptions underneath it. The definition is easy. The conventions around it are where the disagreements live.
The one line, and what it leaves out
The minimum inhibitory concentration is the lowest concentration of an antibacterial that prevents visible growth of an organism after a defined incubation period. Every clause in that sentence is doing work:
- Lowest concentration of the ones tested. It is the first clear well in a dilution series, not a continuously measured threshold.
- Prevents visible growth. Visible. It is an inhibition read, not a killing read. A compound can have a low MIC and kill nothing, which is the difference between a bacteriostatic and a bactericidal agent, and why MBC exists as a separate measurement.
- An organism. Not a species in the abstract. One isolate, with its own genotype.
- A defined incubation. Usually 16 to 20 hours. Read it early and you overestimate potency, read it late and slow growing survivors catch up.
Why the numbers are always powers of two
The plate is made by doubling dilution: 0.03, 0.06, 0.12, 0.25, 0.5, 1, 2, 4, 8, 16, 32, 64. A MIC of 4 means growth at 2 and none at 4. The true value is somewhere between those, and the method cannot say where.
This has a practical consequence that gets forgotten in spreadsheets: the resolution of the measurement is one doubling dilution. Two competent labs testing the same compound and the same strain landing one step apart is considered agreement. Arithmetic means across doubling series are misleading, and geometric means are the conventional choice for a reason.
It also explains why a predicted MIC should be reported as a band. A model that outputs 0.73 µg/mL is claiming resolution the underlying data never had. The MIC reference page goes into the method in more detail.
MIC, MIC50, MIC90 and breakpoint
Four numbers, routinely conflated in conversation:
| Number | Applies to | Typical use |
|---|---|---|
| MIC | One compound, one isolate | A bench result |
| MIC50 | A collection of isolates | Where the middle of the distribution sits |
| MIC90 | A collection of isolates | The number quoted in surveillance papers |
| Breakpoint | A species and an agent | Clinical interpretation, set by CLSI or EUCAST |
A breakpoint is a decision, not a measurement. It takes the wild type MIC distribution, achievable drug exposure and clinical outcome data into account, which is why the same MIC value can be susceptible against one species and resistant against another, and why breakpoints are revised.
What a MIC does not tell you
- Whether the compound kills. That is the MBC, and the ratio between them says something about the mechanism.
- How fast. Time kill curves answer that, and two compounds with identical MICs can behave very differently over eight hours.
- Whether it will work in a patient. That needs exposure, protein binding, site penetration and the PK and PD relationship.
- Why it failed. A high number is a fact without a cause. Whether it was efflux, porin loss or target mutation is a separate investigation, and the reason a read-out that names a mechanism is worth more than one that does not. The antibiotic resistance page sets out the four candidate causes.
Reading a MIC in a paper
Three habits worth keeping when pulling numbers out of the literature:
- Check the units and the relation. A value with a greater than sign is a censored observation, not a value.
- Check the medium and the inoculum. Cation adjusted Mueller Hinton broth is standard for a reason, and results in other media are not comparable without care. Serum content matters enormously for peptides.
- Check the strain. A reference strain and a clinical isolate of the same species are different organisms for this purpose.
All three of these are why assembling a picture from ChEMBL is an afternoon of work rather than a query, and why the roll-up is the part worth automating.
The practical summary
A MIC is a coarse, well standardised, extremely useful number. Treat it as a band, keep the method with it, and never let it travel alone into a decision without the mechanism that produced it. The next note covers how it is measured on the bench, including where the error actually comes from.