A minimum inhibitory concentration is not a property of a peptide. It is a property of a peptide measured in one medium, against one inoculum of one organism, in one kind of plastic, read at one time by one method. The published definition says so directly, and most of the trouble in this field comes from reading the number as though it did not.

What the number is defined to be

Broth and agar dilution methods aim to determine the lowest concentration of the assayed antimicrobial agent that, under defined test conditions, inhibits the visible growth of the bacterium being investigated [1]. Three clauses in that sentence do real work. Lowest concentration means the lowest of the ones you happened to put on the plate. Under defined test conditions means the conditions are part of the result, not a footnote to it. Visible growth means the endpoint is turbidity, so a MIC is an inhibitory measurement and not a killing measurement; a well that looks clear may still contain viable cells.

Broth dilution is often determined in 96-well microtiter plate format: bacteria are inoculated into liquid growth medium containing a range of agent concentrations, and growth is assessed after a defined incubation, typically 16 to 20 h [1]. The protocol as published applies only to aerobic bacteria and can be completed in three days [1]. Killing, as distinct from inhibition, requires a separate readout — subculturing or plating the clear wells, as in the drop plate assay used to obtain bactericidal and fungicidal endpoints alongside microdilution MICs in one peptide study [5].

Two-fold resolution is the coarsest thing about the assay

The concentration series is conventionally two-fold, so consecutive wells differ by a factor of two and the answer can only land on that ladder. The consequence is directional and worth stating carefully: the reported MIC is the lowest rung that inhibited, so it sits at or above the concentration where inhibition actually begins, never below it, and it can exceed that point by up to almost a factor of two. If inhibition truly begins near 6 µg/mL and the ladder runs 4, 8, 16, then 4 fails, 8 is clear, and 8 is reported — about 33% high, taking 6 as the denominator.

Replicate variability sits on top of that quantization. In the BSAC Resistance Surveillance Programmes, repeated annual testing of a panel of control isolates at the central laboratory by the BSAC agar dilution method found that at least 90% of replicate MIC measurements agreed within plus or minus one doubling dilution, and at least 98% within plus or minus two [2]. That figure comes from agar dilution on control strains with conventional antibiotics, not from peptides in broth, and no comparable repeatability figure for antimicrobial peptides is quoted here because none was retrieved. But it sets a floor for expectations: even for well-behaved compounds on reference strains under a maintained method, one doubling dilution is inside the noise. A two-fold difference between two peptides in one table is therefore not, on its own, evidence of a difference.

The medium is a reagent, not a background

Cation-adjusted Mueller-Hinton broth is called cation-adjusted because the calcium and magnesium content was specified deliberately, and the specification has been revised. A multilaboratory study reassessing that adjustment for testing Pseudomonas aeruginosa against amikacin, gentamicin, isepamicin, netilmicin and tobramycin concluded that, to achieve parity with agar dilution tests, cation-adjusted broth should contain 20 to 25 mg of calcium and 10 to 12.5 mg of magnesium per liter rather than the 50 and 25 mg per liter recommended previously, and that quality-control MIC limits should be lowered by at least one doubling-dilution interval as a result [3]. That scope matters: one organism, five aminoglycosides. But it shows that a change in the cation specification of a routine broth was worth at least a full doubling dilution in the control limits.

For cationic peptides, divalent cations are not a quality-control detail. One study compared two symbiosis-specific cationic plant peptides, NCR247 and NCR335, against polymyxin B and streptomycin on three model microbes — Escherichia coli, Bacillus subtilis and Saccharomyces cerevisiae — in three general test media (MHB, MHBII and a low-salt medium), using a validated resazurin microdilution assay [5]. The authors report that the antimicrobial efficiency of these peptides is clearly limited by divalent cations, that NCR247 had a generally high sensitivity for calcium and magnesium in the medium, and that NCR335 was by contrast robust, with efficiency values comparable to polymyxin B [5]. Two peptides of the same family, one assay, sharply different sensitivity to the cation content of the medium.

Monovalent salt behaves the same way, and the pairing matters as much as the peptide. Recombinant human intestinal defensin 5 was assayed for microbicidal activity in the presence of 0 to 150 mM NaCl: activity was reduced in the presence of salt, and at 100 mM NaCl activity against Salmonella typhimurium was abolished, yet at all salt concentrations tested the peptide remained bactericidal to Listeria monocytogenes [6]. Salt sensitivity in that report is a property of a peptide-and-target combination, not of the peptide alone.

How hard it is to settle any of this is visible in the polymyxin case. A joint working group of the Clinical and Laboratory Standards Institute and the European Committee on Antimicrobial Susceptibility Testing examined, among other issues, the effects of divalent cations, binding to laboratory materials, and the addition of polysorbate 80, and recommended broth microdilution without polysorbate 80 as the reference method; other methods, including agar dilution, disk diffusion and gradient diffusion, were reported to show unacceptably high levels of very major errors against that reference, with most data available for colistin and less for polymyxin B [4]. That is the nearest thing to a settled reference method for a cationic peptide antibiotic, it required an international working group, and it covers polymyxins rather than peptides in general.

Inoculum density changes the answer

A systematic characterization measured MICs for 13 peptides and peptidomimetics, including all-D enantiomers, across more than seven orders of magnitude of inoculated cell density [7]. In most cases the inoculum effect was significant for cell densities above the standard inoculum of 5 × 10⁵ cells/mL, while for lower densities the active concentrations remained essentially constant, with values in the micromolar range [7]. For membrane-active peptides the authors rationalize this with a simple model that accounts for peptide-cell association and hypothesizes that a threshold number of cell-bound peptide molecules is required to cause killing — their hypothesis, offered as such [7].

Their conclusion is aimed squarely at screening practice: the observed effect questions the clinical utility of activity and selectivity determinations performed at a fixed, standardized cell density, and they suggest that routine evaluation of the inoculum dependence of activity should be considered [7]. Note what that implies for a table of MICs measured at one density. It is internally consistent and externally fragile.

The concentration you think you added

Two quiet sources of error sit upstream of every number above, and neither is a finding with a primary source behind it. The first is adsorption: cationic amphipathic peptides bind surfaces, so a fraction of the dose can end up on tips, tubes and well walls rather than in solution. Low-binding polypropylene labware and carrier proteins are the usual mitigations, and both are bench convention rather than validated corrections. That the concern is real enough to be examined by a standards body is the one part with a citation — binding to laboratory materials was among the issues the polymyxin working group considered [4].

The second is quantification. A weighed mass of lyophilized synthetic peptide includes counterion and residual water, so weighing overstates the peptide actually delivered unless the net peptide content is known or the stock is quantified spectrophotometrically. Both practices are stated here uncited.

Units compound the problem. Mass concentration and molar concentration rank peptides differently whenever their masses differ. A 2,000 Da peptide at 8 µg/mL is 4 µM; 8 µg/mL is 8 mg/L, and 8 ÷ 2,000 gives 0.004 mmol/L. A 4,000 Da peptide at the same 8 µg/mL is 2 µM. Two peptides quoted at an identical MIC in µg/mL therefore differ two-fold in molar potency when one is twice the mass of the other, with the heavier peptide the more potent on a molar basis.

Hemolysis: what the assay reports and what changes it

The hemolysis assay incubates erythrocytes with the compound, then reads released hemoglobin in the supernatant, expressed as a percentage placed between two controls: the sample signal minus a buffer blank, divided by a full-lysis positive control minus the same blank. That formula is arithmetic rather than a finding, but it explains the sensitivities, because everything that moves the positive control moves every percentage in the column.

An optimization study makes the size of those sensitivities concrete. Protocols in the literature vary substantially; for the compounds tested there, the hemolytic effect varied up to fourfold depending on the species of the blood source, and different types of detergent used to generate the 100% hemolysis positive control produced up to 2.7-fold differences in the calculated hemolysis ratios [8]. The same work reports a substantial increase in the number of hemolyzed erythrocytes with increasing erythrocyte concentration and with prolonged incubation time, which in turn affects the calculated ratios [8].

Anticoagulant choice belongs on the same list. Starting from an observation of high hemolytic activity of LL-37 against human red blood cells collected in EDTA, one group developed a defibrinated human blood assay and reported significant differences between the sensitivity of defibrinated and EDTA-treated red blood cells [9]. Those authors frame the purpose of standardizing the assay as obtaining a more robust calculation of the therapeutic index of the compounds they are developing, a measure they describe as critical in pre-clinical development [9].

Finally, the metric itself varies. Some papers report an HC50, the concentration producing 50% hemolysis, which requires a concentration series and a fitted curve. Others report a single percentage at a single concentration. A single-point percentage cannot be converted into an HC50 without the curve it came from, so the two cannot be placed on the same axis at all.

The selectivity index and why it does not travel

The selectivity index is usually the hemolytic endpoint divided by the antibacterial one, most often HC50 over MIC, with larger values taken to mean greater discrimination between microbial and erythrocyte membranes. Used inside one study it can be informative. In a series of twelve N-methylated gramicidin S analogues, HC50 increased fivefold relative to gramicidin S when the internal hydrogen-bond-forming leucine was methylated, and analogues carrying N-methylleucine in the beta-strand region together with N-methylalanine in the beta-turn region showed a fourfold increase in selectivity index over the parent [10]. That comparison holds because every number came from one laboratory, one pair of assays and one parent compound.

Across papers it collapses, for reasons that are arithmetic before they are biological.

  • It is a ratio, so errors multiply. If a MIC lands one dilution step high, the index halves; if the MIC is two-fold high and the HC50 two-fold low together, the index is off by a factor of four. Neither displacement need be a blunder. Replicate MIC measurements agreed within plus or minus one doubling dilution at least 90% of the time on that BSAC panel of control isolates, tested by agar dilution with conventional antibiotics [2], and the hemolytic endpoint, which is fitted from a concentration series rather than read off a ladder, carries assay-dependent variability of its own [8, 9].
  • The numerator is assay-defined. Blood species, anticoagulant, erythrocyte concentration, incubation time and the detergent chosen for the 100% control each move it [8, 9].
  • The denominator is medium-defined. Divalent cation content, monovalent salt, and inoculum density each move it [3, 5, 6, 7], and how much they move it depends on which peptide and which organism: NCR247 was highly sensitive to calcium and magnesium in the medium while NCR335 was robust across the same three media [5], and defensin 5 lost activity against Salmonella typhimurium at 100 mM NaCl while remaining bactericidal to Listeria monocytogenes across the whole 0 to 150 mM range tested [6].
  • It is organism-specific by construction. A peptide with different MICs against different species has a different index against each. A single headline value has necessarily selected one organism, and comparing two papers means checking that they selected the same one.
  • Units must match. The ratio is dimensionless only when numerator and denominator are in the same units. An HC50 in µM over a MIC in µg/mL is not a selectivity index.

That conditions belong with the numbers is now built into curation practice: a manually curated peptide database records chemical structure, target species, and antimicrobial, hemolytic and cytotoxic activities together with the experimental conditions at which those activities were estimated [11].

What to record, and what remains convention

The recording list follows from everything above: medium and its cation specification; salt concentration; inoculum preparation and the verified cell density; plate and tip material; how the peptide stock was quantified; incubation time and temperature; whether the endpoint was visual turbidity, optical density or a dye such as resazurin; for hemolysis, the blood species, anticoagulant, cell concentration, buffer, incubation and the agent used for the 100% control; and units for both halves of any ratio.

Several common practices have no evidential basis stated here and should be labeled as convention. Reporting the mode or median of replicate MICs rather than an arithmetic mean, because a mean across a log-scaled ladder is not on the ladder. Cutoffs such as a selectivity index above 10, or calling a peptide non-hemolytic below 10% lysis at some concentration: no source retrieved for this guide establishes either threshold, and they should be read as house rules.

What is genuinely unsettled is larger. There is no consensus reference method for antimicrobial peptides as a class; the one cationic peptide family that has a reference method got there through a joint international working group and the recommendation covers polymyxins [4]. Whether inoculum dependence should be characterized routinely is a suggestion from one systematic study, not a standard [7]. And nothing in a broth MIC or a hemolysis percentage speaks to behavior in any more complex system; these assays measure what they measure, under conditions you chose, and the conditions travel with the number or the number means nothing.