Every chromatographic bioanalytical method begins with a curve. A blank, a zero, six or more non-zero calibrators across the validation range, with the lowest at the lower limit of quantification. Plot concentration on the x-axis, response on the y-axis, fit the simplest regression that tells the truth about the relationship. This is calibration in three sentences. The discipline lives in what comes next.
The acceptance criteria are old. Six or more non-zero calibrator levels per validation run. Non-zero calibrators within ±15% of nominal concentration, except at the LLOQ where they are allowed ±20%. Seventy-five percent of calibrators — with a minimum of six non-zero levels — meeting these limits in each run. The same arithmetic has held since the 2001 Crystal City white paper, through the 2018 FDA guidance, into ICH M10. What changes between regulators is not the numbers; it is the framing of the numbers.
/ 01The two regimes: chromatographic and ligand-binding.
Chromatographic methods (CC) and ligand-binding assays (LBA) have different curves and different acceptance criteria. The bioanalytical scientist who treats them as one regime will fail validation on the LBA side. Reading the regulatory text carefully:
Chromatographic methods · CC.
Calibration curve elements.
- A blank (no analyte, no internal standard), a zero calibrator (blank plus IS), and at least six non-zero calibrator levels covering the quantitation range, including LLOQ in every run.
- All blanks and calibrators should be in the same matrix as the study samples.
- The concentration-response relationship should be fit with the simplest regression model that adequately describes the data.
The ±15% / ±20% rule.
- Non-zero calibrators should be ±15% of nominal (theoretical) concentrations, except at LLOQ where the calibrator should be ±20% of the nominal concentrations in each validation run.
- 75% and a minimum of six non-zero calibrator levels should meet the above criteria in each validation run.
Ligand-binding assays · LBA.
LBAs are not linear in the same shape; they fit four- or five-parameter logistic models, and they tolerate looser bounds at the extremes because the underlying biology does. The acceptance criteria reflect this honestly.
Calibration curve elements.
- A blank and at least six non-zero calibrator levels covering the quantitation range, including LLOQ per validation run.
- Calibration curves are usually run in duplicate.
- Anchor calibrators may be used as additional points beyond the validated range to improve curve fit, but they are not part of the validated quantitation range.
- The concentration-response relationship is usually fit with a four- or five-parameter logistic model. Other models may be acceptable with justification.
The ±20% / ±25% rule.
- Non-zero calibrators should be ±20% of nominal concentrations, except at LLOQ and ULOQ where the calibrator should be ±25% of nominal concentrations in each validation run.
- 75% and a minimum of six non-zero calibrator levels should meet the above criteria in each validation run.
- Anchor points are not included in the curve fit assessment against acceptance criteria.
/ 02The side-by-side.
Read together, the two regimes show their character. Chromatographic methods are tighter at the boundaries because the analyte's behaviour through the chromatography is, in principle, deterministic. LBAs are looser at the boundaries because immunoassay binding curves are inherently sigmoidal, and the extremes are where the binding chemistry lives at the edge of what can be quantified. The 75% rule holds across both. The minimum of six calibrators holds across both. The matrix-must-match rule holds across both.
| Element | Chromatographic (CC) | Ligand-binding (LBA) |
|---|---|---|
| Blank | No analyte, no IS | Required |
| Zero calibrator | Blank + IS | Optional |
| Non-zero calibrators | ≥6 levels, single | ≥6 levels, in duplicate |
| Anchor calibrators | Not used | Permitted, outside range |
| Regression | Simplest linear/quadratic | 4PL or 5PL logistic |
| Mid-range tolerance | ±15% | ±20% |
| LLOQ tolerance | ±20% | ±25% |
| ULOQ tolerance | ±15% | ±25% |
| Per-run pass rate | ≥75% & ≥6 levels | ≥75% & ≥6 levels |
/ 03The in-study analysis view.
Validation is the controlled-environment performance review. In-study analysis is the field deployment. The acceptance criteria for in-study runs are the same numbers, but the analyst is now answering a different question: did this run meet the criteria, and if not, what is the disposition?
Same numbers, different question.
- A blank, a zero, and at least six non-zero calibrator levels (in duplicate for LBAs) covering the expected range, including LLOQ per analytical run.
- All blanks and calibrators in the same matrix as study samples.
- The in-study analysis should use the same regression model as used in validation.
- CC: non-zero calibrators ±15%, except at LLOQ ±20% of nominal concentrations.
- LBA: non-zero calibrators ±20%, except at LLOQ and ULOQ ±25% of nominal concentrations.
- CC and LBA: 75% and a minimum of six non-zero calibrator levels should meet the criteria in each run.
/ 04What the curve hides.
The curve passes acceptance criteria. The run is accepted. The analyst stamps the lab notebook. Three months later, an inspector pulls the chromatograms and asks why the LLOQ calibrator was 19.5% above nominal in five consecutive validation runs. The acceptance criterion was met; the trend was not noticed. This is the calibration curve's most expensive failure mode — not failing acceptance, but passing it while drifting toward failure.
Three patterns the experienced analyst watches for:
- Boundary creep. The LLOQ calibrator hovering at +18%, +19%, +20% across consecutive runs is approaching the failure boundary; the validation document is silent on this until ±20% is exceeded. The trend matters more than the number.
- Asymmetric residuals. A linear regression where the residuals are systematically positive at low concentration and negative at high concentration is not actually linear. The simplest-model rule does not mean the simplest model that passes; it means the simplest model that tells the truth.
- Anchor-point dependency in LBAs. When the LBA curve fit only converges with anchor points, the validated range may be effectively narrower than the validation document claims. The validated range is what is between the lowest and highest accepted calibrators that satisfy the per-run criteria, not the chemistry the assay can theoretically measure.
/ 05The regulator is reading the curve too.
An FDA inspector pulling the validation report does not read the regression equation first. They read the failed-run log, the ISR table, the calibrator failures across runs, and the disposition of any out-of-trend points. Then they read the regression. The curve is treated as evidence of the lab's discipline, not as a numerical artefact. ICH M10, since January 2025, harmonises this expectation across regions: the same calibration curve must satisfy FDA, EMA, ANVISA, WHO PQ, and PMDA without regional reinterpretation.
The calibration curve hides nothing in the end. It tells the autobiography of the method to anyone who reads it carefully. The discipline lives in reading carefully.