UV Dissolution Testing: f2, Sink Conditions, Limits

Dissolution is one of the few release tests where the analytics are trivial and the decision is not. Absorbance to percent-released takes a line of code. Deciding whether a post-change batch is still the same product, whether the medium was ever in sink, and whether UV was the right detector at all is where the time goes and where audits find problems.

This walks the full chain: calibration, method validation, profile calculation, f1/f2 comparison, sink-condition check. Every number is output from a live CovaSyn tool call, named inline. The dataset is a constructed 10 mg immediate-release tablet in 900 mL, not a GMP batch - the arithmetic is the tools', the sample is illustrative.

Step 1: the calibration curve

Everything downstream inherits the calibration. Run standards across the expected range and fit before you touch a vessel.

Five standards from 2 to 20 ug/mL, through covauv_quantify:

An r-squared of 0.9999975 looks impressive and means almost nothing on its own; five points on a straight line will do that. The number that matters is whether the intercept is statistically distinguishable from zero, because a non-zero intercept is the fingerprint of a matrix or baseline problem that biases every low timepoint. That question is answered in the next step.

Step 2: validate the method before you trust a profile

covauv_validate runs linearity, accuracy and precision together against ICH Q2(R2) expectations. Same five standards, spiked recoveries at 80, 110 and 140 percent of working concentration, and two days of six replicates:

The useful line is intercept_significant: false. The intercept is 0.00129 AU against a standard error of 0.000806 AU, so it does not clear the significance threshold - evidence that the blank-corrected baseline is behaving. If that flag came back true, the fix is not a better regression; it is finding the excipient, filter or dissolved-gas artefact putting absorbance where there is no analyte.

Repeatability at 0.510 % RSD and intermediate precision at 0.573 % RSD sit close together. A large gap between them signals an analyst, day or instrument effect worth chasing before the method reaches QC.

Step 3: the dissolution profile

covauv_dissolution takes the timepoints, the raw absorbances, the calibration slope and intercept, the label claim and the vessel volume, and returns percent released plus the conventional Q points.

Two batches against the same reference profile, USP Apparatus 2 style sampling at 5, 10, 15, 20, 30, 45 and 60 minutes:

Batch A and Batch B percentages are covauv_dissolution output. Q at 30 minutes: 88.03 % for A, 91.05 % for B. Against a typical Q = 80 % at 30 minutes acceptance criterion, both batches pass the single-point test comfortably. That is the trap.

f1 and f2: where the two batches separate

The single Q point says both batches are fine. Profile comparison says they are not the same product.

f2 is the similarity factor from the FDA and EMA dissolution guidances; similar profiles give f2 between 50 and 100. f1 is the difference factor, accepted band 0 to 15. Both use only the timepoints up to and including the first point at which both profiles exceed 85 percent dissolved, so the plateau cannot inflate similarity.

Computed from the covauv_dissolution percent-released output using the published FDA formula (the tool returns the profile and Q points; the f1/f2 arithmetic here was applied to that output separately):

Batch A fails both criteria: f2 of 45.9 is below 50, f1 of 15.1 is just over the limit. The mechanism is visible in the table - A lags by 15 percentage points at 5 minutes and 16 at 10 minutes, then catches up. Slower early release with the same endpoint is the classic signature of a disintegration or wetting change, and exactly what a single 30-minute Q point cannot see. Batch B, at f2 = 86.5, is indistinguishable from reference.

Two cautions. f2 is a point estimate with no confidence interval, so it is fragile at the boundary; an f2 of 51 is not defensible without bootstrapping. And both guidances expect the coefficient of variation across the 12 units to be no more than 20 percent at early timepoints and 10 percent thereafter. f2 on mean profiles from noisy vessels is not evidence.

Sink conditions: check before, not after

Sink conditions mean the medium can dissolve at least three times the dose (many labs use a stricter factor). Outside sink, the rate becomes solubility-limited and the profile stops describing the formulation.

For a 10 mg dose in 900 mL, the fully dissolved concentration is 0.0111 mg/mL. covasolve_predict on amlodipine free base in water at 310.15 K returns 2.88 mg/mL (log S -2.15 mol/L), model confidence 0.96, inside the applicability domain, 95 % confidence interval -2.62 to -1.68 in log units - roughly 0.97 to 8.6 mg/mL. Even at the pessimistic bound the medium holds about 87 times the dose, so sink is not the constraint here.

The caveat: that is a predicted intrinsic aqueous solubility, not a measurement in your buffered medium with surfactant. Use it to decide whether a sink problem is plausible and whether a measurement is worth doing. Never put a predicted number in a protocol.

Where UV falls down and HPLC does not

UV is fast, cheap, non-destructive and easy to automate inline. It is also non-selective, and every one of its failure modes traces back to that.

• Excipient interference. Povidone, some colourants and several coating polymers absorb in the same 240 to 300 nm window as many APIs. The bias is positive and largest at early timepoints, when analyte signal is smallest. Blank-corrected placebo vessels are the minimum defence.
• Degradants. UV cannot separate a degradation product with a similar chromophore from the parent. A profile can look correct while the number is partly degradant.
• Turbidity. Scattering adds apparent absorbance. Filtration helps, but filter adsorption of the API is its own bias and must be validated.
• Bubbles. Dissolved gas on the cuvette or probe window produces sporadic high readings that look like real data.
• Low-dose products. Below roughly 1 to 2 ug/mL in medium, UV precision degrades and may not reach the sensitivity a 5-minute timepoint needs.

HPLC-UV costs cycle time and solvent and buys selectivity. Practical rule: use UV when the placebo is spectrally clean, the product is stable, and the dose gives a comfortable working concentration. Move to HPLC for absorbing excipients, stress or stability samples, related substances that must be resolved from the parent, or any case where a regulator will ask you to prove selectivity rather than assert it.

Honest limits

What this analysis does not tell you:

• Nothing about in vivo behaviour. f2 similarity is a quality-control comparability tool. It is not a bioequivalence claim, and only supports a biowaiver inside the specific conditions of the relevant BCS guidance.
• Nothing about the 12 units. Everything above is mean-profile arithmetic. Unit-to-unit variability, and the stage 1 to stage 3 USP acceptance tables, are a separate analysis.
• Nothing about selectivity. covauv_validate covers linearity, accuracy and precision. It does not prove the absorbance came from your API rather than an excipient. That needs a placebo study or an orthogonal method.
• Nothing about the medium. The tool takes the volume you give it. Evaporation, sampling replacement volume and degassing state are assumptions you carry in.
• The solubility figure is a prediction. 2.88 mg/mL with an interval spanning nearly an order of magnitude is a triage number, not a specification input.

Frequently asked questions

Related reading

• How accurate is solubility prediction, really
• pH-solubility profiles and salt selection
• Cpk vs Ppk: in control is not the same as capable
• Mass balance in forced degradation studies

Run your own absorbance series through covauv_dissolution and covauv_validate on the free tier and see what the profile says beyond the Q point.