pH-Solubility Profiles and Salt Selection

You have a few milligrams of an ionizable API and a question that decides the next six months: do you formulate the free acid or make a salt? The honest answer needs a pH-solubility profile, and a proper one costs you material and shaker time you may not have yet. A calculated profile will not replace that experiment, but it will tell you where to put your buffers, whether a salt is even worth screening, and which pH points are the ones you actually need to measure.

This article shows how to build that profile, what the numbers mean for salt selection, and - the part most tools skip - exactly where the calculation stops being physically true.

The two numbers that matter: intrinsic and apparent solubility

For a weak acid, only the neutral (unionized) species is in equilibrium with the solid. Its concentration is the intrinsic solubility, S0, and it is constant regardless of pH. What you measure in a buffer is the apparent solubility: S0 plus whatever ionized species has been generated by the pH.

For a monoprotic acid, the Henderson-Hasselbalch form is:

S(pH) = S0 * (1 + 10^(pH - pKa))

Below the pKa the correction term is negligible and the curve is flat. Above the pKa the profile rises with a slope of 1 log unit per pH unit. Bases behave the mirror image (flat above the pKa, rising below it), and ampholytes give the classic U-shape with a minimum near the isoelectric point.

Everything downstream - salt selection, biorelevant dissolution assumptions, buffer choice for a stability study - falls out of where your pKa sits relative to the pH you care about.

A worked example: ibuprofen, real CovaSolv output

Ibuprofen is a monoprotic weak acid, literature pKa 4.9. Two tools were used, both on a CPU, both fast enough to run during a meeting.

That CI is roughly a factor of 2.6 wide in concentration terms. Carry it forward: every pH-corrected number below inherits it.

covasolve_ph_predict at pH 7.4 returned log_s_at_ph = -1.4432, s_mol_l_at_ph = 0.0360, correction_log_units = 2.5014. At MW 206.3 that is about 7.4 mg/mL apparent, against 0.023 mg/mL intrinsic.

Both tools report pka_method in the return. In the run above it reads user_supplied, because the pKa was passed in. That field is the single most important thing to check in the output, for the reason in the next section.

The pKa dominates the answer, so source it deliberately

Run covachem_pka on the same molecule and it predicts an acidic pKa of 4.0 for the carboxylic acid group (dominant species at pH 7.4: anion). Literature says 4.9.

That 0.9 unit gap is not cosmetic. In the region above the pKa the profile is a straight line with slope 1, so a 0.9 unit error in pKa shifts the entire curve horizontally by 0.9 units and therefore misstates apparent solubility by about 0.9 log units - roughly 8x. It also moves your calculated pHmax by 0.9 units.

Practical rule: use a measured or well-attested literature pKa when you have one, and treat a structure-predicted pKa as a triage input only. Rules-based pKa predictors are reliable at classifying *whether* a group ionizes and poor at the second decimal.

From profile to salt decision: pHmax

A salt is not a molecule with higher solubility. It is a different solid with a different lattice, and it only stays the stable solid form over part of the pH range. The crossover is pHmax.

For a weak acid with a counterion:

pHmax = pKa + log(S_salt / S0)

Below pHmax the free acid is the stable solid. Above it, the salt is. Three things follow:

• Your ceiling is the salt's own solubility, not the Henderson-Hasselbalch line. Above pHmax the calculated profile keeps climbing at slope 1 forever. Reality plateaus, because the salt now precipitates and pins the solution concentration at the salt solubility product.
• A salt buys you nothing if pHmax sits outside your route of administration's pH window. For a weak acid with pKa 4.9, gastric pH 1 to 2 is deep in the free-acid region, so an oral salt will convert to the free acid in the stomach. Whether that matters depends on whether the free acid comes down as a fine, fast-redissolving precipitate or as a brick.
• The bigger the gap between pKa and the target pH, the more a salt can theoretically deliver. The calculated 317x gain at pH 7.4 in the table above is the ionization ceiling. That is the number that tells you salt screening is worth material.

The classic screening heuristic still applies: aim for a counterion whose pKa differs from the API's by at least about 3 units (the "rule of three") so the proton transfer is thermodynamically committed and you get a salt rather than a co-crystal or a disproportionating mixture. Use the profile to shortlist; use a wet screen to decide.

What this calculation does not tell you

State these next to the number when you put it in a report.

• It is Henderson-Hasselbalch on top of an ML prediction. Two error sources stack: the intrinsic solubility model (95% CI -4.153 to -3.736 log units here) and the pKa you supplied.
• No salt solubility limit. There is no Ksp in this model, so it over-extrapolates above pHmax. Above roughly pH 7 to 8 for this compound, treat the curve as an upper bound, not a prediction.
• No common-ion or counterion effects. A chloride salt in a chloride buffer behaves differently from the same salt in phosphate. The model does not know your buffer species.
• No ionic strength or activity corrections. Thermodynamic and concentration-based constants are used interchangeably here. At high ionic strength that assumption degrades.
• No solid-state information. Polymorphs, hydrates, amorphous content and salt disproportionation all move real solubility and none are inputs.
• No self-association or micellisation. Ibuprofen and many surfactant-like acids aggregate above their CMC, which bends the real curve away from slope 1.
• No kinetics. This is equilibrium solubility. Dissolution rate, surface pH in the diffusion layer and supersaturation lifetime are separate questions.
• Temperature. The run was at 298.15 K. pKa is itself temperature-dependent and that dependence is not modelled.

Used inside those bounds, the profile is decision-support: it tells you where to put your five experimental pH points instead of twelve.

A practical workflow

1. Get the pKa and its provenance. Measured, literature, or predicted - write it down. 2. Predict intrinsic solubility with covasolve_predict in water and record the 95% CI and the applicability-domain flag. 3. Generate the profile with covasolve_ph_curve over the pH range that matters for your route. 4. Read off the ionization gain at your target pH with covasolve_ph_predict (correction_log_units). 5. If that gain is large, estimate pHmax for candidate counterions and check it sits inside your usable pH window. 6. Confirm with a shake-flask measurement at three or four pH points, chosen from the calculated curve, with solid-form characterisation of the residue.

Step 6 is not optional. The calculation exists to make step 6 cheaper.

Frequently asked questions

Related reading

• ICH Q8 design space, NOR and PAR explained
• Forced degradation study design under ICH Q1A(R2)
• Antisolvent versus cooling crystallization: choosing a route

Run the profile on your own compound on the CovaSyn free tier and see whether a salt screen is worth the material.