Why amorphous solid dispersions crystallize

The argument in brief

1. Crystallization of an amorphous drug depends on a driving force and on molecular mobility, and it proceeds by two dependent steps, nucleation and growth.
2. No bulk glass or thermodynamic property of the neat drug ranked crystallization across drugs.
3. A few percent of polymer barely lowers the driving force and often leaves Tg unchanged, yet slows crystallization strongly. The authors therefore read the inhibition as kinetic, but its proposed routes were not measured.
4. A polymer acts on nucleation and on growth separately, so a polymer ranking holds only for the step measured.
5. The polymer protects while it stays mixed with the drug. A drug-rich amorphous phase decides crystallization, whether it comes from casting, from sorbed water or from the drug loading.
6. Water acts by two routes: it plasticizes, which raises mobility, and it can split the dispersion, which removes the polymer from the drug.
7. Free surfaces, crystalline carriers and the process add routes that can outweigh the formulation.
8. A stability test ranks only inside its window, and a sample counts as amorphous only for the method that looked.

Driving force, mobility and the two steps

The driving force is the free-energy difference between the supercooled liquid and the crystal. It is zero at Tm and grows on cooling. Molecular mobility is how fast the molecules of the amorphous solid move and rearrange; the papers read it from Tg, enthalpy relaxation times, melt viscosity and model relaxation times.

Nucleation and growth each have their own rate and temperature dependence. In classical theory the barrier to a critical nucleus grows with the cube of the interfacial free energy and falls with the square of the driving force, W^* = 16\pi\gamma^3 / (3\,\Delta G_v^2). Growth lies between two opposing factors: near Tm molecules are mobile but the driving force is small, and on cooling the driving force rises while mobility falls.

Nifedipine and felodipine show why both quantities are needed. Their Tg values are nearly the same (onset 42.3 and 43.1 °C), and so are their relaxation times at 23.5 °C (τ 6.3 and 4.2 h). Nifedipine still nucleated about 3.5–6 times faster at every PVP level up to 25 wt%, in line with its larger driving force.

Evidence thermodynamic driving force · molecular mobility · nucleation · Solid state stability F5 · Physical stability F2 · baird2012 C32; vaneerdenbrugh2010a E3; trasi2014a C22; marsac2006a R4.c, R4.m, R4.n, R2, R16

What the neat drug can and cannot predict

The class from a heat–cool–heat cycle orders drugs within one test. In dry storage indoprofen (class I) was fully birefringent after 1 day, while clotrimazole and felodipine (class III) were only at the edges after 3 months. A carrier, the process or the polymer can override this order: lumefantrine, a class III drug, crystallized within 1 week at 50% loading with PVPVA.

The bulk properties that might explain the class failed to rank it. Mean Tg/Tm was 0.69 for class I and 0.73 for class III, a difference that was not significant. Melt viscosity at Tm separated only the extremes, around 0.01 Pa·s. Growth did follow Tg where an additive lowered the Tg of the mixture strongly: a fall of 5–12 °C with 5% PHB went with growth up to ≈10× faster.

Evidence Solid state stability F1 · Solid state stability F2 · baird2010 R23, R25, R14, C37; hiew2022a R6; baird2012 R1, C22; sato2016 R7, R4, R11

A polymer acts kinetically, and the route is not settled

The first few percent of polymer has the largest effect. In dry felodipine films at 22 °C the nucleation rate fell from about 1.3 × 10⁵ to about 2–3 × 10³ m⁻³ s⁻¹ with 3% w/w PVP, HPMCAS or HPMC, and the three polymers gave the same rate within experimental error.

This is not a change in driving force. Flory–Huggins estimates put the drug activity in PVP at about 0.96 at a polymer volume fraction of 0.1 and 0.84 at 0.2, and 3 wt% PVP K-12 moved the felodipine melting point by only 0.8 °C.

It is not simply a change in Tg either. HPMCAS and HPMC left the Tg of felodipine at about 42–45 °C up to 25%, while PVP raised it to about 55 °C, yet all three slowed dry nucleation equally. PVP at 0.5–4.5 wt% left Tg at 43.4–44.1 °C against 43.3 °C and still slowed melt growth up to 35-fold.

The papers therefore attribute the inhibition to kinetics, by three proposed routes: a raised barrier at the nucleus, perhaps from a polymer-rich layer; a higher nucleation activation energy (felodipine below Tg 145 → 201 kJ/mol with 3% PVP); and hydrogen bonds that lower drug mobility. None of these routes was measured directly.

Evidence kinetic stabilization · Physical stability F1 · Physical stability F4 · Physical stability F5 · konno2006 R4, C13, C14, R6, C26, C39, C40; marsac2006a R19, C47, R3, C48; kestur2010b R1, R13, C35, C42

Nucleation, growth and the contact with the drug

A polymer can slow growth without lowering nucleation. In acetaminophen, PAA cut growth at 80 °C from about 5.3 to 0.43 µm/s, yet left 26 nuclei per cooling–reheating cycle against 10.8 for the drug alone. The authors conclude that the effect of a polymer on nucleation is not predictable from simple properties, unlike growth.

Growth is slowed by a factor on the neat rate. Seeded melt growth of 31 neat drugs spans 216 µm/s to 0.02 µm/s, while 10 mol% PVP lowered growth of five donor drugs by a similar 3- to 14-fold. A fast grower therefore stays fast and needs more polymer. The factor is not universal: at 120 °C HPMC slowed indomethacin α ≈2900× and γ ≈2.9×, although the 2900 rests on single, very slow points.

The best polymer has a group complementary to a free group of the drug, so the ranking reverses from drug to drug. At 70 °C felodipine, an NH donor, was slowed 10.0× by 3% of the acceptor polymer PVP and 1.2× by PVAc; bifonazole, an acceptor without a donor, was slowed most by the donor PAA. With chiral HPMCAS the two enantiomers of one compound grew at different rates (1.9–2.5×), while achiral PVP slowed both alike. This is the authors' hypothesis, and the size of an infrared shift did not always follow the outcome.

Evidence Solid state stability F4 · Solid state stability F5 · Solid state stability F6 · trasi2012a R6, R12, C48, C52; trasi2014a R1.a, R1.ae, R6.m, R6.c, C11; tian2017 R12.i; kestur2010a R6; kestur2011 R7; sato2015 R7, R8, C10

Phase state: the polymer protects only where it is mixed

Above the miscibility limit a drug-rich phase forms, and with less polymer in that phase crystallization "would be anticipated" to occur more readily. Indomethacin dispersions above about 50% drug with HPMC and above 30% with HPMCAS crystallized in 2–15 days, while those below stayed amorphous for 13–14 months.

Once a dispersion has separated, the amount of polymer stops mattering. Felodipine–PVP powders with 10, 25 and 50% PVP at 93% RH crystallized at about the same initial rate: 6.4, 5.8 and 5.5 g/day.

A single DSC glass transition does not exclude a drug-rich phase. Itraconazole–HPMC films showed round domains of 0.4–1.4 µm where DSC of the powders gave one Tg, and domains below about 15–30 nm may give one Tg.

Miscibility is hard to predict in advance. The Flory–Huggins parameter χ of indomethacin–PVP K12 from one DSC peak was −6.16, −1.94 or −0.82 depending on the point read. Estimates from different routes agree in sign and not in size, and none predicted whether a dispersion stayed amorphous.

Evidence drug–polymer miscibility · Solid state stability F7 · Solid state stability F8 · Solid state stability F3 · tian2016 C4, R17; rumondor2009b R21, C20; purohit2015a R1, R10, R7; rumondor2009a C6; marsac2009 R6

Water acts by two routes

Route 1: plasticization. Sorbed water lowers Tg, taken in these papers as about 10 °C per 1% water, and raises mobility. At 75% RH and 25 °C pure PVP K29/32 held about 33 g water per 100 g, HPMC about 10–11 and HPMCAS about 6. The dry Tg advantage of 25% PVP was gone at 75–76% RH. Amorphous felodipine films nucleated about 20 times faster at 75% than at 0% RH.

The polymer still helped: at every RH from 0 to 75%, felodipine films with 3–25% polymer nucleated more slowly than the drug alone, although they held more water. Growth of neat felodipine rose only about twofold from 13 to 80% RH and then jumped, with no break in sorption or Tg. The cause is "currently unknown".

Route 2: phase separation. Sorbed water interacts well with PVP and badly with a hydrophobic drug, so a lower free energy is reached by separating. The authors state that this event is thermodynamically driven, with mobility setting only its rate. Pimozide–PVP 50:50 had one Tg of 94 °C when dry and two (−21 and 44 °C) after 94% RH. Separation stayed after a week of drying. Held at 125 °C, a separated felodipine–PVP film regained drug–PVP bonds only slowly over 1500 min.

Which drugs separate followed their affinity for water and not log P or melting data: the drug–water χ was 2.84–3.58 for the four that separated and 2.50–2.71 for the two that did not. Donor type and the PVP carbonyl shift divide the drugs into nearly the same groups, so the cause is not isolated.

Hence the best polymer in humid storage depends on the drug. After about 490 days at 75–93% RH, felodipine was 51–82% crystalline with PVP against 9–28% with HPMCAS. Naringenin and quercetin went the other way: more than 200 to more than 450 days with E100 or PVP, against 1–3 days with HPMCAS, HPMC, CMCAB or PAA when humid.

Evidence plasticization by water · Physical stability F6 · Physical stability F9 · Physical stability F10 · Physical stability F14 · Solid state stability F8 · Solid state stability F9 · Solid state stability F10 · konno2008 R8, R5, C15, R1.a, R3.a, C11, R3, C13; rumondor2010 R1, R2, R8, C31, C33; rumondor2009c R17, C37, C41, R24, C46, R29; marsac2010 R8, C10, R17, C47; rumondor2009b R9, C13; wegiel2013a R9

Surfaces, carriers and the process

A free surface grew felodipine form I about 6 times faster than the bulk at 18 °C, with no difference above 70 °C. PVP protects the bulk more than the surface: at 30 °C the surface/bulk ratio was 5.7 without and 22.8 with 3% PVP. In powders this gives a particle-size effect: about 46% crystalline at 24 h for 150–250 µm against about 12% for 500–710 µm.

A crystalline carrier adds interfaces: a piece of crystalline PEG nucleated supercooled ibuprofen and fenofibrate within minutes, and a crystal grew at the interface and nowhere in the bulk liquid. The route matters too. Of six marketed tacrolimus capsules of similar composition, one generic crystallized to 88% in 3 weeks at 40 °C/75% RH. The authors write that this "strongly suggests" a processing cause.

Evidence Solid state stability F11 · Overview F6 · Physical stability F21 · kestur2013a R8, R9; kestur2013b R12, R3.e, R5; zhu2012 R15, R19; trasi2017 R12, C28

Reading a stability test

A test ranks only inside its window. About 490 days at 0–52% RH could not tell PVP from HPMCAS in felodipine dispersions, while 75% RH and above could. Curcumin dispersions with five polymers all lasted more than 350 days at 25 °C/0% RH and all crystallized within one day at 90 °C/0% RH and at 60 °C/98% RH. High humidity can also change the route: separation at 54 or 75% RH led the authors to warn that the crystallization route "may be altered at high RH".

The detector sets the answer. SHG dated the first crystals of spray-dried flutamide at 11 days against 53 days by XRPD, and crystals that only TEM could find cut the 4 h bicalutamide concentration from about 35 to about 24.5 µg/mL.

Evidence Solid state stability F12 · Overview F7 · rumondor2009b R19, R9; wegiel2014 R13, C23; rumondor2009c C52; kestur2012 R7, R6, R8; moseson2021 R17

What is measured and what is proposed

Checks for a new dispersion

These checks come from the “For your drug” blocks of the findings above. They are not one tested procedure.

1. Classify the neat drug by a heat–cool–heat cycle, as a first flag only. Do not screen by Tg/Tm, fragility or χ alone.
2. List the free hydrogen-bond donors and acceptors of the drug and test complementary polymers against one that is not.
3. Measure nucleation and growth separately, at a temperature near storage.
4. Choose a loading at least one step below the miscibility limit, found on a sample made by the intended route with a method that resolves domains, not by DSC alone.
5. Rank polymers at the intended storage humidity, measure the isotherm and the Tg of samples equilibrated at that humidity, and re-test the phase state after any humidity excursion.
6. Store at the intended condition and at two stress conditions that differ in kind, and pair a sensitive detector with a quantifying one.

Evidence Solid state stability F1 · Solid state stability F2 · Solid state stability F4 · Solid state stability F6 · Solid state stability F7 · Solid state stability F8 · Solid state stability F9 · Solid state stability F12 · Overview F7

Still open

• Why does the best polymer in humid storage reverse between drugs? The drug–water parameter that grouped the separating drugs differs by only a small margin.
• Do interaction parameters or solubility estimates predict whether a dispersion stays amorphous? Estimates agreed in sign but not in size, and 20/80 benzocaine and ibuprofen dispersions below their estimated solubility crystallized within hours.
• By what route does a few percent of polymer slow nucleation? The proposed routes were not measured.
• Why does growth of neat felodipine jump above about 80% RH? The cause is “currently unknown”.

Evidence Overview: open questions · Solid state stability F3 · Physical stability F4 · Physical stability F9 · Solid state stability F10 · rumondor2009c R24, C46; zhu2011 R5; zhu2012 R23.a; rumondor2010 C31, C33