Review: storyline · equations and derivations
Following Moseson D.E., Taylor L.S. (2023) Crystallinity: A Complex Critical Quality Attribute of Amorphous Solid Dispersions. Molecular Pharmaceutics.
What this page is. Moseson and Taylor (2023) reason from the two origins of crystals in amorphous solid dispersions (ASDs), incomplete amorphization and crystal creation, through their detection and their effect on non-sink release, to the conclusion that a limit on crystallinity has to be set product by product. This page follows that line step by step; each step gives the review's argument and then what the library's papers, from 2006 to 2025, have measured, with links to the findings, including the papers after 2023. Where the review states an interpretation or a calculation, and where the library holds a measurement, is marked in each step. Review pages are pages of the PDF (p. 1 is journal page 4802); values marked ≈ and "read from the plot" are this page's readings of the printed figures.
Question. Why does crystallinity need a specification in an ASD, and is 100% amorphous content required?
The review's reasoning (2023). ICH Q6A asks that the solid form be specified, monitored and controlled when it can affect performance, bioavailability or stability [review p. 1]. Crystallinity is one structural attribute in a materials science tetrahedron, and structural attributes change on storage [review p. 2]. The review names three ways in which crystals act: they forfeit the solubility advantage of the amorphous form; they initiate further crystallization, which reduces the rate and extent of release and causes desupersaturation; and they grow during storage [review p. 3]. It separates two origins, incomplete amorphization (crystals "dissolved or destroyed" incompletely) and nucleation and growth (crystals "created") [review p. 2], and states the question of the paper: whether performance relies on 100% amorphous content and what specification is appropriate, given that "one-size-fits-all" cannot be used [review p. 1]. This step needs no earlier step.
What the library shows. Each route has been measured, in different systems. Loss of the solubility advantage alone: physical mixtures of crystalline indomethacin with a 1:1 indomethacin–PVPVA ASD ended 0.5–2.5 µg/mL below the line of amorphous content, with no desupersaturation, at a 50 µg/mL total dose (ASD process F26) [moseson2020a R7, R18, X3]. Further crystallization: posaconazole–HPMCAS extrudates at 10% loading with about 0.3% crystals by SHG fell to ≈18.5 µg/mL by 150 min, where the crystal-free extrudate still held ≈125 µg/mL at 300 min (ASD process F41) [shah2024 R10, R12, R15]. Across the dissolution papers, polymers acted mainly on nucleation: 1 µg/mL HPMC slowed felodipine growth 1.6-fold and nucleation about 1,400-fold, which is the reason crystals already present remove most of the protection (Dissolution F4) [alonzo2012 R5, R16, R17].
Next. The same declared crystallinity can act through different routes, and the route depends on the crystals' origin, so the review turns first to how crystals arise in manufacture.
Pages. residual crystallinity, amorphous solid dispersion, amorphous solubility advantage, desupersaturation.
Question. Below what temperature can crystalline drug not dissolve completely into the molten polymer?
The review's reasoning (2023). Incomplete amorphization matters most for hot-melt extrusion (HME) and other thermal routes, and leaves crystals when temperature or mixing is insufficient [review p. 3]. The review takes the rate of crystal dissolution from the Noyes–Whitney form, with the diffusion coefficient from Stokes–Einstein [review p. 3]:
T_c (also called the solubility temperature T_s) is the minimum temperature for dissolving all crystalline drug into the molten polymer at a given composition, that is, the temperature at which C_s - C can stay positive until the last crystal has gone [review p. 3]. For a miscible pair it lies below the drug's melting point, and Flory–Huggins gives the depression [review p. 4]:
Here T_m and T_c are in kelvin, \Delta H is the enthalpy of fusion, \phi the drug volume fraction, m the ratio of the polymer segment to the drug molecular volume, and negative or small positive \chi indicates a miscible system [review p. 4]. The review considers the DSC offset on heating a fixed composition and isothermal dissolution from an undersaturated state to be thermodynamically equivalent [review p. 4]. In the phase diagram, region 1 (below T_g and left of the solubility line) has thermodynamic and kinetic stability, and region 2 (right of the line) "some degree of kinetic stability" [review p. 4].
What the library shows. Measured T_c values exist for three pairs. Indomethacin–PVPVA at 50% loading gave 132.3 ± 0.8 °C against 160.64 °C for the pure drug, with \chi = −1.97 and a predicted 131.0 °C; bicalutamide–PVPVA 30:70 gave 145 °C against ≈193.8 °C, \chi = −1.57; posaconazole–HPMCAS gave an offset of ≈161.7 °C at 40% drug against ≈168.6 °C, \chi = +0.43, and calculated values of 135, 155 and 165 °C at 10, 25 and 50% (ASD process F8) [moseson2018 R5, R4; moseson2022c R8, R9; shah2024 R1, R4, R5, X7]. Extrusion follows the line. Indomethacin–PVPVA below T_c held ≈26.1, ≈16.0 and ≈7.2% crystals after 2 min at 117, 121 and 126 °C, and still ≈1.5% at 126 °C after 10 min and ≈1.8% at 121 °C after 20 min (limit of quantification about 1%), while at 131 °C the XRPD value fell from ≈2.8% at 2 min to ≈0 at 10 and 20 min (ASD process F9) [moseson2018 R10]. Posaconazole–HPMCAS at 125 °C for 2 min held 0.3, 3.2 and 19.3 wt% by SHG at 10, 25 and 50% loading, and none at 175 °C [shah2024 R5, R10]. For five compounds in PVP K29/32 the predicted crystal solubility is only 0.010–0.134 weight fraction, so most practical dispersions lie in region 2, an inference not measured in those papers (Miscibility F9) [marsac2009 R9, C34, X17].
Corrections and qualifications.
Next. T_c settles the temperature below which crystals remain at any residence time tested. It does not say how long complete dissolution takes above T_c, which is the next step.
Pages. formulation critical temperature, melting point depression, Flory–Huggins interaction parameter, temperature–composition phase diagram, drug solubility in polymer, Flory–Huggins melting point depression, Noyes–Whitney equation, Stokes–Einstein equation.
Question. Above T_c, what sets the time to a crystal-free extrudate, and what limits raising the temperature?
The review's reasoning (2023). From equations 1 and 2 of step 2, dissolution is accelerated by high surface area, high diffusivity, high temperature and intense mixing [review p. 5]. The design space of Fig. 3b maps temperature against kinetics; the time for complete transition shifts with particle-level properties, and the risk runs from residual crystallinity at low temperature and kinetics to degradation at high temperature and kinetics [review p. 4]. The review reports TEM of indomethacin–PVPVA residual crystals, with a crystal of 26 × 70 nm, nanocrystalline domains of 5–10 nm and amorphous channels, and says this "suggests dissolution mechanisms beyond diffusion" [review p. 5]. Its model is that dissolution at defect sites fragments the crystal, and in HME mechanical breakage acts at the same weak sites, so crystals of greater defect density can amorphize faster; it describes this as observed experimentally and confirmed with population balance modelling [review pp. 5–6]. The upper limit appears in the KinetiSol case of LY3009120 (Davis 2021), where degradation of 0.7, 2.0 and 3.1% accompanied crystallinity of 2.3, 0.9 and 0.1% [review p. 12].
What the library shows. Without mixing, the time to dissolve indomethacin crystals into PVPVA on a hot stage rose from below 30 s at 161 °C to ≈0.3 h at 138–141 °C, ≈1.9 h at 131 °C and ≈20 h at 121 °C (ASD process F11) [moseson2018 R13, R14]. Input size acts at fixed temperature. Bicalutamide–PVPVA at 155 °C, 10 °C above T_c, left no XRPD-detectable crystals from D50 3.5 µm input (TEM found 140 ± 10 nm crystals), 3.5% from 38 µm and 30% from 93 µm (ASD process F9) [moseson2021 R6, M1–M3]. In the kinetic series at 145 °C, lots of similar size differed: 37–38 µm lots with 88.6 and 45.5% agglomeration fell below the PXRD limit in 5–6 min, while a 42.2 µm lot with D90/D50 of 4.14 still held ≈2.5% and the 92.7 µm lot ≈15% at 20 min (ASD process F36) [moseson2022c R1, R2, R10]. Fragmentation was observed directly: one indomethacin crystal in copovidone at 130 °C split into at least five pieces within 30 min, with channels, domains of 6.4 ± 2.2 nm and rotations up to 93° (ASD process F12) [moseson2019 R5, R10–R16]. The population balance model reached the bicalutamide data only with breakage: dissolution alone left 26–73% (dissolution constant 0.5 µm/s) or 3–61% (4 µm/s) at 20 min (ASD process F37) [moseson2022c R12, R13]. The window between crystals and degradation can close (a cross-paper synthesis in that finding): bicalutamide–PVPVA extruded for 10 min at 165 and 180 °C gave assays of 95.0 ± 0.4 and 95.0 ± 1.3%, whereas a crystal-free product from small crystals required 180 rather than 155 °C; indomethacin, by contrast, lost below 1% mass at short times up to 200 °C (ASD process F35) [moseson2020 R18, R15, C44; moseson2021 R6].
Corrections and qualifications.
Next. This step explains crystals that were never removed. Crystals can also form from an amorphous state or a supersaturated solution, which needs its own rate terms.
Pages. HME operating design space, crystal dissolution into a polymer melt, diffusion boundary layer, crystal defects.
Question. What terms govern the formation and growth of new crystals from a supersaturated solution or an amorphous solid, and what can a polymer change?
The review's reasoning (2023). The second origin rests on the statement that an amorphous solid or a supersaturated solution will ultimately crystallize, and that nucleation and growth compete for the same supersaturation [review p. 6]. Primary nucleation has a higher barrier than secondary nucleation, which is the reason crystals already present matter [review p. 6]. The nucleation rate is printed as [review p. 6]:
with the driving force printed as
where equation 6 applies when \gamma/\gamma^* \approx 1, in dilute aqueous solutions without solubilizing components [review p. 6]. Higher temperature, higher S and faster agitation raise nucleation; the extent rather than the existence of supersaturation is what matters, which defines a metastable zone whose width polymers change [review p. 6]. Growth is a diffusion and integration process [review p. 6]:
Adsorbed polymer interferes with integration, and the growth rate falls as supersaturation falls. With inhibitors growth may be arrested while supersaturation persists, so the concentration may not return to the crystalline solubility in a relevant time [review pp. 6–7]. In a melt or a glass, molecular mobility is often rate limiting [review p. 7]. This takes up the second origin of step 1.
What the library shows. The contrast between holding a crystal-free solution and holding a seeded one is measured repeatedly. Felodipine with 1 µg/mL HPMC at S 10 went from ≈5.4 to ≈5.0 µg/mL over 200 min unseeded, and from 5.1 to 2.2 µg/mL within 58 min seeded (Crystallization from solution F20) [alonzo2012 R18]. Growth inhibition depends on polymer concentration and supersaturation. For indomethacin at S = 6 PVPVA gave effectiveness ratios of 18 (5 µg/mL) and 56 (50 µg/mL); for bicalutamide at 150 µg/mL polymer, PVPVA gave 8.6-fold and HPMCAS 158-fold, with seeded growth near zero at a dose of 30 µg/mL and "extremely rapid" at 150 µg/mL (ASD process F25) [moseson2020a R3, R4, R5; moseson2021 R4, R5]. Growth arrest above the crystalline solubility is also measured: paclitaxel from HPMCAS ASDs with 18 wt% seeds ended at ≈25.3 (bulk seeds) to ≈7.9 µg/mL (sonicated seeds) at 16 h, against a crystalline solubility of 0.36 µg/mL (Supersaturation F22) [que2018 R1, R2, R13, R14], and posaconazole extrudates ended at 11–16 µg/mL, still 9–13 times C^* of 1.2 µg/mL (ASD process F41) [shah2024 R18, X15]. The paclitaxel authors take this as a reason to revise PBPK models that assume a return to C^* [que2018 C41, C42].
Corrections and qualifications.
Next. These terms are general. In a dispersion they are set by how much drug the matrix carries and how mobile that matrix is, in processing and on storage.
Pages. primary nucleation, secondary nucleation, metastable zone, supersaturation, crystal growth, crystal growth inhibition, classical nucleation theory, activity-based supersaturation ratio, supersaturation ratio, power-law crystal growth rate, crystal growth inhibition effectiveness ratio.
Question. When do crystals form during manufacture or on storage, and what stops them?
The review's reasoning (2023). The review cites the rule of storage 50 °C below T_g (Hancock 1995) and notes that ASDs have remained stable for years under less stringent conditions [review p. 6]. In processing, the review lists dissimilar evaporation rates in cosolvent systems, moisture, delayed secondary drying, insufficient cooling rate and a drug loading that is too high [review p. 7]. Three cases then show drug loading as the failure mode. Naproxen–PVP at 30% was amorphous with no crystallization after 8 weeks at 40 °C, while 60% held 6% (spray dried) and 0.4% (HME) crystals initially and reached 22% and 3%, read by the review as drug loading beyond stabilization with higher mobility or driving force after solvent drying [review p. 11]. In Ivanisevic's series, prepared by solvent evaporation and stored at ambient conditions, miscible dispersions were stable and immiscible ones formed crystals within 1−2 months; nifedipine–PVP at 70% carried residual crystallinity after preparation, which developed in the first 6 months and then stopped at below 10%, and ketoconazole–PVP at 30% showed no crystallization in about 1 year whereas 70% crystallized after about 2 months and progressed [review pp. 13–14]. The review reads the nifedipine plateau as growth that stopped when the solubility of the amorphous drug in the matrix was "presumably reached", and concludes that matrix properties dominate growth [review pp. 14, 16]. This needs region 2 of step 2 and the mobility term of step 4.
What the library shows. Dry storage supports the miscibility link. Indomethacin powders above the miscibility limit read from fluorescence crystallized in 2–15 days at 40 °C over desiccant, and those at or below it stayed X-ray amorphous for 14 months (HPMC, up to 50% drug) and 13 months (HPMCAS, up to 30% drug), while PVP powders at 10–90% drug stayed amorphous for the 3 months tested (Physical stability F30) [tian2016 R16, R17]. A plateau of the kind seen for nifedipine was also measured: felodipine–PVP powders at 93% RH stopped within about 20 days at about 70, 80 and 75% crystallized for 10, 25 and 50% PVP (Physical stability F15) [rumondor2009b R21, R11, C15, C20]. In processing, water in the spray solvent is the best documented variable: ritonavir–PVPVA from 60:40 water:methanol had 59–70% surface drug at 25% loading against 20–21% from methanol (ASD process F3) [li2020 R6, R7]. A droplet model places the onset of separation near 12.5:87.5 water:methanol at 25% loading of ritonavir–PVPVA, with peak supersaturation about 2.4 there and about 320 at 15:85 (ASD process F20) [li2020 R17, R19, R20]; this is calculated, and experiments show subtle separation at 10:90, where the model stays in one phase [li2020 X12].
Corrections and qualifications.
Next. Most of these cases turn on the position of the composition in the phase diagram at the storage condition. Water and temperature move that position, which the next step treats quantitatively.
Pages. drug solubility in polymer, solvent-induced phase separation, crystallization tendency, amorphous–amorphous phase separation.
Question. How far do humidity and temperature move a formulation from its dry position in the phase diagram?
The review's reasoning (2023). Most ASDs sit in region 2, and water acts as a third component [review p. 16]. The review illustrates this with PC-SAFT calculations for naproxen with PVP, PVPVA64 and HPMCAS at 0, 60 and 75% RH (Fig. 17), with 20% and 40% drug marked at ambient temperature; these are calculated diagrams, not measurements [review p. 16]. In them the 40% composition becomes more unstable with humidity and 20% in PVP or PVPVA is metastable at 75% RH. T_g - T runs from 10 to 100 °C when dry, against a rule of more than 50 °C, and from −5 to 25 °C at 60% RH; at 75% RH the T_g of 20% PVP and PVPVA lies below ambient, as does that of 40% PVPVA and HPMCAS [review p. 16]. The review draws from this that packaging is a control, that temperature acts on both T_g - T and relative supersaturation, and that subambient storage may be most appropriate for some products [review p. 16]. For samples already holding crystals it states the rule that follows from step 5: a composition that is not supersaturated is stable and its crystals do not grow, whereas in a metastable composition T_g - T sets the growth rate [review p. 16].
What the library shows. The direction is measured, for other drugs. At 75% RH and 25 °C PVP K29/32 held about 33 g water per 100 g, HPMC about 10–11 and HPMCAS about 6, and the dry T_g advantage of 25% PVP (about 54–55 °C against 43–48 °C) was gone at 75–76% RH, where the felodipine and nifedipine systems sat near 28–36 °C (Physical stability F6) [konno2008 R8, R5, C15; rumondor2010a R1; marsac2008a R6, C17]. Water also divides the matrix: felodipine–PVP showed AFM phase contrast and roughness of 17 nm against 0.22 nm within 1 h at 94% RH, and nine further systems separated after 54–94% RH, whereas felodipine–, quinidine– and indomethacin–HPMCAS kept one T_g (Physical stability F12) [marsac2010 R19; rumondor2010c R3, R3.f, R3.g, R3.h; rumondor2011 R14].
Corrections and qualifications.
Next. A position in the diagram predicts risk. Whether crystals have in fact formed depends on the method used to look for them, which is the next question.
Pages. plasticization by water, hygroscopicity, moisture-induced crystallization, moisture-induced phase separation, drug–polymer–water miscibility, glass transition temperature, sub-Tg nucleation, Gordon–Taylor equation.
Question. Which method sees which crystals, and what does "X-ray amorphous" exclude?
The review's reasoning (2023). Detection needs orthogonal tools, and many do not suit routine QC [review p. 7]. XRPD is the most common method and has relatively poor sensitivity. DSC is relatively ineffective for crystals in miscible pairs because heating dissolves them, and a single T_g may hide inhomogeneity. Polarized light microscopy is highly sensitive, even in X-ray amorphous samples, but diffraction limited [review p. 7]. 19F solid-state NMR reached 0.04% in a tablet, which is not generalizable; in the LY3009120 case it was used because DSC and XRPD lacked sensitivity [review pp. 7, 12]. HME crystals are regularly below 1 µm and as small as 5–10 nm, which TEM can see at low throughput [review p. 8]. SEM reads the surface only, SHG reaches the "parts per billion range" [review p. 8]. Release testing is said to be more sensitive than most techniques, but with variability of ±5% crystallinity below 10% may not be detectable in QC (Purohit 2023) [review p. 8]. The crystal sizes come from step 3.
What the library shows. Each method placed the "amorphous" boundary differently. In indomethacin–PVPVA, birefringence appeared in XRPD-amorphous samples extruded at 141 °C (2 and 10 min) and 131 °C (10 and 20 min), and a DSC endotherm appeared only in four 2 min samples at or below 131 °C; the sample at 131 °C for 20 min was X-ray amorphous with one T_g of 73.8 ± 2.6 °C and held crystals below 100 nm with domains of 5.8 ± 1.2 nm by TEM (ASD process F10, ASD process F9) [moseson2018 R11, R12, R15; moseson2018a R5, R6, R9]. In bicalutamide–PVPVA DSC saw only the 30% sample, XRPD 3.5 and 30%, and only TEM the sample below the XRPD limit [moseson2021 R6, R17, R21]. In posaconazole–HPMCAS, pXRD (LOD 0.3, LOQ 1.2 wt%) read 2.4 against 3.2% by SHG and 23.1 against 19.3%, and nothing at 10% loading where SHG read 0.3% (ASD process F10) [shah2024 R9, R10, P6]. Surface imaging detected crystals earlier on storage: SEM found crystals on uncoated ezetimibe–HPMCAS at 19 days against 3 months by PXRD, and on lumefantrine–PVPVA at 1 week against none through 48 weeks by XRPD (ASD process F10) [duong2022c R16, R18; moseson2022a R11, R13, X14]. Quoted SHG limits are 10 ppm (naproxen in HPMCAS, calculated; lowest imaged level 0.1%) and 0.03–0.047% (flutamide and ezetimibe dispersions, regression-residual limits; a 0.005% standard was visible), against about 0.4 and 0.7% by XRPD there (Solid-state methods F7) [kestur2012 R7, R6, R8, X10; correasoto2017 R5, R8].
Corrections and qualifications.
Next. Detection gives an amount at best. Whether that amount changes performance has to be tested in a release experiment under conditions where crystals can have an effect.
Pages. limit of detection, second harmonic generation, limit of detection and quantification.
Question. Under which conditions can a release test show the effect of crystals?
The review's reasoning (2023). Sink conditions have little relevance; non-sink conditions, and ideally an absorptive compartment, are required [review p. 8]. The dose is placed against both solubilities [review p. 8]:
The amorphous solubility is the concentration at which LLPS may be observed, and it changes with composition and media [review p. 9]. At \mathrm{SI_{cr}} < 1 and \mathrm{SI_{am}} = 1 the review describes two outcomes with crystals: supersaturation proportional to amorphous content with inert crystals (low risk), or reduced rate and desupersaturation with a lower AUC [review p. 9]. \mathrm{SI_{am}} < 1 is the more biorelevant case, owing to low fluid volumes and high doses: drug-rich colloids act as a reservoir, crystallization must be prevented, and crystals may reduce the supersaturation reached [review p. 9]. A single stage can miss failures: with a weakly basic drug and enteric polymers (Nguyen 2023) a pH 3 stage caused matrix crystallization and incomplete release, while a single stage at pH 6.5 was complete [review p. 17]. The growth terms of step 4 explain the desupersaturation.
What the library shows. One tacrolimus series spans the sink indices, with C^* 1.8 and amorphous solubility about 50 µg/mL at a 5 mg dose: (\mathrm{SI_{cr}}, \mathrm{SI_{am}}) were (0.32, 9) at 900 mL, (0.16, 4.5) at 450 mL, (0.04, 1) at 100 mL and (0.01, 0.4) at 40 mL (Bioperformance F16) [purohit2018 R4]. Polymer enters the amorphous solubility itself: HPMC and PVP-VA at 300 µg/mL lowered that of ibuprofen from 177.1 to 96.9 and 90.7 µg/mL without changing the crystalline solubility (LLPS F19) [ueda2020b R1]. At \mathrm{SI_{am}} < 1 the reservoir has been inferred from dose dependence: 10% posaconazole ASDs at 200 µg/mL held ≈22–24 µg/mL to 120 min, whereas a fifth of that dose (≈21 µg/mL) fell to ≈8 µg/mL by 120 min; SHG found crystals after 30 min in acid, so the onset of desupersaturation "does not represent the onset of crystallization" (Supersaturation F26) [elkhabaz2019a R6, R12, R2, R16, C49]. The reservoir itself was not measured [elkhabaz2019a X8]. Two-stage tests changed rankings: delamanid salt dispersions with HPMCP-50 and HPMCP-55 released ≈96 and ≈89% at pH 6.5 alone but ≈41 and ≈80.5% after pH 3.0 (Supersaturation F25) [nguyen2023c R11].
Corrections and qualifications.
Next. A discriminating test still needs crystals of known relevance in order to set or validate a limit.
Pages. sink index, non-sink dissolution, amorphous solubility, liquid–liquid phase separation, pH-shift supersaturation, sink index formula.
Question. Do spiked crystals of a given mass stand in for the crystals that arise in a product?
The review's reasoning (2023). The review distinguishes grown, spiked and residual crystals (Fig. 8) and holds that the relevant properties are surface area and related attributes, not mass [review p. 9]. Residual crystals are generally below 1 µm with high area and defect density, grown and residual crystals are in intimate contact with polymer, and spiked crystals are of defined form, above 10 µm, with low area and defect density [review p. 10]. The "same 10% crystallinity" therefore represents far lower area when spiked, and limits set that way likely underpredict the effect [review p. 10]. In fenofibrate tablets (Theil 2018) two sink methods were "nearly indistinguishable or remarkably different", and spiked crystals poorly represented recrystallized tablets [review p. 13]. This needs the origins of steps 2–5 and the test of step 8.
What the library shows. At a constant amorphous dose of indomethacin, bulk seeds at 20, 50 and 100% of that dose gave C720 of ≈41.4, 40.3 and 39.8 µg/mL, while extrudates dosed to the same amorphous amount peaked lower and fell, from ≈40.7 to 37.3 (131-2) and ≈32.5 to 27.0 µg/mL (117-2); the authors state that bulk seeds "do not provide an equivalent surrogate" (ASD process F26) [moseson2020a R9, R10, R19, X1, C41]. At one seed mass, smaller seeds left less drug in solution: paclitaxel–HPMCAS with 18 wt% seeds ended at ≈25.3 (bulk), ≈19.2 (S16), ≈10.4 (S80) and ≈7.9–9.0 µg/mL (sonicated) at 16 h against ≈31.6 µg/mL unseeded, and percent crystallinity "per se is not a predictive indicator" (Supersaturation F22) [que2018 R13, R14, C5, C54].
Corrections and qualifications.
Next. With test conditions and crystal types defined, the case studies can be compared for the variable that divides them.
Pages. crystal seeding, seeded crystal growth, residual crystallinity.
Question. What separates a system in which crystals cost only their mass from one in which traces cause large losses?
The review's reasoning (2023). Indomethacin–PVPVA (case 1): loss with spiked crystals "fell along the theoretical maximum", read as "the crystal seeds present in the mixture do not grow", with adsorption preventing or slowing growth; extrudates fell slightly lower, by area and energy or XRPD underestimation [review pp. 10–11]. Bicalutamide–PVPVA (case 2): the fully amorphous extrudate already showed matrix crystallization, with incomplete release and no desupersaturation; with crystals supersaturation was distinctly lower and the stable polymorph grew rapidly within the solid, at below 1% by TEM and even under sink conditions [review pp. 11–12]. Follow-up work adds loading above the limit of congruency and mechanical activation as risk factors [review p. 11]. Tacrolimus (case 6): at the highest sink index samples to 20% showed no marked decrease, and at the lowest all lost supersaturation and desupersaturated [review p. 13]. These cases use the sink indices of step 8 and the crystal types of step 9.
What the library shows. Indomethacin at a 50 µg/mL dose gave C720 of 43.1 µg/mL at 0%, 41.8 at 1.9%, 39.4 at 3%, 34.5 at 7%, 32.6 at 16% and 28.8 at 25% (ASD process F26) [moseson2020a R6]. Bicalutamide at 150 µg/mL ended at ≈35 (0%), 24.5 (below 1%), 16.3 (3.5%), 12.3 (10% seeds) and 10.5 µg/mL (30%), and in 10 mg/mL slurries form 1 appeared at about 15, 7 and 0 min (below 1%, 3.5%, 10% seeds) against form 2 at 20–30 min in the crystal-free extrudate (ASD process F28) [moseson2021 R17, R19, R21]. Loading sets this route: 5 and 10% bicalutamide released above 90% within 10 min, and 20, 30 and 50% ≈88, 86 and 73% (ASD process F38) [moseson2022b R6, R7]; cryomilling 30% for 30 s to 5 min lowered release from ≈79–86% to 75–49% (ASD process F39) [moseson2022b R1, R4]. Posaconazole–HPMCAS, published after the review, adds a third pattern: at 10% loading 0.3% crystals caused the loss of step 1, at 25% the hold fell from 1.5 h to 45 min, and at 50% both samples were poor (ASD process F41) [shah2024 R11, R12, R14, R15]. Both indomethacin and bicalutamide used PVPVA, with growth effectiveness of 18–56 against 8.6, and the finding's synthesis is that the larger difference between them is the matrix route (ASD process F29) [moseson2020a C54–C56; moseson2021 C72–C76].
Corrections and qualifications.
Next. No single level follows from these cases, which is why the review ends with a procedure and not a number.
Pages. matrix crystallization, limit of congruency, polymorphism, crystal growth rate.
Question. How should a limit be set, and what does absorption evidence say about it?
The review's reasoning (2023). The review organizes the risk assessment by drug, formulation, physical stability, dissolution, and crystal origin and micromeritics (Fig. 16), as scientific rationale and not a sequence [review pp. 14–15]. Failure modes should be provoked deliberately, and drug products added, since milling and compression can induce demixing or crystallization and dilution hinders detection [review p. 17]. On specification, ICH Q6A accepts surrogate tests with solid-state testing as a "last resort", and the authors urge caution [review p. 17]: Hermans found 6% crystalline detectable only by assuming no additional variability, which does not apply to GMP or QC [review p. 18]. A non-sink QC test is warranted only where a risk assessment justifies it, with solid-state methods linked to bioperformance as the alternative [review p. 18]. The conclusion: in some systems growth is slow and low levels have little impact, in others trace crystals in X-ray amorphous samples cause rapid crystallization, so product-by-product studies should identify failure modes and aim at an IVIVC [review p. 18]. The list uses all earlier steps.
The review prints no absorption simulation. It gives no assumptions, cases, fractions absorbed or residence times. Its quantitative in vivo content is one dog study (Fig. 14c, peak ≈23 ng/mL for the amorphous ASD, ≈2 for the recrystallized and below 1 for the suspension, read from the plot), and it offers three reasons for the 100% result, which it does not test: incomplete crystallization, highly defective crystals, or very small crystals that redissolve quickly [review pp. 13–14]. The following is therefore library evidence, not the review's.
What the library shows: tested. With a sink present, volume determined whether 20% crystals lowered exposure. In a hollow-fibre test at 240, 100 and 40 mL (\mathrm{SI_{am}} 2.4, 1, 0.4), AUC0–120 was 362, 682 and 992 for fresh tacrolimus capsules, 315, 583 and 770 for 20% and 80, 117 and 167 for 100%, against 87 µg/mL·min for suspension; only at 40 mL did 20% lose dissolved drug (48 to 9 µg/mL) and fall to ≈72% of fresh (Bioperformance F28) [hate2019 R9, R10, R11, R12, R13]. Whether growth exceeds absorption depends on supersaturation: S 20 in 100 mL gave AUC0–240 ≈1460 against ≈925 at S 8, with an estimated crossover near S ≈18.5, usable qualitatively (Bioperformance F29) [hate2019 R15, R16, R19, X16]. In six beagles at 1 mg, AUC0–24 was 42.1 (fresh), 22.9 (50%), 21.6 (100%) and 5.29 ng·h/mL (reference suspension), Cmax 25.3, 9.31, 3.84 and 0.69 ng/mL (Bioperformance F30) [purohit2019 R9]. A PBPK simulation (GastroPlus 9.0, 24 virtual subjects, 5 mg, 80–125 limits, no precipitation in the gut) failed bioequivalence from 10–20% crystallinity with 100 mL input and from 40% with 900 mL input (Bioperformance F17) [purohit2018 R13–R17, C59].
What the library shows: untested. No human study of a crystal-containing ASD is in the series, and the simulation over-predicts about 1.4-fold [purohit2018 X3]. The ratio of 100% to fresh differs by test: dogs ≈51%, absorptive ≈17–22%, simulation ≈8% (100 mL) and ≈55% (900 mL), USP I ≈63%; the dog dose of 1 mg in about 95 mL, about a fifth of the amorphous solubility, lies with the large-volume conditions, a weak comparison made from ratios of that finding (Bioperformance F31) [purohit2019 R9, R12, C40; hate2019 R13; purohit2018 R14, R16, R5].
Corrections and qualifications.
Next. The chain ends with a procedure: identify the crystal origin and the failure modes, test at the relevant \mathrm{SI_{am}} with representative crystals, and link the result to exposure.
Pages. in vitro–in vivo correlation, virtual bioequivalence, absorptive sink, oral bioavailability, membrane mass transfer coefficient, hollow-fibre donor depletion, non-compartmental AUC.