Residual crystals in a melt-extruded amorphous solid dispersion: lost potency, seed growth and the role of polymer adsorption

Background and motivation

An amorphous solid dispersion (ASD) is only worth making if the drug it releases becomes supersaturated and stays that way long enough to be absorbed. A polymer in the ASD helps twice: it stabilises the amorphous drug in the solid, and once dissolved it can inhibit crystallization from the supersaturated solution, even at low concentrations. The proposed mechanisms are specific interactions between dissolved polymer and drug, and adsorption of polymer onto the surfaces of drug crystals (Raghavan 2001; Schram 2016) C1·C2.

Crystals in an ASD are a quality problem. They may be left over from manufacture or appear by recrystallization in storage. Because a supersaturating dosage form generates supersaturation quickly, crystals present in it are assumed not to dissolve, so their share of the dose simply does not contribute to the solubility advantage. They may also act as seeds: by growth or secondary nucleation they consume supersaturation and pull the dissolved concentration back down (Sun et al. 2016) C3. Specifications therefore set an allowable crystallinity, usually by mass. The authors argue that percent crystallinity is an arbitrary measure of quality, because drugs differ widely in crystallization tendency, intrinsic seed properties drive crystal growth (Que et al. 2018), and polymer properties matter C4.

Most published work on crystalline content deals with crystals that nucleate from the amorphous matrix during manufacture or storage, a nucleation-based failure. Crystals that survive processing are a different case: for them crystal growth, not nucleation, would dominate, and their effects "may show different trends" C5. Testing their impact is hard. Bulk crystal seeds, or a physical mixture of fully amorphous and fully crystallized material, are commonly used as surrogates. But crystals inside an ASD may differ in size, surface area and defect density, and the group's earlier transmission electron microscopy (TEM) of hot melt extrusion (HME) samples and films found crystals of about 5–100 nm that were highly defective (Moseson et al. 2018, 2019). Surrogates also ignore the polymer: the intimate mixing of HME may alter crystal growth, for example by polymer adsorption on crystal surfaces C6·R20.

The paper asks four questions about indomethacin (IDM) and copovidone (PVPVA): do residual crystals represent a failure mode of an ASD; what are their consequences; are there crystal or formulation properties that allow some residual crystallinity; and what does the polymer do to these crystals. In short, could crystallinity in an ASD ever be "inert"? Two study designs answer them: bulk IDM seeds in supersaturated solutions with and without PVPVA, and non-sink dissolution of HME ASDs made with different levels of residual crystallinity, compared with externally seeded samples. Atomic force microscopy (AFM) and scanning electron microscopy (SEM) were used to look at polymer adsorption and at how seeds grow.

Background concepts

Supersaturation and non-sink conditions. The paper defines the supersaturation ratio S as the solution concentration divided by the equilibrium (crystalline) solubility. In 50 mM pH 4.5 acetate buffer at 37 °C, IDM has a crystalline solubility of about 8 µg/mL and an amorphous solubility of about 80 µg/mL, neither changed significantly by 50 µg/mL PVPVA (p > 0.05) R1·C7. The dissolution tests target 50 µg/mL IDM, so S = 6 with respect to the crystal, a condition the methods call "moderately high" supersaturation P9·E3. Dissolution testing for quality control is normally done under sink conditions, where the concentration never exceeds the crystalline solubility; that hides both the supersaturation an ASD can reach and the kinetics of crystallization. Under non-sink conditions both become visible C27.

Three outcomes for an ASD that contains crystals. Following a commentary by Sun et al., the authors use three idealised profiles (Figure 10b) C27:

• Scenario A: a crystal-free ASD that does not nucleate releases its amorphous drug completely, up to the amorphous solubility.
• Scenario B: the ASD contains crystals, but the polymer "poisons" them so they do not grow; the only cost is the crystals' share of the dose, a loss of solubility advantage (potency).
• Scenario C: the crystals grow or induce secondary nucleation; release is slower and lower, and the concentration falls after a peak (desupersaturation).

Figure 10a sets this in a wider picture: an ASD particle can dissolve and its drug be absorbed, or the drug can crystallize from solution (nucleation and growth), from the solid matrix, or onto seeds dispersed from the particle; all of these reduce the drug available for absorption C26.

How fast a seed grows. The paper writes the crystal growth rate as the mass deposited on crystal surfaces per unit time:

In plain terms: the rate $R_G$ at which mass $m$ is added to the seeds over time $t$ grows with the crystal surface area $A$ (proportional to the amount of seed), with a growth rate coefficient $k_G$, and with the supersaturation $S$ raised to the overall growth order $g$ E1. The authors note that a polymer can change both $k_G$ and $g$, and that mechanical forces such as those in HME or cryomilling "can affect the surface area term" by reducing particle size, creating defects and generating new growth sites C10. The equation frames the study: residual crystals could differ from bulk seeds in $A$ (size, defects), and the polymer acts on $k_G$ and $g$. It is not fitted to any data in the paper E1.

How well a polymer slows growth. The polymer effectiveness ratio compares growth rates without and with polymer:

In plain terms: $R_{G,np}$ is the growth rate with no polymer and $R_{G,p}$ the rate with polymer; a value of $E_G$ above 1 means the polymer disrupts growth, and the larger it is, the stronger the inhibition (Schram et al. 2015) E2.

Surface poisoning and adsorbed layers. An additive that adsorbs on a crystal face can block the sites (kinks and steps) where molecules join the lattice; growth then slows or changes habit. The authors relate their AFM and growth data to kinetic models of growth in the presence of surface-poisoning impurities (Kubota and Mullin) E6·C48, and they cite work describing adsorbed polymer at low concentration as a flat "train" conformation, a thin layer, and at higher concentration as a thicker layer that is a greater barrier to surface diffusion (Patel and Anderson 2015) C51. Step pinning, the halting of advancing growth steps by adsorbed obstacles, is one of the features they measure on SEM images R13.

Size-dependent solubility. By the Gibbs–Thomson relation, which the paper names but does not write out, very small crystals are more soluble than large ones; the authors expect crystals of 5–100 nm to show a "non-negligible" solubility increase E5·C38.

The samples. Eight HME ASDs were made in a small recirculating extruder at product melt temperatures of 117 to 161 °C and residence times of 2 to 20 min, deliberately in the "suspension regime" where temperature or time is too low to dissolve all the drug crystals. Each is named by melt temperature and residence time: 117-2 was extruded at 117 °C for 2 min. Their crystallinity (0% to 25% of the drug, by PXRD against a standard curve made with bulk crystals) comes from the group's earlier paper (Table I) P3. The drug-to-polymer ratio of the extrudates is not printed; a 1:1 ratio follows from Table II and from the authors calling the 1:1 SE ASD/PM blends "compositionally equivalent" X7.

Methods in detail

The study has three parts: what PVPVA does to IDM nucleation and to the growth of bulk seeds in solution; how ASDs with residual crystals dissolve compared with ASDs that carry bulk crystals instead; and what the crystal surfaces look like after PVPVA adsorption, dissolution or growth. All solution work used 50 mM pH 4.5 acetate buffer at 37 °C M3·P9.

Part 1: seeds in a supersaturated solution, with and without polymer

Problem. Before asking what crystals do inside an ASD, the authors needed to know how fast IDM nucleates and grows at the test supersaturation, and how much PVPVA slows each process.

What they did. A 50 µg/mL IDM solution was made by antisolvent addition of a concentrated methanolic IDM solution to buffer. Bulk IDM crystals (sieved below 250 µm; D50 21.9 ± 0.3 µm, BET surface area 0.14 ± 0.08 m²/g) were added at 0–10 mg, the equivalent of 0–50 µg/mL of extra IDM, and the concentration was followed for up to 12 h by a UV dip probe; the same was repeated with 5 and 50 µg/mL PVPVA dissolved in the buffer P10·M1·P1. The 50 µg/mL PVPVA level is the amount of polymer a 50 µg/mL IDM dose of the 1:1 ASDs releases C47. Growth rates (µg/s) were taken from the seeded profiles and compared via $E_G$ at the highest seed level P20·E2. The paper does not say how the rates were extracted from the curves X12.

Why it works. Unseeded solutions report on nucleation (an induction time), seeded ones on growth (the slope of the decline), so the two effects of the polymer can be separated P10.

Part 2: dissolution of ASDs with residual crystals versus bulk-crystal surrogates

Problem. Do residual crystals in HME ASDs consume supersaturation, and do bulk crystals behave the same way?

What they did. Non-sink dissolution used 200 mL of buffer in a jacketed beaker at 37 °C, stirred at 300 rpm, in triplicate, with concentrations read by a UV dip probe using the absorbance at 317 nm minus that at 450 nm to correct for scattering, over 12 h P9. The 106–250 µm sieve fraction of each ASD was used P6. Two designs were run.

• Constant total dose. Each of the eight HME ASDs was dosed to 50 µg/mL IDM P11. The surrogate was a blend of a crystal-free solvent-evaporated 1:1 ASD (SE ASD) with a 1:1 physical mixture of crystalline IDM and PVPVA (PM), mixed to give 50 µg/mL IDM and 50 µg/mL PVPVA with 0 to 100% of the drug crystalline P4·P5·P12.
• Constant amorphous dose. Four crystal-containing HME ASDs were dosed so that their amorphous IDM equalled 50 µg/mL, which means a larger total dose P13. The surrogate was the SE ASD at 50 µg/mL plus bulk seeds equal to 20, 50 or 100% of the amorphous dose (10, 25 or 50 µg/mL) P14. Holding the amorphous dose constant keeps the driving force for dissolution the same across samples, a condition the first design did not meet C18.

Why it works. If crystals do not grow, the concentration reached should fall in proportion to the crystalline fraction: the authors draw a straight line from the all-amorphous to the all-crystalline sample and treat points on it as scenario B, points below it as "some evidence" of scenario C E4·C34. In the second design the line joins the SE ASD alone and the SE ASD with seeds equal to the whole amorphous dose E4·R19.

One detail does not add up. The Figure 6 caption gives the extrudate doses as 50 (161-2), 51.5 (131-2), 60 (126-2), 54 (121-2) and 60 µg/mL (117-2). With the Table I crystallinities, only 131-2 then contains 50 µg/mL amorphous IDM; 126-2 would contain about 55.8, 121-2 about 45.4 and 117-2 about 45.0 µg/mL. The positions at which Figure 12 plots these samples (about 3, 8, 20 and 33% crystals relative to the amorphous dose) do match the 50 µg/mL design, so the caption is most likely misprinted; the SI with the calculations was not available to check P13·X1.

Part 3: what happens at the crystal surface

AFM. Bulk IDM crystals were fixed on an AFM puck and covered with buffer containing 3 µg/mL IDM and 0, 5 or 50 µg/mL PVPVA for 4 days, then imaged in liquid in tapping mode; the centre-to-centre distance between adsorbed globules was measured (30 pairs at 5 µg/mL, 100 at 50 µg/mL) P15.

SEM and Raman. Bulk crystals, cryomilled crystals (about 1 g milled under liquid nitrogen, 2 min at 10 Hz) and the residual crystals of extrudate 117-2 were exposed for 24 h either to "crystal dissolution" (the sample alone, well above the crystalline solubility, so crystals partly dissolve or the ASD dissolves around them) or to "crystal growth" (an extra 50 µg/mL IDM already dissolved), with or without 50 µg/mL PVPVA in the medium (Table II) P2·P16. The particles were recovered on carbon tape, imaged by SEM, and for selected samples identified by Raman microscopy, where peaks at 1700 and 1650 cm⁻¹ identify the γ and α forms of IDM P17·P18.

TEM. Fragments of extrudate 131-2 were imaged by bright-field TEM at 200 keV and lattice periodicity was checked by FFT; the authors stress that, because the technique is slow and still emerging, the images are not a statistically significant sample P19·C24.

Experimental analysis

PVPVA stops nucleation and nearly stops seed growth

Without polymer, the unseeded 50 µg/mL solution held its concentration for about half an hour and then fell steeply, to about 25.6 µg/mL by 120 min; the authors place nucleation "in under an hour", with supersaturation lost after about 40 min (Figure 1a) R3.a·C8. Seeds made the decline start at once and steeper the more seed was added, reaching about 40.7, 31.8 and 26.9 µg/mL at 120 min with 10, 25 and 50 µg/mL seeds R3.b–d. The derived growth rates were about 0.33, 0.71 and 0.99 µg/s (Figure 2) R4.a. The authors call the rate "approximately proportional" to the amount of seed; the rate per unit of seed actually falls from about 0.033 to about 0.020 (µg/s per µg/mL), so the relation is better read as a steady increase C9·X2.

With PVPVA the picture changes completely (Figure 1b, c). Unseeded solutions stayed supersaturated for the whole 12 h at both 5 and 50 µg/mL PVPVA, which the authors read as PVPVA being "a highly effective nucleation inhibitor" at S = 6 R3.e·R3.i·C11. Seeded solutions still declined, but slowly (Table A) R3.f–h·R3.j–l.

Table A. IDM concentration (µg/mL) at the end of the seeded and unseeded runs, starting from 50 µg/mL; read from Figure 1, approximate. Note the different end times R3·X16.

Growth rates fell to about 0.005–0.055 µg/s with 5 µg/mL PVPVA and to about 0–0.018 µg/s with 50 µg/mL R4.b–c. At the highest seed level the polymer effectiveness ratio was 18 for 5 µg/mL and 56 for 50 µg/mL PVPVA, which the authors take as showing "a highly effective crystal growth inhibitor" R5·C12. In the discussion they go further: growth rates were "near zero", so growth "had almost no impact" on the solution concentration C46. That holds for 50 µg/mL PVPVA; with 5 µg/mL PVPVA and the highest seed level the concentration still dropped by about 13 µg/mL in 12 h R3.h·X6.

Residual crystals in the extrudates: lost potency, little desupersaturation

Dosed at 50 µg/mL IDM, all eight extrudates dissolved with profiles of similar shape: a similar rise followed by a near-plateau (Figure 3) R6·C13. Table B gives the numbers.

Table B. Non-sink dissolution of the HME ASDs at a 50 µg/mL total IDM dose, n = 3; crystallinity from Table I (cited from ref 40), concentrations read from Figure 3, approximate P3·R6.

The three extrudates below 2% crystallinity were indistinguishable from the crystal-free one and kept rising slightly to 12 h; from 3% upward the profiles peaked and then fell slightly, and the peak was lower the more crystals the sample held R6·C13. The authors match this to scenario B: the extrudates contained up to 25% crystallinity, yet "near-complete release of the remaining amorphous content" was achieved, with "minimal desupersaturation" C28·C33. Dividing the 720 min values by the amorphous part of each dose gives about 81% (131-2), 74% (126-2), 78% (121-2) and 77% (117-2), against about 86% for the crystal-free 161-2, so "near-complete" is relative X13. Even the crystal-free extrudate did not fully dissolve in 12 h; the authors expect this from the small driving force left near the end, and they note that the 12 h window is longer than gastrointestinal transit but lets dissolution and growth both be seen R6.a·C31·C32. They describe the release of this high-drug-loading ASD as drug-controlled: the polymer "is expected to dissolve quickly", leaving amorphous drug particles that release more slowly C30. They also report, without showing data, that at higher supersaturation "greater rates of desupersaturation were observed" C29·R21·X10.

For the three samples below 2% the authors offer two readings. These crystals "may" follow the same pattern as the others, too little to be seen; or, being nanometre-sized and defective, they "may" have a higher solubility and dissolve, so that "some low level of crystallinity" behaves like amorphous material C37. By Gibbs–Thomson, crystals of the size seen by TEM "would be expected" to be measurably more soluble C38. Neither reading was tested.

Bulk seeds are a poor surrogate

At constant total dose. The SE ASD/PM blends gave qualitatively the same profiles as the extrudates, with the concentration reached at 720 min falling with the crystalline fraction: about 43.2 (0%), 39.7 (5%), 36.7 (10%), 34.9 (20%), 30.4 (30%) and 26.7 µg/mL (40%), and only about 4.4 µg/mL for the all-crystalline mixture (Figure 4) R7. Unlike the extrudates, none of the blends desupersaturated within 12 h, which the authors read as dissolution keeping pace with or outrunning crystal growth throughout C15·C16. The crystal-free SE ASD dissolved like the crystal-free extrudate 161-2 (Figure S3) R8·C17.

Figure 11 puts both sets on one plot of the 720 min concentration against crystallinity. The blends lie close to the theoretical maximum line, which the authors take to mean their seeds "do not grow"; the extrudates with 3–25% crystallinity lie below it, which "suggests that the crystal seeds have grown to some extent" R18·C35·C36. Read from the figure, the blends are not exactly on the line: they sit about 0.5–2.5 µg/mL below it at 5–40% crystallinity, while the 7, 16 and 25% extrudates sit about 4–6 µg/mL below. The 3% extrudate (about 2.5 below) deviates no more than the 10% blend, so the difference is clear only from 7% upward X3. Figure 11 also plots a 25% blend (about 32.0 µg/mL) whose profile does not appear in Figure 4 X4.

At constant amorphous dose. This comparison is sharper (Table C). Adding bulk seeds to the SE ASD at up to 100% of its amorphous dose lowered the 720 min concentration only from about 43.2 to about 39.8 µg/mL, with "no (or minimal)" desupersaturation (Figure 5) R9·C18. The four crystal-containing extrudates, dosed to the same amorphous amount, peaked lower and then declined (Figure 6) R10·C19.

Table C. Dissolution at a constant 50 µg/mL amorphous IDM dose; read from Figures 5, 6 and 12, approximate. Crystal fractions for the extrudates are their x-positions in Figure 12. The printed extrudate doses do not all give 50 µg/mL amorphous IDM R9·R10·R19·X1.

Even the extrudate with the fewest crystals (131-2, about 3% of the amorphous dose) ended below every bulk-seed sample, including the one with seeds equal to the whole amorphous dose R9·R10. The authors conclude that the crystals within the extrudates "are better able to grow in the presence of polymer" than bulk seeds, and that "the bulk crystal seeds do not provide an equivalent surrogate to the residual crystals" C39·C41. The highlights restrict this to "high supersaturation" C41·X17.

What the crystal surfaces show

Polymer adsorption (Figure 7). Without polymer, the bulk crystal surface was generally smooth with occasional features. At 5 µg/mL PVPVA, spheroidal elevations appeared in the height image with matching phase contrast, distinct and non-overlapping, across much of the surface; the authors read these as "likely" adsorbed polymer with moderate coverage. At 50 µg/mL the globules covered most of the surface R11·C20. The spacing between globule centres fell from 160 ± 50 nm to 70 ± 20 nm R12.

Dissolving and growing crystals (Figure 8). Table D summarises what SEM showed after 24 h R13·R14·R15.

Table D. SEM observations after 24 h of crystal dissolution or crystal growth (extra 50 µg/mL IDM dissolved), with or without 50 µg/mL PVPVA; widths and spacings are mean ± SD, n = 20 R13·R14·R15·P16.

All grown crystals tested by Raman, and the residual crystals after both treatments, were the γ form, which the authors call the thermodynamically stable form R16·C50. Their reading: the rounded etch pits indicate that the polymer interacts with the crystal surface during dissolution; the smooth plates grown without polymer "suggest the single crystal origin" of the bulk crystals; with polymer, "the adsorbed polymeric network appears to limit the sites available" for molecules to join, which favours needle-like growth C21·C49·C50. The residual crystals grew like the milled ones, "suggesting underlying disorder in the processed crystal" C23. The globule spacing at 50 µg/mL (about 70 nm) is of the same order of magnitude as the needle widths and the step-pinning distance, which the authors take as "demonstrating the link between polymer adsorption and growth patterns" C52. Which SEM panel the step-pinning distance was measured on is not stated X11.

Crystals inside the extrudate (Figure 9). TEM of extrudate 131-2 showed small crystalline domains below 100 nm embedded in a homogeneous amorphous phase; a 45 × 65 nm domain showed diffraction contrast and lattice planes, with periodic spots in its FFT R17·C25. This fits the group's earlier finding of residual crystals of 5–100 nm, highly defective, some as small as 5–10 nm, and S'ari's report of similar sizes in felodipine extrudates, which the authors call "demonstrating the generalizable nature of this finding" R20·C43.

Discussion

How the authors connect the results. Their argument runs from the competition between dissolution and crystallization to two outcomes for the seeds C26·C27·C34.

In the authors' account the polymer stabilises the supersaturation by two effects, preventing nucleation and poisoning the growth of seeds, "both bulk or intrinsic to the HME samples" C60. The abstract and conclusion state that "extensive polymer adsorption onto residual seed crystals resulted in poisoned crystal growth", which limited the consequences of residual crystals C58. The final claim is framed as "a provocative idea": crystallinity intrinsic to an ASD "could have minimal impact" on non-sink performance under controlled supersaturation, beyond lost solubility advantage C59.

Risk factors. The discussion turns this into three risk factors for residual crystallinity C54·C55·C56:

1. Crystal growth rate. If the polymer inhibits growth, seeds can consume little supersaturation and the drug available is limited only by the amorphous amount; for IDM/PVPVA, crystalline content "poses a fairly low risk" C54.
2. Seed properties. Bulk seeds are "routinely less potent at consuming supersaturation" than residual seeds from HME or seeds with higher surface energy or surface area; residual crystals are small and mechanically damaged C55. Even micron-sized seeds, commonly used to mitigate this, would, the authors "speculate", not reproduce the <100 nm, highly defective residual crystals C42.
3. Supersaturation. At higher supersaturation and growth rates the apparent potency of seeds increased, which the authors "would speculate" makes crystal seeds a greater risk in a low fluid volume in vivo, especially for faster-crystallizing drugs C56.

Performance is drug–polymer specific: IDM/PVPVA sits closest to scenario B because PVPVA inhibits IDM growth strongly, while tacrolimus/HPMC, celecoxib/PVP and IDM/Eudragit EPO are cited as systems that "may trend toward scenario C" C53.

An alternative the authors raise themselves. The extrudates might fall below the no-growth line partly because their crystallinity is underestimated: it was measured by PXRD against a standard curve made with bulk crystals, and peak detection depends on crystal size and defects, so "systematic underestimation of crystalline content may ultimately be common to HME ASDs" C44·X8.

What it suggests for formulation. Stated by the authors: percent crystallinity alone is an arbitrary quality indicator, and whether a given level matters depends on growth inhibition by the polymer, seed properties and supersaturation C4·C54·C55·C56; bulk seeds do not stand in for residual crystals C41.

Note by the AI writer (not discussed in the paper). When a non-sink method is validated for its ability to detect crystallinity in an ASD, spiking a crystal-free ASD with bulk crystals will make the method look less sensitive than it is for process-derived crystals; an under-processed extrudate is the more relevant control. The tolerance shown here also depends on conditions: 50 µg/mL IDM lies below the amorphous solubility of about 80 µg/mL, so the dissolution tests never reached the regime where liquid–liquid phase separation and much higher supersaturation occur R1·X17.

Limitations

Limits the authors state. The TEM findings "do not represent a statistically significant sample" C24. The 12 h monitoring period "is significantly longer than typical GI transit times" C32. The favourable outcome is "consistent at the concentration studied, 50 µg/mL"; at higher supersaturation desupersaturation was faster C29. The authors also allow that the crystallinity values may be underestimated C44.

The AI writer's own judgement. The direct data are solid: PVPVA's inhibition of nucleation and seed growth is large and consistent, and at equal amorphous dose every crystal-containing extrudate ended well below bulk seeds equal to the whole amorphous dose R5·R9·R10. The weaker links are the explanations.

• Polymer adsorption was imaged only on bulk crystals; adsorption on residual crystals, named in the abstract as the reason their impact stayed limited, is inferred X5·C58.
• The proposed causes of the stronger growth of residual crystals, higher surface area and surface energy, were not measured, and growth from solution was not separated from growth of the embedded crystals into the dissolving amorphous particle X9·X15.
• The printed doses of the constant-amorphous test (Figure 6) do not all give 50 µg/mL amorphous IDM with the Table I crystallinities X1.
• The bulk-crystal blends also sit slightly below the no-growth line, so only the 7–25% extrudates differ clearly from them in Figure 11 X3.
• "Near-complete release of the remaining amorphous content" corresponds to about 74–81% of the amorphous dose at 12 h, against about 86% for the crystal-free extrudate X13.
• Growth rates are mass rates for the whole vessel with an unstated extraction method, and are comparable only within this paper X12.
• One drug–polymer pair at one ratio, one buffer and one dose were tested, with no in vivo data; the SI, including the dose calculations and Raman spectra, was not available for this reading.

Conclusion

For indomethacin–copovidone extrudates dissolved at S = 6, residual crystals up to 25% cost mainly the crystals' share of the solubility advantage: the reachable concentration fell from about 43 to about 29 µg/mL, with only slight desupersaturation R6·C28. Dissolved PVPVA explains why: it prevented nucleation for over 12 h and slowed growth on bulk seeds 18- to 56-fold, and it adsorbed more densely on bulk crystals at the higher concentration R3·R5·R12. But bulk seeds understated the problem. At the same amorphous dose, residual crystals drained far more supersaturation than bulk seeds equal to the whole amorphous dose, and under growth conditions they sprouted needles like mechanically damaged crystals R9·R10·R15. The authors attribute this to residual crystals "likely" being smaller and more defective, and they frame residual crystallinity as a risk set by polymer growth inhibition, seed properties and supersaturation rather than by percent crystallinity alone C40·C54·C55·C56.