Choosing a crystallization inhibitor by matching the hydrogen-bond donors and acceptors of drug and polymer

Background and motivation

Many drug candidates dissolve too slowly or too little to be absorbed well, and one answer is to give them in the amorphous state, which can reach much higher solution concentrations than the crystal. The same excess energy that raises solubility also drives the amorphous drug back to the crystal, so it is usually mixed at the molecular level with a polymer to form a solid dispersion. C4 The authors point out that polymer selection for such dispersions was, at the time, "highly empirical", and that their own earlier screening work had found the relative performance of different polymers to depend mainly on the chemistry of the drug (Van Eerdenbrugh and Taylor 2010, ref 19). C4·C5

The paper looks for a rational basis in crystal engineering, the field that relates molecular structure to the way molecules pack in crystals. In that field, the hydrogen bond, which is electrostatic, directional and relatively strong, is treated as the most reliable interaction for building a crystal from its molecules (Steiner 2002, ref 2; Desiraju 1995, ref 3). C1·C2 Crystal engineering also offers a precedent for preventing crystallization: "tailor-made additives", small molecules designed to resemble the crystallizing compound, are recognized and incorporated by a growing crystal and then disrupt its hydrogen-bonding sequence, which alters nucleation, growth, morphology and polymorph (Weissbuch et al. 1991, ref 4, and refs 5–11). C3

From these two lines the authors build a working hypothesis: a polymer should inhibit the crystallization of a given drug better when it can form drug–polymer hydrogen bonds that are more favourable than the hydrogen bonds the drug forms in its own crystal. C6 They add "the important caveat" that a polymer is molecularly very different from a small additive. C7 To test the idea they extend the hydrogen-bond rules of crystal engineering (Etter 1990 and 1991, refs 20, 21) to drug–polymer pairs, rank the expected inhibitors for each drug, and compare these expectations with the observed crystallization of 15 readily crystallizing model drugs combined with 7 polymers. C11·A1·A2 The drugs were chosen from an earlier classification in which all of them crystallized quickly on their own (Van Eerdenbrugh, Baird and Taylor 2010, ref 22), and they carry a wide range of hydrogen-bonding groups. C16

Background concepts

Donors and acceptors. A hydrogen bond links a donor, a hydrogen attached to an electronegative atom (O–H, N–H), to an acceptor, an atom with a lone pair (a carbonyl oxygen, an amine or azole nitrogen, an ether oxygen). The authors rank donors first by the electronegativity of the atom carrying the hydrogen, F–H > O–H > N–H > C–H (Jeffrey 1997, ref 25). They do not consider F–H (rare in drugs) or C–H (weak) donors, regard O–H as "strong" and N–H as "medium", and then adjust for neighbouring groups: a carboxylic acid O–H counts as "very strong", and some N–H groups next to electron-withdrawing groups count as strong or very strong on the evidence of crystal-structure contacts and pKa. C13·P8

The pKBHX acceptor scale. For acceptors the authors use the pKBHX scale of Laurence et al. (2009, ref 26), a thermodynamic scale of hydrogen-bond basicity derived from spectroscopically measured association constants of 1:1 complexes with 4-fluorophenol; the higher the value, the stronger the acceptor. C14 Each functional group was given the pKBHX of a small model compound (for example 1-methyl-2-pyrrolidone, 2.38, for the pyrrolidone of PVP; ethyl acetate, 1.07, for esters) and placed in one of four classes: Weak <0.75 < Medium < 1.5 < Strong < 2.25 < Very Strong. R1·R2·P8 Because no pKBHX value exists for a carboxylic acid carbonyl, it was treated as an ester. C14

Etter's rules. Three empirical rules from crystal engineering guide the analysis: all good proton donors and acceptors are used in hydrogen bonding; six-membered intramolecular hydrogen-bond rings form in preference to intermolecular bonds; and the best donors and acceptors left after intramolecular bonding pair with one another (Etter, refs 20, 21). C11 An intramolecular bond therefore matters because it can take a good donor or acceptor out of play for other molecules, and such bonds are rarely broken (Bilton et al. 2000, ref 23). C12

Self-association and competition. In crystals, carboxylic acids typically form dimers, and amides bond to amides with extremely high frequency; alcohols self-associate less (Infantes and Motherwell 2005, ref 27). C17·C26·C30 Which pattern forms in a crystal containing two kinds of group is largely a competition between the donors and acceptors present, so the authors also count, in the Cambridge Structural Database (CSD), how often an amide bonds to itself or to a competing group. C15·P5

The amorphicity index. The crystallization read-out of the paper is a semi-quantitative score. At each time point a film is placed in one of five classes by polarized light microscopy, written as four letters: "AAAA" (completely amorphous), "AAAC" (slightly crystalline), "AACC" (semicrystalline), "ACCC" (predominantly crystalline) and "CCCC" (completely crystalline). P3 The amorphicity index (AI) counts the A letters over the four time points (day 0, 1, 3 and 7) and divides by 16, the count for a film that stays "AAAA" throughout (SI p.3): E1

In plain terms: $n_A$ is the number of A letters collected over the four inspections, so a film that never crystallizes scores 100% and one that is fully crystalline from the start scores 0%. The AI of a drug–polymer pair is the mean over its drug/polymer weight ratios. E1·P3 A low AI therefore means that the drug crystallizes readily even in the presence of the polymer, and a high AI that the pair is easily kept amorphous. C10

Pattern probabilities in the CSD. For two functional groups A and B, each crystal structure that contains both was assigned to a hydrogen-bonding pattern (none, A–A only, A–B only, B–B only, or a combination), and the probability of a pattern is the share of structures that show it (SI p.4): E2

In plain terms: $N_{\mathrm{c}\mathrm{l}\mathrm{a}\mathrm{s}\mathrm{s}}$ is the number of structures showing a given pattern and $N_{\mathrm{t}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}}$ the number of structures examined after removing duplicates and structures with intramolecular hydrogen bonds. E2·P5 The printed percentages are consistent with whole-number counts; for example 72% of 57 phenacetin/acid structures corresponds to 41 structures (an AI-writer check; the counts are not printed). E2

Methods in detail

The paper has three parts that run side by side and then meet: a crystallization screen of all drug–polymer pairs, a hydrogen-bond evaluation of every functional group of the drugs and polymers, and a CSD survey of competition between amide groups and two polymer-like groups. The expectations from the evaluation are then compared with the screen, drug class by drug class, and a few exceptions are examined with infrared spectroscopy.

Part 1: a crystallization screen of 105 drug–polymer pairs

Problem. To test a rule for choosing polymers, one needs the same read-out for many drugs and many polymers under one protocol. The authors use spin-coated films, which form quickly and thinly so that crystallization, when it happens, is easy to see. P1·C8

What they did. The 15 drugs (Figure 1a) fall into four groups: five carboxylic acids (flurbiprofen, flufenamic acid, 4-biphenylcarboxylic acid, 4-biphenylacetic acid, anthranilic acid), three drugs with acidic N–H groups (chlorzoxazone and the sulfonylureas chlorpropamide and tolbutamide), two alcohols (4-phenylphenol and 4-biphenylmethanol) and five amides and bases (benzamide, phenacetin, lidocaine, bifonazole, dibucaine). A1·M1–M15 The seven polymers (Figure 1b) were Eudragit E100, poly(vinylpyrrolidone) (PVP, K 12), poly(vinylpyrrolidone-vinyl acetate) (PVPVA, K 28), poly(acrylic acid) (PAA), poly(styrene sulfonic acid) (PSSA), hydroxypropylmethylcellulose (HPMC, type 2910, 6 mPa.s) and HPMC acetate succinate (HPMCAS, grade AS-MF). A2·M16–M22

Drug and polymer were dissolved in ethanol, or in a 1/1 (w/w) ethanol/dichloromethane mixture for HPMC and HPMCAS, at drug/polymer weight ratios of 75/25, 50/50 and 25/75. For the drugs of an earlier study, five further ratios (90/10, 60/40, 40/60, 20/80 and 10/90) were added. P1 Some solutions precipitated (bifonazole and lidocaine with PAA or PSSA, in some ratios) and were vortex-mixed before coating. R4 200 µL of solution was spun on an 18 × 18 mm cover slip for 20 s at 8000 rpm, each in triplicate, and the films were stored over phosphorus pentoxide. P1·P2 They were examined by polarized light microscopy within 30 min of coating and after 1, 3 and 7 days, scored by the area of birefringent regions, and summarized as the AI defined above. P2·P3·E1 The authors justify the area-based scoring by the thinness of the films, across which a crystal is assumed to grow through the whole height, and note that birefringence can only come from the drug because none of the polymers crystallizes ("it is reasonable to assume"). C8 Different drugs had different sets of weight ratios; the authors consider this acceptable because they compare polymers for one drug and never compare drugs with each other. C9

Part 2: rating every hydrogen-bonding group

Problem. To apply Etter's rules, the analyst must know which donors and acceptors each drug and polymer has, which of them are free, and how strong they are.

What they did. For each drug group (Table 1) and polymer group (Table 2) the authors recorded whether it can donate and accept, a donor class, an acceptor class with the pKBHX of a model compound, and a calculated pKa (MarvinSketch 5.2.0; only pKa values giving more than 50% ionization between pH 0 and 14 were kept). R1·R2·P7·P8 Intramolecular hydrogen bonds were first screened by ring size, since 5-, 6- or 7-membered rings form in 95% of cases (Galek et al. 2010, ref 24), and then checked against a survey of intramolecular bonds in crystals (ref 23), the drug's crystal structure, and density functional theory (DFT) for the two candidates, flufenamic acid and anthranilic acid. C12·P6·P8 For anthranilic acid, the C(NH2)–C–C–O(H) dihedral angle was stepped from 0 to 180° in 10° steps with the O=C–C angle held at 120°, at the B3LYP/6-311++G(3d2f, 2p) level. P6 Published crystal structures were examined for hydrogen-bond motifs and contact distances, identified by their CSD reference codes. P9

The resulting picture (Tables 1 and 2) is the input to every expectation in the paper. Among the polymers, PVP and PVPVA carry the "Very Strong" pyrrolidone acceptor (pKBHX 2.38) and no donor; E100 has a "Strong" tertiary amine (1.98) and "Medium" esters (1.07) and no donor; PAA and PSSA carry "Very strong" acidic donors (pKa 4.5 and −1.5) with medium or low acceptors; HPMC and HPMCAS carry "Strong" hydroxyl donors and "Medium" ether and ester acceptors, HPMCAS also a small amount of carboxylic acid (1.0 mmol g⁻¹). R2 The same R–C(O)–OH group with pKa 4.5 is called "Very strong" in PAA but "Strong" in HPMCAS, without a stated reason. X4 Among the drugs, the carboxylic acids have very strong donors and medium acceptors, the sulfonylureas very strong acidic N–H donors, the alcohols strong O–H donors, and the amides medium N–H donors with strong carbonyl acceptors (pKBHX of the model compounds 1.69–1.81); lidocaine and dibucaine add strong amine acceptors with basic pKa values of 7.8 and 9.8, and bifonazole has a very strong imidazole acceptor (2.72, pKa 6.7) and no donor. R1

Two footnotes of Tables 1 and 2 need care. Footnote a describes its four classes as a way "to classify donor strength", but the classes are pKBHX acceptor classes; and in Table 2 the footnotes cite ref 18, a pharmaceutical development paper, where Table 1 cites the pKBHX source, ref 26. X2 The column "mmol g⁻¹" is not defined; for the drugs it carries one value per compound, which matches one group per molecule. X14

Part 3: counting competition in the crystal-structure database

Problem. The rules say what should happen; the CSD can show what actually happens when an amide meets a competing group in a crystal. This matters most for the amides, where the expected best partner is not obvious: a strong donor or an extremely strong acceptor might break amide self-association. C32

What they did. The authors queried the CSD (version 5.31) for crystal structures that contain an amide group of the benzamide type (primary) or the phenacetin type (secondary) together with either a carboxylic acid (a stand-in for PAA) or a tertiary amide C(O)NC2 (a stand-in for PVP), using a formula filter to exclude other heteroatoms; for example, structures with a C(O)NC2 and a COOH group were restricted to C_1–500H_1–500N₁O₃. P5 Repeated structures and structures with intramolecular hydrogen bonds were removed, hydrogen bonds were judged with the default criteria of Mercury, and each structure was assigned to a pattern class. P5 The list of classes printed in both the main text and the SI gives "A-A & A-B" twice, as classes 5 and 6; Table 3 shows that the sixth class is A-A & B-B. X1

Experimental analysis

All crystallization results of the paper are in one bar chart, Figure 3. Table A lists the AI of every pair as read from that chart; the values are approximate (about ±1 AI unit), and a 0 marks a pair with no visible bar. R3 The figure has no error bars, and how many weight ratios went into each bar is not stated drug by drug. X9

Table A. Amorphicity index (%) of each drug–polymer pair after 7 days of dry storage, read from Figure 3 (approximate); polymers ordered acceptor-only, cellulosic, acidic donor; 0 = no visible bar. R3

Main result 1: drugs with strong donors favour acceptor-only polymers. For the five carboxylic acids, PVP, PVPVA and E100 gave AIs of about 54–93, while PAA and PSSA gave about 4–30 for four of them. R3.a–e The authors write that the best results "are obtained with polymers bearing strong acceptors and no donors" and that the acidic polymers are "the worst crystallization inhibitors". C20 HPMC and HPMCAS were "more variable": about 76 and 70 for flurbiprofen and 68 and 62 for flufenamic acid, but no visible bar and about 4 for 4-biphenylcarboxylic acid and about 8 for 4-biphenylacetic acid. R3.a–d·C20 The three acidic N–H drugs gave the same split, about 63–100 with PVP, PVPVA and E100 against about 8–31 with PAA and PSSA, while HPMC and HPMCAS were somewhat lower than the acceptor-only polymers except chlorpropamide–HPMC (about 95). R3.f–h·C25 The authors' reading is that a drug with a very strong donor and only a medium acceptor pairs more favourably with a polymer whose acceptor is stronger than its own, provided the polymer brings no donor to compete; their summary is that the observations "suggest a trend towards poorer crystallization inhibition" for polymers with weaker acceptors and stronger or more donors. C19·C21

Main result 2: the alcohols follow, less cleanly. For 4-phenylphenol the AI fell from about 74 (PVP), 71 (PVPVA) and 58 (E100) to 41 (HPMC), 21 (HPMCAS), 4 (PAA) and no visible bar for PSSA; for 4-biphenylmethanol it was about 60, 46 and 27 for the acceptor-only polymers, 6 and 8 for the cellulosics, 4 for PAA and again nothing for PSSA. R3.i–j The authors describe the acceptor polymers as best, HPMC and HPMCAS as intermediate and PAA and PSSA as poor, and they rationalize the last point by "the lack of provision of a competitive acceptor by the drug". C28·C29 The data for 4-biphenylmethanol fit this only in part: HPMC and HPMCAS sit with the acidic polymers, and E100 is well below PVP. X6

Main result 3: for amides and bases the order flips. For benzamide and phenacetin the acidic donor polymers came first (PSSA about 68 and 78, PAA about 64 and 67), PVP and PVPVA followed (about 47–50), the cellulosics were lower (about 34–39) and E100 was lowest (about 30 and 13). R3.k–l The authors relate this to the amide group: its N–H is a poorer donor than an O–H but its carbonyl is a better acceptor than a carboxylic acid, so competition "would be expected" from strong donors or extremely strong acceptors, and the result is "in line with" the donor polymers being best and the pyrrolidone polymers performing well. C32·C33 For the bases, PSSA was again the best inhibitor: about 71 for lidocaine and about 100 for dibucaine, with PAA at about 61 and 71. R3.m·R3.o Bifonazole stayed nearly amorphous with every polymer (about 93–100), which the authors explain by its low crystallization tendency. R3.n·C41

Table B. Probabilities (%) of hydrogen-bond patterns in CSD structures, from Table 3. For drug–drug pairs every hydrogen-bonded structure was counted as A-A. R14

Main result 4: an acid breaks amide self-association more often than a tertiary amide does. Without a competitor, every hydrogen-bonded benzamide structure and 94% of phenacetin structures show amide–amide bonding. R14.a·R14.d With a carboxylic acid partner no structure keeps amide–amide bonding alone: benzamide shows amide–acid bonds only in 40% and mixed with amide–amide bonds in 60%, and phenacetin shows amide–acid bonds only in 72%. R14.b·R14.e With a C(O)NC2 partner the amide–amide-only pattern remains in 14% (benzamide) and 33% (phenacetin) of structures. R14.c·R14.f The authors read this as support for the postulated competition, non-exclusive for the primary amide and at the expense of amide–amide bonding for the secondary amide, and conclude that carboxylic acid functions "compete more successfully" than the C(O)NC2 group, "correlating with the better interference seen for PAA compared to PVP" (benzamide about 64 versus 50, phenacetin about 67 versus 49). C34·C35·R3.k–l The benzamide rows rest on only 10 and 14 structures. R14.b–c

Main result 5: the exceptions are an acid–base pair and a crowded N–H. Anthranilic acid behaved unlike the other acids with PSSA (about 92, against about 60 with PAA). R3.e The authors note that anthranilic acid is also a weak base (calculated pKa 2.0) and PSSA a very strong acid (pKa −1.5), so an acid–base reaction "might occur", much like the zwitterion found in two of its crystal forms; ATR-FTIR spectra "indeed pointed to" such a reaction with PSSA and not with PAA. C22·R7·R8 The much higher pKa of PAA (4.5) and the much lower basic pKa of flufenamic acid (−2.1) "apparently prevent" the same reaction elsewhere. C23·R9 Proton transfer to PSSA, but not to PAA, was also reported for lidocaine and dibucaine, while bifonazole showed no acid–base reaction. R15·R16·R17 For lidocaine the acceptor-only polymers ranked opposite to acceptor strength, about 58 for E100, 42 for PVPVA and 29 for PVP; the authors suggest that the N–H, which sits on a benzene ring with two ortho methyl groups, is sterically hindered ("might explain"). R3.m·C40 For dibucaine, PVP (about 40) scored slightly below E100 (about 42), read as a weaker steric effect, and PVP well above PVPVA (about 15), read as the pyrrolidone being a stronger acceptor than the ester. R3.o·C46

Main result 6: glass transition temperature does not explain the ranking. The polymers differ widely in Tg, about 55 °C for E100 and about 140 °C for HPMC (cited values). M16·M21 If a higher Tg made a better stabilizer, HPMC should beat E100 for every drug; for the five acids it scored lower every time (for example about 8 against 81 for 4-biphenylacetic acid). R3.a–e·C47 The authors make the same point for drug Tg: 4-biphenylcarboxylic acid (69 °C) should be more stable than lidocaine (−31 °C), yet across the seven polymers lidocaine scored about 29–71 and 4-biphenylcarboxylic acid between no visible bar and about 62. M3·M13·R3.c·R3.m·C48

Supporting structural observations. DFT confirmed the intramolecular N–H···O=C bond of flufenamic acid (Figure 2a), so only its carboxylic acid donor is counted. R5·C18 For anthranilic acid the energy along the dihedral scan (Figure 2b) was about 31–33 kJ/mol up to 30°, where the acid O–H bonds to the aniline nitrogen, rose to a maximum of about 50 kJ/mol at 80°, and fell to zero at 180°, where an aniline N–H bonds to the carbonyl oxygen; no bifurcated bond formed, so the second aniline hydrogen remains free, though a weak donor. R6·C18 The acidic N–H groups form unusually short contacts in their crystals: N···O 2.825 Å in chlorzoxazone, 2.727–2.796 Å in chlorpropamide and 2.612–2.714 Å in tolbutamide, against a mean of 2.892 ± 0.112 Å for intermolecular N–H···O=C bonds (Taylor et al. 1984, ref 32). R10·R11 Benzamide uses every donor and acceptor in its crystal, phenacetin forms one N–H···O=C bond per molecule in chains with its ether oxygen unused, and lidocaine's reported intramolecular N–H···N contact has unfavourable angles of 99.0–112.4°. R13·R15·C31·C37

Discussion

The authors' chain of reasoning. The argument can be drawn as one chain that splits by the character of the drug's strongest free group. Each step is the authors' reading; the drug–polymer hydrogen bonds themselves were not measured in the films. C19·C32·X13

The authors present the conclusion firmly: the results "clearly indicate" that hydrogen-bond-directed molecular recognition, adapted from crystal engineering, can be used to design solids in which crystallization is prevented, and the abstract states that the predicted rank order "agreed well" with the observed behaviour of an extensive dataset. C51·C55 They also qualify it: further work "is warranted" to establish the specific role of drug–polymer hydrogen bonding in retarding crystallization, and the rules should be "applied as set of tools providing guidance, rather than a strict set of rules". C52·C53

How hydrogen bonding might act. In the discussion of other factors, the authors list three routes by which drug–polymer hydrogen bonding is "likely to" influence crystallization: it can improve miscibility and slow phase separation of drug and polymer (as hydrogen bonding governs polymer–polymer miscibility, ref 37, and has been suggested for drug–polymer systems, ref 38), it can lower the thermodynamic driving force for crystallization (ref 38), and it can slow crystal growth, as their group found for felodipine (ref 39) and, in unpublished work, for other drugs. C49 None of the three was measured here.

Scope. The evidence covers thin spin-coated films, dry storage over phosphorus pentoxide, 7 days, and a single visual score; the authors note that water, which competes for hydrogen bonds, "might alter crystallization behavior". P1·P2·C50 Within that scope the drug set is broad (four chemical classes, acids and bases, a range of crystallization tendencies) and the polymer set covers acceptor-only, cellulosic and strongly acidic polymers. A1·A2

What it suggests for formulation (note by the AI writer, not discussed in the paper). The practical message is a quick first filter: identify the drug's strongest free hydrogen-bonding group after intramolecular bonds are accounted for, and try a polymer whose groups complement it, an acceptor-only polymer such as PVP or PVPVA for acids, sulfonylureas and phenols, and a polymer with acidic donors for amides and bases. In this dry screen HPMC and HPMCAS were rarely the best and usually fell between the two classes. For basic drugs, an acidic polymer strong enough to protonate the drug added a second, ionic interaction, as PSSA did and PAA did not. The ranking says nothing about humid storage, bulk dispersions or dissolution, which the paper did not test.

Limitations

Limitations stated by the authors. The authors call the difference between polymers and small additives "the important caveat" of their hypothesis. C7 They describe their scoring as semi-quantitative and justify it by the thinness of the films ("it is reasonable to assume"). C8 They state that all studies were done under dry storage and that water "might alter" the behaviour. C50 They write that the specific role of drug–polymer hydrogen bonding still needs investigation and that their approaches give guidance, not strict rules. C52·C53 They also call the polymer differences for bifonazole small and explain the lidocaine and dibucaine orders with steric hindrance ("might explain"; "seems"). C41·C40·C46

The AI writer's own judgement (1): what "ab initio" can and cannot claim. The paper never writes down a predicted rank order of the polymers for each drug before showing the data, and it gives no measure of agreement; the expectations are stated in words, drug class by drug class, in the same paragraphs that report the AIs, and the deviations (anthranilic acid–PSSA, the lidocaine order, PVP against E100 for dibucaine) are explained afterwards by acid–base reaction or steric hindrance. X5 The "agreed well" of the abstract is therefore a qualitative, class-level agreement. Some class statements also fail for single drugs: for 4-biphenylmethanol HPMC and HPMCAS are not intermediate, for lidocaine PAA (about 61) is level with HPMC and HPMCAS, and bifonazole does not separate the polymers at all. X6·X7·X8 The Tg argument is a counterexample built on two polymers and two drugs with cited Tg values, not a correlation analysis. X11

The AI writer's own judgement (2): the read-out and the mechanism. The AI is a visual, categorical score: changing one letter at one of the four time points moves a ratio's AI by 6.25% and a three-ratio mean by about 2.1%, so differences of a few units, several of which carry the discussion (lidocaine PAA against HPMC, dibucaine PVP against E100), are not resolved. X10 The number of ratios averaged per drug, the way the three replicate films were combined, the storage temperature and the solution concentration are not given, and Figure 3 has no error bars. X9 Some lidocaine and bifonazole films with PAA or PSSA were cast from solutions that had precipitated. X12 Most importantly, the mechanism, drug–polymer hydrogen bonds or proton transfer in the films, rests on ATR-FTIR spectra of melts that are "data not shown"; no hydrogen bond was measured in the films whose crystallization was scored. X13 There are also small printing errors: the pKBHX classes are called a "donor strength" classification and Table 2 cites the wrong reference for them, the same carboxylic acid group is rated differently in PAA and HPMCAS, one pattern class is printed twice, and the lidocaine paragraph cites a solid-dispersion review (ref 15) for a hydrogen-bond survey result. X2·X4·X1·X3 None of these changes the main pattern of Figure 3.

Conclusion

The paper contributes three things. First, a single-protocol screen of 105 drug–polymer pairs, scored as an amorphicity index over a week of dry storage. R3·E1 Second, a systematic rating of every hydrogen-bonding group of 15 drugs and 7 polymers, combining crystal structures, a survey of intramolecular bonds, DFT, calculated pKa and the pKBHX acceptor scale. R1·R2 Third, CSD statistics showing that a carboxylic acid breaks amide self-association more often than a tertiary amide. R14·C35 Set side by side, these show a clear class-level pattern: drugs whose strongest free group is a donor stay amorphous best with acceptor-only polymers such as PVP, PVPVA and E100, and amides and bases stay amorphous best with acidic donor polymers, above all PSSA. C20·C25·C33·C38 The authors take this as evidence that crystal-engineering rules can guide polymer selection for amorphous solid dispersions, as tools rather than strict rules; the hydrogen bonds in the films and the predicted ranking itself remain to be measured and scored. C51·C53·X5·X13