How the counterion sets the glass transition of amorphous drug salts

Background and motivation

Many drug candidates are poorly soluble in water, and two classic ways to raise their dissolution rate and apparent solubility are salt formation, for molecules with an ionizable group, and conversion to the amorphous state (Introduction; Lipinski 2000, ref 1; Berge 1977, ref 4). Each route has its own literature. Salt selection for crystalline drugs is routine, and the properties of amorphous drugs, alone or in polymer dispersions, have been studied in depth. The amorphous form brings risks: it tends to be hygroscopic, mechanically poor and physically and chemically unstable, and it can crystallize and lose its solubility advantage (Hancock and Zografi 1997, ref 10). C1 Small amounts of amorphous material also form by accident during lyophilization, milling or spray drying, and can change bulk properties (Ahlneck and Zografi 1990, ref 11). C2

The two routes meet in the amorphous salt, and here the authors found very little prior work. Tong and Zografi had shown that amorphous sodium indomethacin has a much higher glass transition temperature (Tg) than amorphous indomethacin (ref 19). Tong, Taylor and Zografi then varied the alkali-metal cation and found that Tg fell systematically from lithium to cesium, with the neutral acid lowest of all; they attributed this to a stronger electrostatic interaction of the smaller ions with the carboxylate, which lowers molecular mobility (ref 20). Comparable effects are known for inorganic salt glasses (refs 21, 22), and salt formation between loperamide and poly(acrylic acid) was cited as the cause of a high Tg in that solid dispersion (Weuts 2005, ref 23). C3 From this the authors write that "it appears" salt formation will raise Tg relative to the neutral compound. C4

Those studies varied small inorganic cations around an acidic drug. The present paper turns the question round: it takes basic drugs and varies the organic and inorganic acids used to form the salt. The main objective was to establish whether the solid-state properties of amorphous salts, in particular Tg, can be changed by changing the counterion. The secondary objective was to relate the properties of the counterion to the resulting Tg with partial least squares (PLS) regression, borrowing the approach of quantitative structure–activity relationship (QSAR) studies (refs 24–28). Propranolol, a lipophilic β-blocker with a secondary amine of pKa 9.5, and nicardipine, a dihydropyridine calcium-channel blocker with a tertiary amine of pKa 8.6, were chosen because their high pKa makes salt formation "likely" with a wide range of acids (Table 1). M1·M2·C5

Background concepts

Glass transition temperature. An amorphous solid below its Tg is a rigid glass; above it, the material becomes rubbery and its molecules gain enough mobility to crystallize, react or flow. The storage temperature relative to Tg therefore influences physical and chemical stability and mechanical properties, and the authors call Tg "probably" one of the most important properties of an amorphous compound (ref 10). C20 At Tg the heat capacity jumps by ΔCp, which the paper reports for each salt (Tables 2, 3). R4·R5 The general rule the authors take from polymer science (Eisenberg and Shen, refs 36, 37) is that the stronger the interactions between molecules, the more thermal energy is needed to reach the mobility of the transition, so the higher the Tg; free volume and packing also contribute. C21

Counterion pKa. Each salt is described here by the pKa of the acid that forms it: −6 for hydrochloric acid, −3 for sulfuric acid, −1.34 and −1.2 for toluenesulfonic and methanesulfonic acids, 0.7 for benzenesulfonic acid, 1.271 to 3.128 for oxalic, phosphoric, tartaric and citric acids, and 4.19 and 4.756 for benzoic and acetic acids (Tables 2, 3, 6). A lower pKa means a stronger acid. The source of these pKa values is not stated. M5·M6·M7·M8·M9·M10·M11·M12·M13·M14·M15·X15

Ionic versus other interactions. In the language of conceptual density functional theory (DFT), an intermolecular interaction has three contributions: an electrostatic part, which dominates between ionic and "hard" molecules (molecules that resist charge transfer), and covalent and polarization parts, which dominate between "soft" molecules whose valence electrons move easily (refs 38–40). C22 For neutral glasses, hydrogen bonding and nonspecific cohesive forces matter (ref 41); for salts, the authors expect ionic interactions to "be of greater importance" in setting Tg, as the indomethacin salts showed. C23

DFT reactivity descriptors. Three intrinsic electronic properties of a single molecule are used as descriptors of how strongly it tends to interact. The first is the chemical hardness (eq 1):

In plain terms: $\eta$ measures how strongly the energy $E$ of a molecule resists a change in its number of electrons $N$ at fixed external potential $v$ (set by the nuclei); it is approximated by the ionization potential $I$ minus the electron affinity $A$. E1 The inverse of hardness is the softness, $S=1/\eta$. E4 The second is the electronic chemical potential (eq 2):

In plain terms: $\mu$ is the slope of the energy with electron number, the negative of the electronegativity; a more negative $\mu$ means a molecule that holds its electrons more tightly. E2 The third combines the two into the electrophilicity index (eq 3, Parr 1999, ref 46):

In plain terms: $\omega$ is large for a molecule that is both electronegative (large $|\mu |$) and soft (small $\eta$); the authors read it as a tendency to form strong covalent and polarization interactions when charge transfer is involved. E3·C24 These descriptors "are believed" to capture a molecule's inherent tendency to interact, and they trace back to Pearson's hard and soft acids and bases principle (refs 43, 51–54). C24 In Table 7 the electron affinity is printed in a column headed $E$, the same letter used for the system energy in eq 1; read that column as $A$. X16

PLS regression. Partial least squares projection to latent structures compresses many correlated predictors (here, counterion descriptors) into a few latent components that best explain the response (here, Tg). PLS2 handles several responses at once, so one model can relate the same counterion descriptors to the Tg of both the propranolol and the nicardipine salts. The fit is summarized by $R^2$ (variance explained), $Q^2$ (variance predicted in cross-validation) and the root mean square error of estimation (RMSEE). A loading-weights plot shows how each descriptor contributes to each response. P11·E5

Methods in detail

The study has three parts: making a series of amorphous salts of each drug, characterizing them under thermal and humidity stress, and relating their Tg to calculated properties of the counterion.

Contribution 1: a series of amorphous salts of two bases

Problem. To compare counterions, each salt has to be made in the same amorphous form. Both bases are poorly soluble in water, and their hydrochlorides were the starting materials. M3·M4

What they did. The hydrochlorides were dissolved in water at 10 mg/mL and the free base was precipitated with 5 M sodium hydroxide; sodium chloride was washed out with water and the solid filtered and dried. Propranolol came out as a white crystalline powder and nicardipine as a yellow amorphous residue. P1·M1·M2 Each base was then dissolved at 20 mg/mL, propranolol in ethanol and nicardipine in acetone, and the acid was added as an aqueous solution at "the required base to counterion molar ratio" (sulfuric acid as a 10% w/v solution). The organic solvent was removed in a rotary evaporator, 6 mL of water was added and the volume reduced to 5 mL. P2 The solutions were freeze-dried: frozen at −45 °C for 12 h, dried at 100 mT with the shelf at −30 °C for 24 h and −10 °C for 24 h, then dried under vacuum at 5 °C for 12 h and stored over P2O5 at 0% RH. Salt formation was confirmed by Raman spectroscopy, but no spectra are shown. P3·P4 Whether a freeze-dried salt was amorphous was judged by DSC, and salts that were not could be made amorphous by heating through the melt and quench cooling; the conditions of that step are not printed. P5·P6

Counterions. Propranolol was combined with hydrochloric, p-toluenesulfonic, methanesulfonic, benzenesulfonic, oxalic, phosphoric, tartaric, citric, benzoic and acetic acids; nicardipine with hydrochloric, sulfuric, benzenesulfonic, oxalic, phosphoric, tartaric, citric and benzoic acids, plus acetic acid, which appears only in the sorption data (Tables 2, 3, 5). A1·R4·R5·R9 Not every salt could be prepared for each drug because some responded poorly to freeze-drying. P3 The base:acid ratio is never printed, so the stoichiometry of the salts with polyprotic acids is unknown. X8

Contribution 2: three kinds of stress

Heating. About 8 mg of sample in a pinholed DSC pan was first cooled to −40 °C and heated at 20 °C/min to 250 °C to locate Tg, recrystallization and melting (Figure 1 shows these three events for propranolol hydrochloride). For the Tg itself, the sample was cycled several times between −40 °C and about 20 °C above Tg at 20 °C/min, to erase its thermal history and drive off water, and Tg was taken from the last heating. P6·R3 The paper does not say whether Tg, Tc and Tm are onset, midpoint or peak values, or how many samples lie behind each standard deviation. X7

Humidity. Moisture sorption was measured in a symmetric vapor sorption analyzer after a 180 min drying step, raising RH in 10% steps from 5% to 95% with 90 min, or a 0.0001% weight change, allowed at each step; the temperature is not printed. P8 Tables 4 and 5 report the weight change at 55% and 75% RH. R8·R9 A weight gain followed by a loss was taken as the signature of crystallization (Gift and Taylor 2007, ref 33). C15 To check this, samples were stored at 0% RH and then at 79% RH and inspected by polarized light microscopy. The storage time is given as 5 days in the text and as 48 h in the captions of Figures 5 and 6. P9·X4

Contribution 3: counterion descriptors and a PLS model of Tg

Descriptors. Table 6 lists general descriptors for 13 acids: calculated log P (ClogP), molecular weight, molar refractivity, polar surface area, molecular volume, pKa, hydrogen-bond donors and acceptors, rotatable bonds, calculated log D (ClogD) and a flexibility index; how they were obtained is not stated. M5·M6·M7·M8·M9·M10·M11·M12·M13·M14·M15·M16·M17 Two of the acids, maleic and malonic, appear in no salt. X10 The malonic and tartaric rows of Table 6 are shifted by one column, so some printed numbers (for example 46 hydrogen-bond donors for tartaric acid) are typesetting errors. X9 Table 7 adds the DFT descriptors $A$, $I$, $\eta$, $S$, $\mu$ and $\omega$ for 11 acids, calculated on single, fully optimized molecules. P10·R15 Every printed $\eta$, $\mu$, $\omega$ and $S$ follows from the printed $I$ and $A$ through eqs 1–3; for hydrochloric acid, for example, $12.765-(-0.486)=13.251$ eV and ${6.139}^2/(2\times 13.251)=1.422$ eV. E1·E3·R15.a

Model. PLS2 related the descriptors to the Tg of the propranolol and nicardipine salts. With all salts included, the citrate and phosphate salts were outliers, which the authors put down "probably" to their trivalent character, and they were removed. R16·C26 The number of components, the scaling, the cross-validation scheme and the descriptors finally retained are not printed. P11·X12

How it differs from earlier work. The indomethacin studies varied the cation of an acidic drug among the alkali metals, a series in which ion size is the obvious variable. Here the drug is the cation and the counterions are acids of very different strength, size and shape, and the authors turned to multivariate analysis because the literature indicates that several molecular factors influence Tg at once. C3·C21·P11

Experimental analysis

Which salts became amorphous

About half of the freeze-dried salts were amorphous: 11 of the 20 prepared, according to the text. R1 The hydrochloride, phosphate and tartrate salts of both drugs were amorphous, while benzoate and acetate salts never were. R1·C6 For propranolol, the hydrochloride, mesylate, oxalate, phosphate and tartrate came out amorphous and the tosylate, besylate, citrate, benzoate and acetate did not; for nicardipine, all salts in Table 3 except the oxalate and benzoate were amorphous. R4·R5 Propranolol besylate never gave a Tg, because it was crystalline after lyophilization and recrystallized on cooling from the melt. R2 The tables name only 19 distinct salts (10 propranolol, 8 nicardipine in Table 3, plus nicardipine acetate in Table 5), so the 20th salt of the text cannot be identified. X1

Every salt has a higher Tg than its free base

Table A collects the Tg values. Every propranolol salt lies between 9.0 °C (acetate) and 35.9 °C (hydrochloride), against −9.3 °C for the free base; every nicardipine salt lies between 28.5 °C (benzoate) and 81.0 °C (hydrochloride), against 3.2 °C for the free base. R4·R5 The authors state this without a hedge: Tg was higher "in all instances" for the salts than for the neutral compound, in line with the hypothesis drawn from the indomethacin work. C7·C4 The margin is large compared with the scatter: even the lowest propranolol salt (9.0 ± 5.6 °C) is clear of its free base (−9.3 ± 4.0 °C), and the nicardipine free base, whose value is uncertain (3.2 ± 9.6 °C), lies far below its lowest salt. R4.j·R4.k·R5.i·X17

Table A. Tg of the amorphous salts and free bases, mean ± SD as printed (n not stated); "—" = salt not reported (from Tables 2 and 3). R4·R5

The heat capacity jump at Tg also varies with the counterion: 0.08 (benzoate) to 0.65 J/g/°C (tartrate) for the propranolol salts, against 0.82 for the free base, and 0.22 (oxalate) to 1.4 J/g/°C (benzoate) for the nicardipine salts. The authors do not discuss these values. R4·R5

Tg falls roughly as the counterion pKa rises

Figures 2 and 3 plot Tg against counterion pKa. For "the majority" of salts, the authors write, Tg scales with pKa: salts of strong acids have the highest Tg and salts of weak acids the lowest, so that the hydrochloride is highest for both drugs. C8·R6·R7 They add that the relation is "not a linear one, especially" for high-pKa counterions, and that "it is likely" that other factors such as the size and shape of the counterion also matter. C9

Table A shows where the rule holds and where it does not. For propranolol, the hydrochloride (35.9 °C), oxalate (31.5 °C) and mesylate (29.2 °C) lead, and the acetate (9.0 °C) is last, but the tosylate (pKa −1.34, 16.7 °C) and the phosphate (pKa 1.96, 11.3 °C) fall well below salts of weaker acids. For nicardipine, the hydrochloride (81.0 °C) and sulfate (77.2 °C) lead, the besylate (65.8 °C) falls below the oxalate and phosphate, and the citrate drops to 29.3 °C next to the tartrate at 58.3 °C although the two acids have almost the same pKa (3.128 and 3.02). R4·R5·X14 The text names the acetate as the lowest salt for both drugs, but no nicardipine acetate appears in Table 3; the lowest nicardipine salts are the benzoate and citrate. C8·X2 Figure 3 also leaves out the nicardipine benzoate, which is in Table 3. R7

Note by the AI writer (not discussed in the paper): the largest deviation in the propranolol series is a low-pKa salt, the tosylate, not a high-pKa one, and several neighbouring salts differ by less than their standard deviations (propranolol tosylate and benzoate are both 16.7 °C; nicardipine benzoate 28.5 ± 3.1 and citrate 29.3 ± 1.7 °C). The pKa ranking is therefore best read as a broad trend across the series, not a rule for each pair of salts. X14·X17

Crystallization depends on the counterion and on the kind of stress

On heating. Only four propranolol salts recrystallized in the DSC: the hydrochloride (Tc 105.4 °C), besylate (46.5 °C), tartrate (80.1 °C) and acetate (78.4 °C); the free base recrystallized at 31.2 °C. No nicardipine salt recrystallized on heating. R4·R5·C10 Figure 1 shows the full sequence for propranolol hydrochloride: a Tg step at about 30–37 °C, a recrystallization exotherm peaking at about 111–116 °C and a melting endotherm at about 164 °C. R3 The authors write that the Tc values of the propranolol salts are all above that of the free base and "scale with pKa", with the hydrochloride highest (105 °C) and the acetate lowest (78 °C). C11 The besylate, however, recrystallized at 46.5 °C, below the acetate, so the pKa order holds only for the hydrochloride, tartrate and acetate. X3 The tabulated Tc and Tm of the hydrochloride (105.4 and 161.6 °C) lie below the peak positions in Figure 1, which suggests onset values, but this is not stated. X7

In humid air. During vapor sorption, propranolol hydrochloride, tosylate and tartrate gained and then lost weight, the signature of crystallization; no nicardipine salt did. R11 Figure 4 shows the profile of propranolol hydrochloride: the uptake rises to about 2.5% at 45% RH, falls to about 1.7% at 95% RH, and about 1.3% remains after desorption to 5% RH; the authors read the loss as recrystallization of the amorphous material. R10·C15 Storage at 79% RH confirmed crystallization of the hydrochloride and the tosylate; Figure 5 shows the tosylate changing from irregular amorphous particles into needle and plate crystals. R12·R13·C16 The tartrate, which also "apparently" crystallized in the sorption run (at 25% RH), showed no crystals after storage; it took up a large amount of water after that event, which "may indicate" deliquescence, and its image at 79% RH (Figure 6) is "consistent with" a deliquesced sample. R11·R12·R14·C17

Table B. Propranolol salts under three stresses (from Tables 2 and 4 and the text on pp. 949–950). For nicardipine, no salt recrystallized on heating or showed the sorption signature, and the oxalate and benzoate were not amorphous after freeze-drying. R4·R5·R8·R11·R12·R14

The authors summarize that the salts of a given drug had varying tendencies to crystallize, that propranolol salts were overall more prone to it than nicardipine salts, and that the response depended on the stress: no single salt crystallized after freeze-drying, on heating and in humid air. C32·C33 They count the tartrate among the salts that crystallized in response to moisture, which their own storage test does not support. C33·X5 Their conclusion is cautious: the counterion affects crystallization tendency, but "more work needs to be performed" to understand why. C34

Moisture sorption varies widely, with no pKa link

Table C. Weight change (%) at 55 and 75% RH, counterions in order of increasing pKa; "—" = not reported (from Tables 4 and 5). R8·R9

Water uptake at 75% RH ranged from 0.39% (tosylate) to 61.90% (acetate) for the propranolol salts and from 4.2% (oxalate) to 12.4% (besylate) for the nicardipine salts. R8·R9 The authors call many of the salts "extremely hygroscopic" and point to propranolol oxalate (below 1% at 75% RH) against the acetate (above 60%). C12 They note that the nicardipine salts "in general" took up less water than the propranolol salts, and find "no apparent link" between counterion pKa and sorption; crystallization during sorption complicates the comparison. C13·C14 In the conclusions they go further: the sorption differences "cannot be linked" to any property they could calculate for the counterions. C14

Note by the AI writer (not discussed in the paper): the lowest propranolol values at 75% RH belong to salts that crystallized during the run (hydrochloride 1.80%, tosylate 0.39%) or to the oxalate (0.50%), so the table mixes amorphous and partly crystalline samples. Tables 4 and 5 are titled "amorphous" salts, but several of the listed salts were not amorphous after freeze-drying, and the paper does not say how they were prepared for sorption. The two free bases also differ in form: crystalline propranolol, amorphous nicardipine. R8·R9·R11·X6

The PLS model

Without the citrate and phosphate salts, the PLS2 model gave $R^2=0.909$ and $Q^2=0.668$, which the authors call "adequate" predictability; the RMSEE was 3.8 °C for the propranolol salts and 3.4 °C for the nicardipine salts, that is, a calibration error of 3–4 °C. R17·C27 The abstract describes the model as having "successfully correlated" Tg with a number of descriptors. C27 Figures 7 and 8 compare observed with predicted Tg (Table D).

Table D. Observed against PLS-predicted Tg (read from Figures 7 and 8, approximate). R18·R19

The model covers 12 salts: seven of propranolol (including the benzoate) and five of nicardipine. The nicardipine benzoate, which has a Tg in Table 3, is missing from Figure 8 although only the citrate and phosphate are said to have been removed. R18·R19·X11 The largest misfits are the propranolol chloride, benzoate and acetate (each about 4 °C off) and the nicardipine chloride and tartrate. R18·R19

Figure 9 shows which descriptors carry the model. The authors give the relative importance of each, measured as the distance of its loading weight from the origin along the direction of the regression coefficient (Table E). A positive value means Tg rises with the descriptor, a negative value that it falls. R20·C28

Table E. Relative importance of counterion descriptors in the PLS2 model, as stated in the text (p.953). "Not stated" means the authors gave no value, not zero. R20

For nicardipine, Tg depends "primarily" on $\mu$ and $I$, and to some degree on volume, pKa and rotatable bonds; for propranolol, mainly on $\omega$, $\mu$ and log D, and to some degree on pKa and $E$. C28 Because pKa carries a negative weight for both drugs, raising the counterion pKa "should" lower the salt Tg. C29 The authors read the model as confirming that many counterion properties "appear to contribute" to Tg; $\omega$, $\mu$ and pKa "can be related to" the strength of the acid–base interaction, while the others are "less intuitive" and need further study. C30·C31

Discussion

The authors' explanation. The argument runs as a chain. Salt formation adds an ionic interaction between the protonated base and the counterion, and for salts this interaction is expected to outweigh the hydrogen bonding and nonspecific forces that govern neutral glasses. C23 Stronger interactions require more thermal energy to reach the mobility of the glass transition, so Tg rises. C21 The authors "believe" that the Tg of the amorphous salts is "strongly affected" by the interaction between the acid and the base, and they used the descriptors to look for correlations that support this hypothesis. C25 A strong acid (low pKa) with a high electrophilicity index and a low chemical potential would, in this reading, bind most strongly and give the highest Tg; the size, shape and packing of the counterion modulate this, which is why the ranking is not linear. C31·C9·C21

Their summary is deliberately broad: the Tg increase over the free base "can be related to a number of factors", and "Broadly speaking, a good approximation" is that counterions with a low pKa (and a high electrophilicity index) give the highest Tg. C35 The abstract states the same in stronger form: important parameters that influenced Tg "included counterion pKa and electrophilicity index". C35

What is measured and what is inferred. The rise of Tg above the free base, its broad fall with counterion pKa, the counterion-dependent crystallization and the scatter of moisture uptake are all measured. R4·R5·R8·R9·R11·R12 The link to an acid–base interaction is an interpretation: no measurement of the interaction in the amorphous salts is reported, and the descriptors are properties of the isolated acid molecules, not of the salt. C25·P10 The explanation for the citrate and phosphate outliers, their trivalent character, is a "probably", and the salt stoichiometry it depends on is not given. C26·X8 The electrophilicity index, named in the abstract and conclusions, is among the main factors only for propranolol; for nicardipine the listed factors are $\mu$, $I$, volume, pKa and rotatable bonds. X13

Where this applies. The authors place amorphous salts in two practical settings: products where a lyophilized form is required, where combining salt formation and amorphization "may be desirable", and crystalline salts in which some amorphous material forms during production or processing. C18 Since Tg governs storage stability, a counterion that raises Tg is of direct interest in both cases. C20 Looking ahead, they suggest that "it may be possible to select a counterion to impart required properties" and so engineer amorphous properties, as is already done for crystalline salts. C36

Note by the AI writer (not discussed in the paper): the data also show the limits of Tg as a single selection criterion. Propranolol hydrochloride, the highest-Tg propranolol salt, crystallized both on heating and at 79% RH, while no nicardipine salt, all with Tg above 28 °C, recrystallized on heating or during sorption. Moisture uptake, which spans more than two orders of magnitude across the propranolol salts, did not follow Tg or pKa. For a formulator, the paper supports choosing a strong-acid counterion to raise Tg, but it gives no rule for hygroscopicity or crystallization, which still have to be measured salt by salt. R4·R5·R8·R9·R11·R12

Limitations

Limits the authors state. The relation between Tg and pKa is "not a linear one", and other factors are "likely" to contribute. C9 Several descriptors in the model are "less intuitive", and "further investigation is required" to understand why they influence Tg. C31 The moisture sorption differences "cannot be linked" to any property they could calculate. C14 For crystallization, "more work needs to be performed" to understand the factors involved. C34

The AI writer's own judgement.

• The PLS model rests on 12 salts and about 17 descriptors. Two outliers were removed, a third salt (nicardipine benzoate) is absent without explanation, and the number of components, scaling and retained descriptors are not printed; read from Figures 7 and 8, the residuals reproduce the printed RMSEE only if two components are assumed. The model was not tested on salts outside the set used to build it. X11·X12
• Table 6, an input to the model, has shifted rows for malonic and tartaric acids and a garbled entry for oxalic acid; which values entered the model is not stated. X9
• Text and tables disagree in several places: the number of salts prepared (20 in the text, 19 identifiable), the lowest-Tg nicardipine salt (acetate in the text, absent from Table 3), the order of the recrystallization temperatures (the besylate breaks it) and the humidity storage time (5 days or 48 h). X1·X2·X3·X4
• The tartrate is listed in the discussion as crystallizing in response to moisture, but the storage test found no crystals and a deliquesced sample. X5
• The base:acid ratio, the Tg convention (onset, midpoint or peak) and the number of replicates are not reported, and some standard deviations are large (nicardipine free base ± 9.6 °C, tartrate ± 8.7 °C). X7·X8·X17
• The origin of the counterion pKa values is not stated, and benzenesulfonic and toluenesulfonic acids, which differ by one methyl group, are assigned pKa values about two units apart; the positions of the besylate and tosylate in Figures 2 and 3 depend on this. X15
• Several salts in the sorption tables were not amorphous after freeze-drying, and how they were prepared for sorption is not stated. X6

Conclusion

This study of amorphous organic salts made up to ten salts each of propranolol and nicardipine by freeze-drying, measured their glass transition, crystallization and moisture uptake, and related Tg to properties of the counterion. R1·R4·R5·R8·R9 Its firmest result is that every salt had a higher Tg than its free base, by a wide margin. C7·R4·R5 Within each drug, Tg fell broadly as the counterion pKa rose, with the hydrochloride highest, and a PLS2 model on counterion descriptors fitted the Tg of 12 salts with a 3–4 °C error, giving most weight to pKa, the electronic chemical potential and, for propranolol, the electrophilicity index. C8·R17·R20 The authors' explanation, that Tg follows the strength of the acid–base interaction, remains an interpretation, and crystallization and moisture uptake depended on the counterion without any rule they could find. C25·C14·C34