- Email: [email protected]
Rapid Throughput Screening of Apparent KSP values for Weakly Basic Drugs Using 96-Well Format
DRUG DISCOVERY INTERFACE Rapid Throughput Screening of Apparent KSP Values for Weakly Basic Drugs Using 96-Well Format JEREMY GUO,1 PAUL. A. ELZINGA,2 MICHAEL. J. HAGEMAN,2 JAMES N. HERRON1 1
Department of Pharmaceutics and Pharmaceutical Chemistry, University of Utah, Salt Lake City, Utah 84112
2
Bristol Myers Sqibb, Princeton, New Jersey
Received 25 June 2005; revised 31 May 2007; accepted 6 July 2007 Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.21149
ABSTRACT: A rapid-throughput screening assay was developed to estimate the salt solubility parameter, KSP, with a minimal quantity of drug. This assay allows for early evaluation of salt limited solubility with a large number of counter-ions and biologically promising drug leads. Drugs dissolved (typically 10 mM) in DMSO are robotically distributed to a 96-well plate. DMSO is evaporated, and drugs are equilibrated with various acids at different concentrations (typically <1 M) to yield final total drug concentrations around 2.5 mM. The plate is checked for precipitation. Filtrates from only those precipitated wells were subjected to rapid gradient HPLC analysis. An iterative procedure is employed to calculate all species concentrations based on mass and charge balance equations. The apparent KSP values assuming 1:1 stoichiometry are determined from counter-ion and ionized drug activities. A correlation coefficient >0.975 for eight drugs totaling 16 salts is reported. Intra-day and inter-day reproducibility was <10%. Conventional apparent KSP measurements were translated to 96-well format for increased throughput and minimal drug consumption (typically 10 mg) to evaluate at least eight different counter-ions. Although the current protocol estimates KSP from 103 to 107 M, the dynamic range of the assay could be expanded by adjusting drug and counter-ion concentrations. ß 2007 Wiley-Liss, Inc. and the American Pharmacists Association J Pharm Sci 97:2080–2090, 2008
Keywords: solubility; solubility product; counter ion; physiochemical properties; excipients; preformulation; formulation; physical characterization; thermodynamics; solvation; dissolution; high throughput technologies
INTRODUCTION It takes over a decade on average for an experimental drug to travel from lab to medicine chest. Only five in 5000 compounds that enter preclinical testing make it to human testing and only Abbreviations: KSP, solubility product; API, active pharmaceutical ingredients. Jeremy Guo’s present address is 1201 Amgen Ct. West, Seattle, WA 98119. Correspondence to: Jeremy Guo (Telephone: 425-444-0301; Fax: 206-217-0491; E-mail: [email protected]) Journal of Pharmaceutical Sciences, Vol. 97, 2080–2090 (2008) ß 2007 Wiley-Liss, Inc. and the American Pharmacists Association
2080
one of these five ends up being approved by the FDA. It costs millions to develop each new drug.1 Rapid screening assays for bioactivity and toxicity employed at the early stages of drug development have been shown to reduce these development costs.1,2 Application of similar screening approaches to physiochemical property evaluations can potentially improve efficiency of drug lead identification, and increase probability of final success. Drug leads are usually selected and designed around specific interaction with a target molecule leading to bioactivity. However, it is often the physiochemical parameters that influence biological
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
RAPID THROUGHPUT SCREENING OF KSP VALUES
behavior, and possible dosage form strategies to provide adequate levels of drug at the target. Unfortunately, the evaluation of physicochemical properties early enough in development to impact lead selection is often limited by drug availability, and time constraints. Thus, physicochemical screening assays are playing an increasingly important role in early stage pharmaceutical discovery, and development.1–4 Solubility is a critical physicochemical factor that can significantly influence biological behavior, ranging from dissolution of a dosage form, and resulting oral absorption of the drug to safety concerns for intravenous administration.5 In the case of oral drug formulations, both the dissolution, and absorption rates of a drug are directly proportional to the solubility of the drug, hence very low solubility can result in poor, and variable release from a dosage form, coupled with inadequate absorption. Precipitation of drug from solubilized formulation has always been a threat for parenteral drugs because of the high volume of dosage form delivered. Introduction of an ionizable group can significantly enhance drug solubility and enables isolation of solid crystalline salts. The solubility limit of the salt is now controlled by the counter-ion and its solubility constant, KSP. Knowledge of apparent KSP values early in lead development can aid salt selection, as well as interpreting biological data generated using such salts.6 Theoretically, free-base compounds may reach very high solubility as the pH of the dissolving media decreases. In practice, however, solubility is limited by the presence of counterions in the acidic medium, resulting in a solubility plateau as the salt form approaches its apparent KSP. Early in the discovery process, in situ salt formation is often used to obtain highly concentrated drug solutions. Consequently, it is important to understand the limits imposed by various counter-ions on maximum obtainable solubility, thus influencing the selection of the acid for pH adjustment. Similarly, solubility, log P, and pKa, along with KSP values, can all influence salt selection for development.7 Finally, the potential for weak bases to interact with endogenous counter-ions (chloride, citrate, etc.) resulting in drug precipitation is a concern during biological evaluations. KSP values of an ionizable drug in various counter-ions are key parameters in making salt selection decisions. We developed a 96-well based screen that can provide apparent KSP values rapidDOI 10.1002/jps
2081
ly for many compounds with minimal amount of material requirement.
MATERIALS AND METHODS Materials and Equipment Terfenadine (98% pure), triameterene (99% pure), phenazopyridine (95% pure), papaverine hydrochloride (98% pure) were purchased from Sigma (St. Louis, MO), and used without further purification. PHA1, PHA2, PHA3, and PHA4 are weakly basic compounds synthesized by Pharmacia Medicinal Chemistry, listed in Table 1. Acetic acid, citric acid, hydrochloric acid, methanesulfonic acid, sulfuric acid, and phosphoric acid were purchased from Mallinckrodt (Paris, KY). Succinic acid and formic acid were purchased from Aldrich (Milwaukee, WI). A Biomek 2000 laboratory automation workstation (Beckman Instruments, Inc., Fullerton, CA) was used in liquid dispensing. Drug concentration assays were performed by HPLC using an Agilent model 1100 (Santa Clara, CA) equipped with binary pump (G1312A), autosampler (G1313A), column compartment (G1316A), and photodiode array detector (G1315A). A liquid handler (Gilson, model 215, Middleton, WI) was used to inject samples directly from a 96-well plate. DMSO solvent was evaporated by centrifugal evaporation (Genevac Technologies, model HT-4, Valley Cottage, NY). Titer plate shaker (Lab-line Instruments, Barnstead International, Dubuque, IA) accompanied by coated parylene stir bars (V&P Scientific, San Diego, CA) were used in equilibration. Millipore multiscreen 96-well filter plates (0.4 mm, model MAHVN4550) were used for filtration. P250, and P20 pipette tips from Beckman Instruments, Inc., and/or Molecular Bioproducts, Inc. (San Diego, CA). A Gilson injector was used in this study for autosampling from 96-well plates. It exhibited a dispensing variability of 4–8%, which was deemed acceptable for our rapid throughput-screening assay. Drug Plate Preparation Drugs were dissolved in DMSO at a high concentration (typically 10 mM). Drug stocks were robotically (Biomek 2000) distributed into 96-well flat bottom polystyrene plates (Costar 3595, Corning, NY). DMSO was then evaporated (Genevac, typical run time is 2 h) and the drug JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2082
GUO ET AL.
Table 1. Structures of Investigative Compounds, PHA1 to PHA 4 Compound No.
Structure
Mol. Wt.
Estimated pKa
PHA 1
425.91
6.8
PHA 2
376.91
6.15
PHA 3
453.93
6
PHA 4
383.47
6.2
re-dissolved in acid containing solution. For example, spotting 37.5 mL of a 10 mM stock solution, followed by evaporation, and re-dissolution in 150 mL of acid yields a maximum final total drug concentration of 2.5 mM. Because of higher acid concentrations, the amount of weakly basic drugs in each well should exist primarily in their ionized form.
added to each well of a control plate (plate C) and weighed. All three plates were evaporated over a 4 h period and weighed at different time points (0.5, 1.0, 1.5, 2.0, 2.5, 3, and 4 h). The amount of DMSO remaining after each evaporation time point was calculated from the change in weight.
Acid Plate Preparation DMSO Evaporation The extent of DMSO evaporation was determined by mass measurements. In particular, a 150 mL volume of DMSO solution containing 10 mM concentration of a highly soluble drug (antipyrine) was added to each well of a 96-well plate (plate A) and weighed. A 150 mL volume of DMSO solution containing 10 mM concentration of a poorly watersoluble drug (danazol) was then added to each well of a second plate (plate B) and weighed. In addition, 150 mL of DMSO without drug was JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
Selected strong acids included HCl, H2SO4, and methane sulfonic acid; weak acids included acetic acid, succinic acid, citric acid, and phosphoric acid. Since weakly basic drugs screened in this experiment all had pKa values at least two units higher than the pKa of selected acids (except for acetic acid), the majority of drugs were expected to be greater than 99% ionized. Thus, the final KSP was relatively insensitive to the precise pKa value used. Acid solution plates were prepared robotically (Biomek 2000) by serial dilution (1:2) from a high concentration (typically 1M) acid stock. Each DOI 10.1002/jps
RAPID THROUGHPUT SCREENING OF KSP VALUES
acid solution plate may hold up to 16 different acids arranged by row (2 acids per row) and five concentrations, 0.5, 0.25, 0.0125, 0.00625, and 0.03125 M, per acid arranged by column across the plate. Salt Precipitation The five acid concentrations were robotically (Biomek) transferred (150 mL/well) from the acid plate to the evaporated drug plate, yielding a nominal drug concentration of 2.5 mM (10 mM total concentration of drug initially). Mini stir bars (V&P Scientific) were added to each well and the drug plate was shaken (Lab-Line Instruments, speed ¼7 ) for 24 h. After shaking, the stir bars were removed and in some cases, the pH of each well was measured during early validation studies (ATI, Collegeville, PA). Turbidity was used as a semi-quantitative prescreening step to shorten the assay time. Assuming a molecular weight of 400, the 2.5 mM nominal drug concentration would give a solubility of 1 mg/mL if all drug was dissolved. If there was no precipitation with a selected acid, then the apparent KSP was not reached, and HPLC analysis was not necessary, even though samples were still filtered. Turbidity was performed by an absorbance plate reader at 650 nm wavelength. An absorbance reading greater than that observed for blanks indicating precipitation. Each plate was then filtered through a 96-well flat bottom Costar filter plate (Biomek 2000) into a 96-well round bottom plate (Falcon 1190, BD, Franklin Lakes, NJ). The filtrate was then diluted 10 by transferring (Biomek 2000) 20 mL from each well to another 96-well round bottom polypropylene plate (Falcon 1190), and diluting with 180 mL acetonitrile:water (50:50). The plate was then sealed for HPLC analysis.
HPLC Analysis Sample and standard plates were assayed by a generic fast gradient HPLC method using a chromatography system (Agilent 1100) with a diode array detector and a short C8 (Zorbax Agilent Technologies, Santa Clara, CA) column. The flow rate was 1 mL/min with a gradient ramp from 95% mobile phase A (H2O with 0.1% TFA) to 95% mobile phase B (acetonitrile with 0.01% TFA) over a 1 min period, followed by 95% B for 45 s and re-equilibration in 95% A for 30 more seconds. The total assay time was 4 min per sample. Detection DOI 10.1002/jps
2083
utilized a photodiode array with routine quantitation possible at 214, 230, 254, 280, and 295 nm. Drug standards were prepared at 1 mg/mL by dissolving a weighed sample of each drug into a known volume of acetonitrile:water (50:50). Additional standards at 0.5, 0.2, 0.1, 0.05, 0.02, and 0.01 mg/mL were prepared by dilution of the 1 mg/mL standard with acetonitrile:water (50:50). In cases where 1 mg/mL standards were not soluble in acetonitrile:water, DMSO was used instead. Sample drugs that have precipitations as indicated by turbidity assay are diluted in DMSO for HPLC. Each concentration was loaded into a 96-well round bottom plate (Falcon 1190) for HPLC analysis. A standard curve was constructed from the drug standards (linear regression forced through the origin) and used to determine the solution concentration of the sample. The major peak of each injection was selected consistent with retention time of the standards, typically using the response at 254 nm. Sample concentration was determined using the peak area and the standard curve. Apparent KSP Calculation An Excel spreadsheet was used for calculations and to carry out numerical equation solving procedures. The apparent KSP was determined from the saturated solution concentrations of the ionized drug and counter-ion, along with a calculation of solution pH. As shown in Eq. (1), the apparent KSP was determined by activity of ionized drug, and counter-ion assuming 1:1 salt formation DHþ þ Hn1 A ! DHn A; ppt
KSP ¼ aDHþ aHn1 A
(1)
where aDHþ was the activity of species DHþ. The activity was given by: aDHþ ¼ g DHþ ½DHþ
(2)
where gDH was the activity coefficient and [DHþ] was the molar concentration of species DHþ. The activity coefficient was assumed to be primarily a function of molecular charge with neutral species assumed to have g0 ¼ 1. Activity coefficient was calculated using the ionic strength (m) of each solution determined from: m¼
1X 2 ci zi 2
(3)
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2084
GUO ET AL.
where ci was the concentration and zi was the charge of species i, and the summation was carried out over all charged species in the solution. The activity coefficient (g) was estimated from the Davies equation:4 pffiffiffiffi Az2 m log g ¼ (4) pffiffiffiffi þ Cm 1 þ aB m Typical values were used for the constants: A 0.5, aB 1, C 0.2. The ionic strength and activity coefficients were recalculated at each iteration as described below.
pH Calculation An iterative procedure was employed to calculate the various species concentrations. It was initiated with an initial guess of the solution pH from which the concentration of all drug and acid species can be calculated. The OH concentration was then calculated independently from both the water dissociation constant Kw (designated as [OH]Kw) and the charge balance (denoted [OH]bal) as shown in Eqs. (5) and (6), below. Water Dissociation: Kw ¼ aH3 Oþ aOH ;
½OH Kw ¼
Kw (5) g 21 ½H3 Oþ
½OH bal ¼ ½DHþ þ ½H3 Oþ ½Hn1 A 2 (6)
If the two calculated [OH] values vary significantly, then another iteration was required using a new guess for pH given by: ½H3 Oþ nþ1 ¼ ½H3 Oþ n dD
(7)
where D was the difference in [OH] calculated using the charge balance and Kw methods: D ¼ ½OH bal ½OH Kw
The activity of a given protonated drug species (aDHþ ) is given by Eq. (10): aDHþ ¼
aD aH3 O þ KaD
(10)
which can be derived from mass balance Eqs. (11) and (12) Cmeas ¼ Csoln ¼ ½D þ ½DHþ ¼ ½D D D ½H3 Oþ ¼ ½D 1 þ KaD
½D½H3 Oþ KaD (11)
where CD is the total concentration of drug. The concentration of the unprotonated drug species is given by Eq. (12): ½DHþ ¼ Cmeas ½D ¼ Cmeas D D
Cmeas D 1 þ ½HK3 OD
þ
(8)
and d is used as a scale factor to prevent iterative steps from getting too large (typically d 2): ½H3 Oþ nþ1 ½H3 Oþ n d ¼ ð9Þ Dnþ1 Dn JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
(12)
a
Acid Species Since each drug molecule that precipitates also involves a counter-ion, the final solution concentration of the acid is given by: Total Total Cmeas Þ Csoln Hn A ¼ CHn A ðCD D
¼ ½Hn A þ ½Hn1 A þ þ ½An
Charge Balance:
½Hn2 A2 n ½An
Drug Species
(13)
General equations for multibasic drugs and multiprotic acids can be found in existing literature.10 Examples of one strong acid and one weak diprotic acid are illustrated below. Monoprotic Strong Acid (Example: Hydrochloric Acid) In this case all of the acid will dissociate, hence: ½A ¼ Csoln HA
(14)
The initial pH estimate and Ksp are given by: ½H3 Oþ 0 ¼ Csoln HA
(15)
Ksp ¼ aDHþ aA ¼ g 21 ½DHþ ½A
(16)
where aDHþ was obtained from Eq. (10). DOI 10.1002/jps
RAPID THROUGHPUT SCREENING OF KSP VALUES
RESULTS AND DISCUSSION
Diprotic Weak Acid (Example: Succinic Acid) H A
H2 A þ H2 O
Ka 2
! H3 Oþ þ HA aHA ¼
KaHa A aH2 A aH3 Oþ (17)
HA þ H2 O
aA2 ¼
KaHA
! H3 Oþ þ A2
KaHA aHA KaHA KaH2 A aH2 A ¼ aH3 Oþ aH3 Oþ
(18)
Substitutions of Eqs. (17) and (18) for [HA] and [A2] to get mass balance Eq. (19) as shown below 2 CSolu H2 A ¼ ½H2 A þ ½HA þ ½A
KaH2 A ½H2 A KaHA KaH2 A ½H2 A þ g 21 ½H3 Oþ g 2 g 21 ½H3 Oþ 2 ! KaH2 A KaHA KaH2 A ¼ ½H2 A 1 þ 2 þ g 1 ½H3 Oþ g 2 g 21 ½H3 Oþ 2 ¼ ½H2 A þ
(19) Thus, [H2A] can be obtained from Eq. (20) ½H2 A ¼
Csoln H2 A H2 A K H2 A K H2 A Ka1 1 þ 2 a1 þ þ a2 2 g 1 ½H3 O g 2 g 1 ½H3 Oþ
(21)
(22)
Calculations of monoprotic and triprotic acids would be similar to that of diprotic acid. Thus, both drug, and acid species could be set up in an Excel spreadsheet, and the calculations could be automated.
Reproducibility of the 96-well KSP Screening Assay With selected drugs, reproducibility of the assays was described by two different sets of experiments on intra-day, and inter-day basis. Selected drugs were equilibrated in various acid solutions and subjected to HPLC analysis on same day and same plate with the same standard curve. Similarly, selected drugs were subjected to HPLC analysis on different day, and different plate with different standard curves, and freshly prepared standards. DOI 10.1002/jps
Drugs were initially dissolved in DMSO, which was later evaporated. Residual DMSO after evaporation might enhance the overall drug solubility. This possibility was investigated by dissolving a highly soluble drug, antipyrine, and a poorly soluble drug, danazol in 150 mL DMSO, respectively. Highly soluble and poorly soluble drugs were chosen to examine whether the intrinsic solubility of the drug will have any DMSO retention effect. Mass analysis after evaporation indicated that there was less than 1 mL of DMSO remaining in each well (on the average), irrespective of the intrinsic solubility of the drug. Following addition of 150 mL of selected acid, the final volume of DMSO in the reconstituted solution is expected to be less than 0.7% (v/v). The effect of this small amount of DMSO on overall drug solubility was expected to be minimal and acceptable from a screening assay perspective.
pH Validation
Then, KSP can be calculated from Eq. (22) KSP ¼ aDHþ aHA ¼ g 21 ½DHþ ½HA
DMSO Effects on Drug Solubility
(20)
The pH of diprotic acid solutions is estimated by an iterative, numerical procedure based on Eqs. (6) and (17)–(19). The initial pH guess is given by Eq. (21): H2 A H2 A 1=3 ½H3 Oþ 0 ¼ ðCsoln H2 A Ka1 Ka2 Þ
2085
The pH of the equilibrium drug solution is required for apparent KSP calculation as noted in Eqs. (5)–(9) and (21) in Materials and Methods section. While it is a simpler approach, the time, and labor required to measure the pH in every well would have significantly decreased the assay throughput. Hence, an iterative computational approach was developed to determine pH of the equilibrium drug solution after inputting an initial guess of pH. To validate the calculated pH, the pH in all wells of several plates was measured. Figure 1 is a correlation plot between the manually measured pH and the calculated pH for 8 different salts of PHA1. The correlation coefficient for all salts is greater than 0.98 with the exception of acetic acid. This deviation is likely due to the pKa values of the acetic acid, and the PHA1, 4.76, and 6.0, respectively, being close to each other, and the resulting pH being in this range as well, all leading to a high level of sensitivity of the iteration process to the accuracy of the values. The pH deviations were minimal in the other acid solutions because of much lower pH that was far away from the drug’s true pKa. Less extensive examples with other drugs were also evaluated and comparisons made between the manually measured pH and the spreadsheetJOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2086
GUO ET AL.
Figure 1. Correlation between measured and calculated pH values. pH values were measured for eight different salts of PHA1 in precipitated wells. These values were then correlated with pH values calculated using Eqs. (5)–(9). Data were well correlated, giving a correlation coefficient value of 0.98, when all salts except for acetic acid were included in the comparison (pH values calculated for the acetate salt are thought to deviate from measured values because PHA1 and acetic acid have nearly identical pKa values).
calculated pH, reporting the correlations coefficients with similar results. The correlation demonstrated that the derived equations used in the spreadsheet were adequate. The results also confirmed that manual measurements of pH were not necessary if pKa of the acid is expected to be two units below the pKa of the weak base. Turbidity Study Minimizing time and materials were the primary goals of this screening assay. A semi-quanitative approach was adopted to differentiate soluble compounds from compounds with poor solubility so that time-consuming HPLC would only be needed for compounds with poor solubility. Concentration of 2.5 mM or 1 mg/mL (assuming an average molecular weight of 400) aqueous solubility was typically used as the upper limit. Turbid wells were subject to filtration, and HPLC analysis, and non-turbid wells were reported as having solubility greater than 1 mg/mL. Given the flexibility, and easy maneuverability of this 96well format, users could adjust the differentiation point, and expand the dynamic range accordingly. JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
KSP Determination Apparent KSP values were evaluated for wells with precipitation. As shown in Eq. (1), the apparent KSP was determined by the activity of the ionized drug, and counter-ion assuming 1:1 salt formation. Of the eight compounds screened, apparent KSP values were calculated, and compared with either literature KSP values6,8–12 or internally reported KSP values. Reference KSP values were obtained from the literature or Pfizer Corporation research reports. Most literature reported apparent KSP values are determined at a single specified solution concentration. The rapid throughput capacity of the 96-well format allowed calculations of the apparent KSP values for a given salt at five different concentrations of the counter-ion. Table 2 summarizes the apparent KSP values determined from five concentrations (0.5, 0.25, 0.125, 0.0625, and 0.03125 M) of each acid for eight compounds. It was observed that the apparent KSP value increased with increasing concentration of the counter ions. Table 2, Figures 2 and 3A and B showed that the apparent DOI 10.1002/jps
RAPID THROUGHPUT SCREENING OF KSP VALUES
2087
Table 2. Summary of KSP Values Between Reference and Determined by the Screening Assay for 8 Drugs and 16 Salts
Compounds
Acids
Terfenadine
Triamterene Phenazopyridine Papaverine Hydrochloride PHA1
PHA2 PHA3 PHA4
Reference Ksp4,6,8–13
Ksp Range
Ksp Determined at Low Ionic Strength, 0.25 M
Hydrochloric acid Methanesulfonic acid Phosphoric acid Hydrochloric acid Phosphoric acid Hydrochloric acid Hydrochloric acid
5
1.58 10 1.87 105 4.99 106 2.13 104 2.01 105 1.27 104 2.24 103
8.06 106 1.16 105 4.12 106 2.02 104 1.06 105 7.26 105 2.8 103
to to to to to to to
1.95 105 3.01 105 5.16 106 2.68 104 3.72 105 2.56 104 3.6 103
1.36 105 2.47 105 4.21 106 2.2 104 3.71 105 2.5 104 3.6 103
Hydrochloric acid Succinic acid Citric acid Hydrochloric acid Succinic acid Hydrochloric acid Citric acid Hydrochloric acid Hydrobromic acid
5.45 105 7.99 106 7.8 105 1.84 105 1.67 105 4.4 105 2.8 106 2.6 103 1.98 104
3.43 105 4.90 106 2.57 105 2.67 105 1.22 105 1.34 105 2.28 106 1.34 103 1.31 104
to to to to to to to to to
1.19 104 1.93 105 1.23 104 5.80 105 2.39 105 1.05 104 9.44 106 1.76 103 2.53 103
5.75 105 1.44 105 5.45 105 2.77 105 1.47 105 3.41 105 2.63 106 1.36 103 1.76 104
KSP measured at different acid concentrations gave a slightly different apparent KSP value. The highest counter-ion concentration (0.5 M) usually gave the highest value; the lowest counter-ion concentration gave the lowest apparent KSP, thus resulted in a range of apparent KSP reported here. A typical trend was shown in Figure 2 for PHA1, the apparent KSP values increased with increasing acid concentrations. The slope was counter-ion specific, the trend existed over a 15-fold counterion concentration range with less than one magnitude difference between the highest and the lowest KSP. This trend existed even when the
assay was done in vials instead of plates indicating that an experimental bias due to the screening method was not occurring. Serajuddin11 also made similar observations through conventional methodology, in which KSP values did fluctuate with different pH, counter-ion concentration and ionic strength. The systematic trend seemed to disagree with the fact that theoretical KSP should be a constant value, even multiple concentrations of acid should render a constant KSP value. However, the KSP calculated are the apparent values, not the true KSP values, the true KSP value was dependent on
Figure 2. Counter-ion effects on apparent KSP values. The apparent KSP value of was plotted versus counter ion concentration, for the HCl, citric acid, and succinic acid salts of PHA1. Apparent KSP values showed an increasing trend proportional to increasing counter-ion concentration in all three cases. Nevertheless, the assay still allowed the ranking of salts forms by KSP values so that an appropriate salt was selected. DOI 10.1002/jps
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2088
GUO ET AL.
Figure 3. (A and B) Correlation of reference KSP values with apparent KSP values determined at lower acid concentrations. The reference KSP value was compared with apparent KSP values determined at lower acid concentrations (0.25, 0.125, 0.0625, and 0.03125 M), where the Davies equation adequately describes activities between drug species and counter ions. Panel A shows a correlation between reference KSP and KSP determined at lowest counter-ion concentration (0.03125 M) or n ¼ 1, similarly lowest ionic strength, and closest to the true KSP value. A correlation coefficient of 0.979 is returned. Panel B shows a correlation between reference KSP and average KSP values of (0.25, 0.125, 0.0625, and 0.03125 M) or n ¼ 4 determined at low counter-ion concentrations, and low ionic strength. A correlation coefficient of 0.981 is returned. The data shows that KSP values determined from the 96-well format screening assay at different but <0.25 M concentrations are not significantly different from the true KSP value.
activities between drugs and counter-ions, which in turn depend on ionic strength and pH of the acids. All these variables changed simultaneously in our method, it was difficult to pinpoint the contributing variable. The Davies equation used in the calculations might be inadequate to describe these activities because Davies equation theoretically only applies to molar concentrations less than 0.2 M. The highest concentration (0.5 M) JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
used in this assay far exceeded the theoretical limit of the Davies equation, thus giving a higher than expected KSP value. When KSP determined at 0.5 M acid concentration was compared with reference KSP value, a correlation coefficient of 0.91 is returned. Similarly when KSP of the lowest concentration of 0.03125 M is compared, a much greater correlation coefficient of 0.979 is obtained as shown in Figure 3A. To comply with the Davies DOI 10.1002/jps
RAPID THROUGHPUT SCREENING OF KSP VALUES
equation, the highest concentration 0.5 M was not included in KSP average calculations, the average KSP includes only those determined at 0.25, 0.125, 0.0625, and 0.03125 M concentrations, and returns a correlation coefficient of 0.980 (Fig. 3B). The KSP value determined from the low ionic strength was more representative of the true KSP value. As shown in Table 2, the KSP value determined at low ionic strength was closer to the reported literature KSP values, usually below 0.1 M.6,8–12 Literature KSP values always fell within our reported range of apparent KSP values. In fact, the range of the screening apparent KSP values is generally less than 1 order of magnitude. Determination of the true KSP value was not the purpose of this 96-well screening assay, rather, the assay was to allow proper ranking of different salt forms. For instance, Figure 3 clearly demonstrated that the KSP value of PHA1 with succinic acid was much lower than that with HCl, and Citric acid. All reference KSP values were within half an order magnitude of the apparent KSP determined here. From a screening perspective, this discrepancy was small enough to allow adequate KSP ranking at the early development stage. Our results suggest that counter ion concentrations up to 0.25 M are more than sufficient to rank apparent KSP values for different salts. The Davies equation is also valid across this concentration range, leading up to 0.25 M. However, one needed to confirm that KSP was reached at the low concentration by exhibiting precipitation before HPLC analysis. Turbidity studies would provide such confirmation in term of pre-
2089
cipitation observation. In cases with no precipitation, KSP was not reached; a higher counter-ion concentration would be needed to hit KSP. This 96-well format consumes minimal amount of materials, but assists the formulation, and discovery scientists to identify a feasible salt form (from a solubility perspective) in a rapid throughput manner. It should be recognized that salt solubility is only one criteria of salt selection, and that properties such as hygroscopicity, ease of preparation, stability, and mechanical properties. Reproducibility Inter- and intra-day assay reproducibility is assessed in Table 3 for Terfenadine and PHA1. Both are weakly basic drugs with pKa values of 6, and 6.8, respectively. These values are at least two units above the pKa values of the selected acids, hydrochloric acid (Strong acid), methanesulfonic acid (Strong acid), phosphoric acid (1st pKa 2.12), acetic acid (pKa 4.76), succinic acid (1st pKa 4.19), and citric acid (1st pKa 3.15). Additionally, a number of reported KSP values in different salt forms are available for comparison. These drugs have relatively low KSP values, thus only minimal amount drug materials will be needed for KSP determination. These physiochemical characteristics make them ideal candidates for the 96-well format solubility-screening assay. They are representative of the class of insoluble, weakly basic drugs this assay is aimed at. Terfenadine was equilibrated with three different acids and KSP was obtained on different days for respective salts,
Table 3. Reproducibility of the 96-Well Screening Assay on an Inter-Day, Inter-Plate and Intra-Day and Intra-Plate KSP Terfenadine Inter-day and inter-plate Hydrochloride salt Methanesulfonic acid Phosphoric acid
Day1
Day 2
Day 3
Day 4
1.36 105 2.47 105 4.21 105
1.27 105 3 105 4.52 106
1.38 105 N/A N/A
1.51 105 N/A N/A
PHA1
Replicate 1
Replicate 2
Intra-day and intra-plate Acetic acid Citric acid Sulfuric acid Phosphoric acid Hydrochloride Succinic acid
6.17 106 1.53 105 3.15 105 3.1 105 3.89 105 6.21 106
5.98 106 1.62 105 2.01 105 3.38 105 5.7 105 6.19 106
DOI 10.1002/jps
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2090
GUO ET AL.
and KSP for PHA1 was calculated on same day, and same plate in duplicates. On an inter-day, and inter-plate basis, there was less than 10% relative standard deviation for n 3, and less than 5% for duplicates. On an intraday, and intra-plate basis, the reproducibility was also well within 10% for duplicates. Overall, the reproducibility of the KSP screening assay was acceptable from a screening perspective, especially when considering the Gilson injector used in the assay has an intra-day variability of 4–8%.
CONCLUSIONS The 96-well plate format greatly increased throughput of screening assays, and reduced time, and material constraints at the early stage of drug development. Because of the competitive nature of the pharmaceutical industry, early participation of formulation scientists in the drug development process is becoming increasingly important. Physiochemical parameters need to be evaluated in parallel with biological parameters to maximize efficiency and productivity. Our assay evaluates KSP values at multiple concentrations of acid with very little a priori information about the weak base compound. If KSP values for other compounds in a template series are known, the number of acid concentrations can be significantly reduced. As mentioned previously, the best estimates are obtained from the lowest acid concentration that results in precipitation of the salt. Furthermore, by incorporating the turbidity assessment into the workflow, only the lowest acid concentrations with precipitation can be selected for HPLC analysis. Estimation of KSP typically required 10 mg of active pharmaceutical ingredient (API) to screen eight salts using five acid concentrations. The 96-well format offers the flexibility of screening i drugs in j salts at k counter-ion concentrations as long as i j k ¼ 96. Semi-automated data analysis limits assay throughput to be about 25 compounds per week. In the example of the apparent KSP screening, the screening KSP values were compared with the reported KSP result; generally results were within half an order of magnitude. This accuracy was acceptable for a screening study. Apparent KSP values were reasonably constant if determined at modest ionic strengths when Davies equation was used. At higher ionic strengths, a trend towards increasing KSP with increasing ionic strength was noted. JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
ACKNOWLEDGMENTS Financial support for Jermey Guo was provided by a Pfizer internship program. Additional support was provided by University of Utah research development funds.
REFERENCES 1. Michelson S, Joho K. 2000. Drug discovery, drug development and the emerging world of pharmacogenomics: Prospecting for information in a datarich landscape. Curr Opin Mol Ther 2:651–654. 2. Wu G, Irvine J, Luft C, Pressley D, Hodge CN, Janzen B. 2003. Assay development and highthroughput screening of caspases in microfluidic format. Comb Chem High Throughput Screen 6: 303–312. 3. Yu K, Balogh M. 2001. A protocol for high-throughput drug mixture quantitation: Fast LC-MS or flow injection analysis-MS? LCGC 19:60–66. 4. Tong WQ, Whitesell G. 1998. In situ salt screening—a useful technique for discovery support and preformulation studies. Pharm Dev Technol 3:215–223. 5. Alfred Martin JS, Cammarata A. 1983. Phys Pharm 272–313. 6. Lin SL, Lachman L, Swartz CJ, Huebner CF. 1972. Preformulation investigation. I. Relation of salt forms and biological activity of an experimental antihypertensive. J Pharm Sci 61:1418–1422. 7. Kibbey CE, Poole SK, Robinson B, Jackson JD, Durham D. 2001. An integrated process for measuring the physicochemical properties of drug candidates in a preclinical discovery environment. J Pharm Sci 90:1164–1175. 8. Dittert LHT, Reese D. 1964. Phase solubility technique in studying the formation of complex salts of triamterene. J Pharm Sci 53:1325–1329. 9. Kramer SF, Flynn GL. 1972. Solubility of organic hydrochlorides. J Pharm Sci 61:1896–1904. 10. Streng WH, Hsi SK, Helms PE, Tan HG. 1984. General treatment of pH-solubility profiles of weak acids and bases and the effects of different acids on the solubility of a weak base. J Pharm Sci 73:1679– 1684. 11. Serajuddin AT, Jarowski CI. 1985. Effect of diffusion layer pH and solubility on the dissolution rate of pharmaceutical bases and their hydrochloride salts. I. Phenazopyridine. J Pharm Sci 74:142–147. 12. Miyazaki S, Nakano M, Arita T. 1975. A comparison of solubility characteristics of free bases and hydrochloride salts of tetracycline antibiotics in hydrochloric acid solutions. Chem Pharm Bull 23:1197– 1204. 13. Sacchetti M. 2000. General equations for in situ salt screening of multibasic drugs in multiprotic acids. Pharm Dev Technol 5:579–582. DOI 10.1002/jps
Department of Pharmaceutics and Pharmaceutical Chemistry, University of Utah, Salt Lake City, Utah 84112
2
Bristol Myers Sqibb, Princeton, New Jersey
Received 25 June 2005; revised 31 May 2007; accepted 6 July 2007 Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/jps.21149
ABSTRACT: A rapid-throughput screening assay was developed to estimate the salt solubility parameter, KSP, with a minimal quantity of drug. This assay allows for early evaluation of salt limited solubility with a large number of counter-ions and biologically promising drug leads. Drugs dissolved (typically 10 mM) in DMSO are robotically distributed to a 96-well plate. DMSO is evaporated, and drugs are equilibrated with various acids at different concentrations (typically <1 M) to yield final total drug concentrations around 2.5 mM. The plate is checked for precipitation. Filtrates from only those precipitated wells were subjected to rapid gradient HPLC analysis. An iterative procedure is employed to calculate all species concentrations based on mass and charge balance equations. The apparent KSP values assuming 1:1 stoichiometry are determined from counter-ion and ionized drug activities. A correlation coefficient >0.975 for eight drugs totaling 16 salts is reported. Intra-day and inter-day reproducibility was <10%. Conventional apparent KSP measurements were translated to 96-well format for increased throughput and minimal drug consumption (typically 10 mg) to evaluate at least eight different counter-ions. Although the current protocol estimates KSP from 103 to 107 M, the dynamic range of the assay could be expanded by adjusting drug and counter-ion concentrations. ß 2007 Wiley-Liss, Inc. and the American Pharmacists Association J Pharm Sci 97:2080–2090, 2008
Keywords: solubility; solubility product; counter ion; physiochemical properties; excipients; preformulation; formulation; physical characterization; thermodynamics; solvation; dissolution; high throughput technologies
INTRODUCTION It takes over a decade on average for an experimental drug to travel from lab to medicine chest. Only five in 5000 compounds that enter preclinical testing make it to human testing and only Abbreviations: KSP, solubility product; API, active pharmaceutical ingredients. Jeremy Guo’s present address is 1201 Amgen Ct. West, Seattle, WA 98119. Correspondence to: Jeremy Guo (Telephone: 425-444-0301; Fax: 206-217-0491; E-mail: [email protected]) Journal of Pharmaceutical Sciences, Vol. 97, 2080–2090 (2008) ß 2007 Wiley-Liss, Inc. and the American Pharmacists Association
2080
one of these five ends up being approved by the FDA. It costs millions to develop each new drug.1 Rapid screening assays for bioactivity and toxicity employed at the early stages of drug development have been shown to reduce these development costs.1,2 Application of similar screening approaches to physiochemical property evaluations can potentially improve efficiency of drug lead identification, and increase probability of final success. Drug leads are usually selected and designed around specific interaction with a target molecule leading to bioactivity. However, it is often the physiochemical parameters that influence biological
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
RAPID THROUGHPUT SCREENING OF KSP VALUES
behavior, and possible dosage form strategies to provide adequate levels of drug at the target. Unfortunately, the evaluation of physicochemical properties early enough in development to impact lead selection is often limited by drug availability, and time constraints. Thus, physicochemical screening assays are playing an increasingly important role in early stage pharmaceutical discovery, and development.1–4 Solubility is a critical physicochemical factor that can significantly influence biological behavior, ranging from dissolution of a dosage form, and resulting oral absorption of the drug to safety concerns for intravenous administration.5 In the case of oral drug formulations, both the dissolution, and absorption rates of a drug are directly proportional to the solubility of the drug, hence very low solubility can result in poor, and variable release from a dosage form, coupled with inadequate absorption. Precipitation of drug from solubilized formulation has always been a threat for parenteral drugs because of the high volume of dosage form delivered. Introduction of an ionizable group can significantly enhance drug solubility and enables isolation of solid crystalline salts. The solubility limit of the salt is now controlled by the counter-ion and its solubility constant, KSP. Knowledge of apparent KSP values early in lead development can aid salt selection, as well as interpreting biological data generated using such salts.6 Theoretically, free-base compounds may reach very high solubility as the pH of the dissolving media decreases. In practice, however, solubility is limited by the presence of counterions in the acidic medium, resulting in a solubility plateau as the salt form approaches its apparent KSP. Early in the discovery process, in situ salt formation is often used to obtain highly concentrated drug solutions. Consequently, it is important to understand the limits imposed by various counter-ions on maximum obtainable solubility, thus influencing the selection of the acid for pH adjustment. Similarly, solubility, log P, and pKa, along with KSP values, can all influence salt selection for development.7 Finally, the potential for weak bases to interact with endogenous counter-ions (chloride, citrate, etc.) resulting in drug precipitation is a concern during biological evaluations. KSP values of an ionizable drug in various counter-ions are key parameters in making salt selection decisions. We developed a 96-well based screen that can provide apparent KSP values rapidDOI 10.1002/jps
2081
ly for many compounds with minimal amount of material requirement.
MATERIALS AND METHODS Materials and Equipment Terfenadine (98% pure), triameterene (99% pure), phenazopyridine (95% pure), papaverine hydrochloride (98% pure) were purchased from Sigma (St. Louis, MO), and used without further purification. PHA1, PHA2, PHA3, and PHA4 are weakly basic compounds synthesized by Pharmacia Medicinal Chemistry, listed in Table 1. Acetic acid, citric acid, hydrochloric acid, methanesulfonic acid, sulfuric acid, and phosphoric acid were purchased from Mallinckrodt (Paris, KY). Succinic acid and formic acid were purchased from Aldrich (Milwaukee, WI). A Biomek 2000 laboratory automation workstation (Beckman Instruments, Inc., Fullerton, CA) was used in liquid dispensing. Drug concentration assays were performed by HPLC using an Agilent model 1100 (Santa Clara, CA) equipped with binary pump (G1312A), autosampler (G1313A), column compartment (G1316A), and photodiode array detector (G1315A). A liquid handler (Gilson, model 215, Middleton, WI) was used to inject samples directly from a 96-well plate. DMSO solvent was evaporated by centrifugal evaporation (Genevac Technologies, model HT-4, Valley Cottage, NY). Titer plate shaker (Lab-line Instruments, Barnstead International, Dubuque, IA) accompanied by coated parylene stir bars (V&P Scientific, San Diego, CA) were used in equilibration. Millipore multiscreen 96-well filter plates (0.4 mm, model MAHVN4550) were used for filtration. P250, and P20 pipette tips from Beckman Instruments, Inc., and/or Molecular Bioproducts, Inc. (San Diego, CA). A Gilson injector was used in this study for autosampling from 96-well plates. It exhibited a dispensing variability of 4–8%, which was deemed acceptable for our rapid throughput-screening assay. Drug Plate Preparation Drugs were dissolved in DMSO at a high concentration (typically 10 mM). Drug stocks were robotically (Biomek 2000) distributed into 96-well flat bottom polystyrene plates (Costar 3595, Corning, NY). DMSO was then evaporated (Genevac, typical run time is 2 h) and the drug JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2082
GUO ET AL.
Table 1. Structures of Investigative Compounds, PHA1 to PHA 4 Compound No.
Structure
Mol. Wt.
Estimated pKa
PHA 1
425.91
6.8
PHA 2
376.91
6.15
PHA 3
453.93
6
PHA 4
383.47
6.2
re-dissolved in acid containing solution. For example, spotting 37.5 mL of a 10 mM stock solution, followed by evaporation, and re-dissolution in 150 mL of acid yields a maximum final total drug concentration of 2.5 mM. Because of higher acid concentrations, the amount of weakly basic drugs in each well should exist primarily in their ionized form.
added to each well of a control plate (plate C) and weighed. All three plates were evaporated over a 4 h period and weighed at different time points (0.5, 1.0, 1.5, 2.0, 2.5, 3, and 4 h). The amount of DMSO remaining after each evaporation time point was calculated from the change in weight.
Acid Plate Preparation DMSO Evaporation The extent of DMSO evaporation was determined by mass measurements. In particular, a 150 mL volume of DMSO solution containing 10 mM concentration of a highly soluble drug (antipyrine) was added to each well of a 96-well plate (plate A) and weighed. A 150 mL volume of DMSO solution containing 10 mM concentration of a poorly watersoluble drug (danazol) was then added to each well of a second plate (plate B) and weighed. In addition, 150 mL of DMSO without drug was JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
Selected strong acids included HCl, H2SO4, and methane sulfonic acid; weak acids included acetic acid, succinic acid, citric acid, and phosphoric acid. Since weakly basic drugs screened in this experiment all had pKa values at least two units higher than the pKa of selected acids (except for acetic acid), the majority of drugs were expected to be greater than 99% ionized. Thus, the final KSP was relatively insensitive to the precise pKa value used. Acid solution plates were prepared robotically (Biomek 2000) by serial dilution (1:2) from a high concentration (typically 1M) acid stock. Each DOI 10.1002/jps
RAPID THROUGHPUT SCREENING OF KSP VALUES
acid solution plate may hold up to 16 different acids arranged by row (2 acids per row) and five concentrations, 0.5, 0.25, 0.0125, 0.00625, and 0.03125 M, per acid arranged by column across the plate. Salt Precipitation The five acid concentrations were robotically (Biomek) transferred (150 mL/well) from the acid plate to the evaporated drug plate, yielding a nominal drug concentration of 2.5 mM (10 mM total concentration of drug initially). Mini stir bars (V&P Scientific) were added to each well and the drug plate was shaken (Lab-Line Instruments, speed ¼7 ) for 24 h. After shaking, the stir bars were removed and in some cases, the pH of each well was measured during early validation studies (ATI, Collegeville, PA). Turbidity was used as a semi-quantitative prescreening step to shorten the assay time. Assuming a molecular weight of 400, the 2.5 mM nominal drug concentration would give a solubility of 1 mg/mL if all drug was dissolved. If there was no precipitation with a selected acid, then the apparent KSP was not reached, and HPLC analysis was not necessary, even though samples were still filtered. Turbidity was performed by an absorbance plate reader at 650 nm wavelength. An absorbance reading greater than that observed for blanks indicating precipitation. Each plate was then filtered through a 96-well flat bottom Costar filter plate (Biomek 2000) into a 96-well round bottom plate (Falcon 1190, BD, Franklin Lakes, NJ). The filtrate was then diluted 10 by transferring (Biomek 2000) 20 mL from each well to another 96-well round bottom polypropylene plate (Falcon 1190), and diluting with 180 mL acetonitrile:water (50:50). The plate was then sealed for HPLC analysis.
HPLC Analysis Sample and standard plates were assayed by a generic fast gradient HPLC method using a chromatography system (Agilent 1100) with a diode array detector and a short C8 (Zorbax Agilent Technologies, Santa Clara, CA) column. The flow rate was 1 mL/min with a gradient ramp from 95% mobile phase A (H2O with 0.1% TFA) to 95% mobile phase B (acetonitrile with 0.01% TFA) over a 1 min period, followed by 95% B for 45 s and re-equilibration in 95% A for 30 more seconds. The total assay time was 4 min per sample. Detection DOI 10.1002/jps
2083
utilized a photodiode array with routine quantitation possible at 214, 230, 254, 280, and 295 nm. Drug standards were prepared at 1 mg/mL by dissolving a weighed sample of each drug into a known volume of acetonitrile:water (50:50). Additional standards at 0.5, 0.2, 0.1, 0.05, 0.02, and 0.01 mg/mL were prepared by dilution of the 1 mg/mL standard with acetonitrile:water (50:50). In cases where 1 mg/mL standards were not soluble in acetonitrile:water, DMSO was used instead. Sample drugs that have precipitations as indicated by turbidity assay are diluted in DMSO for HPLC. Each concentration was loaded into a 96-well round bottom plate (Falcon 1190) for HPLC analysis. A standard curve was constructed from the drug standards (linear regression forced through the origin) and used to determine the solution concentration of the sample. The major peak of each injection was selected consistent with retention time of the standards, typically using the response at 254 nm. Sample concentration was determined using the peak area and the standard curve. Apparent KSP Calculation An Excel spreadsheet was used for calculations and to carry out numerical equation solving procedures. The apparent KSP was determined from the saturated solution concentrations of the ionized drug and counter-ion, along with a calculation of solution pH. As shown in Eq. (1), the apparent KSP was determined by activity of ionized drug, and counter-ion assuming 1:1 salt formation DHþ þ Hn1 A ! DHn A; ppt
KSP ¼ aDHþ aHn1 A
(1)
where aDHþ was the activity of species DHþ. The activity was given by: aDHþ ¼ g DHþ ½DHþ
(2)
where gDH was the activity coefficient and [DHþ] was the molar concentration of species DHþ. The activity coefficient was assumed to be primarily a function of molecular charge with neutral species assumed to have g0 ¼ 1. Activity coefficient was calculated using the ionic strength (m) of each solution determined from: m¼
1X 2 ci zi 2
(3)
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2084
GUO ET AL.
where ci was the concentration and zi was the charge of species i, and the summation was carried out over all charged species in the solution. The activity coefficient (g) was estimated from the Davies equation:4 pffiffiffiffi Az2 m log g ¼ (4) pffiffiffiffi þ Cm 1 þ aB m Typical values were used for the constants: A 0.5, aB 1, C 0.2. The ionic strength and activity coefficients were recalculated at each iteration as described below.
pH Calculation An iterative procedure was employed to calculate the various species concentrations. It was initiated with an initial guess of the solution pH from which the concentration of all drug and acid species can be calculated. The OH concentration was then calculated independently from both the water dissociation constant Kw (designated as [OH]Kw) and the charge balance (denoted [OH]bal) as shown in Eqs. (5) and (6), below. Water Dissociation: Kw ¼ aH3 Oþ aOH ;
½OH Kw ¼
Kw (5) g 21 ½H3 Oþ
½OH bal ¼ ½DHþ þ ½H3 Oþ ½Hn1 A 2 (6)
If the two calculated [OH] values vary significantly, then another iteration was required using a new guess for pH given by: ½H3 Oþ nþ1 ¼ ½H3 Oþ n dD
(7)
where D was the difference in [OH] calculated using the charge balance and Kw methods: D ¼ ½OH bal ½OH Kw
The activity of a given protonated drug species (aDHþ ) is given by Eq. (10): aDHþ ¼
aD aH3 O þ KaD
(10)
which can be derived from mass balance Eqs. (11) and (12) Cmeas ¼ Csoln ¼ ½D þ ½DHþ ¼ ½D D D ½H3 Oþ ¼ ½D 1 þ KaD
½D½H3 Oþ KaD (11)
where CD is the total concentration of drug. The concentration of the unprotonated drug species is given by Eq. (12): ½DHþ ¼ Cmeas ½D ¼ Cmeas D D
Cmeas D 1 þ ½HK3 OD
þ
(8)
and d is used as a scale factor to prevent iterative steps from getting too large (typically d 2): ½H3 Oþ nþ1 ½H3 Oþ n d ¼ ð9Þ Dnþ1 Dn JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
(12)
a
Acid Species Since each drug molecule that precipitates also involves a counter-ion, the final solution concentration of the acid is given by: Total Total Cmeas Þ Csoln Hn A ¼ CHn A ðCD D
¼ ½Hn A þ ½Hn1 A þ þ ½An
Charge Balance:
½Hn2 A2 n ½An
Drug Species
(13)
General equations for multibasic drugs and multiprotic acids can be found in existing literature.10 Examples of one strong acid and one weak diprotic acid are illustrated below. Monoprotic Strong Acid (Example: Hydrochloric Acid) In this case all of the acid will dissociate, hence: ½A ¼ Csoln HA
(14)
The initial pH estimate and Ksp are given by: ½H3 Oþ 0 ¼ Csoln HA
(15)
Ksp ¼ aDHþ aA ¼ g 21 ½DHþ ½A
(16)
where aDHþ was obtained from Eq. (10). DOI 10.1002/jps
RAPID THROUGHPUT SCREENING OF KSP VALUES
RESULTS AND DISCUSSION
Diprotic Weak Acid (Example: Succinic Acid) H A
H2 A þ H2 O
Ka 2
! H3 Oþ þ HA aHA ¼
KaHa A aH2 A aH3 Oþ (17)
HA þ H2 O
aA2 ¼
KaHA
! H3 Oþ þ A2
KaHA aHA KaHA KaH2 A aH2 A ¼ aH3 Oþ aH3 Oþ
(18)
Substitutions of Eqs. (17) and (18) for [HA] and [A2] to get mass balance Eq. (19) as shown below 2 CSolu H2 A ¼ ½H2 A þ ½HA þ ½A
KaH2 A ½H2 A KaHA KaH2 A ½H2 A þ g 21 ½H3 Oþ g 2 g 21 ½H3 Oþ 2 ! KaH2 A KaHA KaH2 A ¼ ½H2 A 1 þ 2 þ g 1 ½H3 Oþ g 2 g 21 ½H3 Oþ 2 ¼ ½H2 A þ
(19) Thus, [H2A] can be obtained from Eq. (20) ½H2 A ¼
Csoln H2 A H2 A K H2 A K H2 A Ka1 1 þ 2 a1 þ þ a2 2 g 1 ½H3 O g 2 g 1 ½H3 Oþ
(21)
(22)
Calculations of monoprotic and triprotic acids would be similar to that of diprotic acid. Thus, both drug, and acid species could be set up in an Excel spreadsheet, and the calculations could be automated.
Reproducibility of the 96-well KSP Screening Assay With selected drugs, reproducibility of the assays was described by two different sets of experiments on intra-day, and inter-day basis. Selected drugs were equilibrated in various acid solutions and subjected to HPLC analysis on same day and same plate with the same standard curve. Similarly, selected drugs were subjected to HPLC analysis on different day, and different plate with different standard curves, and freshly prepared standards. DOI 10.1002/jps
Drugs were initially dissolved in DMSO, which was later evaporated. Residual DMSO after evaporation might enhance the overall drug solubility. This possibility was investigated by dissolving a highly soluble drug, antipyrine, and a poorly soluble drug, danazol in 150 mL DMSO, respectively. Highly soluble and poorly soluble drugs were chosen to examine whether the intrinsic solubility of the drug will have any DMSO retention effect. Mass analysis after evaporation indicated that there was less than 1 mL of DMSO remaining in each well (on the average), irrespective of the intrinsic solubility of the drug. Following addition of 150 mL of selected acid, the final volume of DMSO in the reconstituted solution is expected to be less than 0.7% (v/v). The effect of this small amount of DMSO on overall drug solubility was expected to be minimal and acceptable from a screening assay perspective.
pH Validation
Then, KSP can be calculated from Eq. (22) KSP ¼ aDHþ aHA ¼ g 21 ½DHþ ½HA
DMSO Effects on Drug Solubility
(20)
The pH of diprotic acid solutions is estimated by an iterative, numerical procedure based on Eqs. (6) and (17)–(19). The initial pH guess is given by Eq. (21): H2 A H2 A 1=3 ½H3 Oþ 0 ¼ ðCsoln H2 A Ka1 Ka2 Þ
2085
The pH of the equilibrium drug solution is required for apparent KSP calculation as noted in Eqs. (5)–(9) and (21) in Materials and Methods section. While it is a simpler approach, the time, and labor required to measure the pH in every well would have significantly decreased the assay throughput. Hence, an iterative computational approach was developed to determine pH of the equilibrium drug solution after inputting an initial guess of pH. To validate the calculated pH, the pH in all wells of several plates was measured. Figure 1 is a correlation plot between the manually measured pH and the calculated pH for 8 different salts of PHA1. The correlation coefficient for all salts is greater than 0.98 with the exception of acetic acid. This deviation is likely due to the pKa values of the acetic acid, and the PHA1, 4.76, and 6.0, respectively, being close to each other, and the resulting pH being in this range as well, all leading to a high level of sensitivity of the iteration process to the accuracy of the values. The pH deviations were minimal in the other acid solutions because of much lower pH that was far away from the drug’s true pKa. Less extensive examples with other drugs were also evaluated and comparisons made between the manually measured pH and the spreadsheetJOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2086
GUO ET AL.
Figure 1. Correlation between measured and calculated pH values. pH values were measured for eight different salts of PHA1 in precipitated wells. These values were then correlated with pH values calculated using Eqs. (5)–(9). Data were well correlated, giving a correlation coefficient value of 0.98, when all salts except for acetic acid were included in the comparison (pH values calculated for the acetate salt are thought to deviate from measured values because PHA1 and acetic acid have nearly identical pKa values).
calculated pH, reporting the correlations coefficients with similar results. The correlation demonstrated that the derived equations used in the spreadsheet were adequate. The results also confirmed that manual measurements of pH were not necessary if pKa of the acid is expected to be two units below the pKa of the weak base. Turbidity Study Minimizing time and materials were the primary goals of this screening assay. A semi-quanitative approach was adopted to differentiate soluble compounds from compounds with poor solubility so that time-consuming HPLC would only be needed for compounds with poor solubility. Concentration of 2.5 mM or 1 mg/mL (assuming an average molecular weight of 400) aqueous solubility was typically used as the upper limit. Turbid wells were subject to filtration, and HPLC analysis, and non-turbid wells were reported as having solubility greater than 1 mg/mL. Given the flexibility, and easy maneuverability of this 96well format, users could adjust the differentiation point, and expand the dynamic range accordingly. JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
KSP Determination Apparent KSP values were evaluated for wells with precipitation. As shown in Eq. (1), the apparent KSP was determined by the activity of the ionized drug, and counter-ion assuming 1:1 salt formation. Of the eight compounds screened, apparent KSP values were calculated, and compared with either literature KSP values6,8–12 or internally reported KSP values. Reference KSP values were obtained from the literature or Pfizer Corporation research reports. Most literature reported apparent KSP values are determined at a single specified solution concentration. The rapid throughput capacity of the 96-well format allowed calculations of the apparent KSP values for a given salt at five different concentrations of the counter-ion. Table 2 summarizes the apparent KSP values determined from five concentrations (0.5, 0.25, 0.125, 0.0625, and 0.03125 M) of each acid for eight compounds. It was observed that the apparent KSP value increased with increasing concentration of the counter ions. Table 2, Figures 2 and 3A and B showed that the apparent DOI 10.1002/jps
RAPID THROUGHPUT SCREENING OF KSP VALUES
2087
Table 2. Summary of KSP Values Between Reference and Determined by the Screening Assay for 8 Drugs and 16 Salts
Compounds
Acids
Terfenadine
Triamterene Phenazopyridine Papaverine Hydrochloride PHA1
PHA2 PHA3 PHA4
Reference Ksp4,6,8–13
Ksp Range
Ksp Determined at Low Ionic Strength, 0.25 M
Hydrochloric acid Methanesulfonic acid Phosphoric acid Hydrochloric acid Phosphoric acid Hydrochloric acid Hydrochloric acid
5
1.58 10 1.87 105 4.99 106 2.13 104 2.01 105 1.27 104 2.24 103
8.06 106 1.16 105 4.12 106 2.02 104 1.06 105 7.26 105 2.8 103
to to to to to to to
1.95 105 3.01 105 5.16 106 2.68 104 3.72 105 2.56 104 3.6 103
1.36 105 2.47 105 4.21 106 2.2 104 3.71 105 2.5 104 3.6 103
Hydrochloric acid Succinic acid Citric acid Hydrochloric acid Succinic acid Hydrochloric acid Citric acid Hydrochloric acid Hydrobromic acid
5.45 105 7.99 106 7.8 105 1.84 105 1.67 105 4.4 105 2.8 106 2.6 103 1.98 104
3.43 105 4.90 106 2.57 105 2.67 105 1.22 105 1.34 105 2.28 106 1.34 103 1.31 104
to to to to to to to to to
1.19 104 1.93 105 1.23 104 5.80 105 2.39 105 1.05 104 9.44 106 1.76 103 2.53 103
5.75 105 1.44 105 5.45 105 2.77 105 1.47 105 3.41 105 2.63 106 1.36 103 1.76 104
KSP measured at different acid concentrations gave a slightly different apparent KSP value. The highest counter-ion concentration (0.5 M) usually gave the highest value; the lowest counter-ion concentration gave the lowest apparent KSP, thus resulted in a range of apparent KSP reported here. A typical trend was shown in Figure 2 for PHA1, the apparent KSP values increased with increasing acid concentrations. The slope was counter-ion specific, the trend existed over a 15-fold counterion concentration range with less than one magnitude difference between the highest and the lowest KSP. This trend existed even when the
assay was done in vials instead of plates indicating that an experimental bias due to the screening method was not occurring. Serajuddin11 also made similar observations through conventional methodology, in which KSP values did fluctuate with different pH, counter-ion concentration and ionic strength. The systematic trend seemed to disagree with the fact that theoretical KSP should be a constant value, even multiple concentrations of acid should render a constant KSP value. However, the KSP calculated are the apparent values, not the true KSP values, the true KSP value was dependent on
Figure 2. Counter-ion effects on apparent KSP values. The apparent KSP value of was plotted versus counter ion concentration, for the HCl, citric acid, and succinic acid salts of PHA1. Apparent KSP values showed an increasing trend proportional to increasing counter-ion concentration in all three cases. Nevertheless, the assay still allowed the ranking of salts forms by KSP values so that an appropriate salt was selected. DOI 10.1002/jps
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2088
GUO ET AL.
Figure 3. (A and B) Correlation of reference KSP values with apparent KSP values determined at lower acid concentrations. The reference KSP value was compared with apparent KSP values determined at lower acid concentrations (0.25, 0.125, 0.0625, and 0.03125 M), where the Davies equation adequately describes activities between drug species and counter ions. Panel A shows a correlation between reference KSP and KSP determined at lowest counter-ion concentration (0.03125 M) or n ¼ 1, similarly lowest ionic strength, and closest to the true KSP value. A correlation coefficient of 0.979 is returned. Panel B shows a correlation between reference KSP and average KSP values of (0.25, 0.125, 0.0625, and 0.03125 M) or n ¼ 4 determined at low counter-ion concentrations, and low ionic strength. A correlation coefficient of 0.981 is returned. The data shows that KSP values determined from the 96-well format screening assay at different but <0.25 M concentrations are not significantly different from the true KSP value.
activities between drugs and counter-ions, which in turn depend on ionic strength and pH of the acids. All these variables changed simultaneously in our method, it was difficult to pinpoint the contributing variable. The Davies equation used in the calculations might be inadequate to describe these activities because Davies equation theoretically only applies to molar concentrations less than 0.2 M. The highest concentration (0.5 M) JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
used in this assay far exceeded the theoretical limit of the Davies equation, thus giving a higher than expected KSP value. When KSP determined at 0.5 M acid concentration was compared with reference KSP value, a correlation coefficient of 0.91 is returned. Similarly when KSP of the lowest concentration of 0.03125 M is compared, a much greater correlation coefficient of 0.979 is obtained as shown in Figure 3A. To comply with the Davies DOI 10.1002/jps
RAPID THROUGHPUT SCREENING OF KSP VALUES
equation, the highest concentration 0.5 M was not included in KSP average calculations, the average KSP includes only those determined at 0.25, 0.125, 0.0625, and 0.03125 M concentrations, and returns a correlation coefficient of 0.980 (Fig. 3B). The KSP value determined from the low ionic strength was more representative of the true KSP value. As shown in Table 2, the KSP value determined at low ionic strength was closer to the reported literature KSP values, usually below 0.1 M.6,8–12 Literature KSP values always fell within our reported range of apparent KSP values. In fact, the range of the screening apparent KSP values is generally less than 1 order of magnitude. Determination of the true KSP value was not the purpose of this 96-well screening assay, rather, the assay was to allow proper ranking of different salt forms. For instance, Figure 3 clearly demonstrated that the KSP value of PHA1 with succinic acid was much lower than that with HCl, and Citric acid. All reference KSP values were within half an order magnitude of the apparent KSP determined here. From a screening perspective, this discrepancy was small enough to allow adequate KSP ranking at the early development stage. Our results suggest that counter ion concentrations up to 0.25 M are more than sufficient to rank apparent KSP values for different salts. The Davies equation is also valid across this concentration range, leading up to 0.25 M. However, one needed to confirm that KSP was reached at the low concentration by exhibiting precipitation before HPLC analysis. Turbidity studies would provide such confirmation in term of pre-
2089
cipitation observation. In cases with no precipitation, KSP was not reached; a higher counter-ion concentration would be needed to hit KSP. This 96-well format consumes minimal amount of materials, but assists the formulation, and discovery scientists to identify a feasible salt form (from a solubility perspective) in a rapid throughput manner. It should be recognized that salt solubility is only one criteria of salt selection, and that properties such as hygroscopicity, ease of preparation, stability, and mechanical properties. Reproducibility Inter- and intra-day assay reproducibility is assessed in Table 3 for Terfenadine and PHA1. Both are weakly basic drugs with pKa values of 6, and 6.8, respectively. These values are at least two units above the pKa values of the selected acids, hydrochloric acid (Strong acid), methanesulfonic acid (Strong acid), phosphoric acid (1st pKa 2.12), acetic acid (pKa 4.76), succinic acid (1st pKa 4.19), and citric acid (1st pKa 3.15). Additionally, a number of reported KSP values in different salt forms are available for comparison. These drugs have relatively low KSP values, thus only minimal amount drug materials will be needed for KSP determination. These physiochemical characteristics make them ideal candidates for the 96-well format solubility-screening assay. They are representative of the class of insoluble, weakly basic drugs this assay is aimed at. Terfenadine was equilibrated with three different acids and KSP was obtained on different days for respective salts,
Table 3. Reproducibility of the 96-Well Screening Assay on an Inter-Day, Inter-Plate and Intra-Day and Intra-Plate KSP Terfenadine Inter-day and inter-plate Hydrochloride salt Methanesulfonic acid Phosphoric acid
Day1
Day 2
Day 3
Day 4
1.36 105 2.47 105 4.21 105
1.27 105 3 105 4.52 106
1.38 105 N/A N/A
1.51 105 N/A N/A
PHA1
Replicate 1
Replicate 2
Intra-day and intra-plate Acetic acid Citric acid Sulfuric acid Phosphoric acid Hydrochloride Succinic acid
6.17 106 1.53 105 3.15 105 3.1 105 3.89 105 6.21 106
5.98 106 1.62 105 2.01 105 3.38 105 5.7 105 6.19 106
DOI 10.1002/jps
JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
2090
GUO ET AL.
and KSP for PHA1 was calculated on same day, and same plate in duplicates. On an inter-day, and inter-plate basis, there was less than 10% relative standard deviation for n 3, and less than 5% for duplicates. On an intraday, and intra-plate basis, the reproducibility was also well within 10% for duplicates. Overall, the reproducibility of the KSP screening assay was acceptable from a screening perspective, especially when considering the Gilson injector used in the assay has an intra-day variability of 4–8%.
CONCLUSIONS The 96-well plate format greatly increased throughput of screening assays, and reduced time, and material constraints at the early stage of drug development. Because of the competitive nature of the pharmaceutical industry, early participation of formulation scientists in the drug development process is becoming increasingly important. Physiochemical parameters need to be evaluated in parallel with biological parameters to maximize efficiency and productivity. Our assay evaluates KSP values at multiple concentrations of acid with very little a priori information about the weak base compound. If KSP values for other compounds in a template series are known, the number of acid concentrations can be significantly reduced. As mentioned previously, the best estimates are obtained from the lowest acid concentration that results in precipitation of the salt. Furthermore, by incorporating the turbidity assessment into the workflow, only the lowest acid concentrations with precipitation can be selected for HPLC analysis. Estimation of KSP typically required 10 mg of active pharmaceutical ingredient (API) to screen eight salts using five acid concentrations. The 96-well format offers the flexibility of screening i drugs in j salts at k counter-ion concentrations as long as i j k ¼ 96. Semi-automated data analysis limits assay throughput to be about 25 compounds per week. In the example of the apparent KSP screening, the screening KSP values were compared with the reported KSP result; generally results were within half an order of magnitude. This accuracy was acceptable for a screening study. Apparent KSP values were reasonably constant if determined at modest ionic strengths when Davies equation was used. At higher ionic strengths, a trend towards increasing KSP with increasing ionic strength was noted. JOURNAL OF PHARMACEUTICAL SCIENCES, VOL. 97, NO. 6, JUNE 2008
ACKNOWLEDGMENTS Financial support for Jermey Guo was provided by a Pfizer internship program. Additional support was provided by University of Utah research development funds.
REFERENCES 1. Michelson S, Joho K. 2000. Drug discovery, drug development and the emerging world of pharmacogenomics: Prospecting for information in a datarich landscape. Curr Opin Mol Ther 2:651–654. 2. Wu G, Irvine J, Luft C, Pressley D, Hodge CN, Janzen B. 2003. Assay development and highthroughput screening of caspases in microfluidic format. Comb Chem High Throughput Screen 6: 303–312. 3. Yu K, Balogh M. 2001. A protocol for high-throughput drug mixture quantitation: Fast LC-MS or flow injection analysis-MS? LCGC 19:60–66. 4. Tong WQ, Whitesell G. 1998. In situ salt screening—a useful technique for discovery support and preformulation studies. Pharm Dev Technol 3:215–223. 5. Alfred Martin JS, Cammarata A. 1983. Phys Pharm 272–313. 6. Lin SL, Lachman L, Swartz CJ, Huebner CF. 1972. Preformulation investigation. I. Relation of salt forms and biological activity of an experimental antihypertensive. J Pharm Sci 61:1418–1422. 7. Kibbey CE, Poole SK, Robinson B, Jackson JD, Durham D. 2001. An integrated process for measuring the physicochemical properties of drug candidates in a preclinical discovery environment. J Pharm Sci 90:1164–1175. 8. Dittert LHT, Reese D. 1964. Phase solubility technique in studying the formation of complex salts of triamterene. J Pharm Sci 53:1325–1329. 9. Kramer SF, Flynn GL. 1972. Solubility of organic hydrochlorides. J Pharm Sci 61:1896–1904. 10. Streng WH, Hsi SK, Helms PE, Tan HG. 1984. General treatment of pH-solubility profiles of weak acids and bases and the effects of different acids on the solubility of a weak base. J Pharm Sci 73:1679– 1684. 11. Serajuddin AT, Jarowski CI. 1985. Effect of diffusion layer pH and solubility on the dissolution rate of pharmaceutical bases and their hydrochloride salts. I. Phenazopyridine. J Pharm Sci 74:142–147. 12. Miyazaki S, Nakano M, Arita T. 1975. A comparison of solubility characteristics of free bases and hydrochloride salts of tetracycline antibiotics in hydrochloric acid solutions. Chem Pharm Bull 23:1197– 1204. 13. Sacchetti M. 2000. General equations for in situ salt screening of multibasic drugs in multiprotic acids. Pharm Dev Technol 5:579–582. DOI 10.1002/jps