Anal. Chem. 1994,66, 368-377
Characterization of Submicrometer Emulsions Using Sedimentation Field-Flow Fractionation with Power Field Programming Shulamit Levin,. Llron Stern, Amlra Ze’evl, and Menashe Y. Levy School of Pharmacy, P.O. Box 12065, The Hebrew University of Jerusalem, Jerusalem 91 120, Israel
Sedimentation field-flow fractionation (MFFF), operated with power-based field programming, was shown to be effective in the characterization of submicrometer fat emulsions. Field programming, in which the decrease of field strength with time gradually increases the average velocity of the sample components, extends the capabilities of sedimentation field-flow fractionation in handling polydisperse samples. Submicrometer fat emulsions were analyzed by three different rates of field decay and two different initial field strengths, using various stop-flow times. Identical size distribution profiles were obtained under all circumstances, using the appropriate stopflow times. Fractions were collected from the SdFFF eluting bands, and diameters were analyzed by photon correlation spectroscopy, showing good agreement with values given by the FFF instrument. The agreement between the two methods indicated also that polydispersity of size dominated band broadening. Accurate and highly reproducible size distribution profiles were obtained in all the cases studied. Sample loadability was examined by studying overloading effects and detection limits. The linearity of detector response was also established, by injecting increasing loads of sample. The accuracy of the FFF peak shape was examined experimentally, by collecting fractions, dissolving the emulsion in them, and measuring the UV absorption of the solutions. The profile formed by the fractions was relatively close to the FFF band, proving that light scattering does not seriously distort the size distribution profile. Sedimentation field-flow fractionation (SdFFF) is a highly selective technique for the separation and characterization of colloids such as emulsion micro sphere^,^ vesicles and microcapsule^,^ and biological species such as cells and v i r ~ s e s .The ~ technique was developed by Giddings and Myers,6q7 Kirkland,8 and C a l d ~ e l l .The ~ history, instrumentation, and methodology of the techniques were described in two reviews.g~l0
Field-flow fractionation has been described as a one-phase analogue of chromatography,1° in which the sample components are partitioned into regions of different mobile fluid velocity in an open ultrathin channel. The SdFFF channel is integrated into the perimeter of a centrifuge rotor; thus, a sedimentation field is constantly applied across the channel faces, perpendicular to the flow axis. The field forces the sample’s components toward one wall, the accumulation wall. If the particles’ density is higher than the mobile fluid, they accumulate at the outer wall of the channel, whereas if their density is lower, such as oil in water (o/w) fat emulsions, the particles accumulate at the inner channel wall. The retention parameter is related to the property of the particle interacting with the field. In the case of sedimentation FFF, this property is the effective mass, or particle diameter and density. Therefore, particle size distribution can be measured from retention data of the sample components by the sedimentation field-flow fractionation system. FFF has evolved into a powerful and flexible technique for the separation and characterization of polydisperse samples, due to the possibility of changing parameters, such as density of the carrier or field strength with time.”-18 Several programs of field strength were developed, among which are linear, parabolic,11J2time-delayed exponential,I3-l5and power field decay,’”I8 which is used in the present work. Emulsions are relatively polydisperse samples that can be conveniently and precisely characterized by sedimentation field-flow fractionation, using a decaying field. At present emulsions are characterized by various techniques, such as light scattering, laser diffraction, electron and light microscopy,1gturbidimetry, { potential, etc. The commonly used particle sizing techniques in studies and assessments of the stability of intravenous fat emulsions have been summarized and evaluated in a recent review by Washington.20 Some of these techniques cannot accurately characterize polydisperse or multimodal populations. Sedimentation FFF, using field ~
(IO) Kesner, L. F.;Giddings, J. C. Field-Flow Fractionation: an HPLC Analogue. (1) Yang, F.-S.;Caldwell, K. D.; Giddings, J. C. J. Colloid Interface Sci. 1983,
93, 115-125.
(2) Caldwell, K. D.; Li, H. J . Colloid interface Sci. 1989, 132, 256-268. (3) Caldwell, K. D.; Karaiskakis, G.; Myers, M. N.; Giddings, J. C. J . Pharm. Sci. 1981, 70, 135G1352. (4) Kirkland, J. J.; Yau. W. W.; Szoka, F. C. Science 1982, 215, 296. (5) Levin, S.Biomed. Chromatogr. 1991, 5, 133-137. (6) Giddings, J. C.; Myers, M. N.; Caldwell, K. D.; Fisher, S.R. In Merhods of Biochemical Analysis; Glick, D., Ed.; John Wiley: New York, 1980; Vol. 26, p 79. (7) Giddings, J. C. Chem. Eng. News. 1988, 66 (41), 34. (8) Kirkland, J. J.; Yau, W. W.; Doerner, W. A.; Grant, J. W. Anal. Chem. 1980, 52, 1944-1954. (9) Caldwell, K. D. Anal. Chem. 1988, 60, 959A.
360 Analytical Chemlstry, Vol. 66,No. 3, February 1. 1994
In Lasers, MoleculesandMerhods; Hirschfelder, J. O.,Wyatt, R. E., Coalson, R. D., Eds.; John Wiley & Sons: New York, 1989; p 601.
F.;Myers, M. N.; Giddings, J. C. Anal. Chem. 1974,46, 19241930. (12) Giddings, J. C.; Smith, L. K.; Myers, M. N. Anal. Chem. 1976, 48, 1587. (13) Giddings, J. C.; Williams, P. S.;Beckett. R. Anal. Chem. 1987, 59, 28-37. (14) Kirkland, J. J.; Rementer, S. W.; Yau, W. W. Anal. Chem. 1981.53, 1730. (15) Yau, W. W.; Kirkland, J. J. Sep. Sci. Techno/. 1981, 16, 577. (16) Williams, P. S.; Giddings, J. C. J. Cbromarogr. 1991, 550, 787-797. (17) Williams, P. S.; Giddings, J. C. Anal. Chem. 1987, 59, 2038-2044. (18) Reschiglian, P.; Pasti, L.; Dondi, F. J. Chromatogr. Sci. 1992, 30, 217-227. (19) Rotenberg, M.; Rubin, M.; Bor, A.; Mcyuhas, D.; Talmon, Y.; Lichtenberg, D. Biochim. Biophys. Acta 1991, 1086, 265-272. (20) Washington, C. InI. J. Pharm. 1990, 66, 1-21. (11) Yang, F. J.
0003-2700/94/03660388$04.50/0
0 1994 American Chemical Society
programming, offers an advantage over current techniques in analyzing complex samples by being a separative technique. Although the analysis requires considerable dilution, as is the case in most of the currently used measuring systems, the SdFFF system, fluid carrier and channel, comprisesa relatively gentle environment that does not inflict extreme risks of shear or stress. Therefore, almost no changes of the original sample components are observed. Two previous works established and validated the feasibility of SdFFF for emulsion characterization,1.2using a constant sedimentation field. The present work utilizes another mode of operation of the SdFFF system, using power-based field programming. The purpose of this work is to explore the consistency of the technique, in giving identical profiles of size distribution under various conditions of field decay. The appropriate and convenient concentration range of sample loads, linearity of the detector response, and effect of the carrier fluid compositionon peak shape were also investigated. The present work further explores the application of field programming in SdFFF for characterization of fat emulsions, which consist of vegetable oil droplets, suspended in aqueous solution, emulsified by phospholipids. In addition to the commercial emulsions, the present work also examines experimental fat emulsionsof comparable overall composition. Among their pharmaceutical functions is the solubilization of highly lipophilic drugs.21 Emulsions are inherently unstable systems, which tend to cream (settle) and coalesce. Both processes depend on the size distribution of the oil droplets, as well as on other physical properties, such as surface tension, (potential, density, and viscosity of the two liquid phases.22 Particles, under isopycnic conditions, cannot be analyzed by SdFFF, since the difference in density is essential to obtain sedimentation (or flotation) of the sample components, which will lead to retention in the channel. Therefore, emulsions prepared from oil, whose density is close to the suspending solvent,cannot be easily analyzed by SdFFF unless the density of the carrier fluid is modified. Soybean oil emulsions (density of 0.9-0.92 mg/mL), suspended in aqueous solutions, can thus be characterized by SdFFF.
EXPER IMENTAL SECTION Materials. The mobile fluid in the SdFFF system was made up of 2.25% (w/v) glycerin in double-distilled water with 0.0125% (w/v) sodium azide added as bactericide (refractive index 1.33, close to that of water). It was filtered through a 0.2-rm filter before use. The density of the solution was determined by pycnometer as 1.004 g/mL. The commercial emulsion was Intralipid 20% (w/w), a soybean fat emulsion, (density 0.92 g/mL at 25 OC, refractive index 1.46) marketed by Kabi Vitrum, date of expiration 1.3.1991 and 1.1.93. A purified soybean oil was purchased from Bertin Co. (Courbevoil, France). Egg lecithin (Lipoid E 80) which is composed mainly of phosphatidylcholine and 8% phosphatidylethanolamine was purchased from Lipoid KG (Ludwigshrafen, Germany). Piroxicam, which conformed with USP requirements, was kindly provided by Agis (Yerucham, ~~
(21) Levy, M. Y.; Benita, S.In?.J. Pharm. 1989, 54, 103-112. (22) Rieger, M. R. Emulsions. In The Theory andPracticeofIndustrialPharrnacy, 3rd cd.;Lachman, L.,Lieberman, H. A., Konig, J. L., Eds.; Lea and Febiger: Philadelphia, PA, 1986; pp 502-533.
Israel). The sources of all the components in the diazepammedicated emulsion are specified in ref 21. All other substances used were of pharmaceutical grade. Instrumentation. A basic unit of particle and colloid fractionator, SedFFF Model S101, equipped with a data station and control of rpm, capable of data acquisition and processingfrom FFFractionation Inc. (Salt Lake City, Utah), was used for the fractionation. A Varian 8500 (Palo Alto, CA) volumetric pump and a Pharmacia detector, Model UV-1 (Bromma, Sweden), or a spectrophotometric detector Model LC-85B from Perkin Elmer (Norwalk, CT) operated at 254 nm completed the fully operating sedimentation field-flow fractionation system. Channel dimensions were 2 cm in breadth, 0.0254 cm in thickness, and 90 cm in length. Radius of the rotor was 15.1 cm. Voidvolume, measured using various small molecular weight substances, was 4.6 mL. Fractions were collected by a Pharmacia Frac- 100 fraction collector. Size analysis of the fractions collected from the SedFFF instrument was done using the submicron particle analyzer Coulter Model N4SD. Methods. Preparation of the Emulsions. The medicated emulsion formulations (% w/w) consisted of the following: (I) diazepam 0.5, oil phase 40.0, egg yolk phospholipids 1.2, Poloxamer 188 2.0, glycerin 2.25, a-tocopherol 0.02, methyl and butylp-hydroxybenzoic esters 0.2 and 0.075, respectively, and distilled water to 100.0 g; (11) piroxicam 0.2, soybean oil 20.0, Lipoid E 80 1.2, glycerin 2.25, a-tocopherol 0.02, and water for injection, made to 100.0 g. The diazepam-loaded emulsion was prepared as described in ref 21. The piroxicam emulsion was prepared as follows: glycerin was dissolved in water for injection. Piroxicam, phospholipids, and a-tocopherol were dissolved in purified soybean oil. The aqueous and oily phases were heated separately to 70 OC, combined, and stirred with a magnetic stirrer. The mixture was further heated to 85 OC. At this temperature a coarse emulsion was obtained, using a high-shear mixer (Polytron, Kinematic, Luzern, Switzerland). After cooling, it was passed through a homogenizer (APV, Gaulin, Hilversum, Holland) at a pressure of 5000 psi for 5 min, with a final pH of 6.05. The emulsion was then packed and stored in 50-mL plastic bottles at 4 O C . Operation of the SdFFF. (1) All the fractograms shown in this work were produced using direct injection through the septum port. Samples of 5 pL of 20% commercial or investigational emulsions were injected, unless otherwise specified. The diazepam-medicated emulsion was diluted to 10% (w/v) before injection. (2) The flow was stopped and resumed by turning the pump off and on. Volumetric measurements of the effluents showed that the Varian syringe pump is capable of instantaneously stopping and starting the flow with minimum disturbances, giving good stability of the flow and consistency of the flow rate with very low pulsation. (3) The accuracy in the operation of the system was routinely checked, using standards of polystyrene latex beads in mixtures of two or three sizes. (4) Thenumericoutputofthedatasystemoftheinstrument, was transferred to an Apple Macintosh format and treated by Igor, a graphing and data analysis software by WaveMetrics AnalvticaIChemistry, Vol. 66, No. 3,February 1, 1994
369
(Lake Oswego, OR). Fractograms and distribution profiles were normalized to the maximum values of the smoothed peaks. ( 5 ) When investigational medicated emulsions are used, it is advisable to inject acetone or ethanol through the injection valve or the septum to clean the area from remaining of oil and/or emulsion additives. From time to time the system was rinsed with a mixture of 5050 alcohol-water, or pure ethanol, or even warm water to clean all the compartments (especiallythe detector) from remains of oil or other emulsion components. Operationof the Photon CorrelationSpectrometer (PCS). Mean droplet diameters in the fractions, collected from the FFF instrument, were measured by photon correlation spectroscopy immediately after collection. Each fraction was vortexed just before measurement. The 2004 measurements of the fractionated samplesshowed reasonable reproducibility. A preliminary assumed value of mean particle diameter of the sample is required by our PCS instrument. The values of the mean diameters of the collected fractions from the FFF instrument were taken from the data system at the corresponding retention times and were used as input of the assumed diameter in the PCS measurements.
I1 I .o-
0.00 000
RESULTS AND DISCUSSION Three types of emulsions were investigated, a commercial Intralipid and two investigational medicated emulsions. The basic mechanism of the normal mode of retention in SdFFF has been described numerous tirnes;l-l8therefore, only a short explanation will be given here for clarification. The cloudsof droplets, which are formed under the influence of the sedimentation field, move downstream at a velocity proportional to their thickness. The more compressed to the wall, the slower they move. Droplets of different sizes form clouds of different thicknesses, which move downstream at different velocities; thus separation takes place. The experimental data appear in the form of the detector response as a function of retention times of sample components, i.e., the fractogram. The currently available SdFFF instruments manipulate the experimental data automatically. Diameters of the sample components are calculated from their retention data, and the relative mass for each diameter is calculated according to the procedure described by Yang et al.23 The result is a profile of size distribution. Figure 1 shows various fractograms, and the corresponding size distribution profiles, as obtained from the SdFFF instrument. The experimental retention times are converted to respective retention parameters, whose exact relationship is described in most of the FFF publications.'-l8 An approximation of the retention parameter can be used when the ratio t!J/tR is smaller than 0.1, X = '/6(to/tR), where void time conventionally is to and t~ is solute retention time. Diameters of the components are then calculated by the system from X at any time, using the following equation: d = ( ~ ~ T / X ~ A ~ G W ) ' / ~ (1) given the parameters Ap, the difference in density between the suspending fluid and the sample components, w the channel (23) Yang, F.-S.;Caldwcll, K. D.; Giddings, J. C. J . Colloid Inferface Sci. 1983, 92. 81-91.
370
AnaMicalChem/stry, Vol. 66,No. 3, February 1, 1994
0 IO
0.20 0 20
0.30 0 30
0 40 040
0.50 0 50
0.60 0 60
0.70 070
0.80 0 80
Dlameter Ilm)
Figure 1. Characterizationof an investlgatbnaidiazepam fat emulsbn, as specified In the graph obtained using various stopflow times (7), (I) normalized fractograms. (11) normalired size distribution profiles. The program used was as follows: inltial field was 108g(SOOrpm) and final field was lg (75 rpm), f1 = 8 mln, t. = -64 min, and flow rate was 1.5 mL/min.
-
thickness, Tthe temperature, and Boltzmann constant k . The sedimentation field G is calculated automatically from the spin rate (rpm), measured by a sensor in the centrifuge at any particular moment in time, using the following equation:
G = ro(2?rw/60)* (2) where rO is the radius of the rotor and o is its spin rate. I. Relaxation Step. It is important to relax the sample before the actual run for a specific period of time, that is calculated from the relaxation time, r , to prevent distortion of the profile of size d i ~ t r i b u t i o n . ~ *This * ~ *time ~ ~ needs to be sufficient, so that all the sample components occupy their equilibrium positions across the channel; however, it is important to ensure that excessive relaxation does not affect the shape of the droplets, because they are deformable species, or even cause phase separation. The smallest particles in the sample are taken into account in the calculation of the stopflow time. The first step is therefore the calculation of 7 for the smallest particles in the sample using the expression r = 18wq/ApGd2 (3) where 7is mobile fluidviscosity and d is the estimated minimal droplet diameter. Figure 1 shows fractograms (I) and size distribution profiles (11) of an investigational diazepam emulsion,*' obtained using various stop-flow periods, as specified in the figure. Distortion and variability of the fractogram and the corresponding size (24) Janca, J.; Chmelik, J.; Pribylova, D. J . Liq. Chromatogr. 1985, 8, 2343. (25) Giddings, J. C. Anal. Chem. 1986, 54, 735-740.
I
0.2J
0.0
,, , , , , , , , , ,, , , , , , 0.2 0.4
0.0
a
I
b
0
, ,, , , 06
0.8
1.0
20
40
12
distribution, due to insufficient relaxation, were observed at relatively low stop-flow periods; the principal peak was shifted to lower elution volumes and broadened, similar to many previously shown examples.8 The void peak contains free drug that was probably released from the oil droplets during the relaxation period. The discharge of drug from the oil droplets is time dependent, hence the large variations of the void peak. Determination of the average diameter and size distribution of this investigational diazepam emulsion was part of a preliminary developmental procedure. The optimal formulation contained 20% oily phase and had quite different characteristics of size propertiesz1than the one studied here, Le., average diameter was below 0.2pm. The emulsion studied here was too polydisperse and the mean size of the droplets was too high for the particular medicinal purpose it was originallyassigned to. It was therefore withdrawn from further pharmaceutical use. Size measurements of the nonfractionated emulsion by PCS gave inconsistent results of average diameters higher than 0.5 pm and/or two populations. When the relaxation time was 15 min and above, highly reproducible and consistent signal shapes were obtained for all types of the soybean emulsions, commercial as well as investigational, as shown in Figures 1 and 2. The stop-flow period used throughout the present study was therefore, 15 or 20 min, using an initial field of 800 rpm (108g). 11. Constant Field. Fat emulsions such as Intralipid are typically polydispersesamples (udm,,/dmtan 0.45); therefore, when a constant field is used, they are spread over a wide range of retention times. An example of a typical fractogram, obtained in our system when a constant field was applied, is shown in Figure 31. A profile of size distribution based on the fractogram obtained until the field was stopped, is shown in Figure 311. Oneof the profiles of size distribution, obtained using field programming (from Figure 2), was superimposed on it to verify their similarity. The operation under field programming may be preferable in case of highly polydisperse samples when a moderately sensitive detection system is available. 111. Rates of Field Decay and Fractionating Power. The field programming used here for Intralipid was power-based field In this program of field strength, the initial field strength SOis held constant for a period of time tl (time lag). Subsequently, field strength is decreased over a chosen period of time until it reaches a constant value (lg in this
100
1 3
T i m e (min)
Diameter (pn)
Figure 2. Size distribution profile of Intralipid, obtained using three different stop-flow times before elution, 15,30, and 60 min. The rest of conditions of the analysis were as in Figure 1.
80
60
0.8
-Distribution (const) Distribution (program)
-
,,**,,.,.,.*
'*..,
%,,
0.0
,,,,,),,,
'.*.... ...*,,,.." ...,,,
.,,,.*......,.,
, I , , , , , , ,
, ( ( , ( , , , ,
, , , , ( , , ( ,
(
,
I
,
,
,
(
,
,
_
study). After tl has elapsed (at t > tl > fa), the field behaves according to the expression (4) where S(t) is the field strength at time t and p is the variable of the power program and ta = -ptl . Curves showing the change of S(t) with time are presented in Figures 41 and 51. The resolving power is characterized by the fractionating power, Fd, the resolution between two close lying particles ( 6 t ~ / 4 u ~divided ) by their relative diameter differences. (5) f R is the retention time and ut is the standard deviation in retention time for particles of diameter d and d + 6d (6d is the difference in diameter between the two close-lying particles). When p = 3n - 1 (n is the exponent of d in the relation between d and the retention parameter A), Fa is uniform over most of the diameters range for a particular rate of field decay." Equation 44 in ref 17 fully describes the fractionating power Fd. In the presently operating SdFFF system where n = 3 andp = 8 it can be reduced to the following equation:
where D is the diffusion coefficient. Three rates of field decay (last term in eq 6 ) were tested for the analysis of Intralipid: method A, $1 = 8 rnin and t a = -64 min; method B, tl = 10 rnin and fa = -80 min; method C, tl = 13 rnin and t, = -104 min. The rest of the experimental parameters, such as the initial (108g)and final (lg) field strengths, stop-flow time, Ana&tIcal Chemistty, Vol. 00, No. 3,February 1, 1994
371
8 1.0 8. B
[ 1.: 02 00
40
20
0
80
60
100
Retentlon Tlme (rnln)
11
Tlme (Inln) 1.2]-.t,Ernl" -1,.lOmln
11
0.0 *
l.2J
I
'*
0.4
0.2
*'
*
I
*
I
' . * * * *I * , .
'
0.6 I*
I .O ' :.:-*&n'i&gk Fd
0.8
--S,=IOBE'
1.0-
11.5
0:O
'
0:2
'
0'4
'
0:6
0:8
1:O
1.2
Dlameter him)
Figure 4. Characterlzatlon of Intralipld uslng three different rates of fieiddecay: method A, ti = 8 mln and t. = -64 min; method B, f, = 10 mln and fa = -80 min; method C, tl = 13 min and t, = -104 min. Initial field was 1088 and final fie6 was lg,and flow rate was 1.5 mL/min. (I) Normalized fractograms on which the experimental (-) and calculated (.e) field strength vs tlme are superimposed. (11) Size dlstributlon profiles, obtained by the three field programs, with superimposed Fd vs diameter.
-
flow rate, and rate of data acquisition, were identical in all three cases. The three normalized fractograms on which the three field decay functions are superimposed are shown in Figure 41. The experimental field strength was calculated from the rpm values, given by the control system for every moment of data acquisition, using eq 2. This experimental field strength is expected to follow eq 4. Figure 41 shows a comparison between the results of the two calculations of field strength with time, the apparently operating field strength [exp S ( t ) ] from eq 2, and the theoretical field strength, expected from eq 4, [calc S ( t ) ] . The two curves overlapped, indicating the precise operation of field programming by the instrument used in the present study. The three size distribution profiles, corresponding to the fractograms in Figure 41 are shown in Figure 411. Three different F d values were calculated according to the programs used, assuming T = 298 K and q = 0.01 P. The value of D1/2(1/b)11ais constant (D 0: l / d a X O ' / ~ )and F d therefore increases with tl, giving the following values of F d , obtained in the three different field programs: 1.56, 2.1, and 3.0 for t l = 8, 10, and 13 min, respectively. The three calculated F d vs diameter plots are superimposed in Figure 411. Another study compared profiles obtained at two different F d , using two different initial field strengths (affecting XO in eq 6), with identical time parameters of field decay t l = 8 min and fa = -64 min. The resulting fractograms are shown in Figure 51, in which the two functions of field decay are superimposed. Again, the experimental and calculated S ( t ) values were in excellent agreement, indicating good performance of the instrument, although the initial field was 372
AnaIyticalChemisfry, Voi. 66, No. 3, February 1, 1994
0.0
0.2
0.4
0.6
0.8
I
.o
Dlameter (Itm
Figure 5. Characterization of Intraiipld, using two dlfferent initial field strengths. The rest of the condltlons were Wentical, fleld-decay parameters: tl = 8 mln, t. = -84 min, and flow rate was 2 mL/min. Initial field strengths were 1088 and 3808. (I) Fractograms and superimposed field strength VI time: (11) size distrlbutlon profiles and Fd vs diameters.
relatively strong, and the rate of field decay was relatively high. The corresponding size distribution profiles, on which the calculated F d values for the entire range of diameters are superimposed, are shown in Figure 511. In this case as well, identical size distribution profile was obtained, in spite of the difference in Fa and rate of field decay. It is noteworthy that an identical size distribution profile was obtained by all these programs, although the rates of field decay were different. IV. Comparison to Photon Correlation Spectroscopy. Various fractions were collected from the eluted bands, displayed in Figure 41, and analyzed immediately by photon correlation spectroscopy (PCS), to verify the droplets' diameters. The values given by the PCS were compared to the corresponding FFF diameters, given by the data system of the instrument, for the particular time of collection. Each fraction contained 3 mL of effluent, taken within 2 min (flow rate was 1.5 mL/min). Panels 1-111 of Figure 6 show the diameter, obtained by the two methods, photon correlation spectroscopy and sedimentation field-flow fractionation, as a function of the time of collection. The values of the FFF diameters were taken from the midpoint of the 2-min fractions. Larger diameters were obtained at increasing retention times, as is expected from FFF theory of the normal mode of elution. Good agreement between diameters calculated by the SdFFF instrument and those measured by PCS for most of the fractions was observed. The error bars in the diameters measured by PCS were taken from the coefficient of variation (CV) given by the
1.2
I
I
- 1.0 p
#
- o.8 - 0.6 8 - 0.4 568 - 0.2 0.04 0
I
20
’... I
I
40
60
B
0.4
0.0 80
Tfme (min) 0.8
1.2
I1
-
..-............
0.0 I
20
0
I
40
1.0
c p
- 0.8 - 0.6 8 iT - 0,4f - 0.2 p 0.0
I
60
80
Time ( m i d 0.8
-
-4
;a,
1.2
I11
0.6-
L4
9
a
0.4-
0.2-
0
20
40
60
80
Tlme (minl
Fburr 6. Average diameter of droplets In the fractions, collectedfrom
the SdFFF instrument during elution, measured by SdFFF ( 0 )and by
PCS (m): (I) method A (I I) method 8; (I I I) method C as speclfled in Figure 4.
instrument. These values indicate a low grade of measurements, rather than the real polydispersity in size of the population in the fractions. The error in the FFF diameter was calculated from the diameters, given by the data system for the two extreme points of collection (2 min apart). It was within the size of the plot symbols. A size distribution profile of the original sampleof Intralipid has been measured by photon correlation spectroscopy ocassionally. The average diameter was around 0.350-0.360 pm each time, close to the result given by the SdFFF system; however, the variance of the size distribution was different for each measurement. Either narrow or bimodal distributions were reported on various occasions, given identical input. It seems that the polydispersity of the nonfractionated sample gave false results of size distribution profile by the photon correlation spectroscopy instrument. In contrast, the reproducibility of the complete profile of size distribution obtained by the SdFFF system was remarkable, as demonstrated by Figure 2. Due to the perfect repeatability, only one or two FFF runs were needed for the analysis of polydisperse samples, rather than multiple repeated measurements as in PCS,or other light scattering methods that are based on statistical treatment of the experimental data. When characterization of an
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Diameter (pnm)
Flgure 7. Characterlzatlon of an lnvestlgational plroxlcam emulsion, usingtwo different field programs,method A and method C as speclfled in Figure 4. Flow rate was 1.5 mL/mln; relaxation time was 30 mln.
unknown sample of a submicrometer emulsion is needed, a relatively high fractionating power can be employed at first, with high initial field strength, and then very low field strength should be applied to test for presence of particles larger than 1 pm. This way the characterization of the unknown sample is enabled over the entire range of diameters, without previous estimationof the averagediameter. The diversity and modality of sizes of the oil droplets’ population will be given straightforward, since this is a separative technique. A size distribution profile of an unknown investigational piroxicam emulsion is shown in Figure 7. The sample was brought in for analysis by SdFFF after determination of average diameter by PCS. The instrument indicated that it contained a bimodal population of -0.3 and -1.5 pm. Method C (section 111) was operated first, for enhanced selectivity at the beginning of the fractionation. Next, the shorter Method A (section 111) was operated. An identical size distribution profile was obtained by the two operations, giving an average diameter of 0.363 f 0.130 pm. No population of large sizes was detected using considerably weaker field strengths. V. Peak Broadening. (a) Contributionsof Nonequilibrium and Polydispersity. The detector response as a function of time (fractogram) describes a large and wide spread band of the soybean fat emulsion. The eluting band of droplets assumes a finite width, which can be used to characterize their size distribution. The extent of dispersion of a sample zone, as a result of migration along the channel length L, under the influenceof the perpendicular field, is termed the plate height, H, or height equiualent to a theoretical plate, HETP. For uniform channels H = a 2 / L(c2is the variance of the profile). The accurate measurement of peak dispersion is complex, but it is necessary to understand it in order to obtain an accurate size distribution profile. The two major contributions to peak broadening,nonequilibriumeffects and sample polydispersity, are given in the following eq~ation:2’-~9
H = 3aqdw2(v)24X3+ 9L( 3)’
(7) kT dP where w is channel thickness, k is the Boltzmann constant, ~~
(26) Hanscn, M. E.;Giddings,J. C.;Schurc,M. R.; Bcckett. R. Anal. Chem. 1988, 60, 1434-1442. ( 2 7 ) Giddings, J. C.; Karaiskalds, G.; Caldwcll, K. D.; Myers, M.N.,J . Colloid Interface Sei. 1983, 92, 66-80.
Ana@tic8lChemlstry, Vol. 66, No. 3, February 1, 1994
373
.... 1.0 mL/m!n mL/m!n
1.OJ
1 1 1
0.8
-
1.5
2.0 mL/mm
0
.
0
5 0.0
0
20
10
50
40
30
60
70
Time (min)
Diameter (pn1
0.61
I
L 1.2
I1
I
Flgure 8. Sire distribution profile of Intrallpld obtalned at three flow rates, 1, 1.5, and 2 mL/min. Field program: tl = 8 min and t. = -84 min; inltlal fleld was lOBg(800 rpm) and final field was lg(75 rpm), t = 20 min.
-
( u ) is the average velocity of the flowing streams, and m/dP is the polydispersity. A typical apparent ud/dp value for Intralipid was 0.45. The contribution of nonequilibriumeffects constitutes the first term in the equation and is highly sensitive to the retention parameter A. When the polydispersity is as high as 0.45, it becomes the major contributor to peak broadening and the contribution of nonequilibrium effects to dispersion of the various sample components does not effect peak shape significantly. The shape of the profile of droplets remains invariant at various flow rates, and no deconvolution of the two contributions (eq 7) is needed. Fractions can be collected from the broad eluting FFF band and analyzed by photon correlation spectroscopy to test whether the sample components are driven away from the center of gravity of the band due to differences in their sizedependent retention or due to diffusive effects. The domination of polydispersity in the profile of Intralipid over nonequilibrium effects is proven in Figure 6. Good agreement between d values of various fractions, collected from the eluting FFF band, analyzed both by SdFFF and PCS was observed at both sides of the peak maximum. Runs at three different flow rates, 1, 1.5, and 2 mL/min, were performed in order toverify the accuracy of the apparent size distribution profiles of Intralipid. The corresponding size distribution profiles are shown in Figure 8. The profiles obtained at 1.5 and 2 mL/min were identical in shape and width, proving that polydispersity dominates peak shape. The profile of size distribution obtained at 1 mL/min was distorted relative to the other two profiles shown in this figure, overestimating the population of larger diameters. Fractions were collected from the eluting bands, and diameters were analyzed by PCS. The results are shown in Figure 91-111, where the normalized fractograms are superimposed to indicate the positions of fraction collection. Good agreement of d values from both techniques was observed when the two flow rates 1.5 and 2 mL/min were used. The overestimation of d values of the larger droplets by SdFFF, indicated in Figure 8 for 1 mL/min, is clearly proven here (91). The diameters of the fractions containing the larger particles, measured by PCS, were significantly smaller than (28) Karaiskakis, G.; Myers, M. N.; Caldwell, K. D.; Gidding, J. C. A w l . Chem. 1981, 53, 1314-1317. (29) Schure, M. R.; Barman, B. N.; Giddings, J. C. A n d . Chem. 1989,61,27352743.
374
Analyticel Chemistry, Vol. 66, No. 3,Februaty 1, 1994
0
. 0
o 10
j
' 20
:
o 30
40
50
60
70
50
60
70
Time (minl
0 0
10
. 20
s 0.0
0 30 40 Time (min)
Flgure @, Mean diemeter of the oil droplets In the fractkns, collected from the eiutlng FFF bands In Figure 8, measwed by SdFFF ( 0 )and byPCSm. Flowrates: (I) 1 mL/mln;(II) 1.5ml/min;(III)2ml/mln.
the corresponding SdFFF values. The distortion of the profile obtained using 1 mL/min can be attributed to effects of secondary r e l a x a t i ~ n .During ~ ~ ~ ~ field ~ decay the sample droplets are expected to continuously shift positions toward the faster streamplanes of the channel. However, if the rate of field decay is high relative to their velocity, the clouds never reach their equilibrium positions and the droplets are overly retained. The result is an overestimation of the larger population due to higher retention than expected. According to guidelines in ref 30, the void time should be reduced as much as possible, when operating under field programming, with high rates of decay. One way of achieving lower void times is increasing flow rates. This is probably the reason for the better results in the operation at the higher flow rates. The agreement between the profiles, obtained using the different rates of field decay and initial field strength, described above, indicates that there were no significant secondary equilibrium errors in the measurements of droplets' diameters by SdFFF at flow rates of 1.5 mL/min and above. (b) Effect of the Initial Injection Volume. Four samples of Intralipid containing the same amount of oil in the sample (1 mg) were analyzed to examine effects of the initial injection (30) Hansen,M. E.;Giddings, J. C.;Schurc, M. R.; Beckett,R. A w l . Chem. 1988, 60, L 434-1 442.
... ,
00
02
04
06
08
10
*.
12
Dlameter (pm)
0
10
20
30
Figure 10. Normalized size distrlbutlon profilgs of four samples of Intralipid, containinga constantamount of olidroplets, invarious injection volumes. Conditions as in Figure 1 using 16min stop-flow.
volume on peak broadening. Samples, such as 5 pL of the original 20% Intralipid, 10 p L of a 2-fold diluted emulsion (lo%), 20 p L of a 4-fold diluted emulsion (5%), and 50 pL of a 10-fold diluted emulsion (2%) were injected. Figure 10 shows the size distribution profiles of these samples. The profiles were almost identical, indicating that the initial injection volumes of 5-50 pL did not contribute to peak broadening significantly, as long as the total amount of oil droplets in the sample was constant. The contribution of the injected volumes to peak broadening was negligible relative to the polydispersity of the sample. It is noteworthy that sample loads in field-programmed SdFFF were more than 1 order of magnitude higher than the loads used in a constant-field operation,2 with essentially no overloading effects. VI. Detector Response. Detection of clouds of particles is based on measurements of the attenuation of the UV signal in the detector. The signal depends both on absorption and on light scattering, which is a size-dependent property; therefore, it might not be linear with concentration. The contribution of absorption to the detector signal in the case of emulsions, especially medicated ones, is considerable, due to the various UV-absorbing additives in them. (a) Linearity of the Detector Signal. The linearity of the UV detector signal to Intralipid was examined using 10-50 pL samples of a 2% (diluted) emulsion. The resulting smoothed fractograms and the integrals of the large peak (beginning from 10 min and on) are shown in panels I and I1 of Figure 11, respectively. The correlation coefficient of the fit to a linear curve was 0.998. Figure 11 indicates that detector response is linear with concentrations; therefore, the fractograms, on which the corresponding size distribution profiles were based, reflected the amount of oil droplets in the sample. The limit of quantification in our detection system, where a noisy fractogram could still be collected and treated, was 5 p L of 2%oil (0.1 mg), which was the maximum sample load used in a constant-field operation2 The detection system described here was less than ideal, and a more sensitive UV detector would provide a order of magnitude lower detection limit if necessary. (b) Sensitivity and Sample Loads. An adequate detector signal, without observable overloading, is one of the most important experimental requirements for an operational and practical FFF instrument. Sample overloading effects, which
00
50
40
Time ( m i d
0
10
20
30
40
50
60
Time (min) Figure 11. Samplesof 10-50pLcontalning2% IntraUpid: (1)smoothed fractograms; (11) integrals. Conditions were the same as In Figure 10.
distort FFF profiles, were described by Caldwell et al.31 and Hansen and G i d d i n g ~ .Overloading ~~ effects are caused by compression of the oil droplets toward the accumulation wall during the relaxation step, which leads to a significant increase in local concentration. At this point the oil droplets occupy a relatively small volume, lying next to the wall of the channel. If too many droplets are introduced and driven into this limited zone, volume exclusion and other repulsion effects force the droplets tochange their size distribution. If overloadingeffects are not significant at the relaxation step, they are less so during the elution process, due to dilution by band broadening processes. The maximum wall concentration co at the end of the elution process is given by31,32
where Knj is the sample injection volume, q n j is sample concentration (5% w/w of oil), L is channel length, V' is the void volume of the channel, u is the standard deviation (in length) of the particles having the particular retention, and X is the retention parameter. Size distribution profiles, obtained from the runs described in Figure 11, are shown in Figure 12, to confirm that essentially no overloading effects were observed. All the profiles except 10 pL were identical. The fractogram obtained from the 10-pL sample run was considerably noisy, since it was close to the detection limit, giving rise to a significant error in the calculations. The identical size distribution profiles of the four samples of constant quantity (Vinjcinj) shown in Figure 10 indicate also (31) Caldwell, K. D.; Brimhall, S.L.; Gau, Y.;Giddings, J. C.; J . Appl. Polym. Sci. 1988, 36,703-719. (32) Hansen, M. E.; Giddings, J. C . J. Colloid InterfaceSci. 1989,132,300-312.
Ana~kalChemlstry,Vol. 66,No. 3, Februsty 1, 1994
375
0 0
0 2
04
06
08
10
12
0
10
20
30
Figure 12. Size distribution profiles, calculated from the fractograms In Figure 11.
that the original samples of oil droplets were not significantly concentrated at the vicinity of the accumulation wall during the elution process under field programming. The use of relatively high loads of samples up to 2 mg of oil was possible, and the maximum wall concentration of the oil did not cause any phase inversion in these emulsion^.^^^^^ The results indicate that relatively high sample loads of submicrometer emulsions can be used (up to 1-mg loads), using initial field strength as high as 360g, and a slowly decaying field strength ( t l = 13, and ta = -104), without considerable overloading. The possibility of using relatively high sample loads facilitates the hyphenation of techniques such as PCS and other spectroscopic methods for the analysis of the collected fractions. (c) Accuracy of the Profile Shape. The accuracy of the size distribution profile depends on the capability of the SdFFF detector in quantification of the components in the sample. The proportionality is quite linear in our case, as shown in the previous section. A simple experiment addresses this question. A sample of Intralipid (10 pL of a 2-fold diluted emulsion) was run, and fractions from the eluting FFF band were collected and analyzed by UV spectrophotometer. Each fraction was divided in two; one portion was measured directly, whereas the other was diluted by ethanol, to obtain homogeneous solutions. The procedure is based on the principle presented in ref 1. Turbidity was measured from the neat fractions, whereas absorption was measured from the dissolved fractions. Figure 13 shows normalized UV absorption of the two series of fractions at 250 nm, one containing ethanol [UV (EtOH)] and the other neat (UV). The measured absorption and/or turbidity form a profile of response vs time, a rough fractogram. The two normalized "profiles" were superimposed on the normalized fractogram. The profile of the fractions containing ethanol was slightly shifted to higher retention times, whereas the direct turbidity measurements of the fractions yielded a profile closer to the FFF band. Figure 13 shows that the problem of peak distortion due to light scattering, which might yield an inaccurate size distribution profile in fat emulsions, is not as serious as anticipated. (a) Compositionof theMobileCamerFluid. The procedure of analysis involved the injection of a small volume of the original sample (5-20 wL) into a channel with a void volume (33)Block,L.H.Emulsions and Microemulsions. In PhurmuceuticolDmageFwms; Lieberman, H. A., Rieger, M. M., Banker, G. S., Eds.; Marcel Dekker, Inc.: New York, 1989;Vol. 2, pp 335-378.
370
Analytcal Chemlstfy, Vol. 86, No. 3, Februaty 1, 1994
40
50
60
70
Time (min)
Diameter ( p n
Flguro 13. Normalizedfractogramof 10 ~ L oaf2-folddlluted Intrallpld, compared to the normalized abswptbn of collected fractlons: dlrect measurement, UV; addition of ethanol. UV (EtOH). . . Conditions were the same as in Figure 10. Time ( m m ) -> "'
123 " '
"""
1.2
' I
I '
441 " '311 ' ~ ' ~ " " ' ,'"'-'
'
538
-
1.0
2.25% glycerol
0.8 0.6
0.4 0.2 0.0
0.0
0.2
0.4
0.6
0.8
1.0
2
Diameter (pm
Flguro 14. Size distrlbution proflle, obtained using water as the carrler fluld, compared to a 2.25% aqueous solution of glycerln as the moblie fluld. Conditions were the same as In Figure 10.
of 4.6 mL, through which a considerable volume of the mobile fluid eluted the sample components. It is possible that the analysis process itself may cause changes in the emulsion if it is not resistant to dilution or if a noncompatible fluid is used as the carrier. Intralipid was not sensitiveto dilution by 2.25% glycerin in water, as can be seen in Figures 11 and 12, where various diluted emulsions were used with no effect on its profile of size distribution. However, it was still suspected that the mobile fluid composition might affect the results; therefore, plain water (with 0.1% sodium azide) was used instead of the 2.25% glycerin solution (included sodium azide as well). The two normalized superimposed size distribution profiles are shown in Figure 14. The difference in density of the mobile fluid was taken into consideration in the calculations by the SdFFF data system. The two profiles obtained with the two different mobile fluids were only slightly different. The difference was close to the experimental error. The profile obtained, using plain water as the mobile fluid carrier, showed a slightly more weighted population of larger droplets than the profileobtained using 2.25% glycerin. This difference may be attributed to coalescence between oil droplets in water, or differences in light scattering, which distort the peak shape in the fractogram. In any case, the mobile fluid composition should be carefully chosen to be as close as possible to the continuous aqueous phase in the emulsions. CONCLUSION Sedimentation field-flow fractionation, operated with field programming, is a very reliable and precise method of
characterization of emulsions, which can be very polydisperse in nature. The basic requirement for convenience of analysis is use of nonisopycnic conditions; i.e., the density of the oil droplets should be significantly different from the suspending solvent. A workable density difference Ap can be as low as 0.05 g/mL for sizes beginning at 150-nm droplet diameter. Higher Ap are preferable, for emulsions with droplet sizes approximately 100-1 50 nm, for convenience of the instrument operation. The operation under gradually decreasing field strength extends the capability of the technique to systems with relatively low detection sensitivities. The gradient in field strength enables the use of higher loads of samples and enhances the sensitivity of the system.
ACKNOWLEDGMENT This work was supported by the following funds of the Hebrew University of Jerusalem: Leonie Emanual Fund, Mary Gordon Fund, and Ernst Chain Fund. We are grateful
to the Committeeof Basic Equipmentat the Hebrew University for the purchase of the SdFFF SlOl system and to Prof. Eli Grushka for the donation of the volumetric pump. We express our deep gratitude to Prof. S. Benita, of the Pharmacy Department at the School of Pharmacy, The Hebrew University, for his financial aid and helpful discussions. The technical assistance of Galia Tawil is greatly appreciated. Many thanks to Prof. P. S. Williams of the Field-Flow Fractionation Research Center at the University of Utah, for his critical prereview of the manuscript. Special acknowledgementsare due to Prof. K. D. Caldwell at the Department of Bioengineering at the University of Utah, for her prereviewing of the manuscript and her useful suggestions and critical comments. Recieved for review M a y 10, 1993. Accepted November 3, 1993.' Abstract published in Aduancc ACS Absrracra, December 15, 1993.
AnalvtkalChernbtry, Voi. 66, No. 3,Febnrary 1, 1994
977
