2360

Ind. Eng. Chem. Res. 1996, 35, 2360-2368

Liquid-Liquid Extraction Using the Composition-Induced Phase

Separation Process
Rajan Gupta, Roberto Mauri,* and Reuel Shinnar
Department of Chemical Engineering, City College of CUNY, New York, New York 10031

This paper describes a new separation process of liquid-liquid extraction. It consists of first
mixing the system to be extracted with a primary solvent, which is soluble with the native solvent,
and subsequently adding a modifier, which is insoluble with either the native or the primary
solvent. This process is similar to the phase transition extraction, which was described in a
previous paper, where the liquid mixture, together with the solute to be separated, is first heated
above its critical temperature, where it forms a uniform solution, and then cooled to the region
below the miscibility curve, where it separates. Both processes have the advantage that the
resulting separation of the solvents is very rapid, even in the presence of emulsion-forming
impurities. In addition, the extraction efficiency of the new process may be 10 times higher
than that of the traditional liquid-liquid extraction. The new process is thought of having
significant advantages in the extraction of products from fermentation broths, plants, and other
natural sources.
1. Introduction
In a previous paper [Ullmann et al. (1995)], one of
the authors developed a novel liquid extraction process,
named PTE (phase transition extraction), which is based
on using solvents that, together with the liquid to be
extracted, form a solution with a critical point of
miscibility. In the new process, the mixture is heated
and cooled across its coexistence curve: in the heating
stage the solvents form a single phase, while in the
cooling stage the solution separates into two phases,
with the desired solute dissolved in the solvent phase.
In Ullmann et al. (1995) the new process was applied
to extract an antibiotic, i.e. ephretomycin, from a
fermentation broth containing fractured cells, obtaining
the two startling and unexpected following results.
(1) Phase separation was very fast (it was completed
within few minutes), whereas conventional extraction
produced a stable emulsion that required a centrifuge
to separate.
(2) The extraction yield was very high and could not
be duplicated even when conventional extraction in
multiple stages was performed. This was surprising,
as the partition coefficient for the conventional solvent
was higher.
More recently, Eliyahu and Ludmer (1995) have confirmed these results using different systems, and in
particular they achieved very high yields when extracting compounds from microbial cells that had not been
fractured. These results are potentially of great relevance, especially in the extraction of biological compounds from fermentation broths, mammalian cells, and
plants.
The present paper has two purposes. The first is to
explain the two results that are mentioned above,
providing a scientific basis for the absence of emulsion
formation and the higher yields observed in the PTE
process. The second objective is to propose and investigate another process, in which the phase transition
of partially miscible mixtures is achieved by adding a
modifier, i.e. changing the mixture composition instead
of its temperature. In the following, we will refer to this
* Email: mauri@[Link]. Fax: (212)650-6660.

S0888-5885(95)00362-9 CCC: $12.00

process, as to the one described in Ullmann et al. (1995),

as PTE, since they both involve phase transition, and
shall distinguish one from the other by denoting the first
process as temperature-induced phase separation (TIPS)
and the second as composition-induced phase separation
(CIPS). CIPS greatly simplifies the separation process
of TIPS, as it can be carried out isothermally, and in
addition, it allows a better control of the final composition of the two phases, thereby improving the separation
yield even further.
To better understand the physical reasons of the
higher yields obtained using PTE, we developed a model
system, consisting of solid particles whose size is similar
to the cells of a typical fermentation broth, using a dye
as our solute. This allowed us to perform a series of
controlled and reproducible experiments and to compare
our results with the theoretical predictions. Our work
on the model system provides a physical explanation of
the experimental results obtained by Ullmann et al.
(1995) and shows under which conditions PTE can lead
to higher extraction yields. This is of special relevance
in the extraction of large and sensitive molecules from
broths, mammalian cells, or plants, where the high
shear required to break the cells could deteriorate the
solute molecules.
The performances of the composition-induced phase
separation process were compared in detail with those
of both conventional liquid extraction and the temperature-induced phase separation process. From our
results it appears that the new, CIPS process could lead
to substantially higher yields and provides a useful tool
for dealing with difficult separation processes, which are
hard to handle using currently-used methods. Patents
for the new processes have been applied for [Ludmer et
al. (1990); Shinnar and Mauri (1995)], and interested
parties should approach the authors.
2. The PTE Process Using the
Temperature-Induced Phase Separation (TIPS)
As mentioned above, TIPS, like all phase transition
extraction (PTE) processes, capitalizes on the properties
of partially miscible solvents with a critical point of
solubility. In this process the liquid mixture, together
with the solute to be separated, is first heated above
1996 American Chemical Society

Ind. Eng. Chem. Res., Vol. 35, No. 7, 1996 2361

Figure 1. Coexistence and spinodal curves of a liquid mixture,

giving the ratio between the temperature of the system and its
critical temperature, as a function of the primary solvent concentration. The vertical arrow describes the temperature quench of
the PTE process; the horizontal arrow describes the addition of a
modifier (which results in a decrease of the primary solvent
concentration) in the CIPS process.

the critical temperature, where it forms a uniform

solution, and then cooled to the thermodynamically
unstable region below the miscibility curve (see Figure
1). Ullmann (1993) described how this process can be
used to design and operate a multistage counter current
extraction column composed of successive heating and
cooling units. More recently, Ullmann et al. (1995)
described an important application of this process to
extract an antibiotic from a fermentation broth, finding,
surprisingly, that the new process had higher yields
than its conventional counterpart. In addition, coalescence was very rapid, and phase separation was completed within a few minutes, despite the fact that the
broth was not filtered and that the suspended cells
normally form a stable emulsion that need a centrifuge
to separate. While the reasons for having a higher yield
will be described in the next section, here we will
concentrate on the phenomenon of rapid coalescence,
presenting some additional observations, which complement those of Ullmann et al. (1995).
The rapid coalescence of critical mixtures was studied
in detail by Ullmann et al. (1995) using as a model
system a dye (i.e. crystal violet) dissolved in water. The
dye allowed an easy visual observation of the coalescence process; in addition, since the dye acted as an
emulsion stabilizer, the timescale of the hindered
coalescence during conventional liquid-liquid extraction
(LLE) could be varied from minutes to hours at will, by
simply increasing the dye concentration from 2 to 50
ppm. As the main solvent, Ullmann et al. used acetonitrile, to which they added 4% (molar) of a modifier,
i.e. toluene, to raise the critical temperature of the
solution from about 0 to 40 C. The water-acetonitrile-toluene solution was put in a vial, heated to 50
C, where it became homogeneous, and then cooled back
to ambient temperature, at which point the system
phase separated very quickly. On the other hand, when
the system was agitated isothermally, separation occurred very slowly (up to several hours), as crystal violet
had caused the formation of a stable emulsion. Very
similar results were obtained for the fermentation broth,
where MIBK, instead of toluene, was used as a modifier.
Ullmann et al.s results, although sufficient to clearly
show the rapid coalescence phenomenon, had one limitation, namely the cooling was too slow (about 1 min)

Figure 2. Direct observation of phase separation of a water/ACN/

MIBK mixture using LLE (pictures at the top of the figure) and
TIPS (pictures at the bottom of the figure). For each process,
pictures were taken with a 3 s interval between each other. We
see that, while in LLE there are two interfaces moving toward
the center of the sample, during TIPS a single interface forms
rapidly and moves only slightly in time.

to allow observations at short timescales. This is

important because, when we perform TIPS, 1 min after
cooling the main phase separation has already been
completed, and the small changes that are observed at
later times are due to secondary nucleation. Therefore,
we repeated Ullmann et al.s separation experiments
using a much thinner (i.e. 1.2 mm thick) glass cell,
where cooling could be completed within 2 s. A mixture
of 52 cc of acetonitrile, 38 cc of water, and 10 cc of MIBK
was used, with a small quantity, 2 ppm, of crystal violet
dye to facilitate visualization and retard coalescence.
Finally, the phase separation resulting after isothermally shaking the system (i.e. performing LLE) was
compared with that after heating and cooling (i.e.
performing TIPS). As one can see from Figure 2, there
is an inherent difference between the two processes. In
LLE we see two, more or less defined, interfaces, which
move toward the center of the cell until they merge. This
behavior was observed also when the composition of the
system was varied, and the time required to complete
the process increased from 5 to 10 s, in the absence of
the dye, to 2 h, with 20 ppm of dye. On the other hand,
during TIPS the interface appears almost immediately
at its final position, and the solution clears up rapidly,
with no emulsion formation, even in the presence of
larger quantities of dye. This fast phase separation in
the presence of emulsion-forming compounds is a special
property of PTE.
These observations are clearly at odds with the
explanation of the phenomenon that was put forward

2362

Ind. Eng. Chem. Res., Vol. 35, No. 7, 1996

Figure 3. Phase separation of a water/ACN/MIBK mixture in a long tube using TIPS. The picture was taken 2 s after quenching and
shows the formation of large drops. The tube inner diameter was 4 mm.

Figure 4. Phase diagram of the water/ACN/toluene system with four different toluene concentrations.

by Ullmann et al. (1995), who assumed (as most people

do) that at the beginning of the process very small nuclei
are formed, which subsequently grow by both diffusion
and coalescence. The fact that coalescence is evidently
not slowed down by the presence of emulsion-forming
compounds was explained by the conjecture that there
are no stable interfaces to which such compounds can
adhere. However, it is well known [Gunton et al. (1983);
Binder (1986); Mauri et al. (1996)] that, following a
temperature quenching, a process, generally referred to
as spinodal decomposition, takes place which can lead
to phase separation without nucleation. In fact, the
process begins with the appearance of a uniformly
dispersed phase that subsequently develops into distinct
regions of coexisting equilibrium phases. To further
investigate the initial stage of phase separation, the
quenching experiment described above was repeated in
a 40 in. long cylindrical tube, with a 4 mm inner
diameter. After a few seconds, large, centimeter-size
drops were formed, as shown in Figure 3, indicating that
during spinodal decomposition large drops can form very
rapidly, without nucleation. This is the reason why
separation during the PTE process is so fast, irrespective of the presence of emulsion-forming compounds.
3. CIPS: Composition-Induced Phase
Separation
When using the TIPS process, we realized that very
few solvents, when mixed with the native solvent, form

a solvent mixture with a critical temperature that is

suitable for simple laboratory experiments. This problem is even more acute when TIPS is used for extractions of biological compounds, as in these cases strong
constraints about the temperature range apply. In
Ullmann et al. (1995) this problem was overcome by
using solvent mixtures with a low critical temperature
and then adding a modifier to change the miscibility
curve. For example, as shown in Figure 4, taken from
Ullmann (1993), the critical temperature of the wateracetonitrile mixture can be raised from 0 to 120 C by
adding 5% (molar) of toluene. In the same way, the
critical temperature of any solvent mixture can be
changed almost at will (obviously within the limitations
of the boiling and freezing points) by adding a suitable
modifier.
This simple observation suggested to us a new way
to perform spinodal decomposition. Let us suppose, for
example, that our system has a 1.3% molar concentration of toluene and is quenched from an initial 50 C
temperature, where it forms a single phase, to 25 C,
where it separates (see Figure 4). Now, the same
separation can be achieved isothermally, at 25 C, by
raising the toluene molar concentration from 0%, where
the system forms a single phase, to 1.3%, where it
separates. In other words, the key of success of the PTE
process is to bring the system to a point of thermodynamic instability below the miscibility curve. As shown
in Figure 1, this can be achieved either by changing the

Ind. Eng. Chem. Res., Vol. 35, No. 7, 1996 2363

Figure 5. Direct observation of isothermal phase separation using

LLE (pictures at the top of the figure, taken within 1 h from each
other) and CIPS (pictures at the bottom of the figure, taken within
15 s from each other). The system consists of a water/ACN/MIBK
mixture in the presence of 20 mg/L of surface-active crystal violet
dye. We see that while CIPS was completed after 30 s, LLE was
still in process after 2 h.

temperature of the system, as in the TIPS process, or

by changing its composition, as in the new process that
we propose here. In this so-called CIPS process [see
Shinnar and Mauri (1995)], the system to be extracted
is first mixed with a primary solvent, which is soluble
with the native solvent, and subsequently a modifier is
added, which is insoluble with either the native or the
primary solvent.
The equivalence between TIPS and CIPS is substantiated in the series of pictures of Figure 5, where
we used the same water-acetonitrile-MIBK mixture
as in Figure 2, only in the presence of a larger amount,
20 ppm, of the crystal violet dye. We see that while
after isothermal mixing a stable emulsion was obtained,
with small 10 m drops that remained suspended for
about 2 h, when we used CIPS, within 30 s the mixture
had completely separated. As in the case of temperature-induced phase separation, during CIPS we do not
observe any moving interfaces, but instead a very fast
phase separation takes place, which is almost unaffected
by the presence of emulsion-forming compounds, with
subsequent cleaning of the cloudy regions around the
interface.
At the moment, we do not have a completely satisfactory theory that can generalize our results to all cases.
So, we do not claim that adding any modifier to an
otherwise single-phase binary mixture will always

result in spinodal decomposition and fast coalescence,

just as we cannot claim that cooling a mixture below
its miscibility curve will always result in fast phase
separation, independent of the rate of cooling and of the
composition of the mixture. However, we have considered a significant number of cases, using different
solvents and modifiers, under conditions in which
isothermal intense mixing would form stable emulsions,
and in all these cases our concentration-induced phase
separation process worked. In fact, separation was fast
even when inorganic salts, instead of liquid solvents,
were used as modifiers, and results identical to those
of Figure 5 were obtained, when 5 g of sodium chloride
was added to a 50 cc water-50 cc acetonitrile mixture.
Clearly, this is not surprising, as salts too modify the
critical temperature of liquid mixtures.
It is important to note that the addition of the
modifier must result in a mixture composition which is
in the unstable, not the metastable, region below the
miscibility curve. However, the miscibility curve is not
essential to perform the CIPS process, while it is so for
TIPS. For example, if we add 9% molar toluene to a
water-acetonitrile mixture, the critical temperature
would raise to 160 C (see Figure 4), so that TIPS could
be performed only by pressurizing the system. On the
other hand, CIPS could be performed easily, even
without knowing how high the critical temperature is.
One difference between CIPS and TIPS is that TIPS
can be performed, at least in principle, by heating and
cooling the binary mixture composed of the native and
primary solvents alone, i.e. without modifiers, provided
that the critical temperature of the system is acceptable.
A modifier is used in TIPS only to bring the critical
temperature up or down to a convenient value and is
not essential to the process, while, on the other hand,
adding a modifier is essential to CIPS. However, in
practice it is very unlikely that one could find a primary
solvent that, together with the native solvents, forms a
mixture with the right critical temperature; in fact,
modifiers were used in most of the experiments performed by Ullmann (1993) and Ullmann et al. (1995).
On the other hand, adding a modifier as in CIPS leaves
us free to choose the operating temperature according
to the process needs (such as low temperatures for
sensitive materials). More importantly, in CIPS we are
free to add the amount of modifier that is necessary to
maximize the extraction efficiency of the process. In
fact, increasing the relative amount of modifier generally means improving the separation between the two
phases, and therefore it increases the partition coefficient of the solute between the two phases. Instead,
in TIPS the amount of modifier is limited by the fact
that the critical temperature of the mixture increases
as more modifier is added. Therefore, the partition
coefficient here can be improved only at the cost of
increasing the temperature excursion of the heating/
cooling process, which is not only expensive (as in some
cases pressurization would be required) but often not
even feasible. For example, in most biological extractions the samples cannot be heated above 60 C. So, in
conclusion, CIPS is equivalent to a TIPS process with
a large temperature differential and rapid cooling. Its
advantage is that it achieves the same results (i.e. high
yield and no emulsion formation) and yet it is performed
isothermally.
Finally, it should be stressed that for modifiers which
are themselves good solvents the operating conditions
can be predicted using the large amount of available

2364

Ind. Eng. Chem. Res., Vol. 35, No. 7, 1996

Figure 6. Triangular diagram for the ternary mixture composed

of water, acetonitrile, and MIBK, where the horizontal and vertical
axes represent the volumetric concentrations of water and acetonitrile, respectively. The curves have been calculated using the
UNIQUAC equations, with parameters taken from Sorensen and
Arlt (1980).

data. For example, one can use the triangular diagrams

given in Sorensen and Arlt (1990) or generate them
using thermodynamic data (see Figure 6 as an example),
to predict the compositions of the phases after separation, and also check whether the operating point is in
the unstable region, therefore ensuring that the separation process is a spinodal decomposition.
4. Extraction from Porous Solids
In the previous Sections we have seen how the PTE
processes (both TIPS and CIPS) work, with rapid phase
separation even in the presence of emulsion-forming
compounds. However, the most important characteristic of the PTE processes is their ability to obtain higher
yields when applied to the extraction of compounds from
porous solid particles or membranes. This was first
shown by Ullmann et al. (1995) for the extraction of
efrotomycin from a fermentation broth containing fractured cells. However, quantitative scientific studies of
the extraction of compounds from fermentation broths,
plants, and other systems containing a solid phase have
one difficulty, namely it is hard to get a completely
reproducible system and in addition, due to the complexity of the system, it is difficult to know how the
extraction process takes place microscopically. Therefore, we decided to develop a well-defined system of
solute and solvents, so that we could perform a series
of controlled experiments.
4.1. The Model System. This model system was
intended to mimic a real fermentation broth, which
typically consists of an aqueous solution of cell debris
containing the solute to be extracted, such as the active
compounds resulting from fermentation. Our model
system consisted of water, a dye (2 mg/L of crystal
violet), playing the role of the solute, and 2% (dry
weight) of silica gel particles, which were intended to
model the cell debris. These particles are S-662 silica
gel macroporous beads, with sizes ranging from 70 to
250 m, purchased from Fisher and commonly used in
gas chromatography.
Using a dye as our solute had the advantage that its
concentration could be readily measured with a spectrophotometer. In fact, using a Lambda 2 Perkin Elmer
UV/vis spectrophotometer, one can see [Ullmann (1993)]

that the absorption spectra of crystal violet show a

maximum at a wavelength of 590 nm and that this
value is unchanged whether the dye is dissolved in
water or in a solvent. Therefore a calibration line was
determined, finding that the crystal violet concentration
is linearly related to the absorbance of the sample at
590 nm, provided that the concentration is smaller than
12 mg/L.
The most interesting aspect of our system (and the
reason why it is a good model of fermentation broths)
is that when the silica gel particles were added to the
water/dye system, after 10 min of stirring 97% of the
dye was adsorbed on the particles. In fact, the absorbance of the water/dye system decreased when the silica
gel particles were added, until, after stirring for 2 and
10 min, it became 15% and 3% of its initial value (i.e.
without particles), respectively. This value did not
change when we stirred the system for longer times.
Now, considering that the diffusion time of the dye in
the particle pores is about 30 s (see section 5.4), these
measurements show that the adsorption/desorption time
of the dye is no longer than a few minutes.
In the following we will denote by model system the
water/particles/dye suspension, with 97% of the dye
adsorbed on the particles. This system is a good model
of real fermentation broths, where the active compounds are often, imprisoned within 100 m-size cell
debris.
4.2. The Solvents. Three solvents were used in our
experiments, namely pure MIBK (methyl isobutyl ketone) and two mixtures of MIBK and ACN (acetonitrile)
in 1:1 and 1:4 volumetric ratios, respectively. In all
cases, the solvents were added to equal volumes of our
model system, composed of 98% water and 2% dyesaturated particles. Now, MIBK is almost insoluble
with water, and when 50 cc of MIBK was added to 50
cc of water containing 2 ppm of crystal violet, then after
agitation, the dye was found to distribute between the
two phases, with a psw ) 5 partition coefficient. Here
and in the following the partition coefficient psw is
defined as the ratio between the dye concentration in
the upper, solvent-rich phase and that in the water-rich
phase.
Studying the second solvent system, we must consider
that the ACN/MIBK mixture is partially soluble with
water. In fact, when a mixture of 25 cc of ACN and 25
cc of MIBK was added to 50 cc of pure water, then after
agitation, the mixture separated immediately into 55
cc of a water-rich phase (raffinate) and 45 cc of a solventrich phase (extract). The compositions of the two phases
were measured using an HP gas chromatograph, finding
79% water, 18% ACN, and 3% MIBK volume composition for the water-rich phase and 6% water, 35% ACN,
and 59% MIBK volume composition for the solvent-rich
phase. As shown in Figure 6 (see tie line A), these
values were within a few percent of the equilibrium
compositions calculated using the UNIQUAC equation,
with parameters given by Sorensen and Arlt (1980). In
addition, when 2 ppm dye was added to the system, the
dye was found to partition between the two phases, so
that at the end of the separation 96% of it was dissolved
in the upper, solvent-rich phase, which corresponds to
a partition coefficient psw ) 25. Comparing this result
with that obtained using pure MIBK, we see that
acetonitrile is a far better solvent of the dye than MIBK.
Finally, we analyzed the third solvent system, by
adding a mixture of 40 cc of ACN and 10 cc of MIBK to
50 cc of water, finding 25 cc of an upper phase with 13%

Ind. Eng. Chem. Res., Vol. 35, No. 7, 1996 2365

Figure 7. Recovery of the dye as a function of time, with the

vertical axis representing the percent of the dye originally dissolved in the model system that is extracted using LLE. The
solvent is pure MIBK, mixed with an equal amount of our model
system.

water, 58% ACN, and 29% MIBK volume composition

and 75 cc of a lower phase with a 64% water, 33% ACN,
and 3% MIBK volume composition, in agreement, within
a few percent, with the equilibrium compositions calculated using the UNIQUAC equation (see tie line B in
Figure 6). In addition, adding the dye, we found that
42% of its original amount was dissolved in the upper
phase and 58% in the lower phase, which corresponds
to a partition coefficient psw ) 2.2.
5. Comparison between CIPS, LLE, and PTE
5.1. The LLE Process. Having saturated the solid
particles with crystal violet dye, we performed various
extraction experiments, to compare the different processes. First, we performed conventional liquid-liquid
extraction (LLE), where the two solvents described in
the previous section were added and stirred isothermally. During this process, at prescribed intervals, the
batch was centrifuged, to break the stable emulsion, and
the dye concentration in the upper phase was measured.
The results are given in Figure 7 for pure MIBK (50%
MIBK and 50% of our model system) and in Figure 8a
for the ACN/MIBK mixture (25% acetonitrile, 25%
MIBK, and 50% of our model system).
Comparing Figures 7 and 8a, we see that at the end
of the extraction using the ACN/MIBK mixture, 95% of
the dye initially embedded in the particles was dissolved
in the upper phase, while with MIBK that amounted
to only 45%. This behavior was expected, since acetonitrile is a far better solvent of the dye than MIBK (see
previous section), and is typical of most processes, where
water-soluble solvents extract far better than nonwater-soluble solvents. What was not expected in these
experiments is that the timescale of the extraction using
the ACN/MIBK mixture was much shorter than that
with pure MIBK, i.e. 10 versus 40 min, where by
timescale we refer to the time nedded to extract 63% of
the dye dissolved in the solvent phase at equilibrium.
As discussed in section 5.4, this is due to the higher
solubility of the dye in the water-rich phase when we
use the ACN/MIBK mixture, so that there is more dye
dissolved inside the pores, and the extraction can be
completed faster.
5.2. The CIPS Process. 5.2.1. First Example. At
this point we performed our composition-induced phase
separation process, using the same ACN/MIBK solvent mixture that we used for LLE. To do that, our
model system (i.e. water, containing 2% of dye-saturated

Figure 8. Recovery of the dye as a function of time, with the

vertical axis representing the percent of the dye originally dissolved in the model system that is extracted using (a) LLE and
(b) CIPS. The solvent is a 1:1 mixture of ACN and MIBK, mixed
with an equal amount of our model system.

Figure 9. Schematics of the pore geometry.

silica gel particles) was first mixed with acetonitrile (25

cc of ACN to 50 cc of water), forming a homogeneous
mixture that was agitated for 10 min. Then we added
the appropriate amount of MIBK (1 part to 3 parts
solution), stirred mildly, and let settle for 2 min,
obtaining two phases that separated without the help
of a centrifuge. (Actually, for sake of precision, before
analyzing the samples we centrifuged them, to eliminate
any residual traces of suspended particles that might
disrupt the reading of the spectrophotometer. However,
the centrifuge was in no way necessary to separate the
two phases.) When the two phases were analyzed, we
obtained the results of Figure 8b, showing that the same
equilibrium state as that using LLE was reached, but
within a much faster timescale. In fact, as the dye
concentration in the water/ACN mixture is about 30
times larger than that in the water phase of the water/
MIBK mixture, a reduction of the timescale from 40 to
2 min was expected.

2366

Ind. Eng. Chem. Res., Vol. 35, No. 7, 1996

Table 1. Equilibrium Data of LLE and CIPS Extraction Processesa

% dye

LLE
LLE/CIPS
CIPSd

global
composition

solvent phase
composition

water phase
composition

50% MIBK,
50% water
25% MIBK,
25% ACN,
50% water
33.3% ACN,
66.6% water

98% MIBK,
2% water
59% MIBK,
35% ACN,
6% water

3% MIBK,
97% water
3% MIBK,
18% ACN,
79% water

ratio, solvent phase

to water phase

solvent
phase

water
phase

solid
phase

ppwb

pswc

44

47

263

0.82

95

22

25

.75

99

a All concentrations are based on unit weight. All partition coefficients were measured separately. b p
pw ) concentration of dye in
particle phase/concentration of dye in water-rich phase. c psw ) concentration of dye in solvent-rich phase/concentration of dye in waterrich phase. d Before modifier addition to homogeneous phase; solvent phase ) water phase.

Table 1 summarizes our experimental results, giving

the compositions of the phases after separation, the
amount of dye in the particles and in the water-rich and
solvent-rich phases, and the equilibrium partition coefficients.
5.2.2. Second Example. CIPS can be performed
using a wide variety of solvents. In the first example
described above we have used as modifier a solvent, i.e.
MIBK, which is insoluble with the native solvent, i.e.
water, and soluble with the primary solvent, i.e. ACN.
However, we could use a modifier which is soluble with
water and insoluble with ACN, such as sodium chloride.
To show that, we added 50 cc of ACN to 50 cc of our
model system, agitated for 2 min, and then added 5 g
of sodium chloride. As a result, the system separated
immediately into 40 cc of an upper phase, with a 33%
water, 67% ACN, and 0% salt volume composition, and
65 cc of a lower phase, with a 58% water, 39% ACN,
and 3% salt volume composition. In this case we found
that the dye would partition between the two phases
with a 1:6 ratio, and 85% of the total quantity of dye
was dissolved in the upper phase.
5.3. CIPS versus TIPS. At this point, a direct
comparison between TIPS and CIPS was in order. To
this purpose, we performed the TIPS process, adding
50 cc of the third solvent system described in section
4.2 to 50 cc of our model system, heating the resulting
mixture above its 40 C critical temperature, and cooling
it down to ambient (25 C) temperature. We saw that,
as expected, two phases formed very rapidly, with 43%
of the original amount of dye dissolved in the upper
phase and 57% in the lower phase, in agreement with
the partition coefficient measurement of section 4.2.
We then performed the same extraction using CIPS
and found the same results; namely we were able to
extract 98% of the original amount of dye from the silica
gel particles, but only 43% of it was dissolved in the
upper phase, due to the relatively low partition coefficient and the fact that the upper phase volume was
only 25% of the total volume. As a comparison, in the
extraction described in section 5.2.1, we were also able
to extract 98% of the original amount of dye from the
particles, but due to the very large partition coefficient,
most of it was dissolved in the upper phase. However,
this is not a direct comparison between CIPS and TIPS,
as we did not optimize the conditions of TIPS. In fact,
we could have increased the amount of dye in the upper
phase by increasing the quantity of solvents and/or the
critical temperature.
Therefore, we may conclude that when we use the
same solvent system, CIPS and TIPS have the same
extraction efficiency. However, comparing the results
described in this section with those of section 5.2.1, we
see that by applying CIPS we can achieve a higher

efficiency than when using TIPS, as we are allowed to

use a larger amount of modifier, so that the solvent
content in the water-rich phase is reduced. Clearly,
we could have achieved the same result by heating
the mixture of section 5.2.1 to about 150 C under
pressure (to prevent evaporation). However, apart from
the obvious cost of heating and cooling the mixture
across such a large temperature gap, many organic
compounds, such as the products of fermentation broths,
cannot be heated to such high temperatures. Therefore,
as in single-stage extractions TIPS has no apparent
advantages over CIPS; the added expense has no
justification.
5.4. Theoretical Model. The large difference between the extraction efficiencies of LLE and CIPS is due
solely to a difference in the transfer rates: eventually,
if we wait long enough, the same equilibrium state must
be reached. A simple theoretical model can help us
understand this phenomenon.
Consider a mass Mp of porous spherical particles of
radius a immersed in a binary mixture composed of a
water-rich phase of mass Mw and a solvent-rich phase
of mass Ms. A thin film of the water-rich phase
surrounds the particles, so that the system can be
schematized as in Figure 8, showing a single capillary
pore of length L a, in contact at the mouth with a
thin film of a well-stirred water-rich liquid phase
followed by a well-stirred solvent-rich phase. Initially,
a mass mtot of dye is adsorbed on the capillary walls.
Then, as time progresses, the water-rich phase removes
the dye from the wall, and in turn the solvent-rich phase
removes the dye from the water-rich phase. In general,
the concentration profile of the dye will depend on the
partition coefficients ppw ) cp/cw, psw ) cs/cw, and pps )
cp/cs ) ppw/psw, expressing the ratios between the
concentrations of dye at equilibrium in the water-rich
phase, cw, in the solvent phase, cs, and in the particles,
cp. At equilibrium, the fraction of dye extracted,  )
ms/mtot, where ms ) csMs is the mass of the dye dissolved
in the solvent phase, can be easily found to be

)

Psw
1 + Psw + Ppw

(1)

where the coefficients Psw and Ppw are defined as

Psw )

Ms
Mp
ms
mp
) psw
; Ppw )
) ppw
mw
Mw
mw
Mw

(2)

From this expression one may conclude that the larger

Psw becomes the better extraction we get. However, this
does not take into account the fact that if it takes the
characteristic time too long to reach equilibrium, the
extraction process becomes inefficient. In fact, this time

Ind. Eng. Chem. Res., Vol. 35, No. 7, 1996 2367

Table 2. Experimental () and Predicted () Values of
the Extraction Relaxation Time
global
composition
LLE
LLE
CIPSa

50% MIBK,
50% water
25% MIBK,
25% ACN,
50% water
33.3% ACN,
66.6% water

ppw(1 mp/mtotal)

40 min

140

70 min

10 min

22

11 min

<2 min

0.75

45 s

a Before modifier addition to homogeneous phase; solvent phase

) water phase.

is inversely proportional to the mean flux in the pores

and so to the amount of dye dissolved in the water
phase, and therefore it is proportional to (1 + ppw). In
addition, is proportional to the amount of the dye that
leaves the pores, (1 - mp/mtot), and to the diffusion time
d ) a2/D that it takes a soluble molecule to exit a pore,
where a is the particle radius, D the diffusion coefficient
of the solute molecules, and the tortuosity factor. This
analysis can be performed rigorously following Froment
and Bischoff (1979), finding that the ratio between the
dye extracted and the total amount of dye as a function
of time is equal to [1 - exp(-t/)], where

) d(1 + ppw) 1 -

mp
mtot

(3)

From this equation, after easy manipulations we find

1 + Psw
a2
(1 + ppw)
D
1 + Psw + Ppw

(4)

This result is valid, provided that the diffusive time d

is much smaller than , which means assuming that the
quasi-steady-state approximation can be applied. In our
case, since d is about 30 s, this approximation is correct
(see Figures 7 and 8). In addition, in eq 4 we have
neglected the desorption time. This approximation,
although appropriate in our case, as we saw in section
4.1, might not be valid in some applications. However,
we will assume that eq 4 is correct, since our goal here
is not to exactly predict the value of the extraction time
but to understand the role that the different physical
parameters play in the diffusion of the dye out of the
particle pores.
In Table 2 we give the measured values of , together
with their predicted values. Clearly, in predicting the
relaxation time, we have assumed a certain value of the
diffusive time, d 30 s, which is consistent with the
particle size a 100 m, a diffusion coefficient D 10-5
cm2/s, and a tortuosity coefficient 3. As we can see,
the theoretical predictions, in particular the ratios
between the relaxation times obtained using different
solvents, are in good agreement with our experimental
results. Note that in the CIPS process the relaxation
time was too short to be measured accurately, and in
addition, the quasi-steady-state approximation cannot
be applied, so that both measurements and predictions
should be taken cum grano salis.
The most interesting aspect of our results is that in
LLE the extraction timescale is several orders of magnitude larger than the diffusion timescale, despite the
fact that the process is actually diffusion controlled. This
analysis can be applied also, for example, to a fermentation broth, even if the transport process in this case is
clearly more complex and may involve diffusion across
membranes. However, if the desorption process which

in our model has been assumed to be instantaneous (or

at least of the same order as d) is slow, with its rate
dependent on the solvent used, then our simple model
should be modified accordingly. In addition, it should
be pointed out that even when the cells are fractured,
the fracturing may be incomplete and uneven, and cells
can form loose aggregates, unless intense agitation
continues during the extraction. Such aggregates will
behave like a porous solid and will hinder extraction,
as the solvent cannot wet the particles easily.
Thus our results explain why in some cases the yield
for PTE (both CIPS and TIPS) can be much higher than
that for conventional LLE (see Ullmann et al. (1995)
and Eliyahu and Ludmer (1995)), especially as the
extraction times are normally limited to a few minutes.
They also explain the results of Eliyahu and Ludmer,
who showed that intense fracturing improves the yield
of conventional extractions. CIPS has the additional
advantage over TIPS that one can reduce the solvent
concentration in the water-rich phase and thereby
increase psw and improve the yield of a single extraction
stage, while TIPS would need additional stages to
achieve the same result. Our results also indicate
another potential advantage of PTE, especially CIPS,
as the ability to choose better, water-soluble solvents
for the extraction. Now, in our case, ACN is a much
better solvent for the dye than MIBK. We know that
this is accidental, as there is no inherent reason why
we could not find a better solvent that is not water
soluble. However, although this would improve the
yield of the process, it would not affect its long timescale.
Now, for natural materials, very often water-soluble
solvents are superior to immiscible solvents. In that
case, CIPS would have a very large advantage, in
addition to the other advantages of a short extraction
timescale and the absence of stable emulsions formed.
It should be pointed out that CIPS can also be performed
using centrifuges, without loosing its advantages of
higher yields.
In conclusion, CIPS has clear advantages over both
LLE and TIPS, since, unlike LLE, it reaches equilibrium
very quickly and, unlike TIPS, it allows us to select
solvents that give large extraction yields.
6. Summary and Conclusions
In this work we have presented a novel separation
method named composition-induced phase separation
or CIPS. The CIPS process is composed of two mixing
stages: first, the system to be extracted is mixed with
a primary solvent, which is soluble with the native
solvent; second, a modifier is added, which is insoluble
with either the native or the primary solvent. Immediately after the addition of the modifier, the system
separates rapidly into two coexisting phases, even in
the presence of emulsion-forming impurities.
CIPS is conceptually similar to the temperatureinduced phase separation, or TIPS, described in Ullmann et al. (1995), where a mixture of native and
primary solvents is heated and cooled across its coexistence curve. In fact, when the same solvent system is
used, CIPS gives the same results as TIPS, provided
that the temperature differential of TIPS is workable.
The similarity of the two methods was demonstrated
experimentally.
The advantages of CIPS over the conventional liquidliquid extraction, or LLE, process can be summarized
as follows.

2368

Ind. Eng. Chem. Res., Vol. 35, No. 7, 1996

(1) Improved extraction yield. When the solute

is adsorbed on solid particles, CIPS has a clear advantage over LLE, since, by using primary solvents that
are miscible with the native solvent, it does not have
the wetting problems that are encountered in LLE,
where insoluble or partially miscible solvents are used.
This is particularly important in the extraction of
natural products and fermentation broths.
(2) Ability to handle emulsion-forming systems.
Unlike LLE, CIPS is almost unaffected by the presence
of surface active agents, and no stable emulsions are
formed.
(3) Equipment savings. Since we do not need to
use centrifuges to break stable emulsions, as we do
when using the traditional LLE process, the equipment
required to perform CIPS is only a tank. Even when a
distillation equipment is required to separate the primary solvents from the modifier, CIPS is still significantly cheaper to perform than LLE.
(4) Lower product degradation. Any possible
shear stress damage to large solute molecules is prevented in the CIPS process, where only a mild mixing
is required, as opposed to the high centrifugation of
traditional LLE. In addition, the fact that in the CIPS
process a small amount of the native solvent is contained in the extract will help prevent the unfolding of
large solute molecules, such as proteins.
(5) Use of water-soluble solvents. The CIPS
process allows the use of water-soluble solvents, which,
despite being in general good solvents of biological
materials, cannot be used in conventional extraction
processes.
Now, most of these advantages over traditional LLE
are also shared by TIPS. Like the CIPS process, TIPS
takes place rapidly and is not much affected by the
presence of surface active agents in the native system.
In addition, since during the heating stage the native
and primary solvents are miscible, the resistance to
extraction of the solute through the cells interface is
greatly reduced; in fact, we found that when we used
the same solvents the extraction efficiency of the CIPS
and TIPS processes was the same and consistently
larger than that of the traditional LLE. Therefore it
appears that TIPS, like the CIPS process, has many
significant advantages over the traditional LLE process,
at least for cases in which either stable emulsions are
formed or the solute is sensitive to shear. Yet, comparing CIPS with TIPS, we may conclude that CIPS has
the following advantages:
(1) Improved extraction yield. Since there is no
concern that high temperatures may be required to
reach the single-phase region, the final compositions of
the water-rich and solvent-rich phases can be adjusted
over a wide range. Consequently, the partition coefficient of the solute to be extracted between the two
phases can be higher, to achieve the desired separation.
(2) No need for separate heat transfer equipments. The extraction can be carried out in a single
vessel, e.g, a fermentation vessel. Alternatively, lowcost settlers may be used for large volume production.
In addition, the difficulty of rapidly cooling large
amounts of liquid is eliminated. In fact, since mixing
is always much faster than cooling, the reemulsification
that takes place during slow cooling cannot occur here.
(3) Larger flexibility in the choice of solvents.
The range of primary and secondary solvents that can
be used is much greater due to the elimination of the
temperature constraint.

Conversely, TIPS has the following advantage over

CIPS:
(1) Possibility of multistage processes. Since the
same system of solvents is alternately heated and
cooled, it is simpler to have multistage extractions.
However, this advantage must be weighted against the
much higher single-stage efficiency of CIPS.
To conclude, we list the areas of applications where
the CIPS process may bring significant advantages.
They include the following:
(1) The extraction of a fermentation broth both
in batch and continuous processes. This avoids the
need to centrifuge and gives better yields due to better
contact during the mixing stage.
(2) The extraction of different compounds from
natural plant or animal materials. Due to the
presence of a single water-rich phase, improved contacting between the solvent and the material to be extracted
is achieved. Here there is no water-solvent interface
and no wetting problems, and the solvent can easily
penetrate the cell walls, thereby increasing the yield of
the process.
(3) Replacement of solvents which are environmentally objectionable, e.g. in extraction processes
using chlorinated solvents. The CIPS process is more
powerful and versatile and also permits the use of
solvents which are environment friendly.
Acknowledgment
This work was supported in part by Grant No. CTS9216133 from the National Science Foundation.
Literature Cited
Binder, K. Decay of Metastable and Unstable States: Mechanisms,
Concepts and Open Problems. Physica A 1986, 140, 35-43.
Eliyahu, N.; Ludmer, Z. Mass Transfer across Microorganism Cell
Walls and Phase Separation during Batch Extraction with
Organic Solvent Mixtures. Abstracts of Papers, 1995 AIChE
Annual Meeting, Miami, FL, 1995; American Institute of
Chemical Engineers: New York, 1995.
Froment, G. F.; Bischoff, K. B. Chemical Reactor Analysis and
Design; Wiley: New York, 1979; pp 174.
Gunton, J. D.; San Miguel, M.; Sahni, P. S. The Dynamics of First
Order Phase Transition. In Phase Transition and Critical
Phenomena; Domb, C., Lebowitz, J. L., Eds.; Academic: New
York, 1983; Vol. 8, pp 267-421.
Ludmer, Z.; Shinnar, R.; Yakhot, V. U.S. Patent, No. 4,954,260,
1990.
Mauri, R.; Shinnar, R.; Triantafyllou, G. Spinodal decomposition
in binary mixtures. Phys. Rev. E 1996, 53, 2613-2623.
Shinnar, R.; Mauri, R. U.S. Patent, 1995 (filed).
Sorensen, J. M.; Arlt, W. Liquid-Liquid Equilibrium Data Collection. Chemistry Data Series, part 2; DECHEMA: Frankfurt,
1980; Vol. V, pp 183.
Ullmann, A. Liquid-Liquid Extraction Using Solvent Mixtures
with Critical Points of Miscibility. Ph.D. Dissertation, The City
University of New York, New York, 1993.
Ullmann, A.; Ludmer, Z.; Shinnar, R. Phase transition extraction
using solvent mixtures with a critical point of miscibility.
AIChE J. 1995, 41, 488-500.

Received for review June 16, 1995

Revised manuscript received February 15, 1996
Accepted February 15, 1996X
IE950362T

Abstract published in Advance ACS Abstracts, June 1,

1996.
X