Removal of soluble palladium complexes from reaction mixtures by fixed-bed adsorption

Article abstract of DOI:10.1021/op800161v

The increasing use of metal-containing catalysts in producing pharmaceutical intermediates and active pharmaceutical ingredients, in conjunction with requirements of low metal content in a drug substance, has motivated the development of efficient separations processes for metals removal. In this paper, fixed-bed adsorption, an attractive alternative to batch adsorption, was investigated using a reaction mixture from a Heck coupling. Adsorption isotherm determinations using three candidate adsorbents revealed that QuadraPure TU had the greatest affinity and identified the optimal adsorption temperature. The bed residence time was key in obtaining good adsorption efficiency in subsequent fixed-bed experiments. Methods for palladium detection by HPLC were developed to monitor adsorption column performance in quasi real-time. A preliminary design methodology was developed in which actual breakthrough time is estimated from the experimentally determined tradeoff between bed efficiency and residence time. Even at the realized 55% bed efficiency, fixed-bed adsorption requires less than one-fourth the adsorbent needed for a single-stage batch adsorption process.

Full text of DOI:10.1021/op800161v

Organic Process Research & Development 2008, 12, 1209–1217
Removal of Soluble Palladium Complexes from Reaction Mixtures by Fixed-Bed
Adsorption
Michael J. Girgis,*,† Laura E. Kuczynski,‡ Sonia M. Berberena,‡ Caitlin A. Boyd,‡ Pamela L. Kubinski,‡ Megerle L. Scherholz,‡
Donald E. Drinkwater,† Xiaoxuan Shen,† Stanislaw Babiak,† and Brian G. Lefebvre‡
Chemical and Analytical DeVelopment, NoVartis Pharmaceuticals Corporation, East HanoVer, New Jersey, U.S.A., and
Department of Chemical Engineering, Rowan UniVersity, Glassboro, New Jersey, U.S.A.
Abstract:
The increasing use of metal-containing catalysts in producing
pharmaceutical intermediates and active pharmaceutical ingre-
dients, in conjunction with requirements of low metal content in
a drug substance, has motivated the development of efficient
separations processes for metals removal. In this paper, fixed-bed
adsorption, an attractive alternative to batch adsorption, was
investigated using a reaction mixture from a Heck coupling.
Adsorption isotherm determinations using three candidate adsor-
bents revealed that QuadraPure TU had the greatest affinity and
identified the optimal adsorption temperature. The bed residence
Figure 1. Schematic diagram of fixed-bed adsorption process
for palladium-containing feed. The filters shown upstream and
downstream of the fixed bed serve as a precaution against solids
formation, respectively, during reaction and due to particle
attrition in the fixed bed.
time was key in obtaining good adsorption efficiency in subsequent
fixed-bed experiments. Methods for palladium detection by HPLC
were developed to monitor adsorption column performance in
quasi real-time. A preliminary design methodology was developed
in which actual breakthrough time is estimated from the experi-
mentally determined tradeoff between bed efficiency and residence
time. Even at the realized 55% bed efficiency, fixed-bed adsorption
requires less than one-fourth the adsorbent needed for a single-
stage batch adsorption process.
adsorption on solids is the most common, in which a solid
adsorbent (e.g., activated carbon) is added, often as fine particles,
to a reaction vessel, and after a contacting period the slurry is
filtered to give a mixture of much lower dissolved metal
content.7
An attractive alternative to adding fine adsorbent particles
to vessels is to place the adsorbent in a fixed bed placed outside
a vessel (Figure 1) and pump the reaction mixture through the
bed. As the reaction mixture contacts the solid in the fixed bed,
dissolved metal species are adsorbed, giving a product at the
bed exit that is very low in metals content. Such an external
fixed-bed process offers several advantages over adding the
adsorbent to a vessel:
• Vessel fouling, especially when the particles are fine, is
eliminated, along with subsequent tedious vessel cleaning,
which typically requires several solvent rinses and may
require manual scrubbing.8 Solvent consumption and disposal
are thus reduced greatly, cutting material and disposal costs
and decreasing waste. With vessel cleaning time being much
less, vessel availability for subsequent reactions is greater,
leading to greater plant productivity.
Introduction
Metal-containing catalysts, used as organometallic complexes
in solution (especially for cross-coupling reactions) and as solid-
supported metals for various reactions (e.g., hydrogenations)
are employed extensively in the production of pharmaceutical
intermediates and active pharmaceutical ingredients (API’s).1,2
The specifications for low metal content in the final product
(e.g., as low as <10 ppm for palladium in an orally administered
drug substance3) require efficient separations processes for
removal of dissolved metals from reaction mixtures, and high-
throughput methods have been reported to identify suitable
palladium scavengers.4-6 Among palladium removal processes,
* Author for correspondence. E-mail: michael.girgis@novartis.com.
† Novartis Pharmaceuticals Corporation.
‡ Rowan University.
• A possibly slow slurry filtration is eliminated, being
replaced with a clarifying filtration that removes much
smaller amounts of solids. The time for filter preparation
and subsequent cleaning is thus reduced.
(1) Larsen, R.D. Organometallics in Process Chemistry; Springer: New
York, 2004; Vol. 6 and references therein.
(2) Blaser, H.-U.; Malan, C.; Pugin, B.; Spindler, F.; Steiner, H.; Studer,
M. AdV. Synth. Catal. 2003, 345, 103–151.
(3) European Medicines Agency, Guideline on the Specification Limits
for Residues of Metals Catalysts, January 2007, http://www.emea.e-
uropa.eu/pdfs/human/swp/444600.pdf.
(6) Welch, C. J.; Albeneze-Walker, J.; Leonard, W. R.; Biba, M.; DaSilva,
J.; Henderson, D.; Laing., B.; Mathre, D. J.; Spencer, S.; Bu, X.; Wang,
T. Org. Process Res. DeV. 2005, 9 (2), 198–205.
(4) Flahive, E.; Ewanicki, B.; Yu, S.; Higginson, P. D.; Sach, N. W.;
Morao, I. QSAR Comb. Sci. 2007, 26, 679–685.
(7) Garrett, C.; Prasad, K. AdV. Synth. Catal. 2004, 346, 889–900.
(8) Bien, J. T.; Lane, G. C. ; Oberholzer, M. R. In Topics in Organome-
tallic Chemistry; Springer: New York, 2004; Vol. 6, Organometallics
in Process Chemistry, pp 263-283.
(5) Flahive, E.; Ewanicki, B.; Sach, N. W.; O’Neill-Slawecki, S. A.;
Stankovic, N. S.; Yu, S.; Guiness, S. M.; Dunn, J. Org. Process Res.
DeV. 2008, 12 (4), 637–645.
10.1021/op800161v CCC: $40.75
Published on Web 11/21/2008
2008 American Chemical Society
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1209

• The amount of adsorbent required can be much less relative
to a slurry adsorption due to a greater average driving force
for adsorption, cutting material and disposal costs signifi-
cantly.8
The goal of this work is to devise a methodology that will
allow estimation of the breakthrough time for fixed-bed adsorp-
tion applied to the removal of soluble organopalladium com-
plexes and thus arrive at preliminary design procedures for such
separations. The approach taken is to use a reaction mixture,
representative of those encountered in practice, to characterize
the adsorption quantitatively using adsorption isotherms and
evaluate subsequently fixed-bed performance. Palladium ad-
sorption is thus characterized in the presence of other species
(e.g., reaction product, solvent, reagents, and possibly uncon-
verted reactant and intermediates), which could impact metals
removal instead of using either an organometallic catalyst or a
model compound (e.g., PdCl2) alone in solution. The Heck
coupling (see scheme above) was selected as a representative
reaction, as it was catalyzed by a soluble palladium complex
and developed and run in-house on a large scale while giving
a large Pd content in the untreated reaction product (see
below).13
While investigation of a single reaction mixture cannot
provide the necessary information for fixed-bed palladium
adsorption from any reaction mixture, a systematic investigation
using a representative reaction mixture should elucidate the key
factors and challenges for process design for adsorption of
organopalladium complexes.
While performing fixed-bed adsorption experiments, two
palladium-detection methods by HPLC/UV were developed,
which provided a quantitative estimate of palladium content at
the fixed-bed exit. These methods were especially useful for
determining breakthrough times much more rapidly (i.e., within
30 min) than typical palladium detection methods (e.g., AA
and ICP/MS), which require tedious sample preparation lasting
several hours as well as costly instrumentation and specialized
analytical chemistry expertise.
• An adsorbent in a fixed bed can be potentially regenerated
in situ, allowing for possible multicycle use and thus
decreasing the amount of disposed solid adsorbent.
Design procedures for fixed-bed adsorbers are well-known.9
For example, assuming that liquid plug-flow is present and that
thermodynamic equilibrium between liquid and solid phases is
established rapidly, the liquid-phase adsorbate concentration is
essentially zero at the bed exit but rises sharply to the feed value
at the “breakthrough time”, when the solid adsorbate concentra-
tion is in equilibrium with the liquid-phase feed value; in such
cases, the breakthrough time can be calculated on the basis of
adsorption thermodynamics, the feed metals concentration, and
adsorber operating parameters:9
FB q*0
ε Ffc0
L
V
tbrkthru
)
1 +
(1)
[
]
where tbrkthru ) breakthrough time, h; L ) length of fixed bed,
cm; V ) fluid interstitial velocity, cm/h; FB ) bed density, g of
solid/cm3 of bed volume; ε ) bed void fraction, cm3 of liquid/
cm3 of bed volume; Ff ) liquid density, g of liquid/cm3 of liquid;
c0 ) feed metals concentration in ppm, (g metal/g of liquid) ×
106; and q*0 ) solids metals concentration in equilibrium with
feed liquid in ppm, (g metal/g of solid) × 106.
Design procedures have also been established for more
complex situations (e.g., presence of axial dispersion, finite
liquid-solid mass transfer resistance).9
Although cartridge filters for fixed-bed operation are com-
mercially available,10,11 little has been published regarding
design procedures for fixed-bed adsorber operation applied to
removal of dissolved metal species from reaction mixtures,
primarily because the adsorption has not been characterized
quantitatively (i.e., in terms of adsorption isotherms). Typically,
examples are given describing metals removal for a particular
set of conditions (e.g., feed metals content, column volumes
processed per hour).12 While such descriptions can be instruc-
tive, it becomes difficult to extend the results to other sets of
conditions (e.g., higher metals feed concentrations, differing
adsorbent amounts) for purposes of breakthrough time estima-
tion. It would thus be helpful to arrive at a systematic method
for fixed-adsorber design, preferably based on physical insights
such as the thermodynamics of the separation and column
operating parameters.
Experimental Section
Chemical Reaction. The reaction was conducted on the
basis of the procedure developed by Slade and Liu.13 Experi-
ments were conducted on various scales, ranging from 60-600
mL using jacketed glass vessels equipped with overhead
(9) Sherwood, T. K. ; Pigford, R. L.; Wilke, C. R. Mass Transfer;
McGraw-Hill: New York, 1975; Chapter 10.
(10) Reaxa.com, http://www.reaxa.com/images/stories/reaxa_quarapureuser-
guide.pdf.
(11) http://www.picacarbon.com/en/markets/health-pharma/purifying-phar-
maceutical-ingredients/.
(12) Hinchcliffe, A.; Hughes, C.; Pears, D.; Pitts, M. R. Org. Process Res.
DeV. 2007, 11 (3), 477–481.
(13) Slade, J.; Liu, H. personal communication, June, 2006.
1210
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

agitators (retreat curve and pitched-blade impellers at the small
and large scale, respectively). In a typical run conducted at a
600-mL scale, 183 g of aldehyde 1 (989 mmol), 1.11 g of
Pd(OAc)2 (5 mmol), 6.02 g of P(o-Tol)3 (20 mmol), 467 g of
CH3CN (594 mL), 189 g of Bu3N (1020 mmol), and 87.9 g of
olefin 2 (1020 mmol) were charged to the reaction vessel. The
reaction mixture was heated under a nitrogen purge to 75 °C
over 20 min and held at that temperature for at least 16 h.
Conversion was monitored by HPLC. Instrument and method
details, as well as sources and purities of materials, are given
elsewhere.14
Batch Adsorption Experiments. Isotherm determinations
were carried out using an apparatus comprising six constant-
temperature small baths, each with independent temperature and
agitation control. Vials containing 15-20 mL of Heck coupling
reaction mixture (transferred to each vial under nitrogen purge)
were placed in a given bath, with varying amounts of adsorbent
(0.025-2 g) added to each vial to determine the isotherm. Care
was taken to minimize exposure of the Heck coupling reaction
mixture to air, as extensive exposure to air sometimes resulted
in formation of palladium black particles. The slurries were
agitated with magnetic stir bars overnight to allow equilibration,
with the vials were sealed to minimize solvent loss and ensure
good material balance closure (typically >95%). The solids
were filtered, and the filtrates were analyzed for Pd content using
ICP/OES by Robertson Microanalytical Laboratories (Madison,
NJ). Solid-phase Pd content was calculated by difference.
Three adsorbents that are commonly used as palladium
scavengers in the pharmaceutical industry were investigated:
(a) P1400 activated carbon (PICA),11 (b) Smopex 110,15
comprising thiourea-grafted polyolefin fiber (Johnson Matthey),
and (c) QuadraPure TU,10 consisting of thiourea bound to 0.5-
mm resin beads (Reaxa Ltd.). All adsorbents were used as
received.
Fixed-Bed Adsorption Experiments. The apparatus used
for the fixed-bed experiments is depicted in Figure 2. The
adsorbent was placed in a jacketed glass column of 1.5-cm
internal diameter and 30-cm length (Ace Glass, part no. 5821-
13). The reaction mixture was pumped through the column with
a Cole Parmer 74900 syringe pump. The column effluent was
collected in a flask placed on a datalogged balance to monitor
the flow rate. The jacket of the column was attached to a
temperature bath (Thermo Neslab model RTE740) filled with
50:50 propylene glycol/water. To avoid crystallization of the
product in the reaction mixture, the syringes and transfer lines
to the column were placed in a 40 °C chamber.
Figure 2. Schematic diagram of laboratory fixed-bed adsorp-
tion apparatus.
inlet; the RTD could be moved inside the 1/8-in. tube to verify
that the axial temperature was constant. With the 0.5-mm
diameter adsorbent particles used, the bed diameter was greater
than 15 particle diameters, ensuring plug flow, even in the zone
with the 1/8-in. tube present. A diagram detailing the fixed bed
is given elsewhere.14
The column and transfer tubing were filled with acetonitrile
solvent before introducing the reaction mixture. Thus, the system
was hydraulically full before reaction mixture introduction to
give a well-defined start of adsorption. Small (ca. 1-mL) samples
of column effluent were taken for HPLC analysis to follow
palladium breakthrough.
HPLC Methods for Palladium Detection. In initial fixed-
bed experiments (see Figure 10), an HPLC method was
developed comprising a modification of the one used for
reaction monitoring.14 This method was based on the finding
from LC/MS analyses that a peak in the reaction-monitoring
chromatograms was associated with palladium-containing spe-
cies. The method is described elsewhere.14 Additional experi-
ence indicated that this method was not sufficiently sensitive
to allow detection of palladium levels below ca. 10 ppm.
A much more sensitive palladium-detection HPLC method
A bed of glass beads (diameter 0.5 mm) was placed at the
column inlet to ensure plug flow and attainment of the desired
temperature at the adsorbent bed entrance; a bed of glass beads
was also placed at the adsorbent bed exit. In some cases, the
resin particles were diluted with glass beads to a give a
sufficiently long bed to minimize axial dispersion (bed length/
particle diameter of at least 200). To obtain an accurate measure
of bed temperature, a thin Pt100 resistance temperature detector
(RTD) was placed inside a 1/8-in. diameter tube, which was
sealed on one end and attached to the bottom of the column
was subsequently developed. The latter was adapted from a
16,17
literature method
and is based on the formation of UV-
detectable complexes between soluble palladium species and
sodium diethyldithiocarbamate (Na-DEDTC). The structure of
the complexes is unknown but may involve coordination of a
palladium cationic species with two diethyldithiocarbamate
anions. Sample preparation involved combining 0.1 mL of Pd-
containing solution with 0.1 mL of a 0.02 M Na-DEDTC
solution and 0.5 mL of acetonitrile. Chromatographic conditions
are given in Table 1. Using this method, a well-resolved
(14) Supplemental Information, available online.
(15) http://www.smopex.com/page-view.php?page_id)122&parent_page_
(16) Mueller, B. J.; Lovett, R. J. Anal. Lett. 1985, 18 (A19), 2399–2419.
(17) Mueller, B. J.; Lovett, R. J. Anal. Chem. 1985, 57, 2693–2699.
id)3.
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1211

Table 1. Second HPLC method for palladium detection
instrument
Agilent series 1100 HPLC, Model G1311A quaternary pump, model G1315B photodiode array detector
column
Zorbax SB-C18, 150 × 3 mm, 3 µm
25 °C
column temperature
mobile phases
A: 0.02 M sodium acetate buffer (pH ) 6)/acetonitrile ) 40/60 (v/v)
B: 0.02 M sodium acetate buffer (pH ) 6)/acetonitrile ) 5/95 (v/v)
flow rate
0.8 mL/min
10 µL
injection volume
detection wavelength
298 nm
gradient elution
% B
time, min
0
0
0
9
100
12
100
20
0
21
0
26
chromatographic peak corresponding to the palladium complex
was obtained.14
ponent. The shape of the isotherm with QuadraPure TU is
consistent with a Langmuir form, with the solid-phase concen-
tration approaching a constant value at high liquid-phase
concentrations; the convex-upward shape of the isotherm with
QuadraPure TU implies that it may be characterized as type I
favorable.18
The much greater palladium affinity of QuadraPure TU
motivated its further evaluation. As shown in Figure 4, even
greater palladium affinity was obtained at higher temperature,
the cause of which is unknown. Two possible reasons for the
greater affinity include (a) a possibly activated adsorption, with
higher temperature needed for a selective interaction between
thiourea functional groups on the resin and soluble palladium
species and (b) a greater availability of adsorption sites because
of resin swelling at higher temperature. The adsorption also
follows the Langmuirian form at higher temperatures, with good
fits obtained when fitting the data to the Langmuir isotherm
(Figure 4):
Results and Discussion
Chemical Reaction and Palladium Content of Untreated
Mixture. Conversion of aldehyde 1, which is the limiting
reactant, was almost complete (i.e., >99%) after 16 h at 75
°C; representative kinetics are shown elsewhere.14
To obtain a base case for product palladium content, coupling
product 3 from one run was isolated without adsorptive
treatment, based on a procedure developed earlier.13 After
cooling the reaction mixture to room temperature, 0.5 N HCl
(104.1 g) and water (680 g) were added to produce a slurry of
product 3. After filtration and drying, the product was recovered
in 89% yield, with a Pd content of 2100 ppm. This large value,
being in excess of drug substance palladium content specifica-
tions by more than 2 orders of magnitude, and the observation
that palladium is preferentially taken up by the solid during
precipitation (as the calculated reaction mixture Pd content is
only 570 ppm) are a further indication of the suitability of this
Heck coupling for investigating palladium removal.
NKc
1 + Kc
q )
(2)
Adsorption Isotherms. The adsorption isotherms at room
temperature for the three adsorbents examined (Figure 3)
indicate that QuadraPure TU had the greatest affinity for
palladium by far, with the equilibrium solid-phase concentration
being 1 order of magnitude greater than those of the other
adsorbents at a given liquid-phase concentration in equilibrium
with the solid. It is likely that palladium exists in solution in
various chemical forms; for purposes of isotherm determination,
all palladium-containing species are lumped into one pseudocom-
where q ) solid-phase Pd concentration in ppm, (g of Pd/g of
solid) × 106; c ) liquid-phase Pd concentration in ppm, (g of
Pd/g of liquid) × 106; N ) constant representing the number
of adsorption sites, (g of Pd/g of solid) × 106; K ) constant
representing the adsorption equilibrium constant, (g of liquid/g
of Pd) × 10-6.
Figure 3. Equilibrium isotherms for three adsorbents with
Heck coupling reaction mixture at room temperature.
Figure 4. Adsorption isotherms for QuadraPure TU at three
temperatures with Heck coupling reaction mixture.
1212
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

Table 2. Fitted Langmuir isotherm parameters
temperature
N (g of Pd/g
K (g of liquid/g
of Pd) × 10-6
(°C)
25
40
60
of solid) × 106
8090
8586
18270
5.96 × 10-2
6.37 × 10-2
7.75 × 10-2
The values of the parameters fitted are given in Table 2.
The fit indicates that both the number of sites and the adsorption
strength increase with temperature but that the N parameter
increases more with temperature (viz., by a factor of >2.2 versus
an increase of only 11% for K). The small change in K (which
might be constant within experimental error) suggests that the
greater adsorption capacity at higher temperature might be due
predominantly to an increase in the concentration of active
adsorption sites.
The selectivity of the adsorption was also characterized. After
equilibration with the adsorbent and subsequent filtration,
selected liquid-phase samples were assayed for content of
product 3. Despite some scatter in the data, the results indicate
that the adsorption was more selective at higher temperature
(Figure 5).
Figure 6. Batch adsorption experiment results.
the observed rate of disappearance by at least 2 orders of
magnitude, consistent with a slow reaction.14
Fixed-Bed Experiments. The results of Figure 6 suggest
that a ca. 6-h bed residence time is needed to approach
fluid–solid equilibrium. However, because such a long bed
residence time gives low throughput (viz., <0.2 bed volumes
processed per hour), it is useful to quantify the tradeoff between
throughput and extent-of-approach to equilibrium, as it may then
be possible to determine if acceptable throughput can be
obtained within a reasonable residence time range.
Approach to Equilibrium Determination. The isotherm
determinations and selectivity data allowed selection of the
optimal adsorbent (QuadraPure TU) and the best temperature
for that adsorbent (60 °C) for the Heck coupling reaction
mixture. To test the hypothesis of rapid equilibrium attainment
between fluid and solid phases in the fixed bed and thus guide
selection of fixed-bed experimental conditions, a batch adsorp-
tion experiment was conducted in which QuadraPure TU was
added to the Heck coupling reaction mixture in a vessel
equipped with an agitator. Several liquid samples were taken
and analyzed for palladium content to quantify the approach to
equilibrium as a function of time.
As indicated in Figure 6, up to 6 h of contact was required
to reach equilibration between the resin and the Heck coupling
reaction mixture. The high agitation in the batch adsorption
experiment suggests that mass transfer resistance between fluid
and solid phases is not high. Thus, the long equilibration time
is consistent with a slow chemical reaction between dissolved
palladium species and thiourea functional groups on the resin
surface. Comparison of the rate of disappearance of palladium
at early times based on Figure 6 with a conservatively low
estimate of the mass transfer rate reveals that the latter exceeds
A systematic methodology was employed in the fixed-bed
experiments in which independent variables (key inputs) were
specified. Based on these variables, outputs, comprising column
operating variables and performance (i.e., breakthrough time),
were calculated. The input and output variables, including their
method of computation, are given in Table 3. The variables
are categorized as either column parameters, thermodynamic
properties, or adjustable parameters. With the apparatus used
(Figure 2), bed density, porosity, and cross-sectional area were
fixed. The palladium feed solution concentration for the Heck
coupling reaction mixture was 570 ppm, and based on the
QuadraPure TU isotherm at 60 °C (Figure 4, Table 2), the
equilibrium value on the solid was calculated to be 1.8 × 104
ppm. A value of 17.3 g for the resin mass was selected to give
a 20-cm long bed.
A suitable residence time based on the batch adsorption results
was then chosen, thus specifying the flow rate and allowing
calculation of a theoretical breakthrough time as shown in Table
3. The actual breakthrough time would then be determined
experimentally. An adsorption efficiency η, defined as the ratio of
experimental to calculated breakthrough times, is calculated. The
tradeoff of efficiency with residence time would be quantified from
experiments at various bed residence times.
A 2-h bed residence time was first selected, as the curve
in the batch adsorption experiment begins to flatten at
about this time (Figure 6). As a result of limitations of
reaction mixture availability, it was necessary to decrease
the flow rate late in the experiment by 21% (i.e., from
0.115 to 0.09 cm3/h) after 55.3 h. This flow rate change
increases the bed residence time by a factor of 1.25.
Examination of Figure 6 shows that increasing the bed
residence time from 2 to 2.5 h does not increase the
approach to equilibrium significantly. Thus, the break-
through time can be reasonably approximated as 70.5 h
Figure 5. Improved adsorption selectivity at 60 °C with
QuadraPure TU.
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1213

Table 3. Process variables in fixed-bed adsorption
value of input variables
set by feed properties, equipment
and adsorption isotherm
symbol/method of determination
and units
parameter
category
bed porosity
bulk density
input: bed property
input: bed property
input: column property
input: feed solution
input: solid-fluid
thermodynamics
input
input
output
output
ε, cm3 liquid/cm3 bed volume
FB, g solid/cm3 bed volume
A, cm2
0.4 cm3/cm3 (measured)
0.50 g/cm3 (measured)
cross-sectional area
feed palladium content
solid Pd concentration in
equilibrium with feed
resin mass in bed
bed residence time
bed length
1.69 cm2 (known)
c0, (g Pd/g liquid) × 106
570 ppm (calculated)
q*0, (g Pd/g solid) × 106
1.8 × 104 ppm (calculated from isotherm)
Mres, g
τ, h
17.3 g (measured)
adjustable
20 cm, set by selection of Mres, FB, and A
set by selection of τ and other input parameters
L ) (Mres/FB)/A, cm
volumetric flow rate
Q ) (Mres/FB)(ε/τ), cm3/h
interstitial velocity
breakthrough time
volume of solution processed
at breakthrough
output
output
output
V ) Q/Aε, cm/h
tbrkthru ) L/V [1 + (FB/ε)(q*0/Ffc0)], h
Vbrkthru ) Q tbrkthru, cm3
(Figure 7). Comparison of the latter with the calculated
breakthrough time of 104 h gives an efficiency of 68%
(Table 4).
Column effluent samples that were collected up to 70.5 h
were combined, and the product was isolated by crystallization
as described above. The isolated product contained 3.9 ppm of
palladium (weighted average from 2 crops), well within
guidelines for palladium content in APIs.3
A residence time of 1.25 h was examined subsequently, as
the batch adsorption results indicate that slightly less product
is adsorbed relative to the 2-h time (Figure 6). The breakthrough
curve for this experiment (Figure 8) shows that samples taken
at early times (viz., e 20 h) contained some palladium, but the
low content (<1 ppm) of these motivated continuing the run.
A late sample terminating at 35.7 h contained 3.9 ppm of Pd,
and thereafter the Pd content increased sharply. Taking the
breakthrough time as 35.7 h, the efficiency in Table 5 is
calculated, further demonstrating the importance of sufficiently
long contact time to obtain good fixed-bed performance. The
palladium-removal capability of the fixed bed was again
evidenced by the low palladium content of the isolated solid
(6.8 ppm, weighted average palladium content from two crops).
The latter results, in conjunction with those from subsequent
experiments, were used to quantify the tradeoff between bed
residence time and efficiency (Figure 9). Despite some scatter
in the data, the figure indicates that the adsorption efficiency is
not improved significantly by increasing the bed residence time
above ca. 0.75 h.
The low residence time runs were based in part on adsorbent
vendor recommendations. Examination of the vendor literature
indicated that a flow rate corresponding to 4-6 column
volumes/h was recommended.10,12 A more conservative value,
namely 1 column volume/h, was used in a representative run
shown in (Table 6). On the basis of the vendor-specified resin
capacity of 20 mg Pd/g resin10 and the Pd feed content of 570
ppm, the resin mass given in Table 6 should have been sufficient
to process about 107 mL of reaction mixture, giving a
breakthrough time of 5.9 h at the flow rate employed.
The result from this representative run is shown in Figure 10
with samples of the column effluent taken periodically and analyzed
for Pd using ICP/OES and the first short-cut HPLC method
Figure 7. Breakthrough profile with 2-h residence time. Zero
time corresponds to the time at which the reaction mixture
reached the bed entrance. Palladium detection in this figure
and subsequent ones is determined using the more sensi-
tive HPLC/UV method involving addition of Na-DEDTC before
sample analysis.
Figure 8. Breakthrough data for fixed-bed adsorption run with
1.25-h bed residence time.
Table 4. Results from fixed-bed adsorption run with 2-h bed
residence time
Table 5. Summary of run with 1.25-h bed residence time
flow rate, cm3/h
6.91
10.2
103.7
70.5
0.68
flow rate, cm3/h
11.0
16.2
65.3
35.7
0.55
interstitial velocity, cm/h
calculated breakthrough time, h
actual breakthrough time, h
η
interstitial velocity, cm/h
calculated breakthrough time, h
actual breakthrough time, h
η
1214
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

like a surface chemical reaction that requires a certain contact
time to obtain sufficient conversion.
Investigation of Axial Dispersion and Bed Residence
Time. Additional experiments were performed to determine if axial
dispersion could adversely impact breakthrough time at the
conditions examined. If axial dispersion effects were significant,
plug flow would not occur, and the bed is not efficient because of
a diffusive flux superimposed on bulk flow along the bed axis.
Toluene tracer tests were performed in which acetonitrile was
pumped through the column filled with either glass beads or resin
particles, with a single 1-mL pulse of toluene injected near the
column inlet at the start of each experiment. Samples were then
taken at the column outlet and analyzed by HPLC to measure the
temporal toluene concentration profile.
Key results of these experiments14 showed that the toluene
temporal profile was similar for various flow rates. Additional
experiments demonstrated that axial dispersion was minimized
with an undiluted bed. Thus, glass beads were not used to dilute
the bed in fixed-bed experiments (see above). Adsorption
column runs were then performed at the optimal conditions
suggested by the toluene tracer experiments. Even under optimal
conditions, the breakthrough time did not increase significantly,
being comparable to that shown in Figure 10 at 0.25-h bed
residence times. These experiments ruled out axial dispersion
as the main cause for the early breakthrough time at such low
bed residence times.
Figure 9. Tradeoff of efficiency with residence time.
Table 6. Representative conditions for initial fixed-bed
adsorption runs
mass of QuadraPure TU resin in bed, g
mass of glass beads in diluted resin bed, g
diluted resin bed volume, cm3
diluted resin bed height, cm
bed void fraction
2.4
20.2
16.9
10
0.48
0.3
28
flow rate, cm3/min
residence time in diluted bed, min
mentioned above. In sharp contrast to expectations, breakthrough
occurred in under 2 h, with only 11.5 g of reaction mixture treated
before palladium was detected at the outlet. Based on the resin
mass in the bed, the resin removed only 2.7 mg Pd/g resin,
representing only 13.5% of the expected value of 20 mg/g.
The data from the fixed-bed adsorption experiments per-
formed at various bed residence times (Figure 9) demonstrate
that a sufficiently long bed residence time, and thus close
approach to fluid-solid equilibrium, is key in obtaining good
fixed-bed adsorption performance (i.e., η > 50%). Considering
that the interaction between dissolved palladium species and
thiourea functional groups on the QuadraPure TU resin is likely
to be selective (versus likely physical adsorption between
adsorbents such as activated carbon), the long required residence
time suggests that the adsorbent/adsorbate interaction behaves
Design for Large Scale. For design purposes, eq 1 is
modified to read
FB q*0
ε Ffc0
L
V
tbrkthru ) η 1 +
(3)
[
]
where η is determined experimentally (Figure 9) and is assumed
to be a function of the residence time alone and scale-
independent. Then the equation above implies that
Figure 10. Representative result of palladium breakthrough at column exit in initial runs. The palladium content was determined
using a HPLC/UV method (based on modification of the reaction monitoring method; see above) and by ICP/OES (Robertson
Micronanalytical Laboratories). The HPLC/UV results are expressed in terms of a relative concentration equal to the actual
concentration divided by the undetermined response factor.
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1215

Thus, with specification of the equilibrium isotherm and an
estimate of column efficiency, a design for a large-scale column
can be made. In practice, solutions containing dissolved metals
might oxidize with time, leading to variations in the isotherm,
which can be difficult to predict. It is thus recommended that
when a large-scale column is operated, the column effluent
should be collected in fractions, each of which can be analyzed
via a shortcut HPLC method, with only those fractions before
breakthrough combined; this approach would compensate for
isotherm variations. It might also be helpful to prepare a second
column at the outset, connect it in parallel to the first, and
redirect the feed solution to the second column in case
breakthrough occurs unexpectedly early. Finally, for the sake
of avoiding large differences in mass transfer resistance between
small and large scales, it is advisable to operate at interstitial
velocities that do not differ significantly on scale-up.
tbrkthru
tbrkthru
)
(4)
plant
lab
L ⁄ V
L ⁄ V
where the subscript “lab” refers to an actual laboratory experi-
ment and the subscript “plant” refers to a large-scale design
case.
If a volume of V liters of solution with a Pd concentration
of c0 is to be processed, one can proceed two ways:
(a) Assume tbrkthru|plant equals tbrkthru|lab
Then the volumetric flow rate Q is fixed and calculated from:
.
V
tbrkthru
Q )
(5)
The resin mass is determined from (Table 3)
Comparison with Batch Adsorption. In the 1.25-h resi-
dence time run, about 394 mL, or 310 g, of solution were treated
with 17.3 g of resin, giving a solution mass L to resin mass S
ratio of 17.9 g/g. Based on the equilibrium isotherm and a mass
balance, it is possible to determine the amount of palladium
removed using batch adsorption at the same L/S value.19
Assuming that no palladium is present on fresh resin, a mass
balance on a single stage of batch adsorption shows that
QFBτ
ε
Mres
)
(6)
(Note that that one would operate at the same η value on both
scales and that Figure 9 implies that selection of η fixes τ).
Once a large-scale column is specified, the cross-sectional
area is fixed, allowing calculation of the required bed length:
Mres ⁄ FB
L )
(7)
L(c0 - c1) ) Sq1
(12)
A
where L ) mass of solution treated; S ) mass of adsorbent; c0,
c1 ) liquid-phase Pd concentration in feed and product,
respectively; and q1 ) solid-phase Pd concentration after single-
stage contacting. Solving for q1 gives
The bed length L must obviously be shorter than the actual
length of the selected column; otherwise, a longer column must
be found. Alternatively, one can use a different breakthrough
time on scale-up (see below).
(b) Assume a different tbrkthru plant
|
L
S
L
S
In this case, the quantity F is defined as the ratio of the large-
scale breakthrough time to that in the laboratory:
q1 ) - c1 + c0
(13)
On c,q coordinates (taking c as the horizontal axis), the mass
balance corresponds to a straight line with slope -L/S and
horizontal intercept of c0. If a batch adsorber is operating with
a feed palladium concentration of 570 ppm and with L/S )
17.9 g/g, the equilibrium concentration after a single-stage of
contacting can be determined graphically by reading the ordinate
at the intersection of the mass balance line (i.e., also called the
operating line) with the isotherm (Figure 11). A more accurate
determination can be done algebraically, giving a value of 15
ppm for the liquid-phase palladium concentration after a single-
stage batch adsorption. Assuming that the palladium concentra-
tion in the isolated solid would be 3.7-fold greater (as seen in
at least two cases; see above), a value of 55 ppm would be
obtained, more than 7-fold greater than the actual palladium
content of the isolated solid (7 ppm, see above).
tbrkthruplant
F )
(8)
tbrkthrulab
The flow rate Q is then specified from
V
Q )
(9)
F * tbrkthrulab
allowing calculation of the resin mass and bed length, respec-
tively, from
QFBτ
ε
FBτ
ε
V
Mres
)
)
(10)
F * tbrkthrulab
To obtain a palladium content of 7 ppm in the isolated solid
(or 1.9 ppm in the liquid-phase reaction mixture, based on the
assumption listed above) with a single batch adsorption stage,
calculations based on the analysis above indicate that a L/S value
of 4.13 is needed, or 4.3 times the resin required in the fixed-
bed with 1.25-h residence time and 55% efficiency.
and
Mres ⁄ FB
V
τ
L )
)
(11)
A
F * tbrkthrulab Aε
F can be increased until L is sufficiently below the actual
length of the column chosen.
If a second stage of batch adsorption is used, it can be shown
that in the optimal case, only a slightly lower resin amount can
1216
•
Vol. 12, No. 6, 2008 / Organic Process Research & Development

the goal of devising a methodology for removal of dissolved metals
species from reaction mixtures. The adsorption was first character-
ized quantitatively by determining adsorption isotherms for three
commonly used adsorbents. The isotherms showed that Quadra-
Pure TU had the greatest affinity for dissolved palladium and
indicated its optimal adsorption temperature. The unexpectedly
short breakthrough times in small residence time experiments were
attributed to insufficient approach to equilibrium in the fixed bed
rather than axial dispersion effects. Subsequent experiments, in
which bed residence times were selected to approach equilibrium
on the basis of batch adsorption results, gave much better
performance, with bed performance exceeding 55% of the theoreti-
cal degree of palladium removal based on the assumption of local
fluid-solid equilibrium. These experiments were also used to
develop a design methodology in which actual breakthrough time
could be estimated from experimental determination of the tradeoff
between bed efficiency and residence time. To arrive at the same
level of palladium removal, fixed-bed adsorption was shown to
require less than one-fourth the resin amount relative to a single
stage of batch adsorption. HPLC methods for detection of soluble
palladium allowed for quasi real-time monitoring of adsorption
column performance.
Figure 11. Graphical depiction of single-stage batch adsorp-
tion.19
be used to obtained the same degree of palladium removal
relative to the fixed-bed case with 1.25-h residence time and
55% efficiency. Summing the material balance equations across
both stages gives
1 + Kc
NKc1
1 + Kc2
NKc2
S1 + S2 ) L(c0 - c1)
1 + L(c1 - c2)
(14)
Acknowledgment
We thank Drs. Joel Slade and Hui Liu and Mr. Mark Davis
for consultations and assistance with the Heck coupling reaction,
Dr. Mahavir Prashad for advice on adsorbent selection, and Drs.
Thomas Blacklock and Oljan Repicˇ for management support.
We are especially grateful to Mr. Lee Alden for his invaluable
assistance with design and construction of the experimental
apparatus.
where the subscripts correspond to the two stages.
An iterative calculation can be performed to estimate
the minimum adsorbent amount required for the two stages
by (a) estimating a value for S1, (b) calculating c1 from
the stage 1 mass balance, (c) calculating S2 from the stage
2 mass balance (with c2 set to 1.9 ppm), (c) calculating
S1 + S2, and (e) varying S1 to minimize the latter sum. In
this manner, we find that the minimum adsorbent amount
for the two stages results when S1 ) 12.7 g and S2 )
4.10 g, giving a total adsorbent amount that is slightly
smaller than the 17.3 g case with the fixed bed adsorber
operating at 1.25-h residence time and 55% efficiency.
Note, however, that the throughput would be decreased
significantly owing to a second batch adsorption stage for
only a slight decrease in adsorbent amount.
Supporting Information Available
Experimental details. This material is available free of charge
via the Internet at http://pubs.acs.org.
Received for review July 8, 2008.
OP800161V
(18) Perry, R. H.; Green, D. W.; Maloney, J. O. Perry’s Chemical
Engineers’ Handbook, 6th ed.; McGraw-Hill: New York, 1984;
Chapter 12.
(19) Treybal, R. E. Mass Transfer Operations, 3rd ed.; McGraw-Hill: New
York, 1980; Chapter 11.
Conclusions
Removal of soluble palladium complexes from a Heck coupling
reaction mixture was investigated using fixed-bed adsorption, with
Vol. 12, No. 6, 2008 / Organic Process Research & Development
•
1217