APPLiED AND ENVIRONMENTAL MICROBIOLOGY, Oct. 1977, p. 382-385
`Copyright 0 1977 American Society for Microbiology
`
`Vol. 34, No. 4
`Printed in U.S.A.
`
`Filtration Removal of Endotoxin (Pyrogens) in Solution in
`Different States of Aggregation
`KATHLEEN J. SWEADNER,t MARK FORTE,tt AND LITA L. NELSEN*
`Millipore Corporation, Bedford, Massachusetts 01730
`Received for publication 23 November 1976
`
`Bacterial lipopolysaccharides are recognized as the major cause of pyrogenic
`reactions from parenteral solutions. Molecular filtration was used to remove
`these pyrogenic molecules (endotoxins) from contaminated parenteral solutions.
`Because bacterial lipopolysaccharides can exist in different states of aggrega-
`tion, depending on the composition of the solution they are suspended in, the
`full range of possible states of aggregation was examined by using filters with
`a wide range of pore sizes. Filters of different pore sizes retained endotoxin
`lipopolysaccharide presumed to be in the vesicle form, the micelle form, or the
`detergent-solubilized form in aqueous solutions. Endotoxins (pyrogens) were
`successfully removed from artificially contaminated solutions of concentrated
`antibiotics by using filters of 10,000-nominal-molecular-weight limit.
`an amphiphile with a large hydrophilic polysac-
`charide chain and a hydrophobic, fatty acid-
`containing tail. When isolated, the LPS aggre-
`gates in aqueous solution as one would expect
`of a major constituent of a biological mem-
`brane. Early attempts in this laboratory to
`remove endotoxin from contaminated solutions
`by using molecular filtration met with a bewil-
`dering variability in results. It became appar-
`ent that the solution properties of bacterial
`endotoxin as an amphiphile would have to be
`understood in detail before its behavior with a
`molecular filter of a given effective pore size
`could be predicted with any certainty. The
`work reported in this paper is the result of
`systematic manipulation of the state of aggre-
`gation of Escherichia coli LPS. This informa-
`tion can be used to predict the state of aggre-
`gation of endotoxin in a given solution, so that
`an appropriate molecular filter can be selected
`for removing it.
`
`The principle method for keeping drugs, vac-
`cines, and parenterals of all kinds free of con-
`taminating bacterial endotoxin is to keep the
`manufacturing process and all subsequent han-
`dling sterile. Because maintaining sterility is
`difficult in many processes, as in the production
`of vaccines and biological drugs such as anti-
`biotics, a reliable method is needed for remov-
`ing endotoxin from an accidentally contami-
`nated product. Conventional heat or chemical
`sterilization, which only kills live bacteria,
`does not alter the toxic activity of the pyrogens;
`conventional sterilization by filtration removes
`whole bacteria but not the endotoxic fragments.
`The ion-exchange, charcoal, barium sulfate,
`and heat incubation procedures now being ex-
`plored by others to remove endotoxin are two
`(or more)-step batch procedures that are often
`unreproducible and accompanied by high losses
`of the substance being purified (7, 10). Asbestos
`fiber beds can be used to remove pyrogens (3),
`but their use in the United States is prohibited
`by Food and Drug Administration regulation.
`Work in this laboratory has concentrated on
`using molecular filtration to remove endotoxin
`from contaminated solutions by separation ac-
`cording to size. It has the advantages of being
`inert, and
`straightforward, highly efficient,
`nondestructive to sensitive biological fluids.
`The endotoxin from gram-negative bacteria
`resides in the lipopolysaccharide (LPS), which,
`with phospholipid and protein, makes up the
`bulk of the outer cell membrane. The LPS is
`t Present address: Department of Neurobiology, Harvard
`Medical School, Boston, MA 02115.
`tt Present address: Department of Chemistry, Boston
`College, Chesnut Hill, MA 02167.
`
`MATERIALS AND METHODS
`Purified E. coli LPS was obtained from Mallinck-
`rodt. Mallinckrodt Pyrogent Limulus amoebocyte
`lysate was used routinely to measure endotoxin.
`Lysate was resuspended in 2.25 ml of pyrogen-free
`100 mM tris(hydroxymethyl)aminomethane (Tris)-
`hydrochloride in 0.9% NaCl, pH 7.1. Limulus lysate
`forms a firmer gel in up to 1.0 M Tris buffer than in
`distilled water, and the buffer ensures that the pH
`of the sample does not affect gel formation (unpub-
`lished data). The Mallinckrodt Limulus lysate re-
`sults correlated well with results obtained with
`Worthington or Difco Limulus lysate and with rab-
`bit tests on samples sent to outside laboratories.
`Pyrogen-free water was produced in the laboratory
`
`382
`
`
`1 of 4
`
`FRESENIUS-KABI, Exh. 1037
`
`

`
`VOL. 34, 1977
`
`FILTRATION REMOVAL OF ENDOTOXIN
`
`383
`
`by distillation or by distillation followed by a Milli-
`pore RO (reverse osmosis) system. Sodium cholate
`and deoxycholate were obtained from Sigma Chem-
`ical Co. All other compounds were reagent grade.
`Filtration experiments were performed at room
`temperature in 25-mm, magnetically stirred molec-
`ular filtration cells (Millipore Corp.) under 25-lb/in2
`air pressure. The cells were rendered pyrogen-free
`by washing with soap and water followed by rinsing
`overnight in a continuous stream of pyrogen-free
`water. All glassware and forceps were sterilized by
`heating for 3 h at 200°C, and all fluid transfers and
`measurements were done with pyrogen-free Plasti-
`pak disposable syringes (Becton, Dickinson). A ster-
`ile Millipore Swinnex-25 filter was inserted between
`the cell and the compressed air supply to prevent
`contamination from dust or droplets. The cell was
`flushed with pyrogen-free water several times after
`the filter was wetted and mounted. A sample of the
`final rinse of each cell was assayed with the Limulus
`lysate test to check that the entire apparatus was
`pyrogen-free downstream from the filter before an
`experiment was begun.
`E. coli LPS from a 500-,ug/ml stock solution,
`stored at 4°C, was made up in the test solution to 1
`,ug/ml, mixed thoroughly, and used that same day.
`One milliliter of the test solution was set aside for
`endotoxin assay, and 9 ml was applied to the filter.
`The first 5 ml of the filtrate was collected and
`mixed thoroughly. Parallel 1:9 serial dilutions were
`performed on the original sample and the filtrate,
`and each dilution was tested with Limulus amoebo-
`cyte lysate. Dilutions were made with pyrogen-free
`portions of the same solution used for the filtration
`experiment so that all samples tested would be of
`identical composition. An exception to this protocol
`had to be made for the experiments using cholate
`and deoxycholate. In the presence of these deter-
`gents, even 100 ,ug of E. coli endotoxin per ml was
`not detected by the Limulus lysate test. The deter-
`gent had to be diluted by a factor of 100 with
`detergent-free water, presumably allowing some
`reaggregation of dissociated LPS subunits, before
`activity could be measured. Consequently, the pro-
`tocol for experiments with cholate and deoxycholate
`became: challenge the filter with 10 Atg of endotoxin
`solution per ml, dilute the sample and the filtrate
`1:100 with water, and then perform serial dilutions
`with a pyrogen-free detergent solution similarly
`diluted.
`Filters were selected that have pore sizes that
`decrease in steps of approximately a factor of 10.
`The filters used were Millipore EGWP (0.22-,um
`nominal pore size), VSWP (0.025-t,m nominal pore
`size), PSVP (10" nominal-molecular-weight limit
`[nmwl]), PTHK (105 nmwl), and PTGC (104 nmwl).
`(The nmwl is defined by the ability to retain approx-
`imately 90% of globular proteins of a given molecu-
`lar weight. The nmwl values are given only to
`indicate the range in pore sizes used; they cannot
`be used to assign an absolute size to an LPS particle
`because the asymmetry of a rod-shaped particle
`could lead to either more or less passage through a
`filter of a given pore size, depending on the flexibil-
`ity of the rod. Samples of the filters were soaked
`
`overnight in 1 ml of pyrogen-free water, and both
`the water and the filters were tested for endotoxin
`by the Limulus lysate assay. All samples were
`negative.)
`
`RESULTS
`Filtration of LPS in different states of
`aggregation. In a representative filtration ex-
`periment, a PTGC (10,000 nmwl) filter was
`challenged with a solution of 1 ,gg of endotoxin
`per ml in saline, and serial dilutions of the
`original sample and of the filtrate were per-
`formed. With the lysate test calibrated with
`the sample solution, the concentration of endo-
`toxin in the filtrate can be calculated from the
`number of 10-fold serial dilutions needed to
`reach the same end point, below which the
`endotoxin is too dilute to be detected by the
`Limulus lysate test. In the experiment reported
`in Table 1, then, the limit of sensitivity of the
`lysate test was of the order of 10-10 g/ml. Based
`on this, concentration of endotoxin in the fil-
`trate must be less than 10-10 g/ml, indicating
`that less than 0.01% of the endotoxin passed
`the filter.
`Similar experiments were performed for the
`combinations of filters and solutions repre-
`sented in Table 2. EGWP (0.22 ,um) is a cellu-
`losic filter suitable for retaining bacteria for
`filtration sterilization. It did not hold back
`endotoxin at all. VSWP, which can be used for
`the concentration of many viruses, retained
`endotoxin when it was in water or salt solu-
`tions, but when ethylenediaminetetraacetic
`acid (EDTA) was added to chelate divalent ca-
`tions, the endotoxin passed the filter. PSVP
`(106 nmwl) and PTHK (105 nmwl) filters re-
`tained endotoxin when it was in solutions with
`chelators of divalent cations, but began to pass
`it when detergents were present to break down
`TABLz 1. Results of a representative experimenta
`Limulus lysate assay
`
`Dilution
`
`Filtrate
`_
`-
`-
`
`-
`-
`-
`
`Sample
`0
`+
`1:10
`+
`1:102
`+
`1:103
`+
`1:104
`+
`1:105
`-
`1:106
`_
`1:107
`_
`1:108
`_
`a A 0.9% NaCl solution was contaminated with 1
`,ug of E. coli endotoxin per ml and filtered through
`a PTHK (100,000 nmwl) filter. Serial dilutions were
`tested for endotoxin with Limulus lysate test. Sym-
`bols: +, visible gel formation; -, no visible gel for-
`mation.
`
`
`2 of 4
`
`

`
`384
`
`SWEADNER, FORTE, AND NELSEN
`
`APPL. ENVIRON. MICROBIOL.
`
`TABLE 2. Molecular filtration removal ofLPS endotoxin from solutions ofE. coli endotoxin
`Sample
`Approx concn (g/ml) of endotoxin in filtrate froma:
`
`EGWP (0.22
`%m)
`10-6
`10-6
`10-
`
`VSWP (0.025
`jsm)
`<10-10
`<10-10
`s10-10
`10-6
`
`Endotoxin
`concn (g/ml)
`10-6
`10-6
`10-6
`10-6
`10-5
`
`10-5
`10-5
`
`Compoio.
`Comjosition
`Water
`0.9% NaCl
`5 mM MgCl2
`5 mM EDTA
`0.5% sodium cho-
`late
`1% sodium cholate
`2% sodium cho-
`late, 5 mM
`EDTA
`1% deoxycholate
`<10-10
`10-5
`10-5
`10-5
`a 100 pg/ml (10-10 g/ml) was generally the lower limit of sensitivity of the Limulus assay in our
`laboratory.
`
`PSVP (106
`nmwl)
`<10-10
`<10-10
`<10-10
`<10-10
`10-5
`
`10-
`10-5
`
`PTHK (105
`nmwl)
`<10-10
`<10-10
`<10-10
`<10-10
`10-7
`
`10-7
`10-5
`
`PTGC (104
`nmwl)
`<10-10
`<10-10
`
`<10-10
`
`<10-10
`<10-10
`
`the endotoxin into smaller particles. The PTGC
`filter
`(104 nmwl) did not pass measurable
`amounts of endotoxin in any solution.
`It was possible to remove endotoxin from tap
`water (Bedford, Mass.) by passage through a
`PTGC filter. The treatment lowered the level
`of Limulus lysate-reacting material to below
`the detectable level. In equivalents of E. coli
`LPS, this was a reduction from about 1 to 10
`ng/ml to less than 100 pg/ml.
`Filtration of antibiotic solutions. Two types
`of antibiotics were artificially contaminated
`with endotoxin and purified by molecular filtra-
`tion. A 10% solution of sodium cephalothin
`(Keflin; Eli Lilly & Co.) was made 1 ,ug/ml in
`E. coli endotoxin. The solution was filtered
`through a PTGC membrane (10,000 nmwl),
`using a 25-mm stirred cell in the protocol de-
`scribed above. Results indicated that PTGC
`membranes removed the endotoxin to levels
`lower than the sensitivity of the Limulus lysate
`assay. Similar results were obtained with sam-
`ples of 33% disodium carbenicillin (Beecham
`Laboratories) in water artificially contami-
`nated with 400 ng of E. coli endotoxin (Difco
`Laboratories) per ml and filtered through
`PTGC membranes. The filtrate was Limulus
`negative. (A larger-scale experiment on a sim-
`ilar solution of artificially contaminated [400
`ng of endotoxin per ml] carbenicillin was run,
`using 0.05 m2 of PTGC membrane. More than
`3 liters of solution was filtered. The filtrate
`was negative to the Limulus lysate assay even
`at the end of the run, indicating that no "leach-
`ing" of pyrogen through the membrane oc-
`curred, even after significant quantities of sam-
`ple had been exposed to the membrane.)
`DISCUSSION
`LPS is believed to be arranged in a bilayer,
`analogous to a bilayer of phospholipid, in which
`
`the hydrophilic components are exposed to the
`aqueous environment while the hydrophobic
`fatty acid tails are sandwiched together in the
`center of the bilayer (1, 2, 4). In the case of
`bacterial LPS, this structure is apparently sta-
`bilized by divalent cations, because removal of
`Ca2+ and Mg2+ causes the bilayer to break
`down (5) into what appear to be micelles of
`300,000 to 1,000,000 molecular weight. These
`appear as small rods or disks in the electron
`microscope, 20 to 70 nm long (rods) or in diame-
`ter (disks) and 3 to 7 nm thick (1, 12). Some
`strains of bacteria appear to give micelles of
`smaller size (8). The micelles can be further
`broken down in the presence of detergent or
`bile salts (6, 9, 13, 14). The particles can then no
`longer be seen in the electron microscope, but
`their size and shape, as calculated from sedi-
`mentation velocity, density, and viscosity mea-
`surements, are 0.8 to 1.2 nm in diameter and 20
`to 70 nm long (2, 11), consistent with what is
`known oftheir chemical structure. The molecu-
`lar weight is 10,000 to 20,000. The best evidence
`that these reductions in size are the result of
`breaking noncovalent bonds is that each step is
`freely reversible. When the detergent is dia-
`lyzed out and divalent cations are added back
`to the LPS, micelles and then membranous
`structures reassemble themselves (1, 2, 6, 8, 9,
`11). The expected aggregation state for LPS
`from wild-type gram-negative bacteria in solu-
`tions of various compositions is summarized in
`Table 3, as gleaned from a number of published
`sources. The data presented in this paper cor-
`roborate the work of these other laboratories
`in describing the existence of bacterial LPS in
`states of aggregation of vastly different sizes in
`solution. All endotoxic activity passed a filter
`of 0.22-,um nominal pore size (EGWP) and was
`retained on a filter of 0.025-,um nominal pore
`size (VSWP), in the presence or absence of
`
`
`3 of 4
`
`

`
`VOL. 34, 1977
`TABLE 3. Aggregation state for bacterial LPS
`resuspended in solutions of various compositions
`Solution
`Aggregation state
`Reference
`1, 4
`Vesicle
`1, 8
`Vesicle
`1, 5, 8, 13
`Micelle
`Micelle or subunit
`2, 12
`
`Water
`5 mM MgCl2, CaCi2
`5 mM EDTA
`1.0% sodium deoxy-
`cholate
`2.0% sodium deoxy-
`cholate + 5 mM
`EDTA
`
`Subunit
`
`2,12
`
`salt. The combination of our data and published
`electron micrographs (1, 4, 13) suggests that
`the highest aggregate state in dilute solution
`is a bilayer sheet or vesicle with a diameter on
`the order of 0.1 ,um. Smaller aggregates and
`possibly single subunits of LPS are seen when
`the compositions of the solutions are manipu-
`lated. The bilayer forms can be retained on
`VSWP, the micelles on PSVP, and all smaller
`forms on PTGC. The ground work is laid for
`the application of molecular filtration to the
`removal of endotoxin from contaminated solu-
`tions of small molecules. For example, all low-
`molecular-weight drugs, salts, and nutritional
`compounds should easily pass a PTGC mem-
`brane, leaving contaminating endotoxin be-
`hind. We have shown that solutions of sodium
`cephalothin and carbenicillin can be separated
`from 400 ng of E. coli endotoxin per ml in a
`single filtration through a PTGC membrane
`with no significant loss of antibiotic. Such a
`procedure, accomplishing a separation on the
`basis of size, avoids the problems of adsorption
`and product losses that are found when ion-
`exchange, charcoal, or barium sulfate methods
`are tried for removing endotoxin, and it does
`not have the problem of introducing a toxic
`element into the solution, as the asbestos filter
`method does. As another example, it might
`prove possible to remove endotoxins from vi-
`ruses by retaining the viruses on a filter such
`as PSVP or VWSP and disaggregating and
`washing through the endotoxin with a solution
`of EDTA. In this way virus concentration and
`endotoxin removal could be accomplished in
`the same step.
`Molecular filtration, then, may be of wide
`applicability for the removal of bacterial endo-
`toxins from solutions. It is necessary to assess
`the state of aggregation of the endotoxin in the
`particular solution to be purified and to choose
`a filter of pore size suitable for getting a good
`separation of endotoxin from the solute being
`purified. If the endotoxin and solute are of
`
`FILTRATION REMOVAL OF ENDOTOXIN
`
`385
`
`similar size, it may be possible in many cases
`to manipulate the effective size of the endo-
`toxin, as outlined here, and facilitate molecular
`filtration separation.
`ACKNOWLEDGMENTS
`We would like to thank Frank Cole for many useful
`discussions, and Steven Emmer and Dolores Govahn for
`assistance in some of the experiments.
`
`LITERATURE CITED
`1. DePamphilis, M. L. 1971. Dissociation and reassembly
`of Escherichia coli outer membrane and of lipopoly-
`saccharide, and their reassembly onto flagellar basal
`bodies. J. Bacteriol. 105:1184-1199.
`2. Hannecart-Pokorni, E., D. Dekegel, and F. Depuydt.
`1973. Macromolecular structure of lipopolysaccha-
`rides from gram-negative bacteria. Eur. J. Biochem.
`38:6-13.
`3. Kaden, H. 1975. The use of asbestos filter beds in the
`production of sterile and pyrogen-free solutions.
`Pharmazie 29:752-733.
`4. Katayama, Y., H. Hattori, and A. Suganuma. 1971.
`Membranous structure of purified Escherichia coli
`lipopolysaccharide. J. Bacteriol. 108:1412-1415.
`5. Leive, L., V. K., Shovlin, and S. E. Mergenhagen.
`1968. Physical, chemical, and immunological proper-
`ties of lipopolysaccharide released from Escherichia
`coli by ethylenediaminetetraacetate. J. Biol. Chem.
`243:6384-6391.
`6. McIntire, F. C., G. H. Barlow, H. W. Sievert, R. A.
`Finley, and A. L. Yoo. 1969. Studies on a lipopolysac-
`charide from Escherichia coli. Heterogeneity and
`mechanism of reversible inactivation by sodium de-
`oxycholate. Biochemistry 8:4063-4067.
`7. Nolan, J. P., J. J. McDevitt, and G. S. Goldmann.
`1975. Endotoxin binding by charged and uncharged
`resins. Proc. Soc. Exp. Biol. Med. 149:766-770.
`8. Olins, A. L., and R. C. Warner. 1967. Physicochemical
`studies on a lipopolysaccharide from the cell wall of
`Azotobacter vinelandii. J. Biol. Chem. 242:4994-5001.
`9. Oroszlan, S. I., and P. T. Mora. 1963. Dissociation and
`reconstitution of an endotoxin. Biochem. Biophys.
`Res. Commun. 12:345-349.
`10. Reichelderfer, P. S., J. F. Manischewitz, M. A. Wells,
`H. D. Hochatein, and F. A. Ennis. 1975. Reduction
`of endotoxin levels in influenza virus vaccines by
`barium sulfate adsorption-elution. Appl. Microbiol.
`30:333-334.
`11. Reynolds, J., S. Herbert, and J. Steinherdt, 1968. The
`binding of some long-chain fatty acid anions and
`alcohols by bovine serum albumin. Biochemistry
`7:1357-1361.
`12. Ribi, E., R. L. Anacker, R. Brown, W. T. Haskins, B.
`Malmgren, K. C. Milner, and J. A. Rudbach. 1966.
`Reaction of endotoxin and surfactants. I. Physical
`and biological properties of endotoxin treated with
`sodium deoxycholate. J. Bacteriol. 92:1493-1509.
`13. Rogers, S. W., H. E. Gilleland, Jr., and R. G. Eagon.
`1969. Characterization of a protein-lipopolysaccha-
`ride complex released from cell walls ofPseudomonas
`aeruginosa by ethylenediaminetetraacetic acid. Can.
`J. Microbiol. 15:743-748.
`14. Rudbach, J. A., and K. C. Milner. 1968. Reaction of
`endotoxin and surfactants. III. Effect of sodium lau-
`ryl sulfate on the structure and pyrogenicity of endo-
`toxin. Can. J. Microbiol. 14:1173-1178.
`
`
`4 of 4