Electrophoretic and mass spectrometric strategies for profiling bacterial lipopolysaccharides

Abstract

Capillary electrophoresis (CE) is a high-resolution separation technique that has been widely used for trace analysis in biological samples. On-line capillary electrophoresis–electrospray mass spectrometry (CE–MS) was developed for the analysis of lipopolysaccharide (LPS) glycoforms from the gram-negative bacteria, Haemophilus influenzae. In this paper, we report on the application of CE–MS to characterize structural differences in O-deacylated LPS samples from H. influenzae strains Rd 11.7 and 375.1. The resolution capability of on-line CE–MS was first demonstrated by analysis of a complex LPS mixture from H. influenzae strain Rd 11.7. This strain contains a mixture of isomeric glycoforms differing in the number and positions of hexose moieties. Sialic acid containing glycoforms were also determined. Structural features of LPS from a lic1 mutant of H. influenzae strain 375 (375.1) were studied using on-line CE–MS/MS. With the separation provided by CE, two isomeric glycoforms differing in the location of phosphoethanolamine substituents were characterized by tandem mass spectrometry.

---

Introduction

The gram-negative bacterium Haemophilus influenzae comprises both capsular and non-typable (acapsular) strains of significant virulence and pathogenicity to humans. Lipopolysaccharide (LPS) molecules make up the outer membrane of this pathogen and their carbohydrate regions provide targets for recognition by host immune responses. Expression of certain oligosaccharide epitopes is known to contribute to the pathogenesis of H. influenzae infections.¹ The structures of H. influenzae LPS have been extensively investigated. H. influenzae LPS is composed of a membrane-bound lipid, a moiety to which an oligosaccharide portion is attached by a single 3-deoxy-D-manno-octulosonic acid (KDO) residue.² A structural model consisting of a conserved tri-L-glycero-D-manno-heptosyl inner-core moiety, in which each of the heptose residues (Hep_I–Hep_III) can provide a point for elongation by hexose containing oligosaccharide chains or for attachment of non-carbohydrate substituents, is now well established (Fig. 1, structure 1).^3–8H. influenzae LPS can undergo phase variation between defined oligosaccharide structures,^9,10 leading to an extensive repertoire of glycoforms within and between strains.^11,12 The addition of phosphate-containing substituents, including monophosphate (P), phosphoethanolamine (PEtn), pyrophosphoethanolamine (PPEtn) and phosphocholine (PCho), as well as glycine and O-acetyl substituents, contribute to the structural variability of these molecules.

Fig. 1 Structure of the lipid A and core region of LPS of H. influenzae (R₁, R₂, R₃ = H or sugar residues).³ It contains the conserved L-glycero-D-manno-heptopyranosyl trisaccharide inner-core region (Hep_I–Hep_III). Hep_I is linked to the lipid A portion of the molecule via a 3-deoxy-D-manno-octulosonic acid (KDO) residue that carries phosphate (P) or pyrophosphoethanolamine (PPEtn) substituents at the O-4 position. The triheptosyl inner-core unit is substituted by a β-D-glucopyranose residue at the O-4 position of Hep_I, by a phosphoethanolamine residue (PEtn) at the O-6 position of Hep_II and, in strain Rd and 375, by oligosaccharides of varying length (R₃) through sequential addition of sugar units at the O-2 position of Hep_III. Hep_III can also carry a PEtn substituent at the O-4 position in strain 375.

The inherent heterogeneity and structural complexity of these molecules pose significant analytical challenges. Previous investigations have demonstrated the application of mass spectrometry (MS) for the structural characterization of trace-level bacterial LPS and complex glycolipids.^13–15 The coupling of high-resolution separation techniques, such as capillary electrophoresis (CE), has enabled the analysis of complex biological extracts with unparalleled resolution and sensitivity. During the last few years our laboratory has developed on-line CE–MS techniques for trace level analysis of LPS glycoform and isoform distributions from a range of bacterial species.^16–20 In this article we report on the application CE–MS to the profiling of isomeric glycoforms and phosphoforms from H. influenzae LPS. The exceptional analytical potential of CE–MS to separate closely related glycoform families is demonstrated for LPS samples obtained from two H. influenzae strains, Rd11.7 and 375.1, that differ in glycoform composition and phosphoethanolamine substitution patterns.

Results and discussion

Profiling microheterogeneity of O-deacylated LPS

CE–MS in the negative ion mode of O-deacylated LPS (LPS–OH) from H.influenzae strain Rd 11.7 revealed the heterogeneity of this LPS sample. The O-acyl groups were first removed from the LPS sample using well established procedures to facilitate its solubility in the aqueous buffer solutions. This produces an O-deacylated lipid A (lipid A–OH) portion (see Structure 1) having a conserved bis-phosphorylated β-1,6-linked glucosamine disaccharide substituted by 3-hydroxytetradecanoamide groups at C-2 and C-2′, to which the triheptosyl inner-core unit is attached.² The total ion electropherogram (TIE) (m/z 500–1500) is illustrated together with a m/zvs. time contour profile in Fig. 2. In the CE–MS spectrum a series of doubly-, triply-, quadruply- and pentuply-deprotonated molecules is observed from which the molecular masses of the different glycoforms can be derived (Table 1). The most abundant glycoforms identified in the Rd11.7 sample contained two or three hexose residues, having compositions of Hex_2–3Hep₃PCho₁PEtn₁P₁KDO₁lipid A–OH. In addition, minor glycoforms containing oligosaccharide chains capped by a sialic acid (Neu5Ac) residue were also readily detected. Sialylated glycoforms have been recently demonstrated to be important virulence factors in H. influenzae disease.¹⁴ The two-dimensional representation of m/zvs. time in Fig. 2b provides a useful method for identifying closely related families of LPS glycoforms. The microheterogeneity of this sample reflected both glycoform distribution and substitution by phosphorylated groups (PCho or PEtn), which is clearly visualized in the contour profile by the families of peaks with incremental changes in electrophoretic mobilities. The structural diversity of H. influenzae LPS arising from phase variation has complicated the study of the molecular features of these molecules. The separation of different glycoforms by using capillary electrophoresis prior to the introduction to an electrospray mass spectrometer provides a sensitive technique for profiling subtle structural changes in the LPS molecules. For example, the extension of one hexose residue to the LPS core unit results in an increase in molecular mass (i.e. the size of the molecule). As a consequence, a decrease in electrophoretic mobility (reverse to the electroosmotic flow) or shorter migration time is observed. Similarly, the presence of LPS phosphoforms that differ in the presence and location of functional groups such as PEtn or PCho is reflected by the changes of electrophoretic mobilities and masses. Thus, from the extracted mass spectra, one can obtain the molecular masses of different components in the LPS extracts. Detailed information concerning the compositions of different glycoforms and migration times in capillary electrophoresis is given in Table 1. The corresponding compositions are in accord with the structural model described previously for H. influenzae Rd.⁸ It is noteworthy to point out that the electrophoretic mobilities of those glycoforms containing sialic acid residues are relatively small (12.7–12.9 min), although the Neu5Ac group should carry a negative charge under the separation conditions (pH 9.0). This indicates that the electrophoretic mobility is not only associated to the charge status of a molecule but also to the size and/or the shape of the molecule.

Fig. 2 CE–MS analysis of mixtures of O-deacylated LPS from H. influenzae strain Rd 11.7 in the negative ion mode. (a) m/z 500–1500. (b) Contour profile of m/zvs. time. Separation conditions: 30 nL injection of 500 μg mL⁻¹ of O-deacylated LPS, bare fused-silica (90 cm x 50 μm id, 185 μm od), 5% methanol in 30 mM morpholine, pH 9.0, +30 kV.

Table 1 CE–MS data and proposed compositions of various O-deacylated LPS glycoforms of H. influenzae strain Rd 11.7

Time/min	Observed ions (m/z)				Molecular mass/Da		Assignment^b
	[M − 5H]^5−	[M − 4H]^4−	[M − 3H]^3−	[M − 2H]^2−	Observed	Calculated^a	PCho	PEtn	Hex	TS^c
a Average mass units were used for the calculation of molecular masses, based on proposed composition as follows: HexN, 161.16; Hex, 162.14; Hep, 192.17; HexNAc, 203.19; KDO, 220.20; Neu5Ac, 291.26; PEtn, 123.05; PCho, 165.13; P, 79.98; lipid A–OH, 953.01.b Glycans and functional groups appended to a core oligosaccharide comprising three Hep, one KDO, one PEtn, a phosphate and a lipid A composed of GlcN2, P2, and two N-linked 3-OH myristic acid (C14H27O2). Hex, GlcN, Hep, KDO, PEtn and PCho designate hexose, glucosamine, heptose, 3-deoxy-D-manno-2-octulosonic acid, phosphoethanolamine and phosphocholine, respectively.c Tetrasaccharide: Neu5Ac–Hex–HexNAc–Hex.
12.7	676.0	845.0	1127.0	—	3384.6	3383.92	1	1	2	1
12.9	643.0	804.0	1072.0	—	3219.2	3218.79	—	1	2	1
12.9	610.0	763.0	1018.0	—	3057.8	3056.65	—	1	1	1
13.0	—	680.0	908.0	1362.0	2727.6	2727.20	1	1	3	—
13.5	—	640.0	854.0	1281.0	2565.4	2565.06	1	1	2	—
13.6	—	650.0	867.0	1301.0	2603.8	2604.15	1	—	3	—
13.8	—	599.0	799.0	1200.0	2402.4	2402.92	1	1	1	—
14.3	—	609.0	813.0	1220.0	2442.5	2442.01	1	—	2	—
14.3	—	558.0	743.0	1117.0	2237.4	2237.91	—	1	1	—
14.6	—	569.0	759.0	1139.0	2280.2	2279.87	1	—	1	—
15.0	—	527.0	704.0	1056.0	2115.0	2114.87	—	—	1	—

---

Characterization of isomeric glycoforms

CE–MS data obtained in both positive and negative ion modes on LPS–OH from strain 375.1 is shown in Fig. 3, together with the extracted mass spectra (XIE) at 12.7 (Fig. 3b) and 13.1 min (Fig. 3c). Triply- and doubly-charged ions corresponding to major glycolipids having masses of 2601, 2724 and 2847 Da were identified at m/z 866/1299, 907/1361 and 948/1423 (negative mode, Fig. 3b and 3c). Their counterparts obtained in positive mode are shown in Fig. 3e and 3f. Based on their molecular masses the three glycolipids were assigned to have the respective compositions Hex₄Hep₃PEtn₁P₁KDO₁lipid A–OH, Hex₄Hep₃PEtn₂P₁KDO₁ lipid A–OH and Hex₄Hep₃PEtn₃P₁KDO₁lipid A–OH. The species with ions at m/z 907 (triply-deprotonated) and 1361 (doubly-deprotonated) appear in both peaks and indicated the existence of isomeric glycoforms.¹³ To obtain structural information, tandem mass spectrometry (MS/MS) experiments were conducted on selected ions in both negative and positive ion detection modes. As reported previously, separations conducted using morpholine buffers generally gave enhanced resolution and better signal intensities of the different LPS glycoforms compared to ammonium acetate or ammonium formate buffers.¹³ The lower conductivity of morpholine solutions is likely to be the reason for the observed higher sensitivity, whereas the enhanced resolution is probably due to interactions between analytes and morpholine.¹³ In the present investigation, morpholine buffer was employed when negative ion detection mode was required. Negative ion MS/MS experiments provide a definitive and straightforward method to identify the size of the lipid A–OH portion of the molecule. However, MS/MS information from negative ion detection of glycolipids is not sufficient for unambiguous identification of glycan sequence and sites of attachment of sugar and non-carbohydrate substituents. To obtain this information, tandem mass spectra in positive ion detection mode are required. Morpholine buffer solutions are not compatible with positive ion detection of multiply-protonated molecular species (data not shown) due to the presence of abundant morpholine adduct peaks. We have found that optimized separation conditions providing moderate resolutions, shorter analysis time, better signal-to-noise ratio and favoring the formation of cationic species with minimum mass interference, are obtained using a 15 mM ammonium acetate buffer adjusted to pH 9.0 with ammonium hydroxide solution. Thus, in the present applications, morpholine buffer was chosen for negative ion mode and ammonium acetate buffer was employed for positive ion mode.

Fig. 3 CE–MS analysis of O-deacylated LPS from H. influenzae strain 375.1. (a) TIE (m/z 800–1500) in negative ion mode. (b) Extracted mass spectrum at 12.7 min and (c) at 13.1 min. (d) TIE (m/z 800–1500) in positive ion mode with 15 mM ammonium acetate as separation buffer. (e) Extracted mass spectrum at 14.8 min and (f) at 15.2 min.

CE–MS/MS (positive mode) on m/z 1425 and 1301 provided evidence for the basic structures shown below. These doubly-protonated ions correspond to molecular weights of 2847 Da and 2601 Da, with the compositions of Hex₄Hep₃PEtn₁P₁KDO₁lipid A–OH and Hex₄Hep₃PEtn₃P₁KDO₁lipid A–OH, respectively. The product ion spectrum obtained from the ion at m/z 1301 is shown in Fig. 4a. The ions at m/z 1649 and 1349 corresponded to losses of lipid A–OH and PKDO from the molecular ion, respectively, clearly indicating that PEtn does not substitute KDO but instead only substitutes Hep_II in accordance with the structural model for the inner-core region of H. influenzae LPS (structure 2). Fragment ions at m/z 670, 832 and 862 corresponded to compositions HexHep₂PEtn, Hex₂Hep₂PEtn and HexHep₃PEtn, respectively.

Fig. 4 CE–MS/MS (positive ion mode) analysis of O-deacylated LPS from H. influenzae strain 375.1. (a) Product ion spectrum of doubly-protonated ion at m/z 1301 (M_r: 2601 Da) and (b) product ion spectrum of doubly-protonated ion at m/z 1425 (M_r: 2847 Da). Separation conditions as in Fig. 3d, E_lab: 100 eV (laboratory frame reference).

The product ion spectrum obtained from the ion at m/z 1425 is shown in Fig. 4b. The ions at m/z 1895, 1772, 1675 and 1472 corresponded to consecutive losses of lipid A–OH, PEtn, P and KDO from the molecular ion, respectively. Furthermore, the fragment ion at m/z 1472 showed subsequent losses of Hep–Hex and Hep–PEtn indicated by ions at m/z 1117 and 802, respectively. The ion at m/z 640 corresponded to Hex₂HepPEtn. The most abundant fragment in Fig. 4b was observed at m/z 1376, corresponding to a doubly-charged ion and resulted from the loss of a phosphate group from lipid A–OH. The fragment ions observed at m/z 855, 468 and 388 are proposed to be derived from the lipid A–OH part of the molecule (structure 3).

These two glycolipids contained either one or three PEtn groups and were well separated in the electropherogram, having retention times of 14.8 and 15.2 min, respectively (Fig. 3d). The glycolipid having the molecular mass of 2724 Da (composition Hex₄Hep₃PEtn₂P₁KDO₁ lipid A–OH) was found to have two isomeric glycoforms. In accordance with the inner-core region of H. influenzae LPS one PEtn group substituted Hep_II. To assign the attachment points for the other two PEtn groups MS/MS experiments were performed. Unambiguous characterization of isomeric glycoforms can be achieved when the isomers are separated prior to MS/MS analyses. The CE–MS/MS electropherogram for the precursor ion at m/z 1363 (doubly-protonated, M_r: 2724 Da) is presented in Fig. 5a. In this experiment, the two isomers were baseline-resolved with apart of 0.4 min in migration time. The product ion spectra extracted at 11.0 and 11.4 are presented in Fig. 5b and 5c, respectively. To identify the structures, several diagnostic fragment ions in the MS/MS spectra could be used. For example, the fragment ions arising from the losses of KDO–PPEtn (423 Da) and/or KDO–P (300 Da) lead to the assignment of one PEtn group. As shown in Fig. 5b and 5c, the molecular ion of both glycolipids at m/z 1363 lost lipid A–OH (M_r: 953 Da) and gave rise to the fragment at m/z 1772, which showed the loss of either KDO–PPEtn at m/z 1349 (Fig. 5b) or KDO–P at m/z 1471 (Fig. 5c). The data indicated that the isomer with lower electrophoretic mobility (shorter migration time or 11.0 min in Fig. 5a) contained a moiety of KDO–PPEtn, whereas the isomer with the structure of KDO–P had a higher electrophoretic mobility. In the spectrum of the low-mobility component (Fig. 5b, structure 4) the ion at m/z 994 corresponded to further loss of HexHep from m/z 1349. The fragment ion at m/z 670 corresponded to a composition of HexHepHepPEtn. In the MS/MS spectrum of the component with higher mobility (Fig. 5c, structure 5) the fragment ion at m/z 640 corresponded to HexHexHepPEtn, to which an additional HepPEtn could be added to give the ion at m/z 955. This provided unequivocal evidence that both Hep_II and Hep_III were substituted with PEtn.

Fig. 5 CE–MS/MS (positive ion mode) analysis of O-deacylated LPS from H. influenzae strain 375.1. Extracted MS/MS spectra for precursor ions of m/z 1363 (doubly-protonated ion with molecular weight of 2724 Da) at 11.0 min (a) and 11.4 min (b). Separation conditions are similar to those in Fig. 3d (different retention times due to the delay for data acquisition). E_lab: 100 eV (laboratory frame reference).

Conclusions

Structural studies of the LPS molecules expressed by a broad range of H. influenzae capsular and non-typeable strains have revealed the presence of a common heptose-containing inner core region to which variable oligosaccharide extensions and phosphate-containing substitutents are attached. We have demonstrated that this microheterogeneity can be characterized by CE–MS and CE–MS/MS. Application of these techniques not only provides structural relationships between families of glycoforms containing oligosaccharide extensions, but also facilitates the visualization of LPS isoforms differing by the location of phosphorylated groups or hexose residues. Applications of on-line CE tandem mass spectrometry open the door to elucidating the structural detail of isomeric oligosaccharides in complex LPS mixtures.

Experimental

Materials

Fused-silica capillaries with 185 μm od × 50 μm id were obtained from Polymicro Technologies (Phoenix, AZ, USA). Methanol and isopropanol (IPA) were from EM Science (Gibbstown, NJ, USA). Anhydrous hydrazine and ammonium acetate were obtained from Fisher Scientific (Fair Lawn, NJ, USA) and formic acid from BDH (Toronto, Canada). The enzymes, proteinase K, deoxyribonuclease I (DNase) and ribonuclease (RNase) were obtained from Sigma (St Louis, MO, USA).

Bacterial strains and growth conditions

The H. influenzae strains used in this study were genetically engineered from either the serotype d-derived strain Rd or nontypeable otitis media isolate 375. Strain Rd 11.7 has been transformed with a heterologous lpsA gene sequence derived from the type b strain Eagan.⁴ Allelic exchange replaced the natural lpsA gene in strain Rd (directs addition of a glucose in β1–2 linkage to Hep_III) with the Eagan derived sequence (directs addition of a galactose in β1–2 linkage to Hep_III).¹² Strain 375.1 has been transformed with a construct comprising a deletion of the lic1 operon (lic1A–lic1D) and insertion of a tetracycline resistance cassette. Lic1 directs the addition of phosphocholine to the oligosaccharide side chains.²¹H. influenzae bacteria were grown at 37 °C in brain–heart infusion (BHI) broth supplemented with haemin (10 µg mL⁻¹) and NAD (2 µg mL⁻¹).

Preparation and extraction of LPS and O-deacylated LPS

LPS samples from Rd 11.7 and 375.1 were extracted by the hot phenol–water method, as described previously, and purified from the dialysed aqueous phase by ultracentrifugation (45 000 rpm, 4 °C, 5 h) after treatment with DNase, RNase and proteinase K.⁴

Purified LPS (∼2 mg mL⁻¹) was dissolved in anhydrous hydrazine and incubated at 37 °C for 50 min with constant stirring to release O-linked fatty acids from the lipid A region of the molecules. The reaction mixture was cooled (0 °C), the hydrazine was destroyed by addition of cold acetone (5 vol.) and the final product was obtained by centrifugation. The pellet was washed with acetone, centrifuged and then lyophilized from water.

CE–MS and CE–MS/MS

The instrumentation and experimental conditions for CE–MS have been investigated and published in our previous reports.¹³ Briefly, a Prince CE system (Prince Technologies, The Netherlands) was coupled to an API 3000 mass spectrometer (Applied Biosystems/Sciex, Concord, Canada) via a microIonspray interface. Sheath solution (isopropanol–methanol, 2 : 1) was delivered at a flow rate of 1.0 μL min⁻¹. An electrospray stainless steel needle (27-gauge) was butted against the low dead volume tee and enabled the delivery of the sheath solution to the end of the capillary column. The separations were obtained on ca. 90 cm length bare fused-silica capillary using 30 mM morpholine (for negative ion detection mode) or 15 mM ammonium acetate (for positive ion detection mode) in deionized water, pH 9.0, containing 5% methanol. A voltage of 30 kV was typically applied at the injection when 30 mM morpholine was used as separation buffer. A separation voltage of 20 kV was chosen for positive ion detection mode.

Acknowledgements

The authors thank Adèle Martin for preparation of O-deacylated LPS and Dr. Pierre Thibault for valuable discussions.

References

1. E. R. Moxon and D. Maskell, Molecular Biology of Bacterial Infection: Current Status and Future Perspectives, ed. C. E. Harmaeche, C. W. Penn and C. J. Smyth, Cambridge University Press, Cambridge, UK, pp. 75–96.
2. I. M. Helander, B. Lindner, H. Brade, K. Altmann, A. A. Lindberg, E. T. Rietschel and U. Zähringer, Eur. J. Biochem., 1988, 177, 483–492.
3. M. K. Landerholm, J. Li, J. C. Richards, D. W. Hood, E. R. Moxon and E. K. H. Schweda, Eur. J. Biochem., 2004, 271, 941–953.
4. H. Masoud, E. R. Moxon, A. Martin, D. Krajcarski and J. C. Richards, Biochemistry, 1997, 36, 2091–2103.
5. N. J. Phillips, M. A. Apicella, J. M. Griffis and B. W. Gibson, Biochemistry, 1992, 31, 4515–4526.
6. D. W. Hood, G. Randle, A. D. Cox, K. Makepeace, J. Li, E. K. H. Schweda, J. C. Richards and E. R. Moxon, J. Bacteriol., 2004, 186, 7429–7439.
7. N. J. Phillips, R. McLaughlin, T. J. Miller, M. A. Apicella and B. W. Gibson, Biochemistry, 1996, 35, 5937–5947.
8. A. Risberg, H. Masoud, A. Martin, J. C. Richards, E. R. Moxon and E. K. H. Schweda, Eur. J. Biochem., 1999, 261, 171–180.
9. C. C. Patrick, A. Kimura, M. A. Jackson, L. Hermanstorfer, A. Hood, G. H. McCracken Jr and E. J. Hansen, Infect. Immun., 1987, 55, 2902–2911.
10. J. N. Weiser, J. M. Love and E. R. Moxon, Cell, 1989, 59, 657–665.
11. D. W. Hood, A. D. Cox, W. W. Wakarchuk, M. Schur, E. K. H. Schweda, S. L. Walsh, M. E. Deadman, A. Martin, E. R. Moxon and J. C. Richards, Glycobiology, 2001, 11, 957–967.
12. D. W. Hood, M. E. Deadman, A. D. Cox, K. Makepeace, A. Martin, J. C. Richards and E. R. Moxon, Microbiology, 2004, 150, 2089–2097.
13. J. Li, A. D. Cox, D. W. Hood, E. R. Moxon and J. C. Richards, Electrophoresis, 2004, 25, 2017–2025.
14. V. Bouchet, D. W. Hood, J. Li, J.-R. Brisson, G. A. Randle, A. Martin, Z. Li, R. Goldstein, E. K. H. Schweda, S. I. Pelton, J. C. Richards and E. R. Moxon, Proc. Natl. Acad. Sci. U. S. A., 2003, 100, 8898–8903.
15. J. Li, R. W. Purves and J. C. Richards, Anal. Chem., 2004, 76, 4676–4683.
16. J. Kelly, H. Masoud, M. B. Perry, J. C. Richards and P. Thibault, Anal. Biochem., 1996, 233, 15–30.
17. S. Auriola, P. Thibault, I. Sadovskaya and E. Altman, Electrophoresis., 1998, 19, 2665–2676.
18. P. Thibault, J. Li, A. Martin, J. C. Richards, D. W. Hood and E. R. Moxon, Mass Spectrometry in Biology and Medicine, ed. A. L. Burlingame, S. A. Carr and M. A. Baldwin, Humana Press, Clifton, NJ, 1997, pp. 439–462.
19. J. Li, P. Thibault, A. Martin, J. C. Richards, W. W. Wakarchuk and W. van der Wilp, J. Chromatogr., 1998, A817, 325–336.
20. P. Thibault and J. C. Richards, Methods Mol. Biol., 2000, 145, 327–344.
21. J. N. Weiser, N. Pan, K. L. McGowan, D. Musher, A. Martin and J. C. Richards, J. Exp. Med., 1997, 187, 631–640.

---