Bordetella holmesii: Lipid A Structures and Corresponding Genomic Sequences Comparison in Three Clinical Isolates and the Reference Strain ATCC 51541

Bordetella holmesii can cause invasive infections but can also be isolated from the respiratory tract of patients with whooping-cough like symptoms. For the first time, we describe the lipid A structure of B. holmesii reference strain ATCC 51541 (alias NCTC12912 or CIP104394) and those of three French B. holmesii clinical isolates originating from blood (Bho1) or from respiratory samples (FR4020 and FR4101). They were investigated using chemical analyses, gas chromatography–mass spectrometry (GC–MS), and matrix-assisted laser desorption ionization–mass spectrometry (MALDI–MS). The analyses revealed a common bisphosphorylated β-(1→6)-linked d-glucosamine disaccharide with hydroxytetradecanoic acid in amide linkages. Similar to B. avium, B. hinzii and B. trematum lipids A, the hydroxytetradecanoic acid at the C-2′ position are carrying in secondary linkage a 2-hydroxytetradecanoic acid residue resulting of post-traductional biosynthesis modifications. The three clinical isolates displayed characteristic structural traits compared to the ATCC 51541 reference strain: the lipid A phosphate groups are more or less modified with glucosamine in the isolates and reference strain, but the presence of 10:0(3-OH) is only observed in the isolates. This trait was only described in B. pertussis and B. parapertussis strains, as well as in B. petrii isolates by the past. The genetic bases for most of the key structural elements of lipid A were analyzed and supported the structural data.


Introduction
The Bordetella genus contains at the moment a dozen species, of which at least five are responsible for respiratory diseases in humans and/or animals. Classical Bordetellae consist of Bordetella pertussis, a strict human pathogen responsible for whooping cough; B. parapertussis responsible for mild whooping-cough symptoms in humans, also described as a sheep pathogen; and B. bronchiseptica, able to infect a broad range of hosts. The Bordetellae virulence factors include toxins-such as pertussis toxin for B. pertussis only, adenylate cyclase-hemolysin, and lipopolysaccharide (LPS)-and adhesins-such as filamentous hemagglutinin, fimbriae and pertactin-, all involved in the binding to ciliated epithelial cells in the host upper respiratory tract. B. bronchiseptica and B. pertussis endotoxins LPS have been shown to be implicated in virulence [1][2][3][4]; therefore, it is important to compare the structures of Bordetella LPSs purified from other Bordetella pathogenic species and, particularly, from the recent B. pertussis relative B. holmesii [5].
B. holmesii was first described in 1995 following its isolation from the blood of a patient with septicemia [6]. At that time, this bacterium was only originating from invasive infections in immunocompromised patients. In the past years, increasing reports of the presence of B. holmesii in the respiratory tract of patients with pertussis-like symptoms have been published [7][8][9][10][11][12][13]. However, it is not known whether this bacterium is an opportunistic or a pathogenic one, able to induce pertussis-like symptoms in humans [14][15][16].
For the moment, it is not possible to differentiate B. holmesii isolates recovered from blood from isolates recovered from respiratory samples [17][18][19]. About 21 genomes of B. holmesii are available on The National Center for Biotechnology Information (NCBI) [20]. First considered as close to B. pertussis on the basis of 16S DNA analysis, B. holmesii is now described in the same clade as B. hinzii, B. avium and B. trematum on the basis of whole genome single nucleotide polymorphism (SNP)-based analysis [21]. Most virulence factors usually produced by the classical Bordetella seem to be missing in B. holmesii except a Bordetella master virulence regulatory system (bvg), a filamentous hemagglutinin (FHA)-like protein, a bvg-intermediate phaseA protein (bipA) ortholog, and an alcaligin operon [19,[22][23][24][25]. B. holmesii LPS have only been roughly studied by Van den Akker in 1998 who found them phenotypically and immunologically distinct from those of B. pertussis [26].
We report here the detailed lipid A structures of three B. holmesii isolates, as compared to those of the B. holmesii reference strain ATCC 51541.

Interpretation of the Main Molecular Species in the Different Lipid a Spectra
The negative-ion spectrum of the di-phosphoryl ATCC 51541 reference strain lipid A was heterogeneous, containing two main molecular ion signals at m/z 1376.9 and 1603.7 as illustrated in Figure 1A. Composition of the corresponding molecular species were attributed on the basis of the overall chemical composition: m/z 1603.7 would correspond to two glucosamine (GlcN), two phosphates, three 14:0(3-OH), one 14:0(2-OH), and one 12:0(2-OH); and m/z 1376.9 corresponds to m/z 1603.7 minus one 14:0(3-OH). A molecular species corresponding to m/z 1404.9 can be explained by some microheterogeneity at the level of the 12:0(2-OH) fatty acid versus the 14:0(2-OH). The same difference was observed between molecular species at m/z 1575.4 and 1603.7. Each of the molecular species presented a twin species at −16, expressing the previously described heterogeneity and peculiarity of the Bordetella genus, the reduced enzymes specificity at different positions carrying, in some species, 2-hydroxylated fatty acids in secondary linkage [1,[27][28][29]. The latter being known as a late structural modification of the structure occurring in the membrane and leading to increased robustness of the bacterial outer membrane barrier [30].

Distribution of the Fatty Acids on the D-Glucosamine Residues of the Clinical Isolates
The negative-ion mode matrix-assisted laser desorption ionization (MALDI) mass spectra of lipids A isolated from the three B. holmesii isolates displayed three major peaks at m/z 1603, 1546 and 1376 ( Figure 1B-D). The first and third peaks were identical to those found in ATCC 51541 lipid A and corresponded, as explained for the reference strain, to penta-and tetra-acylated lipid A molecular species, respectively. For isolate FR4101, the ion at m/z 1546, absent in the ATCC strain spectrum, corresponded to the molecular species at m/z 1376 carrying an additional 10:0(3-OH) (170 units).
The peaks mentioned above were surrounded by smaller ones at −16 u, such as those at m/z 1587, 1360 and 1531. These peaks in which the 16-u difference represented hydroxyl modifications, were attributed to the same structures whose secondary fatty acids linked at C-2 and C-2′, 12:0(2-OH) or 14:0(2-OH) were non-hydroxylated (14:0 or 12:0, respectively). This modification was not observed

Distribution of the Fatty Acids on the D-Glucosamine Residues of the Clinical Isolates
The negative-ion mode matrix-assisted laser desorption ionization (MALDI) mass spectra of lipids A isolated from the three B. holmesii isolates displayed three major peaks at m/z 1603, 1546 and 1376 ( Figure 1B-D). The first and third peaks were identical to those found in ATCC 51541 lipid A and corresponded, as explained for the reference strain, to penta-and tetra-acylated lipid A molecular species, respectively. For isolate FR4101, the ion at m/z 1546, absent in the ATCC strain spectrum, corresponded to the molecular species at m/z 1376 carrying an additional 10:0(3-OH) (170 units).
The peaks mentioned above were surrounded by smaller ones at −16 u, such as those at m/z 1587, 1360 and 1531. These peaks in which the 16-u difference represented hydroxyl modifications, were attributed to the same structures whose secondary fatty acids linked at C-2 and C-2 , 12:0(2-OH) or 14:0(2-OH) were non-hydroxylated (14:0 or 12:0, respectively). This modification was not observed in the FR4020 lipid A mass spectrum and might be a specific trait of this isolate, with biosynthetic relevance. In addition, another type of micro-heterogeneity, regarding the length of fatty acids, was observed in the first major peaks (at m/z 1603 in Figure 1A,B,D, at m/z 1376 in Figure 1A-D and at m/z 1546 in Figure 1C) that were also doubled by minor ones at −28 u or +28 u, i.e., at m/z 1575, 1349 and 1405. The mass difference of 28 u was attributed to 2 × CH2, and related peaks were not found systematically in the three B. holmesii lipid A mass spectra. Of note, in the Bho1 lipid A mass spectrum, this type of heterogeneity was only observed at plus 28 u from the major peak, at m/z 1376. In the high-mass molecular ions region, two minor peaks were observed at m/z 1773 and 1764 in Bho1 (with a really minor contribution for this isolate) and FR4020 lipid A mass spectra and, at m/z 1717 and 1708 in FR4101 mass spectrum. The peak at m/z 1773, as well as the one at m/z 1717, corresponded to 10:0 (3-OH) (170 u), resulting in a hexa-acyl lipid A molecular species in Bho1 and FR4020, and FR4101, respectively. Interestingly, the mass difference between peaks at m/z 1764 and 1708 and penta-acyl lipid A molecular species at m/z 1603 and 1546.9, corresponded to 161 units for GlcN. Moreover, other two small peaks (m/z 1925 and 1869) in FR4020 and FR4101 lipid A mass spectra gave the same mass difference with m/z 1764 and 1708, suggesting that these lipid A molecular species were carrying 2 GlcN, one at each phosphate groups. This modification is described here for the fifth time in lipid A species of the Bordetella genus [27,[31][32][33][34]. Analogous to the FR4020 mass spectrum with the exception of peaks related to GlcN addition at m/z 1538, 1869 and 1925 ( Figure 2) the mass spectrum for lipid A from Bho1 displayed the same peaks as FR4020 mass spectrum. This indicates that both isolates might have similar lipid A structures with the exception that Bho1 lipid A phosphate groups displayed a lower amount of GlcN. The reference ATCC strain also showed a peak corresponding to a very small amount of GlcN, but its presence was confirmed by statistical experiments allowing to obtain persistent and slightly increased signals at this level.

Distribution of the Fatty Acids on the D-Glucosamine Residues of the Isolates and the Reference Strain
The reference strain ATCC 51541 lipid A positive-ion MALDI mass spectrum presented in Figure 3 gave peaks in the lower field at m/z 920.6 and 904.6. The peak at m/z 920.6 was interpreted to correspond to an ion containing one GlcN, one phosphate, three hydroxytetradecanoic acid residues-two 14:0(3-OH) and one 14:0(2-OH)-, the second at m/z 904.6. From this fragmentation pattern, we concluded that these signals corresponded to the GlcN II (non-reducing) part of the lipids A and that the 14:0(2-OH) acid was linked at C-2 in secondary acylation. The presence of the two peaks differing by 16 u illustrates the incomplete hydroxylation at position 2 of the 14:0 branched fatty acid.

Linkage of Fatty Acids in B. holmesii Lipids A
It was previously shown that sequential liberation of FAs was efficient to determine their position in the lipid A structure [32]. As shown in Table 1, the FA in ester linkages at the C-3 position are the first ones to be liberated, followed by those at C-3′, and the acyloxy-acyl secondary esterlinked FAs come as the last ones, this being due to steric hindrance conditions. Mass spectra at different NH4OHtreatment with length times of 15 min, 30 min, 1 h, 2 h and 5 h were recorded and compared to the initial lipid A mass spectra at the starting point, t0 (Figures 2 and 4). The partial Odeacylation pattern, obtained by 28% NH4OH treatment, was different in B. holmesii isolates; it was related to their linkage position. On the one hand, the time release of 10:0(3-OH) at C-3 ranged from 1 to 15 min in FR4020 and FR4101 lipids A and was illustrated by the disappearance of the hexa-acyl lipid A related peaks (at m/z 1773.8 and 1717.7 respectively) which were transformed to penta-acyl lipid A molecular species at m/z 1603.1 and 1546.9, respectively. On the other hand, the 10:0(3-OH) liberation in Bho1 took a longer time, 1-30 min ( Figure 2). Then, we observed by following kinetics of sequential liberation: at the C3 position, 14:0(3-OH) was completely liberated after 1-2 h in all three clinical isolates (Figures 2 and 4). This led to convert penta-acyl lipids A at m/z 1603.1 and 1546.9 to tetra-acyl ones at m/z 1376.6. In fact, two peaks at m/z 1150.2 and 1177.9 appeared after 15 min in all three clinical isolates, corresponding to m/z 1376.6 -226 u and 198 u, respectively. These mass differences were attributed to 14:0(2-OH) and 12:0(2-OH) secondary fatty acids at C-2′and C-2, respectively. Another important peak appeared at m/z 952 after 1 h in Bho1, 30 min in FR4101 and 15 min in FR4020. The structure related to this peak corresponded to a diacyl lipid A with two amide-

Linkage of Fatty Acids in B. holmesii Lipids A
It was previously shown that sequential liberation of FAs was efficient to determine their position in the lipid A structure [32]. As shown in Table 1, the FA in ester linkages at the C-3 position are the first ones to be liberated, followed by those at C-3 , and the acyloxy-acyl secondary ester-linked FAs come as the last ones, this being due to steric hindrance conditions. Mass spectra at different NH 4 OH treatment with length times of 15 min, 30 min, 1 h, 2 h and 5 h were recorded and compared to the initial lipid A mass spectra at the starting point, t 0 (Figures 2 and 4). The partial O-deacylation pattern, obtained by 28% NH 4 OH treatment, was different in B. holmesii isolates; it was related to their linkage position. On the one hand, the time release of 10:0(3-OH) at C-3 ranged from 1 to 15 min in FR4020 and FR4101 lipids A and was illustrated by the disappearance of the hexa-acyl lipid A related peaks (at m/z 1773.8 and 1717.7 respectively) which were transformed to penta-acyl lipid A molecular species at m/z 1603.1 and 1546.9, respectively. On the other hand, the 10:0(3-OH) liberation in Bho1 took a longer time, 1-30 min (Figure 2). Then, we observed by following kinetics of sequential liberation: at the C3 position, 14:0(3-OH) was completely liberated after 1-2 h in all three clinical isolates (Figures 2 and 4). This led to convert penta-acyl lipids A at m/z 1603.1 and 1546.9 to tetra-acyl ones at m/z 1376.6. In fact, two peaks at m/z 1150.2 and 1177.9 appeared after 15 min in all three clinical isolates, corresponding to m/z 1376.6 −226 u and 198 u, respectively. These mass differences were attributed to 14:0(2-OH) and 12:0(2-OH) secondary fatty acids at C-2 and C-2, respectively. Another important peak appeared at m/z 952 after 1 h in Bho1, 30 min in FR4101 and 15 min in FR4020. The structure related to this peak corresponded to a diacyl lipid A with two amide-linked acyl chains [14:0(3-OH)] at C-2 and C-2 on the lipid A core made of two GlcN and two phosphate residues. Of particular note, the ion peak at m/z 1538 arose from addition of GlcN (161 u) to the ion at m/z 1376.6, lasting from 15 min to 5 h in FR4020 and from 15 min to 2 h in the FR4101 lipid A mass spectrum. The peak associated to the glucosamine modification was weak in the Bho1 mass spectrum, but confirmed with repeated experiments. The complete O-deacylation was carried out by releasing secondary ester-linked fatty acids with methylamine treatment (41% CH 3 NH 2 at 37 • C). The negative-ion MALDI mass spectrum from Bho1 [36] displayed two peaks, after 5 h, a major one at m/z 952 (diacyl lipid A), accompanied by a minor one at m/z 872 (a mono-phosphorylated lipid A). No peak at plus GlcN was observed because of the very small amount of this molecular species present in the native molecule. In addition to these two peaks, further ones were observed in FR4101 and FR4020 mass spectra (Figures 2 and 4), such as m/z 1113, which corresponded to 953 plus 161 units (GlcN) in both isolates, and a very minor not indicated peak at m/z 1274 (1113 plus 161 units), thereby demonstrating that both lipid A phosphate groups were carrying GlcN. Of particular note, the ion peak at m/z 1538 arose from addition of GlcN (161 u) to the ion at m/z 1376.6, lasting from 15 min to 5 h in FR4020 and from 15 min to 2 h in the FR4101 lipid A mass spectrum. The peak associated to the glucosamine modification was weak in the Bho1 mass spectrum,

Genomic Analyses of Genes Involved in Lipid A Biosynthesis
We analyzed the sequence of genes involved in lipid A biosynthesis, more particularly focusing on the identification of the genomic basis for the two main differences observed in the lipid A structures of the three B. holmesii isolates in the present study, i.e., the presence/absence of glucosamines and the length of the carbon chain FA at C-3 [14:0(3-OH) or 10:0(3-OH)]. We thus explored the Lipid A GlcN modification locus (lgm locus), the Acyl-[acyl-carrier-protein]-UDP-N-acetylglucosamine O-acyltransferase (lpxA), the Lipid A deacylase (pagL), the LipidA palmitoyltransferase (pagP), two oxygenases lpxO1 and lpxO2 and acyltransferase lpxL1 genes ( Table 2): (i) An orthologous sequence corresponding to the lgm locus was identified in the genome of the three clinical isolates as in the ATCC 51541 reference strain (position 1966909..1970398). We observe no sequence difference for this locus comprising ArnT (position 1967952..1969541) between ATCC type strain and the three clinical isolates. (ii) Bho1 and FR4020 lpxA sequence is identical to that of ATCC 51541 reference strain (position 3231726..3232520) but FR4101 lpxA sequence displays a non-synonymous SNP in position 508 leading to an amino acid modification (S instead of G in position 170 of LpxA protein). We also sequenced lpxA gene from 16 additional B. holmesii isolates collected from blood or from respiratory samples and found that this SNP was not linked to the origin of the isolates (unpublished data). (iii) pagL sequence is identical for the three clinical isolates and the ATCC 51541 reference strain (position 2144078..2144623). We did not observe any sequence difference between the three tested isolates for pagP but a difference was observed in ATCC 51541 reference strain pagL sequence (position 630144..630686) corresponding to an additional G leading to a frameshift. (iv) We found two homologs of lpxO in the genome of the three isolates of B. holmesii as in the ATCC 51541 reference strain (position 2899145..2900041 for the first one and 2402394..2403293 for the second). They respectively displayed 84% nucleotidic identity with the KF214918 lpxO1 of B. avium ATCC 35086 and 84% nucleotidic identity with the KF214919 LpxO2 of B. avium ATCC 35086. We then identified a lpxL1 homolog in the three isolates as in ATCC 51541 reference strain (position 1194726..1195580) with no sequence difference between them.

Discussion
In the present study, we focused our analyses on the lipid A of different B. holmesii isolates originating either from blood or respiratory samples and on the ATCC 51541 reference strain.
Mass spectrometry analyses first led to the observation of a difference in the amount of modification with GlcN on the lipid A phosphate groups. The two clinical isolates FR4020 and FR4101 displayed a higher degree of modification compared to Bho1 and the ATCC strain. Such an hexosamine modification has been described in other Bordetellae species such as B. pertussis, B. bronchiseptica and B. avium and was attributed to a lgm locus encoding a glycosyl transferase ArnT [31,35,39]. An orthologous sequence corresponding to lgm locus was identified in the genome of the three clinical isolates as in the ATCC 51541 reference-strain genome with no sequence difference between them. In B. pertussis and B. bronchiseptica, the lgm locus has been reported to be bvg regulated [33]. Previous studies showed that Bho1 isolate and ATCC 51541 reference strain both have a particular non-functional bvg system because of a non-functional BvgA protein (due to an A insertion within bvgA sequence, leading to a frameshift [19,22]) that could explain why less GlcN is substituting their lipid A phosphate groups on contrary to FR4020 and FR4101. Recent sequencing and annotation of ATCC 51541 strain genome do not report this additional A within BvgA (position 3314780..3315403) suggesting that the mutation can occur frequently at random, according to subcultures or different ways of storage. Moreover, this difference is not a specific trait of respiratory samples as a functional bvg system is found in other B. holmesii isolates collected from blood [19]. All these data show that B. holmesii lipid A can be decorated with GlcN what has already been shown to have consequences on the modulation of host immunity via a different Toll-like receptor 4 (TLR4) activation for B. pertussis.
In addition, other specificities and differences deduced from the structural analysis seem to be supported by the genomic analysis: (i) The differences in length of the carbon FA at C3', 14:0(3-OH) for Bho1, FR4020 and ATCC 51541 or 10:0(3-OH) for FR4101. LpxA is the first enzyme of the lipid A biosynthesis pathway catalyzing the addition of an acyl chain onto the C3' carbon [40]. We identified lpxA B. holmesii gene and observed a non-synonymous SNP in FR4101, leading to an amino acid change, as compared to ATCC 51541, FR4020 and Bho1 lpxA sequences. Thus, we concluded that the difference observed in the length of the carbon FA at C3' between FR4101 and the other isolates could be due to this difference within lpxA gene. Analysis of additional French B. holmesii isolates also led to the conclusion that this was not related to the blood or respiratory origin of the sample which was also confirmed by the lpxA sequence analysis of isolates with available genomes on NCBI. (ii) The presence of 10:0(3-OH) at C-3, like in B. pertussis [1] and B. petrii [41] is also a common trait between the three clinical isolates. In B. pertussis, this is the consequence of the lack of activity of the C-3 de-O-acylase PagL, which has been shown to result in a lower cytokine induction capacity of the Bordetella human pathogens [29,42]. The presence of this fatty acid was interpreted as resulting in the facility of such bacteria, with short-chain fatty acid structures, and low acylation pattern, to escape the human host defense through the MD2-TLR4 complex [42,43]. We found that the pagL gene sequence is the same for the three isolates and the ATCC 51541 strain. Both H and S residues in the C-terminal part of PagL protein, described as essential for catalytic activity are present in B. holmesii [44]. In B. pertussis, pagL is a pseudogene because of a frameshift due to a deletion of "CA" bases. Such a deletion is not found in the B. holmesii sequence. Further investigations are necessary to understand why PagL is inactive in these isolates. (iii) The presence of a secondary palmitoyl chain at two positions of lipid A (3 and 2). We previously described the presence of such palmitoyl chains at C-3 in B. avium, and in both C3 and C2 in B. parapertussis [27,45] as a result of PagP action. In the present study, we observed no pagP sequence difference between the three tested isolates. They nevertheless display an additional G insertion as compared to ATCC 51541 reference strain, leading to a frameshift that probably explains why no palmitoyl acid is observed in these isolates. (iv) The hydroxylation of 12:0 and 14:0 at position 2 of the two secondary acylated residues was observed in the lipid A of the three isolates ( Figure 5). In Novikov et al. [27], such hydroxylations have been observed in B. avium, B. hinzii and B. trematum and were attributed to two homologs of the LpxO enzyme. We found two homologs of lpxO in the genome of the three isolates of B. holmesii as in the ATCC 51541 type strain that support the degree of hydroxylation observed. (v) The presence of a 12:0(2-OH) residue in a secondary position at C-2 on the amide-linked 14:0(3-OH) is a common trait between all the B. holmesii structures. LpxL is a lipid A lauroyl acyltransferase. Geurtsen et al. [46] showed that LpxL1, a homolog of the classical LpxL2 lauroyl acytransferase, is present but usually poorly expressed in B. pertussis, compared to LpxL2 [40,46]. However, through overexpressing it in B. pertussis B213 strain, they showed that this enzyme led to the presence of an extra secondary 12:0(2-OH) chain at the C-2 position, like we describe here in the B. holmesii isolates, and they reported that this type of acylation was required for efficient infection in human macrophages and could help host infection. We found a lpxL1 homolog in the three isolates as in ATCC 51541 type strain supporting this substitution.
All these results allowed identification of differences within lipid A structures in the different B. holmesii strain and isolates tested supported by both structural and genomic analysis. As illustrated in Figure 5, lipid A structures differ in their fatty acid substitution. The difference in the amount of free amino GlcN derivatives on the phosphate groups is described herein but not shown in the figure as only the peak intensity can be illustrating such differences.

Strains and Isolates
Strain ATCC 51541 was originating from the National Research Council (NRC) collection and was grown as previously described [47]. The three human isolates were selected from the French National Reference Center for Whooping Cough and other Bordetelloses [19]. Bho1 was isolated in 1996 from the blood of a 20 years old man with sickle-cell anemia [48]. FR4020 and FR4101 were isolated, respectively; from nasopharyngeal swabs of an adult in 2008, and of an adolescent in 2009, both presenting Pertussis-like symptoms [19,49]. Bacteria were grown on Bordet-Gengou agar (Difco by Becton Dickinson, NJ, USA) supplemented with 15% sheep defibrinated blood (Biomerieux, Marcy l'Etoile, France) at 36 • C for 72 h, plated again for 18 h and then grown in enriched Stainer Scholte medium [50].

Lipid A Isolation from Whole Cells
Briefly, cells (10 mg) were washed twice with 400 µL of a fresh, single phase mixture of

Lipid A Isolation from LPS as Performed by the Triethylamine Citrate (TEA) Method
A concentration of 0.01M TEA-citrate (1:1 molar ratio, pH 3.6) was used. Samples were suspended in Eppendorf ® tubes (Eppendorf AG, Hamburg, Germany) in the above-mentioned reagents at concentrations of 5 µg/µL, or 10 µg/µL depending to their solubility. After agitation and homogenization using an ultrasonic bath, the tubes were incubated for 1 h in a Thermomixer system (Eppendorf AG) under stirring at 1000 rpm and 100 • C [53].

Fatty Acid
Fatty acids were analyzed after hydrolysis of the LPS or lipids A with 4 M HCl for 2 h at 100 • C, and neutralization, followed by treatment with 2 M NaOH 2 h at 100 • C [55], extraction with ethyl acetate, methylation of the extract with a mixture of anhydrous methanol and acetyl chloride (10:1.5 v/v) [56]. They were identified by GC-MS for the fine characterization and confirmation of FA composition as visualized by MALDI-MS (mass spectrometry).

O-Deacylation
Ester-linked fatty acids were likewise characterized after treating the lipids A with NH 4 OH during 5 h at 50 • C [32] for liberation of primary ester-linked fatty acids and with methylamine for 5 h at 50 • C for the secondary ester-linked fatty acids. GC-MS analysis of the methylated fatty acids was performed as previously described. Arachidonic acid (C 20 , Sigma), ,was used as a standard [56].

Mass Spectrometry
MALDI negative-ion mass spectrometry analyses were performed on a PerSeptive Voyager-DE STR model time-of-flight mass spectrometer of Applied Biosystem, (I2BC, Université de Paris Sud XI). The apparatus is equipped with a 337 nm nitrogen laser and spectra were recorded in the linear negative-ion mode with delayed extraction.
The ion-accelerating voltage was set at 20 kV. Dihydroxybenzoic acid (DHB) (Sigma) suspended in 0.1 M citric acid was used as a matrix. A few microliters of lipid A were dissolved in a mixture of chloroform:methanol:water (3:1.5:0.25 v/v/v) at 1 mg/mL and desalted with a few grains of ion-exchange resin (Dowex 50W-X8, Sigma, St. Louis, MO, USA) (H + ), in an Eppendorf tube. A 1 µL aliquot of the solution (50 µL) was deposited on the target and covered with the same amount of the matrix suspended at 10 mg/mL in the same mixture of solvents [39]. Different ratios between the samples and DHB were tested when necessary. B. pertussis lipid A was used as an external standard.

Conclusions
The B. holmesii lipid A structures established in the present study differ from one to the other: (i) Presence of free amino GlcN on the phosphate groups of lipid A structure: in Bho1 as in ATCC 51541, the failure of BvgA to function leads to a weaker substitution than in the two other isolates.
(ii) Presence of 10:0(3-OH) at C-3, which we already observed in other Bordetella species such as B. pertussis [31], is thought to be the consequence of the lack of activity of the 3 O-deacylase PagL, even if genomic explanation could not be found to support this hypothesis. (iii) The 14:0(2-OH) and 12:0(2-OH) fatty acids found in all the presented structures were previously shown to be important structural features for the infection of human macrophages [42].
The combined structural virulent traits, characteristic of the B. holmesii lipid A structures, demonstrate the potential strength of this bacterium in its capacity to adapt for escaping to the host immune defenses. The efficiency of these traits, and especially the presence of the short-chain 10:0(3-OH) is already attested by the virulence of the whooping-cough pathogens B. pertussis and B. parapertussis [31,52] and were also recently characterized in B. petrii lipid A isolates [41]. However, no real difference of lipid A structure was observed between isolates from blood or respiratory origins which correlates with data from other studies [18,19]. This also correlates with the fact that, in cases of bacteremia, nasal carriage should be assessed as shown in a previous study [13].
The lipid A structures we characterized along the years on numerous strains and species always gave important information confirmed by genomics; lipid A is a powerful taxonomic tool and this work is another example of the kind. We demonstrated here the correlation between B. holmesii and B. hinzii, B. avium and B. trematum also recently shown to belong to the same clade as based on whole genome SNP-based analysis.
As already mentioned, lipid A are good bacterial markers, and their characterization in small biological samples can now help to differentiate the human Bordetella pathogens.