All strains and plasmids used in this study are listed in Table 1. Bacteria were grown as described before (Ifill et al., 2021). As needed, media were supplemented with nalidixic acid (Nal; 30 μg/mL), gentamicin (Gm; 15 μg/mL), kanamycin (Kan; 75 μg/mL), diaminopimelic acid (DAP; 250 μg/mL) and/or anhydrous tetracycline (aTC; 12.5 ng/mL).

A markerless clean deletion protocol was used to delete the genes of interest: lgmA-D, waaL, wbmA-E, pagP and pagL in B. parapertussis as described previously (Marr et al., 2010a; Shah et al., 2013; Ifill et al., 2021). Due to the sequence similarity in the lgm locus between B. pertussis and B. parapertussis, the same plasmid (pSS4245 ΔlgmA-D) used to delete the locus in B. pertussis (Shah et al., 2013), was used in B. parapertussis as well. pIG02 was used for the clean deletion of other genes by cloning ~600-bp of the upstream and downstream regions of the genes of interest [separated by an SpeI restriction enzyme (RE) site] between the KpnI and BamHI RE sites in its multiple cloning site (MCS). This construct was then transformed into E. coli RHO3. These allelic exchange plasmids were then conjugated into B. parapertussis using the di-parental mating protocol as previously published (Ifill et al., 2021) with the noted absence of nalidixic acid (B. parapertussis strain is not Nal^r). In the case of double mutants, the allelic exchange protocol was repeated with the generated single gene deletion mutant and the RHO3 strain containing the pIG02 construct for the clean deletion of the second gene of interest.

The B. parapertussis waaL, wbmA-E, pagP and pagL deletion mutants were complemented with their respective genes using pIG10, an anhydrous tetracycline (aTC) inducible expression plasmid created for Bordetella species by Gyles Ifill (Ifill and Fernandez, manuscript in preparation). The gene of interest was amplified and cloned into the MCS of pIG10 between the SpeI and BamHI RE sites. In the case of B. parapertussis ΔpagP ΔpagL, both genes were cloned into pIG10 separated by a HindIII RE site and a ribosome-binding sequence, 5′-GGCAAGTCTAAAGCCATAGAAGGATAC-3′ to ensure the expression of both genes. The construct was introduced into the respective mutants using di-parental conjugation as described above.

The lgm locus, along with ~1,000-bp of the upstream region (to include its native promoter), was introduced into the chromosome using the mini-Tn7 transposon delivery plasmid, pUC18T-mini-Tn7T-Km-FRT, at the attTn7 site located downstream of the highly conserved and essential glmS genes (Choi et al., 2005). The construct, and its transposase vector, pTNS2, were introduced into the ΔlgmA-D mutants using tri-parental conjugation. This method, similar to di-parental mating described above, involved mixing E. coli RHO3 containing the pUC18-miniTn7T-lgmA-D construct, E. coli RHO3 carrying pTNS2 and the B. parapertussis mutant strain in the ratio of 1:1:2. Kanamycin was used as the selection antibiotic. Successful integration of the transposon was confirmed by PCR.

B. pertussis and B. parapertussis strains were grown in liquid culture inoculated at an initial OD₆₀₀ of 0.001. They were grown to an OD₆₀₀ of 0.6–0.8 (up to 72 h under agitation). 1.5 mL of bacterial suspension (concentrated to an OD₆₀₀ of 2) was digested with DNase I, RNase and proteinase K, and the resulting lysate was separated using tricine-SDS-PAGE and visualized using silver staining (Marolda et al., 2006).

To prepare cells for MALDI-TOF analysis, 100 mL of B. pertussis or B. parapertussis liquid culture was grown as stated above. The bacteria were harvested, and lipid A was extracted using the ammonium-isobutyrate method (El Hamidi et al., 2005) and analyzed in the Applied Biosystems MALDI-TOF spectrometer as described previously (Fathy Mohamed and Fernandez, 2024). Data was acquired and analyzed using the Data Explorer software and graphed using GraphPad Prism 10 (RRID:SCR_002798).

HEK-Blue^™ hTLR4 cells (InvivoGen Cat# hkb-htlr4) and HEK-Blue^™ Null2 cells (InvivoGen Cat# hkb-null2) were cultured as described previously (Shah et al., 2013). HEK-Blue^™ hTLR4 cells are engineered from HEK293 cell line to stably express human TLR4, MD2, CD14 and an inducible secreted embryonic alkaline phosphatase (SEAP) reporter gene to measure NFκB activation. HEK-Blue^™ Null2 is the parental cell line of HEK-Blue^™ hTLR4 expressing the SEAP reporter alone to exclude NFκB responses induced by the activation of endogenously expressed pattern recognition receptors, including TLR3, TLR5 and RIG-1-like receptors.

THP1-Dual^™ cells (InvivoGen Cat# thpd-nfis) were cultured in RPMI 1640 (GIBCO) containing 10% heat-inactivated (30 min at 56°C) fetal bovine serum (Sigma), 2 mM GlutaMAX (GIBCO), 25 mM HEPES (GIBCO), 100 μg/mL Normocin (InvivoGen), and Penicillin-Streptomycin (100 U/mL-100 μg/mL; GIBCO) in the presence of selection antibiotics: 100 μg/mL zeocin (InvivoGen) and 10 μg/mL blastocidin (InvivoGen). They were passaged after reaching densities of 1–2 × 10⁶ cells/mL. THP1-Dual^™ cells are engineered from human THP-1 monocyte cell line to express two inducible reporter genes, SEAP to measure NFκB activation and Lucia luciferase to measure IRF3 activation.

THP-1 cells (ATCC) were cultured as described previously (Fathy Mohamed and Fernandez, 2024) and passaged after reaching a density of 1 × 10⁶ cells/mL.

All cell lines were incubated in a CO₂ incubator at 37°C with 5% CO₂.

All Bordetella strains, and the positive control, E. coli DH5α, were grown to an OD₆₀₀ of 0.6–0.8. They were concentrated to an OD₆₀₀ of 5 in phosphate-buffered saline (PBS) and heat-killed at 60°C for 1 h. The lack of viability was confirmed by spotting a small aliquot (2 μL) on agar plates and checking for the lack of bacterial growth after incubating the plates for up to 5 days. The heat-killed samples were stored at −20°C.

E. coli K12 LPS (InvivoGen) was resuspended as recommended and stored in aliquots at −20°C. When needed, an aliquot was thawed, placed in a sonicating water bath for 10 min and then used to prepare required dilutions for the respective assays.

HEK-Blue^™ hTLR4 cells and HEK-Blue^™ Null2 cells were grown to ~70–80% confluency. Then the reporter assay was carried out as described previously (Shah et al., 2013) using the indicated dilution of heat-killed bacterial suspension as stimulants or media for negative control. The alkaline phosphatase reporter activity was quantified by measuring the absorbance after the indicated incubation period with the QUANTI-Blue reagent (Invivogen) at 650 nm in the Molecular Devices SpectraMax 190 microplate reader or the Thermo Scientific VarioSkan Flash multimode plate reader. Readings were converted as a percentage of B. parapertussis WT. One-way ANOVA with Tukey’s multiple comparison test was performed using GraphPad Prism 10.

One hundred and eighty microliters of THP-1 Dual^™ cells (~100,000 cells/well) were aliquoted per well of 96-well flat-bottomed, tissue culture-treated plates (Corning, Cat# 353072). They were differentiated into macrophages by treating them with 50 ng/mL phorbol 12-myristate 13-acetate (PMA; Sigma-Aldrich) for 48 h at 37°C in a CO₂ incubator. The cells were then washed with fresh media twice to remove traces of PMA and rested for 72 h. On day 4, the cells were washed again, and 180 μL of fresh media was added per well. Subsequently, 20 μL of stimulant prepared in media was added per well to obtain the desired final concentration of stimulant (1:100 dilution of heat-killed bacterial suspension; endotoxin-free water for negative control; 1 μg/mL E. coli K12 LPS as positive control). After incubation at 37°C for 24 h, 10 μL of the supernatant was used to determine the luciferase activity using the QUANTI-Luc 4 Lucia/Gaussia reagent (InvivoGen) as per manufacturer’s flash detection protocol in the Perkin-Elmer Victor X5 Multilabel reader. Readings were converted as a percentage of B. parapertussis WT. Mixed-effects analysis with Tukey’s multiple comparison test was performed using GraphPad Prism 10.

The THP-1 stimulation assay was performed as previously described (Marr et al., 2010a). THP-1 cells were differentiated using PMA and stimulated with the desired dilution of stimulant (1:10 dilution of heat-killed bacterial suspension in complete RPMI 1640; sterile media for negative “no stimulation” control; 10 μg/mL of E. coli K12 LPS) in duplicates. At 4 h post-stimulation, the supernatant was removed, and the THP-1 cells were washed with sterile PBS. The cells were then scraped in 100 μL PBS and stored in Eppendorf tubes at −20°C.

When performing western blot experiments, the cells were denatured and proteins were separated by 12% SDS-PAGE as described before (Fathy Mohamed and Fernandez, 2024). Three such gels were prepared for each stimulation assay. The first gel was stained with PageBlue^™ protein staining solution (Thermo Scientific). Image Lab Software (RRID:SCR_014210) was used to detect total protein content of the sample lanes. Proteins from the second and third gels were transferred to Immobilon-P polyvinylidene difluoride (PVDF) membranes (Sigma) and immunoblotted for p-IRF3 and p-STAT1, respectively (Fathy Mohamed and Fernandez, 2024). Primary antibodies anti-IRF3 (phosphor S386) antibody EPR2346 (Abcam Cat# ab76493, RRID:AB_1523836) and phospho-Stat1 (Tyr701) (58D6) Rabbit mAb (Cell Signaling Technology Cat# 9167, RRID:AB_561284), as well as, secondary antibody Peroxidase AffiniPure^™ Goat anti-Rabbit IgG (H + L) (Jackson ImmunoResearch Labs Cat# 111–035-144, RRID:AB_2307391) were used. The proteins were detected using chemiluminescence (ECL^™ Prime Western Blotting Detection reagent; Cytivia) in the BioRad Imaging System. The membrane was exposed for 360 s, and the image obtained was analyzed using Image Lab Software to calculate band intensities. The band intensities, normalized to the total protein content, were converted as a percentage of B. parapertussis WT. One-way ANOVA with Tukey’s multiple comparison test was performed using GraphPad Prism 10.

To further understand how TLR4 differentiates structurally distinct Bordetella LPS, B. pertussis and B. parapertussis strains were genetically engineered to encode LPS with different structural features. First, to investigate if the O antigen biases signaling towards the TRIF pathway, LPS structures with and without the O antigen were engineered. Of the two parent strains, B. parapertussis expresses an O antigen due to the presence of an intact wbm biosynthesis locus, while, B. pertussis lacks an O antigen as the locus was replaced by an insertion sequence (Di Fabio et al., 1992; Preston and Maskell, 2001). The O antigen from B. parapertussis was removed by deleting either the ligase (WaaL) that attaches the pre-formed O antigen to the lipid A + core moiety (Mulford and Osborn, 1983; Kalynych et al., 2014) or by deleting the first five genes of the wbm locus (wbmA-E), which has been shown sufficient to prevent O antigen synthesis (Preston et al., 1999) as indicated in Figure 1A. Thus, hexa-acylated LPS with (B. parapertussis WT) and without (B. parapertussis ΔwaaL or ΔwbmA-E) an O antigen were generated. Additionally, by altering the number of acyl chains in B. parapertussis as outlined below, we generated a penta-acylated LPS expressing an O antigen (B. parapertussis ΔpagP ΔpagL) which could now be compared to the penta-acylated B. pertussis WT LPS that lacks the O antigen.

To study the influence of the number of acyl chains on TLR4 signaling, strains encoding lipid A with 4 to 7 acyl chains were created by deleting pagP and/or pagL from hexa-acylated B. parapertussis (El Hamidi et al., 2009). PagP is a palmitoyl transferase that adds a secondary palmitate group to the acyl chains present at the C2 and C3′ positions (Hittle et al., 2015) (Figure 1A). On the other hand, PagL is a lipid A deacylase that removes the 3-hydroxydecanoic acid moiety from the C3 position (Geurtsen et al., 2006). Hence, deleting pagP, pagL or both together led to LPS species with 4, 7 or 5 acyl chains respectively, that could now be compared to B. parapertussis and B. pertussis LPS containing 6 and 5 acyl chains, respectively.

Both B. pertussis and B. parapertussis encode the lgm locus responsible for the GlcN modification of the phosphates of the lipid A backbone (Marr et al., 2008; Geurtsen et al., 2009). To study their role in TLR4 signaling in conjunction with other LPS structural features, the entire locus, lgmA-D, was deleted from both B. pertussis and B. parapertussis to create LPS with and without the GlcN modification.

The genes stated above were either deleted singly or in combination, deleting one gene at a time, using a markerless clean deletion protocol. In total, 10 Bordetella mutants were created that expressed eight different LPS structures that differed in the presence or absence of the O antigen, the number of acyl chains, and/or the presence or absence of the GlcN modification as indicated in Figure 1B. All Bordetella mutants were complemented with the respective deleted genes, either on an aTC inducible pIG10 plasmid (for waaL, wbmA-E, pagP and pagL) or the mini-Tn7 transposase system (for lgmA-D) (Choi et al., 2008). These strains, their respective WT strain as well as E. coli K12 LPS were used in subsequent experiments to study TLR4 recognition and signaling.

After confirming the deletion of the respective genes in the mutants using PCR, evidence for the presence or absence of the O antigen in the engineered LPS was obtained using tricine-SDS-PAGE and visualized using silver-staining. Bands corresponding to the lipid A + core moiety (bottom band) and the O antigen-containing LPS (top smear) were noted (Figure 2A). The homopolymeric O antigen-containing LPS of B. parapertussis WT is seen as a smear at the top of the gel (Lane 1). The mutants lacking the O antigen (ΔwaaL, ΔwbmA-E, ΔlgmA-D ΔwaaL and ΔlgmA-D ΔwbmA-E; Lanes 3 to 6) lack the smear at the top, confirming the loss of O antigen. All other mutants containing modifications to the lipid A alone show no change in the expression of the O antigen as expected (Lanes 2 and 7 to 10). B. pertussis WT and ΔlgmA-D mutants naturally do not express the O antigen as confirmed by the lack of the O antigen-containing LPS smear (Lanes 11 and 12). However, the lipid A + core band of B. pertussis migrates slower than that of B. parapertussis, presumably due to the addition of the distal trisaccharide unit to the core by the wlb locus, which is absent in B. parapertussis (Allen et al., 1998). The B. parapertussis O antigen mutants (ΔwaaL, ΔwbmA-E, ΔlgmA-D ΔwaaL and ΔlgmA-D ΔwbmA-E), when complemented with the respective deleted genes, demonstrate successful complementation of waaL or wbmA-E as indicated by the reappearance of the O antigen-containing LPS smear (Supplementary Figure S1).

Evidence for the absence of the GlcN moiety (in the ΔlgmA-D mutants) and of the number of acyl chains present (in the ΔpagP and ΔpagL mutants) was acquired via MALDI-TOF. Lipid A was extracted using mild acid hydrolysis and analyzed using negative, linear ion-mode MALDI-TOF. While MALDI-TOF was performed for all mutants, Figures 2B,C illustrate the spectra of select mutants. Peaks of interest are labeled in red or green. Other m/z in the spectra correspond to micro-heterogeneity due to changes in hydroxylation and acyl chain length. The top panel of Figure 2B shows the spectra of B. pertussis WT LPS. The peak at m/z 1,558 represents the penta-acylated LPS species, followed by tetra-acylated species at m/z 1,332 and 1,234. The addition of a single GlcN moiety is observed as an addition of m/z 161 resulting in a peak at m/z 1,720 (green) (Marr et al., 2010a). The ΔlgmA-D mutant lacks this peak confirming the loss of this moiety (Figure 2B lower panel).

The central panel of Figure 2C shows the spectra of B. parapertussis WT LPS. Similar to what was previously reported for B. parapertussis (El Hamidi et al., 2009; Hittle et al., 2015), our data shows peaks at m/z 1,332, 1,571, and 1,811 that correspond to LPS species with 4, 5 and 6 acyl chains, respectively. The corresponding GlcN modified species, with m/z 161 higher, are highlighted in green (m/z 1,493, 1,733 and 1,972 respectively). The peaks corresponding to the GlcN modification are absent in the ΔlgmA-D mutant (top left). The ΔwaaL and ΔwbmA-E mutants, which did not have any modifications to their lipid A structure, had the same mass spectra as B. parapertussis WT (Supplementary Figure S2). Likewise, the ΔlgmA-D ΔwaaL and ΔlgmA-D ΔwbmA-E had mass spectra similar to the ΔlgmA-D mutant (Supplementary Figure S2). PagP adds two palmitate groups to B. parapertussis LPS, with m/z 238.4 each. Correspondingly, the ΔpagP mutant (top right) lacks the peaks at m/z 1,571 and 1,811 corresponding to the penta-and hexa-acylated species. Deletion of PagL prevents the deacylation of a C₁₀-OH group with m/z 170.25. Hence, the ΔpagL mutant (bottom right) is seen to have an additional peak at m/z 1,983 indicative of a hepta-acylated species, followed by the addition of a single GlcN at m/z 2,144. Also, the ΔpagP ΔpagL double mutant (bottom left) corresponded to the loss of two palmitate groups at m/z 238.4 each and the addition of C₁₀-OH group at m/z 170.25, resulting in a final loss of m/z 306.55. Thus, the double mutant has a major peak corresponding to a penta-acylated species with m/z 1,502, that is m/z 309 less than the hexa-acylated WT LPS at m/z 1,811.

Next, the ability of the engineered Bordetella LPS to activate TLR4-mediated NFκB signaling was investigated. Heat-killed bacteria (at 1:100 dilution) from each strain were introduced to an NFκB reporter cell line, HEK-Blue^™ hTLR4 cells, that expresses human TLR4/MD-2 and CD14. HEK-Blue^™ Null2 was used as a control to rule out NFκB activation by endogenously expressed pattern recognition receptors. The degree of NFκB activation was measured 15 min post-mixing with the QUANTI-Blue reagent as an absorbance readout at 650 nm (Figure 3). Hexa-acylated E. coli DH5α was used as a control (Lane C1). B. parapertussis WT (Lane 1), with a hexa-acylated lipid A, did not induce as much NFκB as E. coli LPS (Lane C1), but it activated NFκB stronger than the penta-acylated B. pertussis LPS (Lane 11). The most prominent phenotype observed across the board was the significant reduction in NFκB activation upon the deletion of the lgm locus when compared to the WT or their respective single mutant parent in the case of a double mutant (Lanes 2, 4, 6, 8 and 12). Furthermore, the reduction in NFκB activation on deleting the lgm locus in B. parapertussis WT is much more striking than that observed in B. pertussis (Lanes 1 and 2 vs. Lanes 11 and 12).

Secondly, the O antigen had no impact on NFκB activation. B. parapertussis LPS lacking O antigen either by the deletion of the ligase (ΔwaaL; Lane 3) or the biosynthesis locus (ΔwbmA-E; Lane 5) showed equivalent NFκB activation as the WT strain (Lane 1). Also, penta-acylated Bordetella LPS that expresses the O antigen (B. parapertussis ΔpagP ΔpagL; Lane 10) and one that does not (B. pertussis WT; Lane 11) activated NFκB to equal degrees. Lastly, the number of acyl chains had a nuanced impact on NFκB signaling. Having an extra acyl chain (ΔpagL; Lane 9) compared to the B. parapertussis WT (Lane 1) was neither beneficial nor detrimental to NFκB activation. However, reducing the number of acyl chains (to 4 in ΔpagP mutant or 5 in ΔpagP ΔpagL double mutant; Lanes 7 and 10) reduced NFκB activation. Any reduction of NFκB activation was restored to WT levels upon complementation of the respective mutants (Supplementary Figure S3).

TLR4-mediated NFκB activation (as measured above) is triggered by both the MyD88 and the TRIF pathways. Of these, the MyD88 pathway is key for the immediate and strong activation of NFκB, while the TRIF pathway, though majorly involved in activating IRF3 and Type I interferons, triggers the late-phase activation of NFκB (Lu et al., 2008). Hence, we investigated if these structural changes in LPS biased the TLR4-mediated signaling toward the TRIF pathway.

To assess TRIF-mediated IRF3 activation, two techniques were employed. First, heat-killed bacteria were used to stimulate THP-1-Dual cells, which consist of a Lucia luciferase reporter under the control of the interferon-stimulated response element (IRSE) which is induced by phosphorylated IRF3 or Type 1 interferon-mediated STAT signaling. Luciferase activity was then measured at 24 h using the flash detection method and expressed as relative light units. E. coli K12 LPS was used as a positive control.

B. parapertussis WT showed high levels of IRF3 activation which was comparable to B. pertussis WT (Figure 4A; Lanes 1, 11). Both O antigen mutants (ΔwaaL and ΔwbmA-E; Lanes 3 and 5) showed a significant reduction in IRF3 activation. Altering the number of acyl chains down to 4 or 5 also abolished IRF3 activation (Lanes 7, 10), while increasing it to 7 had no impact (Lane 9). Intriguingly, the penta-acylated B. parapertussis ΔpagP ΔpagL mutant (Lane 10) activated IRF3 to a significantly lower extent than B. pertussis WT (Lane 11) despite it having an O antigen (whose presence seemed to increase IRF3 activation in B. parapertussis WT compared to its O antigen mutants). Upon the absence of the GlcN modification in respective single or double mutants, negligible IRF3 activation was observed (Lanes 2, 4, 6, 8, 12). Like that observed in NFκB activation, the drop in IRF3 activation in the B. parapertussis ΔlgmA-D mutant when compared to its WT is more dramatic compared to that observed in B. pertussis (Lanes 1 and 2 vs. Lanes 11 and 12).

The above assay was not sufficiently sensitive to detect immediate IRF3 activation at 4 h post-stimulation. Thus, western blot was used to detect the presence of phosphorylated IRF3 (p-IRF3) and phosphorylated STAT1 (p-STAT1) 4 h after the stimulation of PMA-differentiated THP-1 cells with heat-killed bacteria. A representative western blot image for p-IRF3 and p-STAT1 is shown in Figure 4B,i. The band intensity of p-IRF3 and p-STAT1 was quantified using Image Lab software and represented as a percentage of B. parapertussis WT in Figures 4B,ii,iii respectively. Except for B. parapertussis ΔpagL mutant (Lane 9) and B. pertussis WT (Lane 11), the western blot data at 4 h replicated the luciferase reporter assay results observed at 24 h. B. parapertussis had the highest intensity of p-IRF3 and p-STAT1 bands (Figures 4B,ii,iii; Lane 1). Deleting the O antigen caused a moderate reduction in p-IRF3 band intensity (Lanes 3, 5), while deleting the lgm locus (Lanes 2, 4, 6, 8, 12) or altering the number of acyl chains (Lanes 7, 9, 10) caused a significant reduction (Figure 4B,ii). Contrary to the luciferase assay, the hepta-acylated ΔpagL mutant (Lane 9) had significantly less p-IRF3 band intensity than the WT (Lane 1) at 4 h post-stimulation. Similarly, B. pertussis also had minimal IRF3 activation (Figure 4B,ii; Lane 11). Additionally, any structural changes to B. parapertussis WT LPS led to significantly lower STAT1 activation in all strains when compared to the WT (Figure 4B,iii; Lane 1). B. pertussis also had minimal STAT1 activation at 4 h (Figure 4B,iii; Lane 11).