Expression level of human TLR4 rather than sequence is the key determinant of LPS responsiveness

To address the role of Toll-like receptor 4 (TLR4) single nucleotide polymorphisms (SNP) in lipopolysaccharide (LPS) recognition, we generated mice that differed only in the sequence of TLR4. We used a bacterial artificial chromosome (BAC) transgenic approach and TLR4/MD-2 knockout mice to specifically examine the role of human TLR4 variants in recognition of LPS. Using in vitro and in vivo assays we found that the expression level rather than the sequence of TLR4 played a larger role in recognition of LPS, especially hypoacylated LPS.


Introduction
Two coding human Toll-like receptor 4 (TLR4) single nucleotide polymorphisms (SNP; rs4986790 and rs4986791) resulting in Asp!Gly or Thr!Ile substitutions at amino acids 299 and 399, respectively, have been described [1]. In Europeans, the allele frequency of the double TLR4 D299G+T399I allele is~7% [2]. The earliest study on these coding TLR4 SNPs suggested that they resulted in decreased responsiveness to inhaled lipopolysaccharide (LPS) [1]. Another report demonstrated prevalence of the single TLR4 D299G polymorphism in Africa (5-9% allele frequency), whereas the TLR4 D299G+T399I allele was found to be prevalent in Europe (3-9% allele frequency) [3]. The authors proposed that the TLR4 D299G variant initially arose to protect against cerebral malaria in Africa and that this allele was hyperresponsive to LPS, in contrast to the previous study. They proposed that the compensatory T399I mutation that reduced LPS responsiveness arose in Europe to protect from sepsis. That study also showed that the single TLR4 T399I variant is very rare. Other in vitro studies also demonstrated reduced responsiveness of TLR4 D299G+T399I to hexa-or penta-acylated LPS [4,5].
Multiple association studies have linked these SNPs to a wide variety of diseases (early studies reviewed in [6]). Published reports describe association with sepsis, meningococcal disease, cerebral malaria, Crohn's disease, ulcerative colitis, respiratory syncytial virus, atherosclerosis, rheumatoid arthritis, malignant melanoma, and metabolic syndrome, although others show no association for the same diseases suggesting complex genetic and epigenetic interactions [1,4,[7][8][9][10][11][12][13][14][15][16][17]. Two additional reports suggested that the TLR4 D299G variant is protective against a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 increased systolic blood pressure with increased obesity as well as against periodontitis, though both failed to examine the co-segregating rs4986791 SNP and thus actual genotype cannot be determined [18,19]. Since the data in the literature is association data and even this data in some instances is incomplete and conflicting, we sought to develop a model system that would more clearly isolate the role of the coding TLR4 SNPs in response to LPS and disease states.
Thanks to the success in generating mice that display human-like responses to LPS through the expression of human TLR4 and MD-2 from human genomic BACs [20], we generated transgenic mice expressing either the TLR4 D299G or the TLR4 D299G+T399I human allele and compared them to mice expressing human TLR4 WT . Although our comparisons were limited by the copy number of lines randomly generated, we found that the expression level of human TLR4 was the largest determinant of LPS response and that the actual sequence of human TLR4 did not reveal obvious differences in LPS recognition.

Reagents
Ultrapure O111:B4 EC LPS was purchased from InvivoGen (San Diego, CA), and ODN1826 from Coley Pharmaceuticals (prior to acquisition by Pfizer, Düsseldorf, Germany). Monophosphorylated lipid A (MPL) was purchased from Avanti Polar Lipids Inc (Alabaster, AL). Antibodies for flow cytometry were purchased from Becton Dickinson (Franklin Lakes, NJ), eBioscience (San Diego, CA) or BioLegend (San Diego, CA). Anti-huTLR4 clones HTA125 and TF901 as well as anti-muTLR4 clone MTS510 were used for surface expression analysis.

PA and YP LPS preparation
Bacterial growth conditions and strains. The strains used in this study are P. aeruginosa (PAO-1) and Y. pestis (KIM6-). P. aeruginosa was grown in Lysogeny broth (LB) supplemented with 1mM MgCl 2 at 37˚C with shaking and Y. pestis was grown in brain heart infusion (BHI) supplemented with 1mM MgCl 2 at 37˚C with shaking.
LPS and lipid A isolation. LPS was extracted by the hot phenol/water method [21]. Freeze-dried bacterial pellets were resuspended in endotoxin-free water at a concentration of 10 mg/ml. A volume of 12.5 ml of 90% phenol (Fisher Scientific, Pittsburgh, PA) was added and the resultant mixture was vortexed and incubated for 60 minutes in a hybridization oven at 65˚C. The mixture was cooled on ice and centrifuged at 12,096 x g at room temperature for 30 minutes. The aqueous phase was collected and an equal volume of endotoxin-free water was added to the organic phase. The extraction was repeated and aqueous phases were combined and dialyzed against Milli-Q purified water to remove residual phenol and then freezedried. The resultant pellet was resuspended at a concentration of 10 mg/ml in endotoxin-free water and treated with DNase (Qiagen, Venlo, Limburg) at 100 μg/ml and RNase A (Qiagen, Venlo, Limburg) at 25 μg/ml and incubated at 37˚C for 1 hour in a water bath. Proteinase K (Qiagen, Venlo, Limburg) was added to a final concentration of 100 μg/ml and incubated for 1 hour in a 37˚C water bath [22]. The solution was then extracted with an equal volume of water-saturated phenol. The aqueous phase was collected and dialyzed against Milli-Q purified water and freeze-dried as above. The LPS was further purified by the addition of chloroform/ methanol 2:1 [vol:vol] to remove membrane phospholipids [23] and further purified by an additional water-saturated phenol extraction and 75% ethanol precipitation to remove contaminating lipoproteins [24]. For Mass Spectrometry (MS) structural analysis 1 mg of purified LPS was converted to lipid A by mild-acid hydrolysis with 1% sodium dodecyl sulfate (SDS) (Sigma, St Louis, MO) at pH 4.5 as described previously [25].

MALDI-TOF mass spectrometry. Lipid
A isolated by small-scale lipid A isolation procedures was analyzed on a AutoFlex Speed MALDI TOF mass spectrometer (Bruker Daltonics, Billerica, MA). Data was acquired in reflectron negative and positive modes with a Smartbeam laser with 1 kHz repetition rate and up to 500 shots were accumulated for each spectrum. Instrument calibration and all other tuning parameters were optimized using Agilent Tuning mix (Agilent Technologies, Foster City, CA). Data was acquired and processed using flexControl and flexAnalysis version 3.3 (Bruker Daltonics, Billerica, MA).

Mice
Ethics statement. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All protocols were approved by the Institutional Animal Care and Use Committee of the University of Washington.
Husbandry. All mice were bred in-house at the University of Washington. C57BL6/J mice were originally obtained from the Jackson Laboratory (Sacramento, CA). Mice were maintained in a specific pathogen free (SPF) facility in standard ventilated cages on autoclaved corn cob bedding with an automatic watering system and fed irradiated standard rodent chow with a 12 hr on/12 hr off light cycle. Several methods of euthanasia were used, depending on experimental manipulation. Mice were euthanized by CO 2 narcosis, followed by a secondary method, when only bones were collected. However, for experiments using splenocytes (including those where bones were also collected), mice were euthanized by sharp cervical dislocation. Finally, for experiments where mice had received intraperitoneal LPS prior to euthanasia, mice were injected IP with an overdose of Beuthanasia-D (Schering-Plough Animal Health Corp., Union, NJ).
Generation. Human genomic BAC RP11-150L1 was modified by introducing point mutations using lambda Red mediated recombination and galK positive/negative selection to generate the D299G and T399I variants as described [26]. Primers used for recombineering are shown in Table 1. The entire TLR4 coding region was sequenced in the modified BACs prior to injection into fertilized B6C3 x B6 oocytes by the Transgenic Core at the University of Washington. Founders were backcrossed to B6 mice for 5 generations followed by backcrossing to TLR4/MD-2 DKO mice on a C57BL/6J background for at least another 5 generations. Throughout the figures and text, KO or TLR4 KO refers to a functional TLR4 KO, i.e. it is a TLR4/MD-2 DKO mouse that also expresses human MD-2, but not human TLR4. Only males are used in these studies because the human MD-2 BAC is integrated on the Y-chromosome limiting our ability to study females [20].
BAC copy number. Copy number of the TLR4 BAC was determined using real-time PCR as previously described [20]. Primers used for amplification of TLR4 amplified both mouse and human TLR4 with equivalent efficiency allowing us to use a WT B6 mouse as a calibrator (copy number = 2) in the Pfaffl method to calculate the relative copy number of the transgene [27].

Flow cytometry
Bone-marrow-derived macrophages (BMDM) were generated and stained as described for surface TLR4 expression using anti-human TLR4 clones HTA125 and TF901 and anti-mouse TLR4 clone MTS510 and appropriate isotype controls (mouse IgG2a and IgG1, and rat IgG2a respectively) [20]. Whole splenocyte stimulation and intracellular cytokine analysis was performed as described [20] with macrophage/monocyte population identified as CD11b + CD11c low/negative (neg) CD3 neg CD19 neg cells.

RT-PCR
Relative TLR4 expression levels were determined by real-time PCR as described [20] using an ABI 7300 Real Time PCR System and the Brilliant II SYBR Green QPCR reagents (Agilent Technologies Inc., Santa Clara, CA). Total RNA was isolated from cells or tissues using Trizol, DNase treated, and reverse transcribed using oligo dT priming and Superscript II (Invitrogen, Carlsbad, CA). A no-RT control confirmed RNA amplification. ß-actin was used to normalize the data between samples.

Statistics
Prism software (GraphPad, La Jolla, CA) was used for all statistical analyses. Multiple comparisons were performed using 1-way ANOVA followed by Bonferroni's Multiple Comparison test.

Generation of human TLR4 SNP mice
Initially, we modified the human TLR4 BAC to introduce the D299G mutation using BAC recombineering methods [26]. We then further modified this BAC to introduce the T399I mutation and generate the double D299G/T399I allele (often labeled as 399 in the figures). The BACs were then injected into fertilized oocytes at the University of Washington Transgenic Core (http://depts.washington.edu/compmed/transgenic/). Chimeric pups were identified by coat color and backcrossed to B6 mice and subsequently to TLR4/MD-2 double-KO mice, such that mice differed only in the sequence of human TLR4. The mice were also bred to humanized MD-2 mice [20]. Thus, in addition to human TLR4, these mice also express human MD-2 and lack mouse TLR4 and mouse MD-2. We generated and further characterized two lines for each TLR4 allele.

Expression of human TLR4 variants
We first examined copy number using methods as described in [20] and found that our new lines consisted of 4 and 7 copies of huTLR4 D299G and 1 and 8 copies of huTLR4 D299G+T399I , as compared to our previously generated 2-and 4-copy lines of huTLR4 WT . While increasing BAC copy number from 1 to >48 correlates with increased expression (we did not have lines with identical copy numbers for comparison), when independent lines of same copy number were examined differences in staining intensity could be seen [28]. Therefore, we next examined expression of TLR4 in bone-marrow derived macrophages (BMDM) by staining for surface expression (Fig 1A-1C) and by RNA expression (Fig 1D). Increased copy number clearly resulted in increased RNA (Fig 1D) although protein expression appears to plateau by 7 copies, with no increase in the 8-copy line (Fig 1A and 1C). For this analysis, we also stained cells with a second clone (TF901; S1 Fig) that confirmed our results with clone HTA125, due to potential differences between antibodies directed at human TLR4 [29].

LPS responsiveness in primary splenic macrophages/monocytes
To determine whether we had restored E. coli (EC) LPS responsiveness in the splenic macrophage/monocyte population expressing the variant huTLR4 alleles, we used primary splenocyte intracellular cytokine staining (Fig 2) as described [20,30]. An example of the gating strategy and TNF production in the macrophage/monocyte population is shown in S2 and S3 Figs. Compared to the percentage of cells producing TNF in TLR4 KO splenocytes, all humanized TLR4 lines showed an increase in TNF+ cells following EC LPS stimulation (Fig 2A). The   absolute percent positive varied from day to day, as did the response to the positive control CpG (Fig 2B; each symbol is from a different mouse spleen on a different day). Therefore, we normalized the results by dividing the percentage of TNF+ cells in response to LPS by the percent TNF+ in response to CpG (Fig 2C). This correction helped reveal decreased TNF production by the 1-copy compared to 8-copy TLR4 D299G+T399I line. We also examined the primary splenic responses to hypoacylated LPS or monophosphorylated lipid A (MPL; Fig 3). We used a penta-acylated preparation of Pseudomonas aeruginosa (PA) LPS [31], as well as tetra-acylated Yersinia pestis (YP) LPS [20]. Unlike the response to EC LPS, all humanized TLR4 mice, including huTLR4 WT , showed reduced activity in response to hypoacylated LPS, as compared to WT B6 mice. In addition, following correction by the response to CpG, all lines except the 1-copy TLR4 D299G+T399I line showed significantly increased response to PA LPS or MPL compared to the KO responses. The 1-copy TLR4 D299G+T399I line did not respond to either PA or YP LPS even though we observed a response to EC LPS (Fig 2).

In vitro cytokine responses
We next generated bone-marrow-derived macrophages (BMDM), and stimulated cells with either PA, YP LPS, or EC LPS, and measured IL-6, CCL5, and IL-10 cytokines in the supernatants 24 hr later by Luminex (Fig 4). All humanized BMDM secreted very low levels of cytokines in response to PA or YP LPS, in contrast to WT B6 cells. In fact, the responses were not significantly different from TLR4 KO responses, although there was a trend towards higher values in the 7-copy TLR4 D299G and 8-copy TLR4 D299G+T399I lines compared to lower copy numbers. All humanized lines and the KO line were significantly reduced for IL-6 production, as compared to WT B6 cells, when stimulated with EC LPS. However, all humanized lines with the exception of the 1-copy TLR4 D299G+T399I line had increased IL-6 production compared to the KO. In contrast to the previous assays, the 1-copy TLR4 D299G+T399I line did not respond to EC LPS. Percent of mac/mono cells that are TNF+ in response to 1000 ng/ml PA LPS, YP LPS or MPL, top row, or percent TNF + corrected by the percent TNF+ in response to CpG, bottom row. * above genotypes shows results compared to B6 whereas † is compared to KO. Data for huTLR4 WT in response to MPL were combined for 1-way ANOVA since the 4-copy line was only tested once and our previous studies did not reveal any differences between the 2-and 4-copy lines. In vivo cytokine responses To determine whether differential recognition of hypoacylated LPS could be measured in vivo, we extended our studies into mice, and injected mice intraperitoneally (IP) with either EC or PA LPS (Fig 5 and Fig 6). In response to EC LPS, IL-6 and TNF were increased at 1 hr and decreased by 6 hr, although B6 maintained high IL-6 levels at 6 hr ( Fig 5). IL-10 was also increased at 1 hr in all genotypes compared to KO though absolute levels were vey low. At 6 hr, IL-12p40, CCL5 and CXCL9 were all increased. However, in all cases no significant differences were observed between the humanized lines indicating that the sequence of TLR4 did not alter responses to EC LPS.
Mice were also injected with hypoacylated PA LPS (Fig 6). When comparing the values of IL-6 at 1 hr to the EC response, it is obvious that PA LPS induced less IL-6 at 1 hr, although a small response was also observed in KO mice, potentially indicating that non-TLR4 contaminants were present in the preparation despite repurification [32]. By 6 hr, the levels of IL-6 were comparable between PA and EC LPS in B6 mice; however, the humanized TLR4 lines all had reduced levels that were not significantly different from KO mice. There is a trend, however, towards higher levels in the TLR4 SNP mice compared to human TLR4 WT for both IL-6 and IL-12p40 at 6 hr, though the results are not significant.

Discussion
There is a large body of literature that continues to expand examining the association of TLR4 SNPs with disease. However, no study has clearly, reproducibly, and mechanistically linked the common coding TLR4 SNPs (rs4986790 and rs4986791) with disease. TLR4 and MD-2 are coreceptors that bind directly to the lipid A component of LPS [33,34]. The structure of lipid  A varies between bacterial species and is also regulated in response to culture/environmental conditions or infection of a host [35,36]. Highly endotoxic lipid A found in E. coli and Salmonella is predominantly hexa-acylated and bisphosphorylated. This lipid A can be detoxified by the removal of one phosphate resulting in mono-phosphoryl lipid A (MPLA), a known vaccine Mice of the indicated genotypes were injected IP with PA LPS and then bled from the retro-orbital sinus 1 hr after injection. At 6 hr, the mice were euthanized for a terminal sample collection. Shown are combined data from 2 experiments. Data are plotted on a log scale for easier visualization due to large variation. * above symbols shows results for that genotype compared to B6 whereas † is compared to KO. adjuvant that results in a selective decrease in MyD88-dependent responses while TRIFdependent responses are maintained [37,38]. These results suggest that the structure of the lipid A molecule alters signaling. Pseudomonas aeruginosa contains lipid A that is penta-acylated although cystic fibrosis isolates produce hexa-acylated molecules [39]. Similarly, Yersinia pestis produces hexa-acylated lipid A at room-temp but a tetra-acylated molecule at 37˚C [40,41]. These hypoacylated molecules induce reduced inflammatory responses selectively through human TLR4/MD-2 compared to mouse TLR4/MD-2 again demonstrating that the structure of lipid A alters signaling through TLR4/MD-2 [20,31]. Similarly, TLR4 and MD-2 modification (by targeted mutagenesis) alters recognition of lipid A [42].
The crystal structure of TLR4 D299G+T399I in complex with E. coli LPS demonstrated conformational changes at the site of the D299G polymorphism though not at the site of LPS binding [43]. One in vitro study that showed equivalent LPS binding between the D299G variant and WT TLR4 suggested that decreased inflammatory cytokine production mediated by huTLR4 D299G is due to decreased recruitment of MyD88 and TRIF [44]. Our studies here do not reveal any obvious mechanistic differences in natural LPS recognition, including hypoacylated LPS, by the TLR4 variants, although one study suggests that the TLR4 D299G+T399I variant is impaired in dimerization in response to MPLA [29]. In fact, our results confirm what others have clearly demonstrated in vitro and in mouse studies [45][46][47][48], that increased expression of TLR4 results in increased responsiveness to LPS. Our results also suggest that non-coding TLR4 SNPs (e.g. rs2770150, rs7873784, rs10759932) that could result in altered expression levels of TLR4 are likely more important in disease association than the coding SNPs.