Engineering of LPS biosynthesis in E. coli
We began our construction of LPS-free E. coli K-12 using the Kdo-depleted strain KPM22 L11 [12]. This strain contains deletions of kdsD and gutQ, which encode d-arabinose 5-phosphate isomerases essential for the biosynthesis of Kdo [13,14], and a C:G to T:A transition at position 52 of msbA, which acts as a suppressor of the normally lethal ∆Kdo phenotype [12]. To produce a strain that contains lipid IVA as the only LPS outer membrane component and that cannot revert to synthesize endotoxic derivatives, we sequentially generated mutant strains with unmarked deletions of the genes lpxL, lpxM, pagP, lpxP and eptA. These genes encode enzymes acting downstream of Kdo incorporation into lipid IVA, and are either part of the constitutive pathway of Kdo2-hexaacyl lipid A biosynthesis (the Kdo2-lipid IVA lauroyl-ACP acyltransferase LpxL and the Kdo2-lauroyl-lipid IVA myristoyl-ACP acyltransferase LpxM), modify lipid A with additional acyl chains (the phospholipid:lipid A palmitoyl transferase PagP and the Kdo2-lipid IVA palmitoleoyl-ACP acyltransferase LpxP), or append phosphoethanolamine (P-EtN) under certain conditions (the lipid A P-EtN transferase EptA) [15] (Figure 1). Analysis of the LPS isolated from this strain, designated KPM318, using electrospray-ionization Fourier-transformed ion cyclotron mass spectrometry (ESI FT-ICR) revealed a single primary peak at 1404.85 u, consistent with the structure of the unmodified tetraacyl-1,4′-bisphosphate LPS precursor lipid IVA (Figure 2A).
The successful construction of KPM318 demonstrated the viability of E. coli containing only lipid IVA. However, like other E. coli K-12 strains with partially defective outer membranes such as the ∆Kdo prototype strain KPM22 [11], KPM318 displayed growth defects at temperatures above 40°C. To overcome this, we isolated a series of stable temperature-resistant derivatives of KPM318 capable of growing exponentially at 42°C. KPM335 was the most robust of the spontaneous mutants. Whole-genome sequencing of KPM335 identified a single de novo mutation in comparison to the temperature sensitive KPM318 parent strain, a G:C to T:A transversion at base number 181 of the frr gene (Additional file 1: Table S1). The frr gene encodes an essential ribosome recycling factor which has been described to play multiple cellular roles, such as in disassembly of the post-termination complex [16], prevention of translation errors [17], promoting translational coupling [18], and increasing cell viability through polyamine stimulation of stationary-phase E. coli cultures [19]. The derivation of KPM335 argues strongly for a direct correlation between the development of the frr181 allele and the ability of the strain to tolerate the ∆Kdo phenotype at 42°C. However, elucidation of the underlying mechanism must await further investigations.
ESI FT-ICR analysis of the LPS isolated from KPM335 revealed no significant changes in the LPS composition, with lipid IVA remaining the predominant LPS-related molecule as in the parent KPM318 strain (Figure 2B). However, in contrast to KPM318, spectra of KPM335 displayed a minor peak with a molecular mass of 1484.82 u, consistent with the structure of 1-diphosphate lipid IVA. Touzé and coworkers have shown previously that transfer of a second phosphate group to the 1 position of lipid A is catalyzed by LpxT, an inner membrane protein of the undecaprenyl-pyrophosphate phosphatase family, which is able to phosphorylate Kdo2-lipid IVA
in vitro, but not lipid acceptors lacking Kdo [20]. Thus, the presence of a minor tris-phosphorylated lipid IVA fraction in KPM335 argues against an absolute requirement of LpxT for Kdo-glycosylated lipid A acceptors under in vivo conditions. It remains unclear, however, why phosphorylation of lipid IVA, albeit with a very low efficiency, can occur in KPM335, but apparently not in its KPM318 parent strain.
Apart from the frr181 mutation in KPM335, a total of 12 mutations were specific to both KPM318 and KPM335, including ∆kdsD, ∆gutQ, ∆lpxL, ∆lpxM, ∆pagP, ∆lpxP, ∆eptA and msbA52 as a prerequisite for synthesis of lipid IVA as the predominant LPS-related outer membrane component. For another four mutations, we believe that they arose spontaneously during the generation of mutant strains, namely a silent mutation located in yqiI, two missense mutations in gor and waaY, and a point mutation in the non-coding region upstream of the deleted eptA gene. The latter mutation is most likely a result of KPM274 construction as a donor strain of the ΔeptA::kan cassette. The mutation is located within the sequence of the homology region of primer ECOeptAH1 (Additional file 2: Table S2) and suggests an error in PCR amplification of the kanamycin resistance cassette targeting the eptA gene and ultimately integration into the genome of KPM318. Compared to the reference genome sequence of the common E. coli MG1655 progenitor, the strains BW30270, KPM318 and KPM335 share sequence variations at six locations. Of these, both a silent nucleotide substitution at position 114 and a single nucleotide insertion at base number 253 of ylbE, and deletion of nucleotide 151 within the glpR gene have been described recently as genetic variations in common E. coli MG1655 stock cultures [21]. Furthermore, while E. coli MG1655 has been shown to express a defective ribonuclease PH due to a frameshift by deletion of the first nucleotide of codon 223 of the pseudogene rph-1 [22], insertion of a single nucleotide as the first base of codon 224 of rph-1 predicts reconstitution of RNase PH function and elimination of the polar effect of the rph-1 mutation on the downstream pyrE gene in BW30270, KPM318 and KPM335.
Next, we decided to replicate the unique set of unmarked genomic deletions in the genetic background of the popular E. coli expression strain BL21 (DE3). As a first step towards the construction of an LPS-free E. coli BL21 (DE3) derivative, we replaced the wild-type msbA gene with the msbA148 suppressor allele of strain KPM22 L1 [12]. This allowed the resulting strain, MWB03, to tolerate null mutations in otherwise essential genes of the LPS biosynthesis pathway. We then sequentially deleted the same genes that had been deleted during the creation of KPM318. Whole-genome sequencing of the final strain, KPM404, confirmed the presence of the msbA148 suppressor mutation (Additional file 3: Table S3), and verified the absence of the kdsD, gutQ, lpxL, lpxM, pagP, lpxP and eptA genes. In comparison to the genome sequence of E. coli BL21 (DE3), we further identified a silent mutation in yceJ, three missense changes within the coding sequences of YadG, ECD_00513 and RstA, and a point mutation in the intergenic region between nudC and hemE. Finally, a total of five single nucleotide substitutions were counted in the region between nucleotides 46 and 222 of the basR gene downstream of ∆eptA. These substitutions perfectly matched the sites of basR sequence variations in E. coli B and K-12, indicating that about one third of the BasR-encoding gene of KPM404 was replaced by the corresponding basR sequence of E. coli K-12. Just as for the construction of KPM318, the E. coli K-12 strain KPM274 served as the donor for transfer of the ΔeptA::kan cassette via P1vir transduction to yield strain KPM403, which well explains the generation of a basR hybrid sequence by co-transduction of the ΔeptA::kan cassette and adjacent basR sequences. In fact, as described above, the use of KPM274 as a ΔeptA::kan donor strain also explains why KPM404 carries a point mutation at the same position upstream of the deleted eptA gene as in KPM318.
Mass spectrometric analyses of the LPS profiles from mutants established in the BL21 (DE3) genetic background underscored the need for radical modification of LPS biosynthesis to accomplish synthesis of only lipid IVA in KPM404 (Figures 3 and 4). While disruption of the gutQ, kdsD and lpxL genes in the intermediate mutant strain KPM396 resulted in synthesis of non-glycosylated lipid IVA precursors lacking the secondary lauroyl and myristoyl chains, mass spectrometry revealed a rather heterogeneous mixture of differently modified lipid IVA species. The spectra displayed four prominent peaks with molecular masses of lipid IVA substituted with one P-EtN group (1527.86 u), lipid IVA modified with two P-EtN moieties (1650.87 u), and palmitoylated lipid IVA molecules carrying either one (1766.09 u) or two (1889.10 u) P-EtN residues. Since lipid A palmitoylation seems to be an indication for an adaptive response to aberrant translocation of phospholipids to the outer leaflet of the outer membrane [23], we suspect that PagP-mediated transfer of palmitate to lipid IVA is triggered by perturbations of outer membrane lipid asymmetry in Kdo-depleted strains of the KPM series. As shown for the LPS sample of KPM400, complete loss of the palmitoylated lipid IVA fraction was achieved by deletion of the pagP gene, leaving lipid IVA molecules modified with either one or two P-EtN groups unaffected.
Due to the block in Kdo biosynthesis and the lack of LpxL, Kdo2-lipid IVA and Kdo2-(lauroyl)-lipid IVA, the preferred substrates for LpxP and LpxM, respectively [24-26], cannot be synthesized in strains derived from KPM396. In addition, previous work has shown that expression of LpxP is induced under conditions of cold shock (12°C) to incorporate an unsaturated C16:1 acyl chain at the expense of a laurate (C12:0), perhaps reflecting the demand to adjust membrane fluidity in the cold [25,27]. It was therefore not surprising that deletion of the lpxP and lpxM genes did not exhibit an obvious effect on lipid IVA composition of KPM400 and KPM402, respectively. There is no data indicating that LpxP and LpxM are capable of using lipid IVA as an acceptor substrate. It seems to be quite possible, however, that both enzymes show low levels of activity under specific conditions. Contrary to the proposed physiological role of LpxP in adaptation of E. coli cells to low growth temperatures, limited induction of LpxP expression has been demonstrated as a potential compensatory mechanism even at 30°C in lpxL and lpxL lpxM mutants of E. coli W3110 [28]. Likewise, LpxM was able to transfer a myristoyl chain directly to Kdo2-lipid IVA in E. coli W3110 strains lacking lpxL and lpxL lpxP [28].
Like PagP and LpxP, EptA-dependent modification of lipid A with P-EtN is part of the complex regulatory network associated with the structural redesign of LPS upon exposure to changing environmental conditions or envelope stress factors. As shown for E. coli K-12, modification of lipid A with P-EtN occurs under certain conditions such as in response to external stimuli like ammonium metavanadate [29] or mild acid pH [30]. Although P-EtN appears to be transferred predominantly to the 1-phosphate group of lipid A [31], double P-EtN substitutions at the 1- und 4′-phosphate positions were evident in lipid A of E. coli K-12 lacking LpxT activity [32] and lipid IVA of a mutant strain defective in MsbA-dependent translocation of LPS across the inner membrane [12]. Based upon ESI FT-ICR analysis presented here, deletion of the eptA gene was clearly necessary as well as sufficient to prevent lipid IVA of KPM404 from being substituted with one or two P-EtN residues. Thus, our data not only corroborate previous findings on the ability of EptA to transfer P-EtN to both the 1- and the 4′-phosphate group of lipid A [32], but also provide experimental evidence for its ability to use lipid IVA as a substrate for single and double P-EtN modifications.
In contrast to the E. coli K-12 strain KPM318, integration of lipid IVA into the outer membrane of KPM404 did not result in a temperature-sensitive phenotype of the BL21 (DE3)-derived mutant (data not shown). Although closely related at the genomic level [33], combined analysis of the genomes, transcriptomes, proteomes and phenomes of E. coli K-12 and BL21 (DE3) revealed significant differences in their cellular physiology and metabolism [34], which may explain the differences in the ability of KPM318 and KPM404 to maintain outer membrane integrity in the presence of lipid IVA at temperatures above 40°C.
Biological activity of engineered E. coli cells and LPS
To test the endotoxic potential of the engineered E. coli K-12 and B strains, we performed TLR4/MD-2 activation assays using HEK-Blue hTLR4 cells. Stimulation of these cells, which express human TLR4, MD-2 and CD14 on their surfaces, induces the production of the NF-κB- and activator protein-1 (AP-1)-dependent reporter secreted embryonic alkaline phosphatase (SEAP). The phosphatase levels can be determined by reading the absorbance at 655 nm using a colorimetric substrate. In order to address the question of whether NF-κB is specifically induced via the hTLR4/MD-2 signaling pathway, HEK-Blue Null2 cells, the parental cell line of HEK-Blue hTLR4 cells lacking the hTLR4/MD-2 receptor complex, were used as a control in all hTLR4/MD-2 activation assays. The strains KPM318, KPM335 and KPM404 as analyzed by challenging HEK-Blue hTLR4 cells with an increasing number of colony forming units (cfu) of up to 106 cfu/ml were virtually free of hTLR4/MD-2-stimulating activity, whereas their parental strains BW30270 and BL21 (DE3) elicited a substantial hTLR4/MD-2 activation already at 103 cfu/ml (Figures 5A, B, 6A and B). When the extracted LPS of the strains was subjected to the TLR4-specific assay, we could confirm the lack of endotoxic activity of the samples isolated from KPM318, KPM335 and KPM404 (Figures 5C, D, 6C and D). The data also demonstrated that palmitoylation of lipid IVA (when PagP was expressed) and/or modification of lipid IVA with one or two P-EtN groups (when EptA was expressed) in KPM396, KPM400 and KPM402 are capable of conferring certain hTLR4/MD-2 stimulatory activity on the otherwise endotoxically inactive tetraacylated lipid A precursor (Figure 6). Our results allow us to draw the major conclusion that inactivation of regulated lipid IVA modifications as demonstrated herein to be present in BL21 (DE3)-based intermediate mutant strains is a crucial prerequisite to yield consistently endotoxin-free E. coli strains.
As one of the critical mediators induced in response to endotoxin, we assayed the release of TNF-α upon stimulation of human macrophages by lipid IVA samples of KPM318, KPM335 and KPM404. The samples exhibited very low biological activity as demonstrated by their low capacity to provoke TNF-α production in human macrophages even at concentrations of 0.1-1 μg/ml (Figure 7A). Compared to the LPS from the parental strains, which induced maximal TNF-α release at 0.01 μg/ml, TNF-α induction was reduced by about 80-95% even at 100-fold higher levels of mutant extracts. The well-documented ability of lipid IVA to act as an antagonist for the hTLR4/MD-2 signaling pathway [35,36] prompted us to examine the inhibition of the agonistic activity of S-form LPS from Salmonella enterica subspecies enterica serovar Abortusequi (S. Abortusequi) by lipid IVA samples from KPM318, KPM335 and KPM404. As shown in Figure 7B, pre-exposure of macrophages to 0.1 μg/ml and 1 μg/ml of the lipid IVA extracts resulted in 72.0 ± 11.2% and 75.9 ± 2.0% (mean percent inhibition ± SD) inhibition of TNF-α production induced by LPS from S. Abortusequi, respectively. Thus, lipid IVA from KPM318, KPM335 and KPM404 displayed potent antagonistic activity against biologically active wild-type LPS.
Endotoxin-free expression of ApoA-1 and Hsp70
Previous studies have shown that a BL21 (DE3) ∆lpxM::cat mutant synthesizing a non-myristoylated LPS can be used to express heterologous proteins with reduced stimulatory activity in human LPS-responsive cells [37]. To test the ability of KPM318, KPM335 and KPM404 to serve as hosts for the production of endotoxin-free recombinant proteins, we selected the heterologously expressed human proteins ApoA-1 and Hsp70 as model systems. ApoA-1, a major component of high-density lipoprotein and important mediator in maintaining cholesterol homeostasis [38], is particularly challenging because the 28-kDa protein is known to be directly involved in neutralization of LPS toxicity and, therefore, difficult to separate from endotoxic activity [39,40]. Yet another challenging protein able to associate with LPS is Hsp70 [41]. Moreover, the 70-kDa molecular chaperone has been suggested to function as an endogenous damage-associated molecular pattern protein for activation of the TLR4 signaling pathway upon tissue injury [42]. As Hsp70-induced stimulation of innate immune system cells in many respects resembles the effects of LPS, removal of endotoxin contamination, as frequently present in recombinant Hsp preparations [43], remains a key issue to discriminate between LPS- and Hsp-induced effects.
For production of ApoA-1, we used the T5 promoter-based plasmid pApo404 in E. coli strains BW30270, KPM318 and KPM335, whereas Hsp70 was expressed from pHsp70His under the control of a T7 promoter in BL21 (DE3) and KPM404. SDS-PAGE analysis of the protein samples following minimal purification of each soluble protein extract, by immobilized metal affinity chromatography (IMAC), revealed that the endotoxin-free strains KPM318/pApo404, KPM335/pApo404 and KPM404/pHsp70His produced recombinant ApoA-1 and Hsp70 in approximately equal quantities and had similar impurity profiles as their parental strains, respectively (Figure 8). Since ApoA-1 has been described to also associate with proteins of the host cell [44], it was not surprising to detect relatively high levels of protein contaminants in the IMAC-purified ApoA-1 samples. Regardless, we did not perform any further protein purification or endotoxin removal steps to investigate the biological activity of the ApoA-1 and Hsp70 samples.
The first test of biological activity employed was the Limulus amebocyte lysate (LAL) assay, an FDA-approved method that is based on activation of a coagulation cascade in the LAL by trace amounts of endotoxin [45]. In this assay, the endotoxin unit equivalents determined in ApoA-1 samples from KPM318/pApo404 and KPM335/pApo404 were significantly reduced by 92.7 ± 1.3% and 82.2 ± 3.9% (mean percent inhibition ± SD) compared to those found in the ApoA-1 sample from BW30270/pApo404, respectively (Figure 9). Furthermore, when KPM404/pHsp70His was used as a host for production of Hsp70, the LAL response to the protein decreased by 97.2 ± 0.5% in comparison to the response elicited by Hsp70 obtained from BL21 (DE3)/pHsp70His. The LAL assay, though widely used for the detection and quantitation of endotoxins, is an inappropriate method to discriminate between endotoxically active hexaacylated LPS and endotoxically inactive tetraacylated lipid IVA due to the presence of the LAL-activating 4′-monophosphoryl-diglucosamine backbone in both lipid structures [46,47]. As such, the LAL clotting cascade is activated by a wider spectrum of LPS/lipid A variants than LPS-responsive cells of the human immune system. The residual LAL reactivity of the proteins from the lipid IVA host strains most likely reflects the non-specific nature of the assay, giving rise to false positive endotoxic results.
In order to specifically address the endotoxic activity of the ApoA-1 and Hsp70 samples, we utilized the HEK-Blue hTLR4/MD-2 cell activation assay. The ApoA-1 and Hsp70 samples derived from the endotoxin-free strains did not trigger an endotoxic response in HEK-Blue hTLR4 cells, even when present in the assay at 10 μg/ml, whereas the proteins produced in the parental strains showed substantial NF-κB activation already at concentrations in the range between 0.1 μg/ml and 1 μg/ml (Figure 10). These results were in excellent agreement with the inability of KPM318, KPM335 and KPM404 cells and LPS to stimulate the hTLR4/MD-2 signaling pathway (Figures 5 and 6).
As exemplified herein by heterologous expression of ApoA-1 and Hsp70, proteins prepared from engineered E. coli strains are innately free of endotoxic activity in human LPS-responsive cells, consistent with earlier observations that KPM335 is also a suitable host for production of functional endotoxin-free inclusion bodies of the aggregation-prone fluorescent fusion protein VP1GFP [48]. However, it must be taken into account that lipid IVA may act agonistically in other mammalian hosts such as mouse [49], Chinese hamster [50] or equine [36] cells, which reflects species-specific lipid IVA sensing by the TLR4/MD-2 complex [51,52]. The application of proteins in cells other than human cells may therefore also require lipid IVA depletion. Due to the lack of specificity, the LAL assay cannot be used to assess endotoxic activity but is ideally suited to detect residual lipid IVA in recombinant proteins prepared from LPS-free E. coli strains. We are aware that we currently cannot respond to the question of whether further purification of ApoA1 and Hsp70 would result in complete loss of lipid IVA. However, our data on significantly decreased LAL reactivities of minimally purified ApoA1 and Hsp70 proteins from endotoxin-free bacteria give reason to assume that lipid IVA, being homogeneous, is much easier to remove from downstream products than mature LPS. Despite a common backbone structure, LPS is synthesized as a heterogeneous mixture of structurally related molecules that are decorated with various substituents, usually present in non-stoichiometric amounts [53]. These substitutions, which may vary significantly depending on growth conditions, contribute to considerable physico-chemical heterogeneity of LPS molecules, posing a major challenge to the development of a generally applicable method for endotoxin removal from proteins produced in common E. coli expression strains [5].