|Home | About | Journals | Submit | Contact Us | Français|
Toll-like receptor (TLR)-expressing cells, for the first time, detected and identified a microbial contaminant in a product made in Escherichia coli using an old manufacturing process. It was suspected of having a microbial contaminant(s) because, although it tested negative by standard pyrogen assays, it was associated with adverse events in early clinical trials. The assay readout is the induction of NF-κB and/or cytokines in response to TLR activation. Four coded samples, labeled A to D, including a sample prepared by the older manufacturing process, were submitted. The cell lines were activated only by samples B and D. Sample D stimulated only Mono-Mac 6 and HEK-human TLR4 (hTLR4) cells and was later identified as lipopolysaccharide. Except for TLR3 cells, sample B stimulated cells bearing the different TLRs (TLRs 2, 4, 5, 7, 8, and 9) and nontransfected HEK293 cells. These data suggested that flagellin was the microbial contaminant, since TLR5, the receptor for flagellin, is known to be expressed constitutively on HEK293 cells. Moreover, purified flagellin from Salmonella enterica serovar Typhimurium behaved like sample B, stimulating HEK293 and HEK-hTLR5 cells but not HEK-hTLR3 cells, and this stimulation by flagellin and sample B was blocked by an anti-hTLR5 neutralizing antibody. Western blots showed bands positive for flagellin and sample B with the molecular sizes expected for the flagellins from S. Typhimurium and E. coli, respectively. Mass spectrometry data were consistent with the presence of flagellin in the manufacturer's sample B. Taken together, these data indicate that the microbial contaminant in sample B was flagellin and may have been associated with adverse events when the recombinant product was administered.
Biological products, including cellular and acellular vaccines, cells used in gene therapy, and plasma-derived and recombinant proteins, can become contaminated with many different types of organisms, e.g., gram-positive and gram-negative bacteria, fungi, viruses, parasites, and their by-products, during manufacturing. When the product is administered to a patient, these microbial contaminants may cause unwanted side effects, such as by inducing inflammation due to the release of cytokines or by acting as adjuvants that can potentially enhance the immunogenicity of a therapeutic protein.
Traditionally, biological products are tested during the manufacturing process and at the time of lot release by the in vivo rabbit pyrogen test (RPT) and by the in vitro bacteria endotoxin test, commonly referred to as the Limulus amebocyte lysate (LAL) test. The requirement for final container product testing is stated in the U.S. Code of Federal Regulations, Title 21 (2, 3). These tests are designed to detect lipopolysaccharide (LPS), which is a constituent of gram-negative bacteria, or endotoxin and rely on nonhuman systems to predict human responses. Plasma-derived products and acellular vaccines can be sterile filtered before they are filled, and therefore, intact microorganisms can be removed. However, other microbial constituents, such as those derived from cell walls and nucleic acids (DNA and RNA), can evade filters and still end up in the final product. Recombinant proteins made in Escherichia coli, yeasts, or cell lines may also contain trace levels of host impurities, such that the final product may contain microbial components. These constituents may be difficult to detect by traditional methods.
Recent studies have revealed that microbial pathogens possess specific pathogen-associated molecular patterns (13). The host innate immune system recognizes these pathogen-associated molecular patterns by using germ line-encoded pattern recognition receptors to elicit immune responses. Toll-like receptors (TLRs) are well-known pattern recognition receptors. Cells of the innate immune system utilize TLRs to detect cell wall components of bacteria, mycoplasma, fungi, and protozoa at the cell surface, whereas bacterial and viral nucleic acids are recognized by TLRs in a specialized intracellular endosomal compartment (12, 15, 26).
Recent efforts have focused on the development of an in vitro test system that combines the sensitivity of the LAL test with the wide range of pyrogens detectable by the in vivo RPT. The Interagency Coordinating Committee on the Validation of Alternative Methods (ICCVAM) evaluated the validation status of five in vitro test methods for assessment of the potential pyrogenicity of pharmaceuticals and other products proposed as potential replacements for the in vivo RPT (10). The ICCVAM proposed in vitro pyrogen tests that use enzyme-linked immunosorbent assays (ELISAs) for interleukin-1β (IL-1β) or IL-6 to measure increased levels of cytokine release when human blood cells, i.e., whole blood (WB), isolated monocytes, and a Mono-Mac 6 (MM6) cell line, are stimulated by endotoxin. On the basis of the findings of two published international validation studies, these five proposed in vitro pyrogenicity test methods are WB/IL-1β, WB/IL-6, peripheral blood mononuclear cell/IL-6, MM6/IL-6, and cryopreserved WB/IL-1β (4, 9, 11, 19).
Armed with the knowledge of the critical role of TLRs in microbial detection as well as the need to avoid such issues as donor variability and the hazards associated with human blood products, our approach was to develop an assay that utilized cell lines expressing different TLRs. This was accomplished by using a panel of HEK293 cells transfected with different human TLRs. Initially, a screening test is performed by using the MM6 cell line. Previously, it was shown that MM6 cells respond to ligands to TLR2 and TLR4 (27). On the basis of the data presented in that paper, MM6 cells also respond to the TLR5 ligand flagellin but not to the TLR3, TLR7, TLR8, or TLR9 ligand. If a product sample tests positive with MM6 cells, then cell lines with more restricted TLR expression are used as detector cells to characterize the microbial ligand and, in turn, its microbial origin.
Here we show the utility of this approach for the detection of a microbial contaminant in a sample obtained from a process used to make a recombinant product that had passed the standard lot release testing but that was associated with adverse events in humans. Not only was the panel of TLR-bearing cell lines able to detect a strong proinflammatory signal in the product, but also its use facilitated the identification of the microbial constituent at the molecular level.
Unless otherwise specified, human TLR1 to TLR9 ligands (TLR1L to TLR9L, respectively) were purchased from InvivoGen (San Diego, CA). These included purified lipoteichoic acid (LTA) from Staphylococcus aureus (TLR6/2L); Pam3CSK4 (triacyl lipopeptide, TLR1/2L); poly(I-C) (synthetic double-stranded RNA [dsRNA]; TLR3L); ultrapure LPS from E. coli K-12 (TLR4L); recombinant or purified flagellin from Salmonella enterica serovar Typhimurium (recFLA and flagellin, respectively; TLR5L) (8); imiquimod and gardiquimod (TLR7L); ssRNA40 as a single-stranded RNA (ssRNA) control, single-stranded poly(U) [ss-poly(U)], and E. coli RNA (TLR8L); and ODN2006 (type B), ODN2006 control, ODN2216 (type A), and ODN2216 control (TLR9L) (1, 15, 26). An internationally harmonized reference standard endotoxin (RSE; lot EC-6) containing 10,000 endotoxin units (EU) of E. coli endotoxin per vial (2,000 EU/ml), was obtained from CBER, FDA (Bethesda, MD).
Preliminary dose-response analyses were conducted with LTA at concentrations ranging from 10 ng/ml to 10 μg/ml, RSE at concentrations ranging from 0.125 EU/ml to 2 EU/ml, and flagellin at concentrations ranging from 1 ng/ml to 1 μg/ml. Optimal cytokine-producing conditions were determined with various cell lines and were indicated in each experiment.
InvivoGen's neutralizing immunoglobulin A (IgA) antibodies are chimeric monoclonal antibodies (MAbs) in which the constant domains of the human IgA molecule are combined with murine variable regions. Anti-human TLR5 (anti-hTLR5) IgA is a neutralizing MAb to human TLR5 which blocks flagellin-induced cellular activation. Isotype-matched MAbs, anti-hTLR4 IgA and anti-hTLR2 IgA, are neutralizing MAbs to human TLR4 and human TLR2, respectively. The neutralizing ability of anti-hTLR5-IgA antibodies was tested with HEK-hTLR5 cells by using concentrations ranging from 1 ng/ml to 10 μg/ml.
Unless otherwise stated, cell culture reagents were purchased from Gibco Invitrogen (Carlsbad, CA). Cells of the MM6 human monocytic cell line (28) were purchased from DSMZ (Braunschweig, Germany). MM6 cells were cultured as described previously, with a few modifications (9, 25). Briefly, MM6 cells are maintained in RPMI complete medium. The RPMI complete medium is prepared with RPMI 1640 medium containing 10% heat-inactivated low-pyrogen fetal bovine serum (FBS; HyClone), 2 mM GlutaMAX-I supplement, 0.1 mM minimal essential medium nonessential amino acids, 1 mM sodium pyruvate, 20 mM HEPES, 1% insulin-transferrin-selenium X, penicillin-streptomycin, 1 mM oxaloacetic acid (Sigma-Aldrich, St. Louis, MO), and normocin (InvivoGen). For use in the ELISA procedure, the concentration of FBS was reduced to 2%. After stimulation with various microbial stimuli, the cell supernatants were collected and the level of IL-6 and/or IL-8 production was measured by ELISA.
The HEK293 cell line is a stable cell line derived from primary human embryonic kidney transformed by adenovirus type 5 DNA (7, 16, 21). HEK293 cells were obtained from ATCC and cultured in complete Dulbecco minimal essential medium, a high-glucose medium supplemented with low-endotoxin 10% heat-inactivated FBS, GlutaMAX-I supplement, penicillin, streptomycin, and normocin. HEK-hTLR9 cells were kindly provided by Ken J. Ishii (Osaka University, Osaka, Japan) (24).
The other HEK293 cells used in this study were stably transfected with a plasmid that constitutively expresses one TLR gene and/or TLR-related genes (CD14, MD2) of human origin and were maintained in complete Dulbecco minimal essential medium with selective antibiotics, as described by the manufacturer's instructions (InvivoGen). Briefly, HEK-Blue-2 cells (termed HEK-hTLR2 cells in this paper) and HEK-Blue-4 cells (293-hTLR4/MD2-CD14, HEK-hTLR4) were transfected with TLR2 and TLR4, respectively, and with an NF-κB-inducible alkaline phosphatase (AP) reporter gene system. On interaction with the appropriate ligand, the TLR transduces a signal which results in NF-κB activation and the expression of secreted AP, which can be detected by using Quanti-Blue (a medium used for the detection and quantification of secreted AP; InvivoGen) and which can be quantified by reading the absorbance at 620 to 655 nm by use of an ELISA plate reader. In the case of the other TLR-expressing cell lines, 293-hTLR3 (HEK-hTLR3), 293-hTLR5/CD14 (HEK-hTLR5), 293-hTLR7 (HEK-hTLR7), 293-hTLR8 (HEK-hTLR8), and HEK-hTLR9, IL-8 production was measured as the readout of the TLR-induced NF-κB activation.
For cell stimulation, MM6 cells were cultured at a final concentration of 5 × 105 cells/ml; all HEK cells were cultured at a final concentration of 2.5 × 105 to 106 cells/ml. These cells were cultured for use in cytokine induction assays in a total volume of 250 μl per well (96-well plates) in the presence or the absence of different microbial stimuli. For the neutralization experiments, the cells were pretreated with anti-hTLR antibodies (antibodies blocking TLR5, TLR4, or TLR2) for 30 min to 1 h before the addition of microbial stimuli. Following stimulation, the culture supernatants were collected after overnight incubation (17 to 20 h) at 37°C with 5% CO2 and stored at −20°C until they were assayed. The levels of cytokines IL-6 and IL-8 were measured by using commercial ELISA kits from R&D Systems (Minneapolis, MN). The concentration and absorbance values are expressed as the means of duplicate or triplicate samples (with the standard deviation). All data shown are from reproducible experiments.
Several biologics manufacturers were approached with the request to submit for testing samples that were suspected of being contaminated with bacterial components. One manufacturer identified one such contaminated protein product sample which we describe here. In order for us to test this sample without bias, the manufacturer submitted a blinded test panel which consisted of four samples, labeled samples A, B, C, and D. Among these four samples, two protein product samples, one LPS-positive sample, and one negative control sample were made by the use of two different manufacturing methods, process I and process II. Both protein product samples had passed lot release testing, which consisted of the LAL test, the total residual DNA assay, and a host cell protein ELISA. However, the sample from process I was associated with an adverse event in a human clinical trial. Batches prepared by process I have been withdrawn by the manufacturer. To this panel, we added a phosphate-buffered saline (PBS) control and various TLR ligands as controls, depending on the cell line being tested.
Product samples, recFLA from S. Typhimurium, and LPS preparations from E. coli K-12 were subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) analysis with 10% polyacrylamide minigels (Invitrogen) under denaturing conditions. The separated proteins were electroblotted onto polyvinlyidene difluoride membranes (pore size, 0.45 μm; Invitrogen) and were then incubated with anti-E. coli flagellin MAb (MAb 15D8; 1:1,000; Bioveris, Gaithersburg, MD) (5, 18, 22), followed by incubation with goat anti-mouse IgG (H+L)-horseradish peroxidase conjugate (1:5,000; ZyMax; Invitrogen). Immunodetection was revealed with an ECL Plus chemiluminescent detection system, according to the manufacturer's instructions (GE Healthcare UK Ltd., Buckinghamshire, United Kingdom). Prestained molecular mass markers (range, 3.5 kDa to 260 kDa) were used as references (SeeBlue Plus2; Invitrogen). The relative amount of protein was estimated by densitometry analysis with ImageJ software (National Institutes of Health, Bethesda, MD).
For mass spectrometry (MS) analysis, a sample was prepared by immunoprecipitation with an anti-FliC (flagellin) MAb (BioLegend, San Diego, CA), separation of antigen and antibody, and subsequent trypsin digestion. Liquid chromatography (LC)-tandem MS (LC-MS/MS) analysis was performed by using an Agilent 1200 high-pressure liquid chromatograph equipped with a C18 reverse-phase low-trifluoroacetic acid MS Grace Vydac column (2.1 mm by 150 mm) coupled to a Thermo Fisher Scientific LTQ XL electrospray linear ion trap mass spectrometer equipped for electron transfer dissociation. Peptide assignments were performed with Bioworks software and de novo sequencing (Thermo Fisher Scientific, Waltham, MA).
The results are expressed as the means ± standard deviations. The data were analyzed by the one-tailed unpaired t test at the 95% confidence interval by using Prism (version 5.01) software for Windows (GraphPad Software, San Diego, CA). Differences were considered to be statistically significant when the P values were <0.01 and <0.001.
When IL-6 was used as the readout for the activation of TLR in MM6 cells (Fig. (Fig.1A),1A), LTA (the ligand for TLR2/6), Pam3CSK4 (the ligand for TLR1/2), RSE/LPS (the ligand for TLR4), and flagellin (the ligand for TLR5) induced cytokine levels far above the background levels obtained with PBS. Other known TLR ligands for TLR3 [poly(I-C)], TLR7 (imiquimod, gardiquimod), TLR8 [ssRNA40, ss-poly(U), E. coli RNA], and TLR9 (ODN2006, ODN2216) were nonstimulatory. When IL-8 was used as a readout, similar results were seen (Fig. (Fig.1B).1B). Again, TLR ligands 2, 4, and 5 were stimulatory but TLR ligands 3, 7, 8, and 9 were not.
On the basis of these results, MM6 cells were shown to be capable of detecting microbial components that act via TLRs 2 (complexed with either TLR1 or TLR6), 4, and 5 but not via the other TLRs.
In all assays, RSE (0.5 EU/ml) and PBS were used as positive and negative controls, respectively. The threshold value of 0.5 EU/ml of RSE was chosen on the basis of historical data from RPT. A sample with 0.5 EU/ml may be considered pyrogenic for humans, depending on the dose administered. We analyzed the differences between the levels of RSE- and PBS-induced IL-6 and IL-8 production from eight experiments by a one-tailed paired t test. The t test showed significant results for IL-6 (P = 0.001) and IL-8 (P = 0.0004).
When the set of blinded product samples (detailed in Materials and Methods as samples A, B, C, and D) were added to the MM6 cells, samples B and D elicited significant amounts of IL-6 and IL-8 release compared to the background levels (Fig. 1A and B), whereas samples A and C did not produce a response. The response produced by sample B was considered pyrogenic because it exceeded the 0.5-EU/ml RSE response. Since MM6 cells express multiple TLRs, no deduction as to whether any specific TLR(s) was stimulated or whether a TLR-independent pathway was stimulated by samples B and D can be made. However, this screening test clearly yielded positive results that warranted further testing by the panel of TLR-bearing cell lines.
To determine which specific TLRs were being activated by samples B and D, a panel of TLR-expressing cell lines was used. For the TLR2- or TLR4-transfected HEK293 cells, the assay readout is based on the expression of a reporter gene, that for AP, secondary to activation of the NF-κB pathway. As expected, the known ligand of TLR2, LTA, was associated with a dose-dependent stimulation in the HEK-hTLR2 cells, whereas the TLR4 ligand, RSE, was not stimulatory (Fig. (Fig.2A).2A). When the blinded samples were tested with these cells, they did not produce a response, with one notable exception, sample B, which induced a response similar to that induced by the highest concentration of LTA. This result suggests that sample B was acting via TLR2. The fact that these cells were nonreactive to sample D indicated that it was not a TLR2 ligand but that the MM6 cells were activated via another pathway.
The blinded samples were also tested with HEK-hTLR4 cells (Fig. (Fig.2B).2B). RSE is a known ligand of TLR4 and was stimulatory in a dose-dependent fashion. The TLR2 ligand, LTA was not stimulatory, as expected. Samples A and C had background levels of activation. The results supported the conclusion that sample D was LPS, since it did induce a signal with the HEK-hTLR4 cell line, although not at a pyrogenic level (less than the amount activated by RSE at 0.5 EU/ml). This was confirmed by showing that polymyxin B, an inhibitor of LPS, blocked the stimulation of TLR4 by sample D (data not shown). Sample B was again highly stimulatory with the HEK-hTLR4 cells. Thus, sample B appeared to activate both ligands TLR2 and TLR4. Alternatively, it may stimulate HEK293 cells via a TLR-independent pathway.
The HEK293 cells transfected with the TLR3, TLR5, TLR7, TLR8, and TLR9 gene constructs were stimulated by their respective TLR ligands and responded appropriately (data not shown). Samples A and C were again nonstimulatory in all the cells tested. Sample B activated cells transfected with TLR5, TLR7, TLR8, and TLR9 but not those transfected with TLR3. These results are summarized in Table Table1.1. Of the four blinded samples, sample B remained the prime candidate for having been manufactured by the old manufacturing process, process I, which was associated with adverse events. Sample B's pattern of activation was interesting because it induced signals in all the TLR-expressing cells tested, except for TLR3 cells. This was a major clue to the molecular identity of the stimulus in sample B because it is known that HEK293 cells express constitutive levels of TLR5 (22, 23).
We speculated that HEK-hTLR3 cells expressed low levels of TLR5 or were defective in their functional response to flagellin. In turn, we hypothesized that sample B contained a TLR5 ligand, and we investigated this possibility further. As shown in Fig. Fig.2D,2D, only flagellin and sample B induced the release of IL-8 from HEK293-hTLR5 and other TLR cells (HEK293-nontransfected and TLR 2, 4, 7, 8, and 9 cells; data not shown). As noted above, the HEK293-hTLR3 cell line was responsive only to poly(I-C) and not to flagellin or sample B (Fig. (Fig.2C).2C). The concordant results between flagellin and sample B strongly suggest that flagellin is the active component of sample B.
Flow cytometry experiments revealed that, unlike other TLR cells, TLR3 cells lacked the detectable expression of TLR5 (data not shown). Interestingly, HEK293 cells constitutively expressed TLR3 and responded to poly(I-C); however, dose-response experiments showed that the TLR3- and TLR5-transfected cells were more sensitive to poly(I-C) than other TLR cells (data not shown).
To further determine the identity of the TLR ligand in sample B, sample B and flagellin were added to the HEK293-hTLR5 cell line in the presence or absence of an anti-hTLR5 neutralizing antibody. Figure Figure33 shows the dose-dependent inhibition of both flagellin (Fig. (Fig.3A)3A) and sample B (Fig. (Fig.3B)3B) activation by the anti-hTLR5 antibody. This confirms that sample B contains a TLR5 ligand which is necessary and sufficient to cause the release of IL-8 from HEK293 and MM6 cells. Furthermore, the flagellin- and sample B-induced IL-8 production was not blocked by isotype-matched anti-hTLR2 and anti-hTLR4 antibodies (data not shown).
To determine whether the TLR5 ligand in sample B was indeed flagellin, a Western blot was performed with an anti-flagellin-specific antibody (MAb 15D18) which has been shown to recognize an epitope in the flagellins of flagellated E. coli and S. Typhimurium (5, 18, 22). Figure Figure44 shows a positive band for recombinant flagellin from S. Typhimurium (lanes 1 and 2) and sample B (lane 5) probed with the antiflagellin antibody but no bands in lanes with LPS (lane 3) or sample A (lane 4). Flagellin has a known molecular mass of 40 to 60 kDa, depending on the bacterial species and strains of origin (6, 18, 20). The molecular mass of the flagellin in sample B (lane 5) was slightly less than that of the flagellin used as a control, because they are derived from different bacterial sources, i.e., S. Typhimurium and E. coli, respectively. The relative amount of flagellin was estimated by densitometry to be approximately 12 ng/ml. Thus, the results of the Western blot experiment confirmed the presence of flagellin, a TLR5 ligand, in sample B, which agrees with the stimulation pattern observed with the TLR-bearing cell lines.
Confirmation of the presence of flagellin was performed by immunoprecipitation with an antiflagellin antibody, digestion with trypsin, and LC-MS/MS analysis. The observed peptide sequences are provided in Table Table22 and are mapped onto the E. coli flagellin sequence in Fig. Fig.5.5. The collision-induced dissociation tandem mass spectrum of a unique E. coli flagellin signature peptide, T27-28, detected in the tryptic digest is shown in Fig. Fig.66.
From our experience with this investigation, we would like to propose the following testing algorithm for the detection of microbial proinflammatory components in plasma-derived and recombinant biological products (see the flowchart in Fig. Fig.7).7). Lot release samples that tested negative by standard tests (the LAL test or RPT) but that are associated with adverse events could be tested by the use of MM6 and TLR-expressing cells. Testing with nontransfected HEK293 cells and cells expressing TLR2 and TLR4 (panel 1) is used to identify bacterial, fungal components, and mycoplasma, whereas testing with nontransfected HEK293 and cells expressing TLRs 7, 8, and 9 (panel 2) is used to detect viral, fungal, and bacterial nucleic acids. If all the cells from panels 1 and 2 are positive, the ligands suspected are dsRNA or flagellin. Testing with cells expressing TLR3 and TLR5 can distinguish between these ligands, as shown in the algorithm (Fig. (Fig.7).7). Depending on the results, the causative agent could be identified as flagellin (TLR5); LPS (TLR4); fungal, mycoplasma, or gram-positive bacterial components (TLR2); or nucleic acids from viral, bacterial, or fungal sources. The presence of LPS and flagellin can be confirmed by blocking stimulation with polymyxin B and anti-TLR5 antibody, respectively. The resolution of suspected fungal or gram-positive bacterial components is more complex. Fungi possess ligands that can stimulate TLR2 (zymosan) and TLR4 (mannans), whereas gram-positive bacteria lack LPS and do not have ligands that stimulate TLR4. Confirmation would require going back to the batch records and checking prefiltered samples for fungal and bacterial growth by standard culture methods. PCR methods can be used to confirm the presence of viral, fungal, and bacterial nucleic acids.
Since the first mammalian TLR (now known as TLR4) was identified by Charles Janeway and colleagues (17), 10 human TLRs and 13 murine TLRs have been identified (14). TLRs 1, 2, 4, 5, and 6 are expressed on the cell surface, while TLRs 3, 7, 8, and 9 are found in intracellular compartments, such as endosomes and lysosomes (15).
We reasoned that the TLR system, which has evolved to recognize microbes, could be exploited to detect and identify residual microbial components in plasma-derived or recombinant biologic products more efficiently than available tests which utilize nonhuman systems. To achieve this objective, we started with a human cell line, MM6, the cells of which express multiple TLRs. If they react to the product by secreting a cytokine (IL-6 or IL-8), further testing is performed with human HEK293 cells that have been transfected with a single TLR or multiple other TLR-related genes (MD2, CD14). The pattern of response by the different TLRs can help deduce the identity of the microbial contaminant. This can be valuable information for manufacturers, which can apply this during process development.
To determine whether this system had any utility for detecting microbial contaminants, we asked a manufacturer for a set of blinded samples, one of which had tested negative for bacterial contaminants by standard testing as well as ELISA testing for host cell proteins in nonhuman systems. This product sample was made from an old purification process, process I. The sample made by process I was a highly purified recombinant protein that was derived from E. coli and that had been sterile filtered. Similarly, a sample from the improved purification process, process II, was included as a blinded sample. The remaining samples consisted of PBS as a negative control and diluted LPS as a positive control. The four samples were added to a human cell line, MM6, the cells of which express multiple TLRs. Only samples B and D induced IL-6 and IL-8 cytokine responses at levels above the background level in MM6 cells.
When the product samples were tested with the panel of HEK293 cells transfected with human TLRs 2, 3, 4, 5, 7, 8, and 9 (Fig. (Fig.22 and Table Table1),1), only sample D stimulated HEK-hTLR4 cells (Fig. (Fig.2B)2B) and was inhibited by polymyxin B. Since LPS is the known ligand for TLR4, it was simple to deduce that this sample contained LPS. Since LPS should be detected by standard LAL testing, it was unlikely that this was the implicated product sample but was probably added to the sample set by the manufacturer as a positive control. Samples A and C were nonstimulatory with all the cells tested. Upon breaking of the code, sample C was listed as a negative control containing PBS. Sample A was revealed to be the sample made by process II, the manufacturing process that was improved to further reduce the possibility of contamination. Sample D was listed as a diluted LPS (0.33 EU/ml).
The response pattern that emerged with sample B was complex, since it stimulated all the TLR-bearing cells except TLR3. This implied that sample B contained multiple TLR ligands and/or stimulated HEK293 cells by TLR-independent pathways. After the information about TLRs and these cell lines was reviewed, it became apparent that HEK293 cells express TLR5 constitutively; therefore, it is possible that all of the cell lines tested, except the TLR3 cell line, were triggered by flagellin, a known TLR5 ligand that was present in sample B. Indeed, as little as 1 to 10 ng/ml of flagellin was able to induce IL-8 production from HEK293 cells and even induced a higher degree of IL-8 production from HEK-hTLR5 cells, which overexpress TLR5. The HEK-hTLR3 cell line that we obtained was not responsive to flagellin (Fig. (Fig.2C)2C) and was found not to express detectable TLR5 by flow cytometry (data not shown).
To further test the possibility that sample B contained flagellin, we showed that anti-hTLR5 antibody (Fig. (Fig.3)3) but neither anti-hTLR2 nor anti-hTLR4 antibody could block the flagellin- and sample B-induced IL-8 production in various cell lines that express TLR5. Moreover, sample B was shown to contain flagellin when it was stained with antiflagellin antibody by Western blot analysis (Fig. (Fig.4).4). This confirmed the results obtained with TLR-bearing cell lines. In addition, the presence of flagellin in sample B was confirmed by immunoprecipitation and MS analysis. That analysis revealed flagellin sequences, including a peptide unique to the flagellin from E. coli.
In conclusion, we have developed a testing algorithm using a cell-based approach to identify subcellular microbial contaminants. Employing this algorithm, we have shown that a panel of TLR-expressing human cell lines is able to detect a bacterial contaminant missed by standard product testing procedures. Furthermore, the results with the cell lines enabled the molecular nature of the contaminant to be identified. This testing algorithm may aid manufacturers with identifying and tracking the source of the contamination in one or more of the manufacturing steps and, in turn, with devising methods that may be used to avoid such contamination in the future. More importantly, tests based on human cells are more likely than tests based on nonhuman systems to detect microbial contaminants that can cause adverse effects in humans.
We are grateful to Hana Golding and Maria Luisa Virata-Theimer for review of the manuscript, Ewa Marszal for mass spectrometry, and Jessica Kim for advice with statistical analyses.
This project was supported in part by DHHS NVPO funds for 2007-2008 (to L.-Y. Huang and B. Golding).
The findings and conclusions in this article have not been formally disseminated by the Food and Drug Administration and should not be construed to represent any agency determination or policy.
Published ahead of print on 2 September 2009.