Optimising the use of commercial LAL assays for the analysis of endotoxin contamination in metal colloids and metal oxide nanoparticles.

Abstract Engineered nanoparticles (NP) are generally contaminated by bacterial endotoxin, a ubiquitous bacterial molecule with significant toxic and inflammatory effects. The presence of endotoxin, if not recognised, can be responsible for many of the in vitro and in vivo effects attributed to NPs. The Limulus Amoebocyte Lysate (LAL) assay, the test requested by regulatory authorities for assessing endotoxin contamination in products for human use, is not immediately applicable for testing endotoxin in NP preparations, mainly due to the possible interference of NPs with the assay readouts and components. In this study, we have compared different commercially available LAL assays for detecting endotoxin in gold, silver and iron oxide NPs. Different NP chemistry, concentrations and surface coatings could differently interfere with the LAL assays’ results. After accurate testing of the possible interaction/interference of NPs with the various assay components, the modified chromogenic LAL assay proved the most suitable assay for measuring endotoxin in NP samples, provided the appropriate controls are performed. Thus, endotoxin determination can be performed in NP preparation with commercial LAL assays only after assay validation, i.e. once possible interference of NPs with the assay components and readouts has been excluded.


Introduction
A vast number of studies in the last years has dealt with the problem of the possible adverse effects of engineered nanoparticles (NPs) on human health, mostly based on in vivo exposure of animal models or in vitro exposure of relevant cells (Geraets et al., 2012;Yazdi et al., 2010). Besides overt toxic effects that lead to cell death, it is particularly important to address the possibility that NPs alter the normal cellular functions at the molecular level. In this regard, it is of major interest how NPs interact with immune effectors and functions. Innate immunity/ inflammation is an immediate defence reaction able to identify and destroy potentially dangerous foreign agents, including nanoparticulate entities (Boraschi & Duschl, 2012;Hussain et al., 2012;Medzhitov, 2001). If not properly controlled, an excessive innate/inflammatory reaction may lead to tissue damage and pathology (e.g. inflammatory diseases). Studies showing the ability of NPs to trigger inflammation have been published using different experimental models with different types of nanomaterials, suggesting a possible risk for human health (Gosens et al., 2010;Hussain et al., 2009;Yazdi et al., 2010). Since the innate immune system is exquisitely able to recognise and react to foreign agents, in particular microbial products, it is of utmost importance, when evaluating the inflammatory/innate effects of NPs, both in vivo and in vitro, to exclude that these effects may arise from unrecognised contamination of the NPs with bacterial products such as endotoxin (Lieder et al., 2013;Malyala & Singh, 2008;Morrison & Ulevitch, 1978).
Gram-negative endotoxin (lipopolysaccharide, LPS) binds to the TLR4 receptor on mammalian cells and triggers the gene upregulation, synthesis and secretion of several inflammatory factors, including the classical inflammatory cytokines IL-1b, TNF-a and IL-6, thereby initiating a complex cascade of sequential events that constitute the inflammatory reaction and immune activation. As little as 0.1 EU/ml of endotoxin can significantly upregulate inflammatory gene expression in human monocytes (Oostingh et al., 2011), whilst excessive LPS stimulation causes deregulated inflammation and toxicity.
The effects of endotoxin exposure in humans have been well documented. The maximum permissible endotoxin levels in pharmaceuticals and in clinical procedures have also been well regulated by the United States Food and Drug Administration (FDA) in 1985. Endotoxin acquired from environmental exposure can be associated with respiratory complications such as pulmonary inflammation and asthma (Liu, 2002;Perros et al., 2011). High circulating levels of endotoxin (endotoxemia) as it may occur in some systemic infections, cause fever, hypotensive shock, impaired organ function and eventually multiple organ failure and death (Danner et al., 1991).
LPS is a heat-stable ubiquitous contaminant, usually present in all chemicals and glassware and resistant to sterilisation procedures. Therefore, NP preparations are therefore likely contaminated with LPS, unless endotoxin-free synthesis is performed (Vallhov et al., 2006). In this context, it is of major importance to have reliable methods for monitoring the presence of endotoxin in NP preparations. These methods would allow qualitative control of NP synthesis procedures and to distinguish NP effects from those of contaminating endotoxin. Batch-to-batch variability in endotoxin contamination of the same type of NPs, even when produced by the same lab, may be one of the causes for huge variability and conflicting reports regarding toxic and inflammatory effects of NPs (Oostingh et al., 2011).
Different methods are currently used for measuring endotoxin. The original FDA-approved rabbit pyrogen test (RPT) was replaced in 1973 with the Limulus Amoebocyte Lysate (LAL) assay, which provides faster and more sensitive endotoxin assessment. The FDA issued its ''Guideline on Validation of the Limulus Amebocyte Lysate Test as an End-product Test for Human andAnimal Parenteral Drugs, Biological Products, andMedical Devices'' in 1987 (FDA, 1987). In 2000, the US and the European Pharmacopoeias have issued a harmonised document for LAL testing. In July 2011, the FDA has withdrawn its guideline on LAL testing, as it did not any longer reflect the Agency's current thinking on the topic. Drug manufacturers should now refer to the US Pharmacopeia General Chapter 85 ''Bacterial Endotoxins Test'', which provides information on the performance and acceptance criteria for LAL-based endotoxin testing (USP, 2005). The regulatory agencies of other countries, such as the European Medicines Agency (EMA), accepted these rules and criteria for endotoxin testing. Thus, the methods for testing endotoxin are well established, officially included in the regulatory guidelines, and routinely applied worldwide. The European Centre for the Validation of Alternative Methods (ECVAM) also approved various highly sensitive and reliable bioassays for assessing pyrogens, such as the human PBMC activation assay and the human monocyte activation test. However, these bioassays are not specific for endotoxin, because any pyrogen (putatively including nanomaterials) can trigger the same biological effects (e.g. production of inflammatory cytokines). Furthermore, reports have shown that some NPs can trigger the TLR4 signalling pathway, similarly to LPS (Bastús et al., 2009;Qu et al., 2013). Others reported that NPs could inhibit the LPS-induced inflammatory effect in macrophages . Therefore, for specific detection of endotoxin in nanomaterials, the LAL assay is recommended, as opposed to bioassays based on PBMC or monocytes, because it could discriminate endotoxin from intrinsic nano bio-effects. The ISO 29701:2010 regulation ''Nanotechnologies-Endotoxin test on nanomaterial samples for in vitro systems'' (ISO, 2010) aims at providing guidelines for the use of the LAL assay for measuring endotoxin in NP preparations. This regulation mentions the possibility that NPs can interfere with the LAL assay optical readouts and recommends the use of appropriate controls. Despite this warning, there is no information available on potential interference of NPs with the LAL assay components and readouts, causing uncertainty about the validity of the endotoxin measurements. Assessing NP interference with the LAL assay components and readouts is therefore a necessary requirement for the reliable evaluation of endotoxin contamination in NP preparations (Dobrovolskaia et al., 2009(Dobrovolskaia et al., , 2010Neun & Dobrovolskaia, 2011). Very few studies have addressed this issue. Smulders et al. compared different endotoxin detection assays for an accurate evaluation of LPS contamination in NPs, but they did not examine the possible interference of the NPs on the assay readouts (Smulders et al., 2012).
In this study, we have investigated the interference of gold, silver and iron oxide NPs with commercial LAL assays. Gold and iron oxide NPs have wide use in biomedicine and many other applications (gold for drug and gene delivery, photovoltaics, sensors, conductors, dyes, etc.; iron oxide in contrast agents, dyes, hard drives, catalysis, sensors, etc.; Liu et al., 2013;Sperling et al., 2008). Silver NPs are widely applied in photovoltaics, biological and chemical sensors and for antimicrobial coatings of textiles, wound dressings, and biomedical devices (Kim et al., 2007). In addition to their extensive use, we have also chosen gold NPs (metal NPs) as they are chemically stable, iron oxide NPs (metal oxide NPs) because they are chemically unstable and silver NPs because they are a model of both metal and metal oxide NPs, due to their plasmonic and catalytic properties, respectively. We have compared different commercial LAL assays, based on different methodological principles, in a highly controlled series of tests, in order to assess their ability to reliably detect endotoxin in NP batches.

Nanoparticle synthesis
The metal and metal oxide NPs used in this study (Au, Ag and Fe 3 O 4 ) were synthesised by wet chemistry methods under conditions that enabled the collection of stable and narrowly dispersed NPs (Figure 1). Synthesis was carried out starting with metallic salt precursors either decomposed or reduced in the presence of stabilisers. Reagents were purchased from Sigma-Aldrich (St. Louis, MO) and used as received. In detail, citrate coated 10 nm Au NPs were obtained based on Turkevich et al. (1955), i.e. by the fast addition of 1 ml hydrogen tetrachloroaureate (HAuCl 4 , 0.025 M) to 150 ml of boiling trisodium citrate (SC, 2.2 Â 10 À3 M) under vigorous stirring. After 3-5 min, NPs developed a characteristic red colour of the colloidal gold and the solution was cooled down to room temperature (RT). The resulting NPs were loosely coated with the negatively charged citrate ions acting as stabilisers. Citrate coated 14 nm Ag NPs were prepared based on Jana et al. (2001), by reduction of silver nitrate (AgNO 3 ) with sodium borohydride (NaBH 4 ) in the presence of SC. Practically, 1 ml of ice-cold freshly prepared NaBH 4 (0.1 M) was rapidly added to 50 ml of an aqueous solution containing AgNO 3 (2.5 Â 10 À4 M) and SC (2.5 Â 10 À4 M), under vigorous stirring. The appearance of a yellow colour in the solution, after adding NaBH 4 , indicates particle formation. In this synthesis, citrate serves only as a capping agent since it cannot reduce the silver salt at RT. Ag NPs coated with 11-mercaptoundecanoic acid (MUA) and 11-amino-1-undecanethiol (AUT) were obtained through ligand exchange of previously synthesised citrate-coated Ag NPs, thus obtaining cationic and anionic Ag NPs functionalised with two identical alkanethiolated molecules terminating in either positive (NH þ 3 ) or negative (COO -) charged groups at physiological pH. Ammonium coated 7 nm Fe 3 O 4 NPs were synthesised following Massart's method (Jolivet et al., 1983;Massart, 1981). Amounts of 1 mmol iron (II) chloride (FeCl 2 ) and 2 mmol iron (III) chloride (FeCl 3 ) were dissolved in 10 ml of deoxygenated water and then added dropwise to 10 ml of deoxygenated tetramethylammonium hydroxide (TMAOH, 1 M). After 30 min of vigorous stirring under a N 2 stream, the Fe 3 O 4 precipitate was washed by soft magnetic decantation, re-dissolved in TMAOH 0.1 M and diluted 100 times to obtain the final stable colloidal solution.
Two sets of solvents were prepared and used as controls. Solvents are intended as the aqueous solution in which NPs are dispersed. The first set was identical to the synthesis protocol except for the absence of precursor reagents, whilst the second set was obtained by removing NPs from the solution by high-speed centrifugation. It is worth noting that molecules present in the solvents of this study, i.e. citrate ions for Au NPs and Ag NPs and TMAOH ions for Fe 3 O 4 NPs, were previously proved biocompatible (Pfaller et al., 2009). The solvents are required for the stability of the colloids, and therefore their presence in the NP formulation cannot be avoided.

Transmission electron microscopy
Transmission electron microscopic images were acquired with a JEOL 1010 electron microscope operating at an accelerating voltage of 80 kV. Samples for TEM were prepared by drop casting on carbon-coated cooper TEM grids. Grids were dried at RT. Different areas of the grid were observed with different magnifications, and more than 400 particles were computer-analysed and measured for the size distribution.

Z-Potential measurement and dynamic light scattering
Measurements were made with a Malvern ZetaSizer Nano ZS operating at a light source wavelength of 532 nm and a fixed scattering angle of 173 , on 0.8 ml aliquots of the colloidal NP suspensions. The software was set with the specific parameters of refractive index and absorption coefficient of NP material and solvent viscosity. This was required to obtain the correct value for each type of NP. Z-Potential (surface charge) measurements are commonly used to determine the stability of a colloidal suspension of electrostatically stabilised NPs. Conversely, DLS allows the determination of the hydrodynamic diameter of colloidal particles and conjugates, i.e. the diameter of the sphere with the same Brownian motion as the particle under analysis.

UV-visible spectrophotometry
UV-Vis spectra were acquired with a Shimadzu UV-2400 spectrophotometer, for 1 ml aliquots of the NP suspension, and performed the spectral analysis in the 300-800 nm range. This technique is widely used for metallic NPs that exhibit a characteristic absorbance maximum in the visible range (the Surface Plasmon Resonance, SPR), which changes depending on the size and surface alterations. However, all the materials used absorbed light in the visible or UV range, making these measurements appropriate in all cases.

ICP-MS to determine the free ions in NP solution
The release of free ions by NPs was assessed by inductive coupled plasma mass spectrometry (ICP-MS). NP stocks were centrifuged, and the supernatants were collected and examined for ion content. Free ions in solution are expressed in absolute values (mg/l) and in relative values (%) compared with the total amount of precursor reagents added initially to the synthesis.

LAL assays
Three different LAL assays were compared in this study: the traditional chromogenic LAL assay (QCL-1000Ô, cat. no. 50-647U, Lonza Group Ltd., Basel, Switzerland), the new chromogenic assay (Pyrochrome Õ , cat. no. CD 060, Associates of Cape Cod, Inc., East Falmouth, MA), with and without Glucashield Buffer and the fluorescent assay (PyroGeneÔ Recombinant Factor C Endotoxin Detection Assay, cat. no. 50-658U, Lonza). The endotoxin standard curve was run by preparing a fresh standard curve from the endotoxin stock for each test with twofold dilutions (in duplicate).
''Spiking'' controls, meant to identify the ability of the samples to inhibit/enhance the detection of the endotoxin in the assay, consisted of NP dilutions to which a known amount (0.5 EU/ml) of standard endotoxin was added. NP and solvent samples were serially diluted from the stock suspensions with LAL Reagent Water and distributed in endotoxin-free microplates (25 ml/well for chromogenic assays; 50 ml/well for the fluorogenic assay). Standard endotoxin was prepared at 1 EU/ml and added to each well in an equal volume (25 or 50 ml/well). Duplicate wells were set up for each NP or solvent dilution. Endotoxin evaluation was then performed following the manufacturers' instructions. Each experiment was repeated twice. As a preliminary indication of the interference of samples with the LAL assay components, we have calculated the recovery rate, i.e. the amount of endotoxin measurable in the spiked sample relative to a known input spike. The recovery rate was calculated with the following formula.
It should be noted that according to regulatory guidelines, the recovery rate is considered acceptable when in a range between 50 and 200% of the endotoxin spike.
The traditional chromogenic LAL assay is based on the detection of the endotoxin-stimulated LAL end-product 4-nitroaniline (pNA) at 405 nm. The new chromogenic assay was used in its modified version, with readout shifted from 405 to 540 nm. Briefly, after sample incubation with enzyme and substrate, the diazo reagents (provided by the kit) were added sequentially: 6 mM sodium nitrite in 0.48 N HCl (reagent 1), 26.3 mM ammonium sulfamate in water (reagent 2) and 3.76 mM N-(1-naphthyl)ethylenediamine dihydrochloride in water (reagent 3). The reagents modify pNA to turn from yellow to deep purple, thus allowing detection of the azo dye product at a wavelength of 540 nm.

Measurement of interference with the LAL readout products at 405 and 540 nm
Twofold dilutions of 4-nitroaniline (pNA, Sigma-Aldrich) were distributed in flat-bottomed 96-well plates in a volume of 50 ml/ well. For each pNA dilution, different concentrations of NPs and corresponding solvents were added in 50 ml aliquots in triplicate wells and mixed. Volumes selected correspond to those used in the LAL assays. For measuring interference at 405 nm, another 100 ml of water (in place of the substrate) and 100 ml of stop solution (sodium dodecylsulfate solution) were added to bring the final volume to 300 ml, i.e. the same volume as in the QCL-1000 LAL assay. Optical density was measured with a microplate reader at 405 nm. For measuring interference at 540 nm, diazo reagents were added to the wells containing pNA and NPs in rapid sequence (1, 2 and 3), and the optical density was immediately measured with a microplate reader at 540 nm. All chemicals were supplied by Sigma-Aldrich.

Measurement of interference with the LAL readout fluorescent product
The reagents of the PyroGeneÔ assay were activated with endotoxin (100 ml of a concentration of 0.5-1 EU/ml of standard endotoxin for 1 h), to generate the fluorescent product. Then 100 ml of different concentrations of NPs and solvents were added to the wells, and fluorescence was immediately detected at excitation/emission wavelengths of 380/440 nm. Each sample was tested in duplicate.

Measurement the catalytic properties of NPs on LAL substrate
The substrate of the Pyrochrome LAL assay, Boc-Leu-Gly-Arg-pnitroanilide (Tianjin YongSheng Biotechnology Co. Ltd., Tianjin, China) was prepared in 0.1 M Tris-HCl (pH 8.0) and 25 mM MgCl 2 as described (Harada-Suzuki et al., 1982). Threefold dilutions of the substrate (ranging from 0.01 to 1 mM) were distributed in flat-bottomed 96-well plates in a volume of 50 ml/well. For each substrate dilution, 5 nM Au NPs, 0.4 nM Ag NPs and 10 nM Fe 3 O 4 NPs (highest concentration without readout interference) were added in 50 ml aliquots in triplicate wells, mixed and incubated at 37 C for 30 min. Selected volumes and time-points correspond to those used in the LAL assay. Optical density was measured with a microplate reader at 405 nm. Diazo reagents were added, optical density measured again at 540 nm.

ECVAM PBMC-based pyrogen test
Human peripheral blood mononuclear cells (PBMC) were isolated by density gradient centrifugation using Ficoll-Paque PLUS (GE Healthcare, Bio-Sciences AB, Uppsala, Sweden). PBMC were cultured at a density of 1 Â 10 5 cells/well in 96-well culture plates in 200 ml RPMI 1640 + Glutamax-I medium (GIBCO Õ by Life Technologies, Paisley, UK) supplemented with 1% of heat-inactivated human AB serum at 37 C in moist air with 5% CO 2 . Fifty ml/well of LPS (from E. coli 0113:H10; Associates of Cape Cod, Inc., East Falmouth, MA) and different concentrations of NPs were added to the cells. After 16 h, supernatants were collected and centrifuged, and the production of IL-6 was assessed by ELISA (R&D Systems, Minneapolis, MN).

Statistical analysis
Data are presented as mean ± SD of technical replicates and/or replicate assays, as indicated. Each assay was repeated 2-4 times. Statistical significance was calculated by Student's t-test.

Characteristics of NPs
Three metal and metal oxide NPs were used in this study, such as Au NPs, Ag NPs (with three different coatings) and Fe 3 O 4 NPs. The physico-chemical characteristics, concentration, surface coatings and solvents in which NPs are suspended are reported in Figure 1 and Table 1. Synthetic procedures are described in the ''Methods'' section. The release of free ions from NPs was measured by ICP-MS, and found negligible in Au and Ag NPs (sodium citrate coated), whereas about 9% iron was released from iron oxide NPs (Supplemental Table S1). In this study, NP concentrations are shown as molarity of NPs. Correspondence with other units of measurement (mass, surface area and NPs number) could be found in Table 1.

Selection of LAL assays
After evaluation of the characteristics and performance of the commercially available LAL assays (Supplemental Table S2), and based on the data already available for endotoxin evaluation in NPs (see below), two LAL assays have been selected for experimental evaluation in this study, the end-point chromogenic assay and the fluorescent assay. The endpoint chromogenic assay can be performed easily with no need for specific equipment or software. Furthermore, the assay readout can be detected at two different wavelengths (405 and 540 nm). Therefore, if the samples interfere with the readout at one wavelength, the other detection wavelength could be used. Among the several commercial endpoint chromogenic assays available, we have chosen one for detection at 405 nm, and another one for detection at 540 nm (Table 2). We have excluded the kinetic chromogenic assay because of the need for expensive equipment and software that would necessarily limit its general use. The fluorescent assay is highly specific for endotoxin, as it is not affected by b-glucan interference, which on the other hand is recognised similarly to endotoxin in the chromogenic assay. In addition, the fluorescent assay should not be prone to interference by non-fluorescent NPs.
The gel-clot and the turbidimetric LAL assays were not included in this study for different reasons. The gel-clot LAL assay is generally of low sensitivity (although the newest commercial assays have improved sensitivity) and is not quantitative. Furthermore, several publications have already shown that NPs can significantly interfere with the gel-clot-based detection of endotoxin (Dobrovolskaia et al., 2010(Dobrovolskaia et al., , 2013Kucki et al., 2013;Smulders et al., 2012). We have also excluded the turbidimetric assay because it needs specific equipment, a requirement that will limit its use, and because the assay readout is based on light scattering, a method that it generally avoided when using NPs. In fact, objects larger than 50 nm are able to scatter light, and samples containing objects larger than 200 nm become visibly turbid. Thus, many colloidal NP suspensions and NP aggregates that fall into this range are not compatible with the turbidimetric assay for endotoxin testing (Bastús et al., 2011;Gumprecht & Sliepcevich, 1953). Moreover, the interference of NPs with the coagulation process induced by LPS or with the coagulated proteins still needs to be addressed.
The traditional endpoint chromogenic LAL assay uses a synthetic chromogenic peptide substrate that is cleaved by the clotting enzyme contained in the LAL reagent, resulting in the release of a product that exhibits a yellow colour (4-nitroaniline, pNA, detected at an OD at 405 nm). The clotting enzyme is activated not only by endotoxin, through activation of Factor C, but also by b-glucan, through activation of Factor G. All these factors are present in the LAL reagent. In the case of samples optically responsive at 405 nm, the chromogenic assay can be modified by the addition of diazo reagents that shifts the readout wavelength from 405 to 540 nm (pink). In this study, we will define these two variants as traditional and modified chromogenic assay, respectively. The principle of the fluorescent LAL assay is the activation by LPS of recombinant Factor C, which in turn cleaves a fluorogenic substrate liberating fluorescence. Thus, the fluorogenic assay is specific for endotoxin, with no crossactivation by b-glucan.
We planned the study as follows: (1) Evaluation of interference of NPs with the assays' readouts.
Different NP concentrations were mixed with different concentrations of the assays' readout products (pNA at 405 nm, azo dye at 540 nm, fluorescence). Only NP concentrations that do not affect the readouts can be used in the LAL assays.
(2) Evaluation of the recovery rate. Non-interfering NP concentrations were tested for their interference with measurement of a known amount of endotoxin in the different LAL assays.
Only NP concentrations that do not affect endotoxin measurement (i.e. do not interfere with the assay components) can be considered for reliable evaluation of endotoxin contamination.
(3) Evaluation of other possible interfering factors. In the specific case of the modified chromogenic assay, for instance, we have evaluated the possible interference by solvents, the possible co-contamination with b-glucan, and the possibility that NPs may have direct catalytic activity on the assay substrate.

NP interference with the optical readout at 405 nm
Application of the LAL assay to endotoxin detection in nanomaterials has been recently addressed in a few recent publications (Dobrovolskaia et al., 2009(Dobrovolskaia et al., , 2010Neun & Dobrovolskaia, 2011;Smulders et al., 2012). Our study intended to improve the accuracy of endotoxin detection in NPs by examining all the possible levels of non-specific interference that NPs may have on the assay parameters. We did therefore started by evaluating the NP effects on optical density readouts, a parameter that has never been appropriately considered. As a first step, the interference of NPs with the assay readout at 405 nm has been evaluated, using commercial pNA to represent the pNA liberated from the substrate by LPS-induced enzymatic cleavage in the LAL assay. Scalar concentrations of pNA were set up, to encompass the entire range of pNA released in the  Figure 2, the dose-dependent increase of optical density of pNA alone is not affected by the presence of Au NPs even at the highest concentration tested (Figure 2A), in agreement with the fact that Au NPs are optically unresponsive at 405 nm ( Figure 1D and Supplemental Figure S2A). Similar results were obtained for Fe 3 O 4 NPs (Supplemental Figure S1A), despite the fact that these NPs are optically responsive at 405 nm ( Figure 1D and Supplemental Figure S2G). On the other hand, the addition of Ag NPs did significantly increase the pNA readout at 405 nm ( Figure 2B), in agreement with the fact that Ag NPs are optically responsive at 405 nm ( Figure 1D and Supplemental Figure S2D). This suggests that the traditional chromogenic LAL assay would overestimate the presence of endotoxin in Ag NP preparations, when Ag NPs are tested at concentrations above 0.4 nM, whilst Fe 3 O 4 NPs cannot be tested above 10 nM. Based on these results, we conclude that the traditional LAL chromogenic assay can be used for testing endotoxin contamination of Ag NPs only when at low concentrations (50.4 nM), whilst it can be applied to Au NPs and Fe 3 O 4 NPs concentrated up to 5-10 nM.

NP interference with the optical readout at 540 nm
We have also evaluated the interference of NPs with the chromogenic assay readout at 540 nm. The optical readings at 540 nm of the azo dye generated by pNA + diazo reagents (i.e., the product of the new modified chromogenic assay) is essentially unaffected by the presence of increasing Figure 2. NP interference with the LAL assays' readouts. Interference by Au NPs (panels A, C and E) and by Ag-SC NPs (panels B, D and F) on the reading of pNA at 405 nm (panels A and B), azo dye (obtained by adding diazo reagents to pNA) at 540 nm (panels C and D) and fluorescence at excitation/emission 380/440 nm. In panels E and F, the dotted lines are the positive fluorescence control (generated by 1 EU/ml endotoxin in the fluorogenic assay), and shaded areas represent the SD above and below the mean. Data are the mean ± SD of triplicate determination within one test representative of three performed. *p50.05 versus control; **p50.01 versus control.
concentrations of Au NPs ( Figure 2C), despite the small optical response of these NPs at this wavelength ( Figure 1D and Supplemental Figure S2B). The same lack of interference was observed for Fe 3 O 4 NPs up to a 10 nM concentration (Supplemental Figure S1B). On the other hand, Ag NPs inhibited in a dose-dependent fashion the pNA + diazo reading at 540 nm ( Figure 2D). These results indicate that it is likely that of the modified chromogenic assay could underestimate endotoxin contamination of Ag NPs when concentrated 40.4 nM, whereas the same assay could be applied to Au NPs (up to 5 nM) and to Fe 3 O 4 NPs (up to 10 nM). The Ag NPs used in this assay, which are coated with sodium citrate (SC), do not directly show any absorbance at 540 nm ( Figure 1D and Supplemental Figure S2E). By changing the Ag NPs coating from SC to mercaptoundecanoic acid (MUA, negatively charged like SC) or to aminoundecathiol (AUT, positively charged), the Ag NPs lost their capacity to inhibit of the azo dye reading at 540 nm (Figure 3). When testing in parallel Au-SC NPs, Ag-SC NPs or SC alone ( Figures 2C, D and 3A), only the Ag-SC NPs showed inhibition of the azo dye reading at 540 nm, suggesting that SC is not responsible for inhibition. An additional control with Ag ions showed that these do not interfere either ( Figure 3B). Thus, only Ag-SC NPs seem to be able to interfere with the reading at 540 nm, which is neither due to the SC (Au-SC NPs and SC alone do not interfere), nor to the charge (Ag-MUA NPs, same negative charge, do not interfere), nor to the chemical nature of the NPs (Ag-MUA and Ag-AUT NPs do not interfere).
Thus, judging by the interference with the assay readouts, the modified chromogenic assay appears to be suited for testing endotoxin in Au NPs, Fe 3 O 4 NPs, Ag-MUA NPs and Ag-AUT NPs at all tested concentrations, and for Ag-SC NPs only at concentrations below 0.4 nM.
The interference caused by Ag-SC NPs was further explored. Two reasons were identified that could explain this behaviour, surface adsorbability and surface activity. The SC coating on Ag NPs is the weakest coating compared to MUA or AUT. In fact, the method for obtaining MUA-and AUT-coated Ag NPs consisted in mixing Ag-SC NPs with MUA or AUT, which readily displaced SC from the particle surface (''ligand exchange'' method, see ''Methods'' section). Thus, it is often considered that SC-stabilised NPs expose ''naked'' surfaces, since SC can readily exchange with other molecules (Daniel & Astruc, 2004;Sellers et al., 1993). We did therefore check whether the pNA that is released during the assay reaction could be adsorbed on the Ag-SC NP surface. By mixing Ag-SC NPs with pNA, we found that the absorbance peak was indeed shifted from 387 to 390.5 nm, which indicates a significant adsorption ( Figure 4B). In contrast, a red-shift did not occur with Au-SC NPs or with Ag NPs with a different coatings ( Figure 4A and C). Furthermore, we have measured the Z-potential of the particles after incubation with pNA and found that it changed from À50 to À26.8 mV, further confirming the binding of pNA to the particle surface.
Even though SC coating of Au NPs is weak (Daniel & Astruc, 2004), the above data confirm that it is stronger than in Ag NPs (Bastús et al., 2014). Furthermore, the interaction of the pNA amino groups with silver is stronger than that with the carboxylic group of SC (Sperling & Parak, 2010). This would imply that the Au NP surface is less active than the Ag NP surface, and therefore unable to bind pNA. On the other hand, this also implies that the MUA and AUT coatings on Ag NPs are quite stable, as they can prevent the adsorption of pNA. The binding of pNA to Ag-SC NPs also affects the reaction of pNA with the diazo reagents and the azo dye formation in the modified chromogenic assay. We hypothesize that, once the pNA is directly absorbed onto Ag NP surface, the latter could provide electrons that reduce the diazo bond in the p-nitrohyenyl diazonium molecule, preventing its reaction to form the final azo dye. This would occur when pNA forms p-nitrohyenyl diazonium (the precursor of the final azo dye detectable at 540 nm). To strengthen this hypothesis, we could inhibit the azo dye formation from pNA in the presence of the reducing agent sodium borohydride ( Figure 4D). The inhibition of azo dye formation depended on the surface activity, since larger Ag-SC NPs (50 nm, lower surface energy) showed a reduced inhibition compared to smaller Ag-SC NPs (10 nm, high surface activity) at the same molar concentration ( Figure 4E). Overall, these results show that the optical absorbance of NPs is not sufficient for predicting their interference with an optical density-based assay, since chemical reactions between the dyes and NPs can also interfere. Consequently, different coatings of the NPs can significantly affect their capacity to interfere with the assay reagents and final readouts.
Moreover, one should bear in mind that NPs, especially if their surface is not efficiently passivated and made compatible with the experimental environment, are prone to aggregation. Aggregation, even if mild and limited, can produce significant colour changes in the NP sample. In the case of 14 nm Ag NPs, they may potentially absorb from 400 to 800 nm, depending on their aggregation state. This underlines the necessity of preliminary testing each NP batch both for optical responsiveness and for optical interference, to select the most suitable conditions for reliable endotoxin detection.

NP interference with fluorescence
The fluorogenic LAL assay has the advantage of avoiding the problems of optical interference by non-fluorescent NPs, although it is prone to interference by fluorescent NPs. The three NPs tested (Au, Fe 3 O 4 and Ag-SC NPs) do not show auto-fluorescence at the excitation/emission wavelengths of 380/440 nm used in the assay (Supplemental Figure 2C, F and I).
When testing the ability of NPs to interfere with/quench the fluorescence liberated upon LPS action, Au NPs did not show interference ( Figure 2E), whilst strong inhibition was caused by Ag NPs at all concentrations tested ( Figure 2F), and by Fe 3 O 4 NPs at concentrations over 10 nM (Supplemental Figure S1C). Therefore, based on the lack of interference with fluorescence, the fluorogenic LAL assay appears to be suitable for testing endotoxin in Au NPs and Fe 3 O 4 NPs at concentrations up to 5-10 nM, whilst it cannot be applied to Ag-SC NPs. It should be noted that, due to limitations in synthesis procedures, only Fe 3 O 4 NPs could be tested at concentrations higher than 10 nM.

NP interference with endotoxin recovery
Several molecules and compounds (e.g. EDTA and heparin) are known to affect the results of the LAL assay when present in high concentrations. This occurs mainly by interacting with the substrate or by interfering with the enzymatic activity. In a study published 30 years ago, among 333 drug products tested, 236 (71%) interfered with the LAL test if not previously diluted or treated (Twohy et al., 1984). The LAL methodology uses samples diluted in endotoxin-free water. Thus, all components of the sample (except endotoxin-free water) may potentially interfere with the assay, thereby affecting endotoxin measurement. ''Recovery'' of endotoxin, i.e. measuring the amount of endotoxin in a sample deliberately spiked with a known amount of standard endotoxin, provides information on the possible interference of the sample with the LAL assay components (see ''Methods'' scetion for the recovery rate calculation). This test (also termed Positive Product Control, PPC) should be performed on all samples, to identify possible non-specific inhibitory or enhancing effects of the sample components (USP, 2005). The proper recovery of endotoxin spiked in the sample is among the FDA requirements for validating LAL-based endotoxin measurement in a given product.
The interference of sample components with an enzyme or substrate activation in the LAL assay can be often overcome by diluting the sample to a point at which the interfering factors cease to affect the test, but at which the presence of endotoxin is still detectable. This dilution is called ''Maximum Valid Dilution'' (MVD;USP, 2005). The MVD depends on the sensitivity of the LAL test and on the endotoxin limit set for the specific use of the samples (e.g. intravenous injection), a limit that differs from case to case. In our case, since the NP use is not defined a priori, we intended to set up a general procedure aiming at obtaining reliable results, by identifying the concentrations of NPs in which endotoxin can be reliably measured without nonspecific interference. Thus, we will not consider MDV as a relevant parameter. Data in Figure 5 and Supplemental Figure S3 compare the detection of endotoxin in NP samples with the three different LAL assays. For Au NPs and Fe 3 O 4 NPs, concentrations from 0.5 to 5-10 nM were tested, whilst for Ag-SC NPs the highest concentration used was 1 nM, due to the significant interference of higher concentrations with the readouts of all three assays. The results show that the modified chromogenic assay measured significant endotoxin levels, which however could not be detected by the other two assays ( Figure 5A, D and Supplemental Figure S3A). In parallel, the recovery of endotoxin was examined by spiking the different NP dilutions with a fixed concentration of 0.5 EU/ml of standard endotoxin. The results in Figure 5B show that endotoxin recovery is full in the presence of Au NPs when using either the traditional or the modified chromogenic assays, whilst Au NPs appeared to inhibit dose-dependently the measurement of the Figure 5. Endotoxin detection and recovery in Au and Ag NPs by three LAL assays. The presence of endotoxin was measured with the traditional chromogenic assay (black circles), the modified chromogenic assay (blue squares), and the fluorogenic assay (red triangles), in Au NPs (panels A, B and C) and in Ag-SC NPs (panels D, E and F), in the absence (panels A and D) and in the presence (panels B and E) of 0.5 EU/ml of exogenously added endotoxin. Endotoxin recovery rates (panels C and F) are calculated based on results in panels B and E. The inset in panel D reports the reading of 1 nM Ag-SC NPs in the LAL assays in the absence of enzyme (colour blank control). Data are the mean ± SD of duplicate determinations within one test representative of two performed. The horizontal dotted line represents the endotoxin value measured in the absence of NPs (0.515 EU/ml in panels B and E; mean of 10 data; 100% in panel C and F), and the shaded area the SD above and below the mean. *p50.05 versus control; **p50.01 versus control. endotoxin standard in the fluorogenic assay. By calculating the endotoxin recovery rate ( Figure 5C), it is evident that the two chromogenic assays, which do not show any significant enhancement or inhibition, can be used to test the endotoxin contamination in Au NPs at concentrations up to 5 nM. On the other hand, the significant inhibition of the recovery rate rules out the use of the fluorescent assay for detecting endotoxin in Au NPs at concentrations !1 nM. Based on assay sensitivity, the modified chromogenic assay can provide more precise and more sensitive results than the traditional chromogenic assay and is therefore the choice assay for endotoxin measurement in Au NPs. Thus, in the light of all above tests, we can say that the endotoxin contamination in this batch of Au NPs was 90.6 ± 11.3 EU/nmol (corresponding to 0.45 ± 0.06 EU/ml; average ± SD of values obtained with four NP concentrations tested in duplicate in two separate replicate tests).
The endotoxin contamination in Au NPs, assessed after LAL assay optimisation, was confirmed in the PBMC pyrogen test, in which the same amount of endotoxin contamination could be measured (0.45 ± 0.07 EU/ml, average ± SD of quadruplicate determinations). It should be noted that this finding suggests that Au NPs do not have an intrinsic pyrogenic (inflammatory) effect, since the PBMC pyrogen test is not specific for endotoxin, but it measures in general the capacity of inducing the production of an inflammatory cytokine. In this case, LPS is apparently responsible for the totality of the pyrogenic effect of the Au NPs.
Also for Fe 3 O 4 NPs, the comparison of the three LAL assays excluded the use of the fluorogenic assay and led to the selection of the modified chromogenic assay as the most sensitive method (Supplemental Figure S3). The endotoxin contamination measured in Fe 3 O 4 NPs was 84.5 ± 9.1 EU/nmol (corresponding to 0.88 ± 0.09 EU/ml; average ± SD of results obtained with six dilutions tested in duplicate in two separate replicate tests).
In the case of Ag-SC NPs ( Figure 5D-F), interference with the assay components and readouts limited the possibility of reliably using the chromogenic assays for testing the endotoxin contamination at concentrations above 0.4 nM ( Figure 2B and D for the interference of Ag-SC NPs with the two chromogenic readouts). In addition, the fluorescent assay turned out to be unsuitable at any NP concentration. The inset in Figure 5(D) shows that, in the traditional chromogenic assay, Ag NPs at a concentration of 1 nM can simulate (in the absence of enzyme; colour blank control) an endotoxin contamination of about 0.23 EU/ml. This is in agreement with what we found with pNA readings (Supplemental Figure S2D). The same NP concentration is inactive in the colour blank of the modified chromogenic assay, but this cannot be taken as lack of interference, as we already know that Ag-SC NPs do strongly interfere with the azo dye readings ( Figure 2D). Thus, although the data in Figure 5 (panels D-F) seem to show an acceptable recovery rate (similarly to data with Au NPs), from the additional controls of optical responsiveness and interference we know that only results at concentrations of Ag-SC NPs below 0.4 nM are reliable. With these restrictions, we can say that the endotoxin contamination in this batch of Ag-SC NPs is of 408.5 ± 74.6 EU/nmol as measured by the modified chromogenic assay (corresponding to 0.82 ± 0.15 EU/ml; average ± SD of results obtained with three dilutions tested in duplicate in two separate replicate tests).
Thus, these results underline an important issue that we need to consider, i.e. the validity and accuracy of the endotoxin measurement. According to FDA rules (which is considered acceptable value between 50 and 200%), the recovery rates we have observed for Ag-SC NPs at 1 nM with the two chromogenic assays are acceptable (about 120% with the traditional assay and about 100% with the modified assay; Figure 5F). However, we know that the Ag-SC NPs increase the readout of the traditional assay and decrease that of the modified assay, and also that they can adsorb pNA on their surface and change its reactivity. Therefore, the results obtained with high Ag-SC NP concentrations are inaccurate. Although these variations in the recovery rate can be acceptable from the regulatory point of view, the inaccurate estimation of endotoxin in NPs may pose a significant problem when testing NP safety in vitro and in vivo, given the important inflammation-inducing effects of endotoxin even at non-toxic trace concentrations. Thus, when testing endotoxin in NPs, it is important that, in parallel to the recovery rate data, also the readout interference results are considered.

Effects of solvents and contaminants on endotoxin detection in NPs
For NPs obtained by liquid synthesis, the ''solvent'' in which NPs are suspended may cause interference or have a direct effect on the assay components. Centrifugation of NPs and resuspension in endotoxin-free water would reduce the interference due to chemicals in the solvent (or indeed reduce the overall endotoxin contamination), but often cause NP aggregation or agglomeration. Therefore, before performing the LAL assay on suspended NPs, the chemicals used to keep the NPs in a monodispersed suspension, or those to shape their surface characteristics and charge, should be tested.
To determine the effect of NP solvents on endotoxin detection, freshly synthesised Au, Ag and Fe 3 O 4 NPs were centrifuged at high speed and supernatants collected and tested with the three LAL assays. Their lack of interference with the assays' readouts was first assessed (Supplemental Figure S2). In the LAL assays, none of the solvents did affect endotoxin recovery (data not shown), thereby excluding non-specific solvent-dependent interference with the results.
High concentrations of EDTA and heparin can also affect LAL-based endotoxin evaluation, and in this case, the best way to eliminate the interference is diluting the NP samples. Anyway, no EDTA or heparin was present in the NP samples tested in this study.
An important and common contaminant, which may affect endotoxin detection, is b-glucan. The chromogenic assays (as well as the gel-clot and the turbidimetric assays) are based on LAL, a cell lysate that contains a series of enzymes, including Factor C, activated by endotoxin, and Factor G, activated by b-glucan. Both activated Factor C and Factor G cause clotting in the gel-clot and turbidimetric assays, and pNA release in the chromogenic assays. Thus, these assays do not distinguish between endotoxin and b-glucan. On the other hand, the fluorogenic assay, which is based on recombinant Factor C, is specific for endotoxin. Since the fluorogenic assay is not suitable for NP testing, we have used a specific blocker of b-glucan (the glucashield buffer) in the modified chromogenic assay and found that endotoxin measurement was not affected (data not shown), indicating that NPs were not contaminated with b-glucan and that the reaction measured can be attributed solely to endotoxin contamination.

Catalytic properties of NPs on LAL substrate
An important issue, when validating the use of LAL assays for measuring endotoxin contamination in NPs, is the possibility that NPs possess catalytic activity and directly cleave the LAL substrate, thereby causing a reaction that could be erroneously attributed to endotoxin. Therefore, we examined the catalytic properties of the Au, Ag, and Fe 3 O 4 NPs for Boc-Leu-Gly-Arg-pnitroanilide (the substrate in the modified chromogenic assay). NPs were incubated with different concentrations of substrate for the same time and at the same temperature as in the chromogenic assay, but no release of pNA could be detected (data not shown).
This result further supports the conclusion that the reaction measured in the modified chromogenic assay can be specifically attributed to an endotoxin contamination.

Conclusions
Evaluating endotoxin contamination in NP preparations is a key element for valid nanosafety assessment, because of the powerful biological effects that an endotoxin contamination, even in trace amounts, may cause both in vitro and in vivo. However, the LAL assays that are generally used for testing the presence of endotoxin in NPs need to be validated, given the possibility of interference of the nanomaterials with the assay components and readouts. In this study, validation of commercially available LAL assays for endotoxin testing in metal and metal oxide NPs suggests that the modified chromogenic LAL assay is suitable for testing endotoxin in NPs, showing good sensitivity and limited interference by NPs with the assay components. However, to confirm the reliability of the results obtained, for each NP type it is important to perform an accurate evaluation of the possible interaction of NPs with the assay components and interference with the assay readouts, as well as assessing the presence of interfering contaminants (e.g. solvents, b-glucan). Thus, the validity of endotoxin measurement in NP preparations with a LAL assay relies on the following steps.
(1) Evaluation of NP optical activity or autofluorescence at the wavelength/parameters of the LAL assay readouts. This provides a first rough estimate of the possible interference of the NPs with the readouts of the different assays.
(2) Evaluation of the interference using different concentrations of NPs (and of the solvent in which they are dispersed) with the reading of the assay readout. This will reveal at which concentration the NP suspension can be reliably tested in the LAL assay. (3) Calculating the recovery rate of endotoxin in the presence of different concentrations of NPs/solvent. This should provide an indication of the assay reliability, once the previous criteria are met. It is important to remember that recovery rate alone is not a proof of assay reliability. (4) Assessing the possible non-specific effects of NPs on the assay components (the overall effect is already suggested by the recovery rate experiment), and the possible contamination with interfering agents (such as b-glucan).