Immunity induced by a broad class of inorganic crystalline materials is directly controlled by their chemistry

Because alum adjuvants act on DCs, at least in vivo (Kool et al., 2008a,b; Marrack et al., 2009; McKee et al., 2009), we initially evaluated the immunological properties of LDHs in culture by assessing their ability to stimulate responses of human monocyte-derived DCs (Sallusto and Lanzavecchia, 1994). We measured the capacity of DCs to secrete proinflammatory cytokines (IL-1β, IL-6, and TNF), as well as cytokines that polarize Th1 responses (IL-12p70), maintain memory T cell responses (IL-15), and promote antiviral responses (IFN-α2). We also investigated DC production of chemokines (IL-8, MIP-1β, and MCP-1) and of membrane molecules involved in T cell activation (CD86), DC–T cell cross talk (CD40), and suppression of CD8 T cell proliferation (CD274). To study this systematically, we designed, synthesized, and characterized a series of endotoxin-free LDHs, with systematic substitutions of metal cations and anions (Table S1). The LDHs were observed to comprise crystalline hexagonal platelets (Fig. 1). DCs were co-cultured with LDHs, and the cellular responses stimulated by each compound were evaluated and compared with those elicited by the commercially available alums Imject and Alhydrogel. We found that individual LDHs stimulated distinct DC responses across many immunological outputs but with donor to donor variation. We therefore undertook a statistical repeated measures analysis, with a random effect to account for donor variability, and fitted a multilevel model to obtain estimates of the overall responses. Several compounds were found to stimulate robust DC responses, frequently much greater than the commercially used Alhydrogel adjuvant (Fig. 2 and Tables S2 and S3). In Fig. 2 and subsequently, LDHs are denoted M_xM_y′-X, where M and M′ are the metal cations in the LDH layers, x/y is the ratio of these ions, and X is the interlayer anion.

We then hypothesized that the varying immunological activities of different LDHs may be determined by their respective physicochemical characteristics. We therefore applied a regression model between all of the DC response datasets and the physicochemical properties of these LDHs obtained from published studies or measured directly (Table S4 a). Essentially, this approach considered that the physicochemical properties were causative of the immune responses and sought to find the closest fit between any given subset of properties and all the immunological responses. Surprisingly, this revealed that all in vitro human DC responses were very highly correlated with a linear combination of three LDH properties: the radius of the spherical M⁺ or M²⁺ metal cations; the distance between the LDH layers (interlayer spacing); and the zeta potential, which defines the magnitude of the electrical charge at the interfacial double layer around the LDH particle (Fig. 1; Attwood, 2007). Correlation coefficients between this combination and the observed responses ranged from 0.78 to 1.0, with one exception (IL-8) at 0.65. These correlations can be expressed by a simple linear equation:

In this equation, A, B, and C_i are coefficients for any given immunological response, and P_i is the value of each respective physicochemical property. (Values for LDH properties, which were standardized to ensure equal variance, and the coefficients are shown in Tables S4 b and S5.)

We next investigated the model’s ability to predict a priori the immunological properties of newly synthesized LDHs, using the robust and relatively high-throughput DC assays we had previously used. We synthesized two new LDH compounds, LiAl₂-NO₃ and Mg₂Al-Cl (Table S1), and made blind predictions of the multiple (n = 12) DC responses they might induce based purely on their physicochemical properties. We found the correlation between predicted and actual observed responses to be remarkably high, with a median coefficient of variation of 5.14% (Fig. 3 and Table S6). We observed that 22 of the 24 predicted values fell within the 95% confidence intervals (CIs); the probability of this occurring by chance is P < 0.0002. Conversely, just 14 out of 24 predicted values for responses elicited by one compound fall within the CIs for the responses elicited by the other, showing that the model’s predictions are composition specific. Because the predictive power of our model is proportional to the size of the datasets, its accuracy should be increased by further calibration across a wider series of LDHs with a broader range of properties.

To explore the potential in vivo adjuvant activities of LDHs, we next evaluated the capacity of some of our compounds to elicit antibody responses in C57BL/6 mice, based on a vaccination protocol to study adjuvant-driven responses against the model antigen OVA. Mice were primed i.p. with OVA admixed with each LDH, or alum, and boosted i.p. with soluble OVA 1 wk later (Eisenbarth et al., 2008; Kool et al., 2008b). Because this leads to primary antigen presentation in the mediastinal lymph nodes, which also drain the lungs, we then challenged the mice with a final, intranasal boost of soluble OVA 1 wk later. The resultant data confirmed that chemically different LDHs can elicit in vivo antibody responses that are at least as potent as those of the alums tested (Fig. 4 and Tables S7 and S8). Notably, vaccination with alum adjuvants resulted in OVA-specific IgG1 and IgE antibody production alone, reflective of a Th2-dominated response as expected, whereas some LDH compounds (e.g., Mg₂Fe-CO₃) stimulated production of OVA-specific IgG2c antibodies, indicative of a Th1-polarized response (Gavin et al., 2006; Eisenbarth et al., 2008). We then applied our regression model (above) to all of these in vivo antibody datasets and found that these, too, were very highly correlated with the same subset of LDH properties as above, conforming to Eq. 1. (Values for the properties and coefficients are shown in Tables S4 and S9.)

Finally, we have investigated the capacity of LDHs to induce human macrophage responses in vitro. Our initial data indicate that proinflammatory responses of human monocyte-derived macrophages and dermal macrophages are highly correlated with the same LDH properties as above, also conforming to Eq. 1 (Table S10). Together, our findings suggest that the same “chemical–immunology rule” for these materials may apply at other levels of immunity, such as T cell responses. In the latter context, we also have preliminary data suggesting that certain LDHs, notably Mg₂Al-NO₃ and Mg₂Al-CO₃, can elicit expansion of mouse OVA-specific effector CD8⁺ T cells in vivo (not depicted), and the former compound has been reported to elicit cytotoxic T lymphocytes in vivo (Li et al., 2011).

This study suggests a highly discriminatory mode for the sensing of crystalline materials by the immune system. Whether this is caused by the capacity of the plasma membrane of DCs to act as a general sensor for solid structures (Flach et al., 2011), the regulated production of DAMPs, or other mechanisms (Dostert et al., 2008; Hornung et al., 2008) is under investigation; so too is the nature of the interaction between LDHs and DCs (e.g., phagocytosis) and the impact of systematically changing particle dose, size, and/or morphology (Hu and O’Hare, 2005) on immunological responses. The regression model we developed (Fig. 1) derives from a systems vaccinology approach (Plotkin, 2008; Pulendran et al., 2010a; Pulendran and Ahmed, 2011; Nakaya and Pulendran, 2012), one aspiration of which has been to predict which materials will act as potent adjuvants to the immune system. Until now, lack of systematically variable adjuvants has made this impossible to achieve. In contrast, our chemical–immunology rule enables accurate prediction of the multiple DC, and potentially other, responses that can be induced by any given LDH, the first time such an approach has been possible for any material. Ultimately, advances in computational chemistry should permit pure in silico screening of vast numbers of LDHs to select those with the most desirable properties. Hence our findings may comprise a paradigm shift in vaccinology by enabling the selection and synthesis of designer adjuvants that can be incorporated into bespoke vaccines, tailored for use in different disease settings.

LDHs were synthesized according to previously reported protocols (Khan et al., 2009). LiAl₂-Cl and LiAl₂-NO₃ were prepared by the direct reaction of LiCl or LiNO₃ with gibbsite. Mg₂Al-NO₃, Mg₂Al-Cl, Mg₂Fe-Cl, Mg₂Fe-NO₃, Ca₂Al-Cl, and Ca₂Al-NO₃ were generated via a coprecipitation synthesis. LiAl₂-CO₃, Mg₂Al-CO₃, and Mg₂Fe-CO₃ were produced by reaction of the nitrate- or chloride-containing LDHs with an excess of sodium carbonate. Endotoxin was not detectable in any compound by Limulus amebocyte assay. Characterization was performed by x-ray diffraction, infrared spectroscopy, elemental microanalysis, and dynamic light-scattering measurements. The positions of the 00l reflections in diffraction patterns were used to calculate the interlayer spacing (the distance between the LDH layers). The zeta potential was measured for 10 mg/ml solutions of the LDHs in phosphate buffered saline, initial pH 7.2. The M⁺ or M²⁺ ionic radius was taken from the literature (Lide, 2000).

Human monocyte-derived DCs were generated according to standard procedures (Sallusto and Lanzavecchia, 1994). Monocytes were isolated from human blood products (Buffy coats or leukocyte cones) using density gradient centrifugation (Lymphoprep; Axis-Shield) followed by negative selection (EasySep Human Monocyte Enrichment kit; STEMCELL Technologies) and cultured for 6 d in complete RPMI supplemented with recombinant human (rh) GM-CSF and rhIL-4 (DC-RPMI). Cytokines were supplied by Gentaur Belgium BVBA or PeproTech. On day 6, DCs were suspended in DC-RPMI at 5 × 10⁵ cells/ml and transferred to flat-bottomed, 96-well plates with 200 µl of cells per well. DCs were then co-cultured for ∼24 h with the LDHs or commercial alums (Alhydrogel [Sigma-Aldrich] and Imject alum [Thermo Fisher Scientific]) at 476 µg/ml final concentration. The expression of surface proteins by the DCs was measured by flow cytometry (FACSCanto [BD], with analysis using FlowJo software). Cytokines were measured either by ELISA (Insight Bioscience Ltd.) or by Luminex technology (EMD Millipore) according to the manufacturers’ instructions. Macrophages were generated from monocytes isolated in an identical process to that used for generating DCs. Subsequently, monocytes at 5 × 10⁵ cells/ml were seeded in 96-well plates (200 µl/well) and cultured for 6 d with 10 ng/ml rhM-CSF alone to generate macrophages (Becker et al., 1987) or with 10 ng/ml rhM-CSF, 10 ng/ml rhIL-10, and 10 U/ml rhGM-CSF to prepare dermal macrophages (Kwan et al., 2008). Cells were co-cultured with LDHs or commercial alums as above, and cytokine production was measured by ELISA in accordance with the manufacturer’s instructions.

For antibody assays, C57BL/6 mice (Harlan Laboratories) were immunized with the model antigen OVA (Worthington Biochemical Corporation) with or without each LDH compound or commercial Imject alum (Thermo Fisher Scientific) as a potential adjuvant (n ≥ 5 per experimental group). Mice were sensitized i.p. with 500 µl OVA at 20 µg/ml admixed with 2 mg/ml LDH (a dose of 10 µg OVA and 1 mg LDH/mouse). 7 d later they were boosted i.p. with 10 µg OVA alone. 1 wk later (day 14) mice were bled via the tail vein, and the serum levels of OVA-specific IgG1, IgE, and IgG2c were measured by ELISA (eBioscience). The prime-boosted mice were then challenged with a 1% OVA aerosol for 30 min on each of days 17, 18, and 19. On day 20 the mice were sacrificed with an overdose of Avertin. Blood was collected to determine the serum levels of OVA-specific IgE, IgG1, and IgG2c antibodies as above. All experiments were approved by the Ethical Committee for Animal Experimentation, Faculty of Science, University of Ghent and VIV site Ardoyen; site license number LA1400091.

The observational data comprised repeated measures of immunological responses from several experiments. Therefore, to obtain estimates of the mean response for each compound, it was necessary to fit a series of two-level linear mixed-effects models. This was done using the lme function from the nlme library in the R statistical package with experiment ID as the second level. To obtain normally distributed residuals and stability of variance, the responses were log-transformed, after which all of the usual diagnostic tests proved to be entirely satisfactory. This analysis provided the estimates of the mean responses given in Tables S2 (shown in Fig. 2), S7 (shown in Fig. 4), and S10 a and the p-values detailed in Tables S3, S8, and S10 b.

Investigation of possible relationships between the immune responses and the physicochemical properties of the LDHs given in Table S4 was performed by computing the canonical correlation coefficients between subsets of the properties and the set of mean responses for each adjuvant. Because the magnitude of the different “as-measured” properties varied considerably (Table S4 a), the properties were first standardized to have mean 0 and unit variance (Table S4 b). This was performed to ensure that each property had equal weighting at the start of analysis. All possible subsets of three or four properties were investigated, and the correlation coefficients were calculated, along with their associated p-values. The subset [ionic radius of M⁺ or M²⁺, interlayer spacing, and zeta potential] consistently produced the smallest p-values and very high correlations across all response types (median = 0.863), all of which were statistically significant; note that all p-values for this subset were <0.025 (with a correlation of >0.6075). It should be noted that the physicochemical properties are often internally correlated with one another, so the immune responses were correlated with several subsets of properties. The selected subset was chosen because it gave the highest correlations and lowest p-values. This resulted in the formula

where the C_i values were calculated to maximize the correlation between the properties and the mean responses. The constants A and B were then determined by treating the relationship as one of cause and effect, in other words to regard the LDH properties as responsible for the responses; thus, A and B were obtained as regression coefficients. Values for the coefficients A, B, and C_i for the respective immunological datasets are shown in Tables S5, S9, and S10 c.

A priori predictions of the responses to newly synthesized LDHs were calculated using the formula in Eq. 1 with the values for A, B, and C_i determined from experiments previously undertaken with other LDHs. To obtain a conservative estimate for the likelihood of the prediction accuracy being attributable to chance, we first calculated the full range of estimated mean ln response across all compounds, then determined the mean 95% CI width and divided this by the range to give the largest plausible upper bound to the proportion of the range covered by the CI. The mean of this proportion across all DC responses was then taken as an over-conservative estimate of the probability of a single prediction falling within the 95% CI by chance. This was 0.6272, and therefore the overall probability of the accuracy of prediction observed being caused by chance (22 of 24 responses within the 95% CI) is 0.00023.

Table S1 shows chemical details and size data for the LDHs used in this study. Tables S2 and S3, included in a separate PDF file, respectively, show actual and ln values and significance data for DC responses to LDHs and alums. Table S4, included in a separate PDF file, shows the actual (a) and standardized (b) physicochemical properties of the LDHs used in this study. Table S5, included in a separate PDF file, shows coefficients for DC responses according to Eq. 1. Table S6, included in a separate PDF file, shows the actual DC response to newly synthesized LDHs and that predicted from Eq. 1. Tables S7, S8, and S9, included in a separate PDF file, respectively, show the actual and ln values, significance data, and coefficients according to Eq. 1 for mouse OVA-specific antibody responses to LDHs and alums. Table S10, included in a separate PDF file, shows the actual and ln values (a), significance data (b), and coefficients (c) according to Eq. 1 for macrophage responses to LDHs and alums.

We thank Dr. Jan Digby (Department of Cardiovascular Medicine, University of Oxford, Oxford, England, UK) for help with Luminex measurements; Dr. Kerstin Jurkschat and Dr. Alison Crossley (Department of Materials, University of Oxford) for assistance with zeta potential measurements; and Dr. Mark Suter (University of Zurich, Zurich, Switzerland), Dr. Gordon McPherson (Oriel College, Oxford, England, UK), and Dr. Richard Dillon (King’s College London, London, England, UK) for helpful discussions.

D. O’Hare was supported by a Medical Research Council Discipline Hopping Grant (with J.M. Austyn as sponsor), and J.M. Austyn was supported by the EU FP6 Programme (via the DC-THERA Network of Excellence, Project No. LSHB-CT-2004-512074). B.N. Lambrecht is supported by a European Research Council grant and by a University of Ghent MRP-GROUP-ID grant.

The authors declare no competing financial interests.

Author contributions: G.R. Williams designed and conducted in vitro experiments, synthesized LDHs, performed statistical analysis, and wrote the manuscript; K. Fierens conducted in vivo experiments and wrote the manuscript; S.G. Preston designed and conducted in vitro experiments and wrote the manuscript; D. Lunn designed and performed the systems vaccinology modelling and undertook statistical analysis; O. Rysnik performed in vitro experiments; S. De Prijck performed in vivo experiments and in vitro analysis of immunoglobulin levels; M. Kool designed and performed in vivo experiments; H.C. Buckley prepared and characterized LDH and alum samples; B.N. Lambrecht designed in vivo experiments and wrote the manuscript; D. O’Hare synthesized the LDH compounds, designed experiments, and wrote the manuscript; J.M. Austyn designed in vitro experiments and wrote the manuscript. D. O’Hare, J.M. Austyn, and B.N. Lambrecht jointly conceived of the use of LDHs as adjuvants, and B.N. Lambrecht proposed the systems vaccinology approach.