key: cord-0726869-h1eyfeio
authors: Rothen, Dominik A.; Krenger, Pascal S.; Nonic, Aleksandra; Balke, Ina; Vogt, Anne‐Cathrine S.; Chang, Xinyue; Manenti, Alessandro; Vedovi, Fabio; Resevica, Gunta; Walton, Senta M.; Zeltins, Andris; Montomoli, Emanuele; Vogel, Monique; Bachmann, Martin F.; Mohsen, Mona O.
title: Intranasal administration of a VLP‐based vaccine induces neutralizing antibodies against SARS‐CoV‐2 and variants of concern
date: 2022-04-15
journal: Allergy
DOI: 10.1111/all.15311
sha: b2df328b803c0807070b4bb7b8cd3721eb0e83b2
doc_id: 726869
cord_uid: h1eyfeio

BACKGROUND: The highly contagious SARS‐CoV‐2 is mainly transmitted by respiratory droplets and aerosols. Consequently, people are required to wear masks and maintain a social distance to avoid spreading of the virus. Despite the success of the commercially available vaccines, the virus is still uncontained globally. Given the tropism of SARS‐CoV‐2, a mucosal immune reaction would help to reduce viral shedding and transmission locally. Only seven out of hundreds of ongoing clinical trials are testing the intranasal delivery of a vaccine against COVID‐19. METHODS: In the current study, we evaluated the immunogenicity of a traditional vaccine platform based on virus‐like particles (VLPs) displaying RBD of SARS‐CoV‐2 for intranasal administration in a murine model. The candidate vaccine platform, CuMV(TT)‐RBD, has been optimized to incorporate a universal T helper cell epitope derived from tetanus‐toxin and is self‐adjuvanted with TLR7/8 ligands. RESULTS: CuMV(TT)‐RBD vaccine elicited a strong systemic RBD‐ and spike‐IgG and IgA antibodies of high avidity. Local immune response was assessed, and our results demonstrate a strong mucosal antibody and plasma cell production in lung tissue. Furthermore, the induced systemic antibodies could efficiently recognize and neutralize different variants of concern (VOCs). CONCLUSION: Our data demonstrate that intranasal administration of CuMV(TT)‐RBD induces a protective systemic and local specific antibody response against SARS‐CoV‐2 and its VOCs.

Till date, COVID-19 caused by SARS-CoV-2 is still considered a global pandemic that has wreaked havoc globally and put a heavy toll on public health and economy. The marketed vaccines such as mRNA, viral vector, and inactivated viruses have greatly reduced the number of COVID-19 mortality and hospitalization and continue to provide different levels of protections against the emerging variants of concern (VOCs). 1 Viral tropism depends, among other factors, on the susceptibility of a specific host cell. COVID-19 patients often present with respiratory illness that can progress to severe pneumonia. 2 These observations suggested that the lung is the primary organ infected by SARS-CoV-2. The fact that lung epithelial cells express the angiotensin converting enzyme 2 (ACE2), the viral receptor, substantiates this observation. 3 The primary port of entry to the body is alveolar epithelial cells, but vascular endothelial cells also express ACE2 and are a prominent place of viral replication. 4, 5 These cells may be considered the base for early infection and viral replication as well as long-term viral persistence in some cases. 6 The currently available marketed vaccines are administered intramuscularly (i.m.) producing systemic spike and RBD-specific antibodies (Abs) that can recognize and neutralize the virus. 7 Given the tropism of SARS-CoV-2, recent research efforts have also been de- Efficient induction of mucosal immunity can be best achieved by vaccines that mimic mucosal pathogens. Virus-like particles (VLPs) constitute an efficient and safe vaccine platform as they lack genetic material for replication in vivo. Their particulate and repetitive surface structure enables them to stimulate innate and adaptive immune response and target the mucosa as well as the underlying dendritic cells (DCs). 14 We have previously assessed the efficacy of Qß-VLPs as an i.n. vaccine platform. Our results indicated efficient induction of specific-IgGs in serum and lungs besides robust local IgA production. 14 

In this study, we describe a COVID-19 vaccine based on virus-like particles (VLPs) for intranasal administration. We demonstrate that the vaccine candidate (CuMV TT -RBD) is highly immunogenic in mice and is capable of inducing mucosal and systemic RBD as well as spike specific antibody responses. The induced antibodies are capable of neutralizing SARS-CoV-2 and variants of concern (VOCs).

including high levels of IgG and IgA, plasma cell (PC) formation as well as broad viral neutralization. Taken together, our vaccine constitutes an efficient candidate for the generation of Ab-based vaccine that can be administered mucosally in a needle-free manner.

All in vivo experiments were performed using (8-12-week-old) wildtype (wt) female BALB/cOlaHsd mice purchased from Harlan. All animal procedures were conducted in accordance with the Swiss Animal Act (455.109.1-5 September 2008) of University of Bern. All animals were treated for experimentation according to the protocols approved by the Swiss Federal Veterinary Office. 

Expression and production of CuMV TT was described in detail in Zeltins et al. 15 The level of LPS is 10 endotoxin per mg of CuMV TT measured using LAL test (pierce).

RBD wt was conjugated to CuMV TT using the cross-linker succinimidyl 6-(beta-maleimidopropionamido) hexanoate (SMPH) (Thermo Fisher Scientific, Waltham, MA, USA) at 7.5 molar excess to CuMV TT for 30 min at 25°C. The coupling reactions were performed with molar ratio RBD/CuMV TT (1:1) by shaking at 25°C for 3 h at 250 g on a DSG Titertek (Flow Laboratories, Irvine, UK). Unreacted SMPH and RBD proteins were removed using Amicon Ultra 0.5, 100 K (Merck Millipore, Burlington, MA, USA). VLP samples were centrifuged for 2 min at 12,000 g for measurement on ND-1000. RBD wt . SDS-PAGE was stained with InstantBlue TM Coomassie stain and image was obtained with Azure Biosystem using visible channel. Coupling efficiency was calculated by densitometry (as previously described for the IL17A-CuMV TT vaccine 15 

Physical stability and integrity of the candidate vaccine CuMV TT -RBD were visualized by transmission electron microscopy (Philips CM12 EM). For imaging, sample-grids were glow discharged and 10 μl of purified CuMV TT -RBD (1.1 mg/ml) was added for 30s. Grids were washed 3× with ddH 2 O and negatively stained with 5 μl of 5% uranyl acetate for 30s. Excess uranyle acetate was removed by pipetting, and the grids were air dried for 10 min. Images were taken with 84,000× and 110,000× magnification.

To test if the vaccine can bind the relevant human receptor ACE2, ELISA plates were coated with 2 µg/ml of ACE2 in PBS at a volume of 

Wild-type BALB/cOlaHsd mice (8-12 weeks, Harlan) were vacci- 

To determine the total IgG Abs against the candidate vaccine To detect IgA Abs, ELISA plates were coated with 1 µg/ml RBD protein and goat anti-mouse IgA POX (ICN 55549, 1:1000 dilution) was used as a secondary Ab. IgG depletion was performed prior to serum incubation. 10 µl of Protein G beads (Invitrogen, USA) were transferred into a tube and placed into a magnet. The liquid was removed, and 75.6 µl diluted sera in PBS-Casein 0.15% was added to the beads and mixed. The tube was incubated on a rotator at RT for 10 min. The tubes were placed back into the magnet, and ELISA was carried out as described above.

To test the avidity of IgG and IgA Ab against RBD, the abovedescribed protocol was expanded by an additional step as previously described. 16 Following serum incubation at RT for 1 h, the plates were washed once in PBS/0.01% Tween, and then washed 3× with 7 M urea in PBS-0.05%Tween or with PBS-0.05% Tween for 5 min.

After washing with PBS-0.05%Tween, goat anti-mouse IgG conjugated to Horseradish Peroxidase (HRP) (Jackson ImmunoResearch, West Grove, Pennsylvania) was added (1:2000) and incubated for 1 h at RT. IgA Abs were detected by using a goat anti-mouse IgA POX (ICN 55549, 1:1000 dilution) detecting Ab. Plates were developed with TMB as described above and read at OD 450 nm.

Bronchoalveolar Lavage (BAL) samples were collected as described in Sun et al. 17 

Lungs were perfused with 10 ml of 1 mM EDTA in PBS via the right ventricle of the heart to remove blood cells from the lung vasculature. Lungs were dissected and digested for 30 min at 37°C using RPMI media (2% FBS+Pen/Strep, glutamine, 10 mM HEPES) containing 0.5 mg/ml Collagenase D (Roche). The digested fragments were passed through a 70 μm cell strainer (Greiner bio-one, Art. Nr. 542070), and RBCs were lysed using ACK buffer. Lymphocytes were isolated using 35% Percoll gradient.

The spleen was collected from mice and transferred into 5 ml RPMI media (2% FBS+Pen/Strep, glutamine, 10 mM HEPES). A single cell suspension was prepared by passing the spleen through a 70 μm cell strainer. The suspension was collected and transferred into a falcon tube. The tube was centrifuged for 8 min at 4°C and 300×g. ACK lysis was performed, media added and centrifuged for 8 min at 4°C and 300×g. The pellet was resuspended in media.

Tibia and femur were collected from mice and transferred into 5 ml RPMI media (2% FBS+Pen/Strep, glutamine, 10 mM HEPES). The BM cells were isolated using a syringe by rinsing the bones to flush out the cells.

A single cell suspension was prepared by passing the spleen through a 70 μm cell strainer on a petri dish. The suspension was collected and transferred into a falcon tube. Petri dish was washed with 5 ml media and also added to the falcon tube. The tube was centrifuged for 8 min at 4°C and 300×g. ACK lysis was performed, media added and centrifuged for 8 min at 4°C and 300×g. The pellet was resuspended in media. After, they were loaded with sera from mice, diluted 1:20 in BLI assay buffer. Next, association with 50 nM of human receptor ACE2 (Sino Biological, USA) diluted in BLI assay buffer was measured. To regenerate the tips, two additional steps with regeneration buffer (0.1 M glycine, pH 1.5) and neutralization buffer (BLI assay buffer) were performed. 

The neutralization assay was performed as previously reported by Manenti et al. 18 

Data were analyzed and presented as mean ± SEM using Student's t-test or one-way ANOVA as mentioned in the figure legend, with GraphPad PRISM 9. The value of p < .05 was considered statistically significant (*p < .05, **p < .01, ***p < .001, ****p < .0001).

In order to generate a vaccine-candidate against SARS-CoV-2 for i.n. administration, we have utilized our optimized plant-derived VLPs (CuMV TT ) as a vaccine platform. 15,20-22 RBD amino acid sequence (a.a.

Arg319-Phe541) of SARS-CoV-2 was chemically coupled to CuMV TT using SMPH bifunctional cross-linker ( Figure 1A) . The generated vaccine candidate CuMV TT -RBD is self-adjuvated with prokaryotic ssRNA (TLR7/8 agonist) which is packaged during expression and assembly in the bacterial E. coli system ( Figure 1B) . Efficiency of RBD coupling to CuMV TT was confirmed by SDS-PAGE ( Figure 1C ) and

Western blot ( Figure 1D ). The integrity of the VLPs following the coupling process was checked by electron microscopy and showed no signs of aggregation ( Figure 1E ). Finally, to confirm the antigenicity of CuMV TT -RBD as well as the correct folding and confirmation of the RBD displayed on the particle`s surface, a receptor binding assay was performed. To this end, the human receptor ACE2 was coated on ELISA plate. CuMV TT -RBD, CuMV TT , and RBD were added. Anti-CuMV TT antibodies were used as a secondary antibody to detect receptor bound VLPs. The results revealed that CuMV TT -RBD can bind to ACE2 receptor indicating correct folding of RBD on the surface of the VLP while the control did not show any binding ( Figure 1F ).

To test the immunogenicity and the induction of a humoral immune response in murine models, BALB/c mice were i.n. primed on day 0 and boosted on day 28 with 40 µg of CuMV TT -RBD vaccine or with 40 µg of CuMV TT as a control without addition of adjuvants. Vaccination and bleeding regimen are shown in Figure 2A . Total systemic RBD and spike-specific IgG were measured by ELISA. Systemic RBD-specific IgG response was detected in the group receiving CuMV TT -RBD seven days after the priming dose. Furthermore, the induced response increased by about 1000-fold following the booster dose on Day 35

( Figure 2B and C) . Full-length spike protein responses remained low after priming but increased significantly by 30-folds upon a booster injection resulting in a stable IgG antibody titer ( Figure 2D and E) . ELISA plates were coated with 1 μg/ml of spike or RBD protein as mentioned in the method section, however, from a molar point of view, about 8fold more RBD is coated than spike. This is a likely explanation for the higher reactivity of RBD compared with spike. No IgG response has been detected in the control group which received CuMV TT only.

The avidity of an Ab is defined as the binding strength through points of interaction. It can be quantified as the ratio of K d for the intrinsic affinity over the one for functional affinity of a multiple point interaction. 20 High avidity Abs are formed upon affinity maturation in germinal centers (GCs) and are associated with protective immunity against SARS-CoV-2 infection. 23 To assess the avidity of the induced IgG Abs against RBD, we carried out an avidity ELISA using day 49 sera. The obtained results indicated that about 40% of the systemically induced RBD-specific IgGs are of high avidity following the i.n. vaccination ( Figure 2F and G) . 

IgG subclasses are of major importance in the immunological response against viruses because of enhancing opsonization as well as immune effector funcitons. 24 In additionally, IgA plays an important role in protection against respiratory viruses as it is found in mucosal tissue, the main entrance site for these kind of viruses. 10 The ability of the CuMV TT -RBD vaccine to induce serum IgA and IgG subclasses was evaluated by performing ELISA against RBD with sera collected at day 42. All IgG subclasses were induced in vaccinated mice with IgG1, IgG2a, and IgG2b being the dominant ones. In contrast, no IgG subclasses were detected in the control group ( Figure 3A) . The vaccine was also able to induce isotype switching to IgA as shown in Figure 3B . Approximately, 20% of serum IgA Abs were of high avidity as shown in Figure 3C , D.

The mutation potential of SARS-CoV-2 is considered a burden in vaccine design and development, especially in terms of prolonged protection. Accordingly, we studied the capability of the induced RBD-specific IgG Abs in recognizing mutant RBDs of the different VOCs. Specifically, we have expressed and produced the following mutated RBDs: RBD K417N , RBD E484K , RBD N501Y , RBD K417N/E484K/N501Y , and RBD L452R/E484Q . 25 Compared with RBD wt , RBD VOCs specific IgG levels were slightly lower ( Figure 4) . However, the difference observed 

To test the ability of the CuMV TT -RBD vaccine to induce a humoral immune response in the lung mucosa; BAL was collected two weeks after the booster injection (Day 42) and assessed for RBD and spike protein specific IgG and IgA ( Figure 5A-F) . RBD-and spike-specific IgG Abs were detected at equal levels in the BAL (Figure 5A-C) .

However, IgA Abs in BAL were more abundant against RBD than against spike protein ( Figure 5D-F) . We have also assessed the induced RBD-specific IgG subclasses in BAL. Interestingly, the local mucosal IgG response was less balanced than the serum response and dominated by IgG2b ( Figure 5G ). Next, we tested the quality of the induced RBD-specific IgA in BAL samples, the results confirmed that about 60% of detected IgA Abs in BAL were of high avidity ( Figure 5H-J) . 

In order to characterize the humoral immune response upon i.n.

CuMV TT -RBD vaccination, specific plasma blasts were quantified in lymphoid organs and lung tissue. To this end, spleen, BM and lung were collected on day 42 and analyzed for the presence of RBDspecific IgG and IgA secreting plasma cells. As shown in Figure 6 and Figure S1 , IgG secreting plasma cells were detected in all investigated tissues. Around 25 IgG secreting plasma cells were found per two million cells in spleen and BM. In lung, this ratio was ten-fold higher because the same amount of IgG plasma cells was observed while ten times less cells were seeded. IgA secreting cells were detected in all three tissues; however, at a lower level compared with IgG secreting plasma cells. RBD-specific IgA producing plasma cells in lung were thereby about ten times more abundant compared with spleen or BM ( Figure 6A -C). In overall term, i.n. vaccination with CuMV TT -RBD induced a systemic humoral immune response which was accompanied by a potent local humoral immune response in the lung.

To test the ability of the sera of immunized mice to inhibit binding of RBD to ACE2, a biolayer interferometry competition assay was performed. Accordingly, RBD was immobilized onto anti-His biosensors and binding capacity of ACE2 to RBD in the presence of serum samples was quantified. As depicted in Figure 7A , the binding of ACE2 to RBD was reduced in the presence of sera from CuMV TT -RBD vaccinated mice. In contrast, no binding inhibition was observed in the presence of sera from CuMV TT control mice. Percentage of ACE2 to RBD binding inhibition of individual mice is shown in Figure 7B . Interestingly, the binding inhibition correlated with RBD-specific IgG titers in serum (R value = 0.78) ( Figure 7C ), indicating higher RBD specific IgG titer in serum are more efficient at blocking RBD-ACE2 interaction.

For the evaluation of viral neutralization capacity, sera from CuMV TT -RBD and CuMV TT immunized mice were tested in a CPEbased neutralization assay. To this end, sera from immunized mice were assessed for their ability to prevent cytopathic effects of wt SARS-CoV-2 as well as VOCs on Vero cells in vitro. As shown in Figure 7D , all sera from CuMV TT -RBD vaccinated mice were able to neutralize the wt SARS-CoV-2 as well as SA and UK variants with high neutralization titers reaching to (1:600). In contrast, no neutralizing capacity was determined for sera from CuMV TT immunized mice.

The immune responses of the mucosal compartments are considered an early and essential line of defense against harmful pathogens such as SARS-CoV-2. 26 The majority of mucosal vaccines have been administered via oral or nasal routes, with the rectal, ocular, sublingual, or vaginal routes being less often used. 27 form, serum IgA is usually present monomeric form. 33 In addition to IgA, IgG in the lung might also mediate protection. 9 By passive transudation across alveolar epithelium, IgG can pass from blood into the lower lung and from there by the mucociliary escalator further be carried to the upper respiratory tract and nasal passages. However, only at high serum concentrations local protection through IgG is achieved in the lung. 9

While RBD-specific IgG subclass response in serum was well balanced, IgG2b was the most abundant subclass in BAL. Besides IgG2a, IgG2b is the only subclass that binds all three activating Fc receptors (FcγRI, FcγRIII, and FcγRIV) and the only inhibitory receptor (FcγRIIB) in mice. 34 IgG2b Abs therefore mediate a wide variety of effector functions, which is of key importance in the maintenance of immune protection.

SARS-CoV-2 has the ability to mutate, albeit a proof-reading system is in place to keep the large genome of almost 30 kD genetically stable. 35 Previously, we have described that a single N501Y mutation increased the binding affinity to ACE2, but could still be detected by convalescent sera. Contrary were the results with the E484K mutation, where no enhanced binding to ACE2 was shown but much lower recognition by convalescent sera. Triple mutant RBD (K417N/E484K/N501Y) exhibited both features: stronger affinity to ACE2 and much lower detection by convalescent sera. 36 Since vaccines optimally mediate protection for many years, vaccine induced Abs should therefore be able to recognize new virus variants as well.

In the present study, we could show that i.n. applied CuMV TT -RBD induced serum IgG Abs that are able to recognize wt RBD as well as numerous RBD VOCs.

A crucial milestone in vaccine development is effective neutralization of the virus. Sera induced after i.n. administration of CuMV TT -RBD could completely inhibit the cytopathic effect of wt SARS-CoV-2 as well as other VOCs, specifically SA and UK variants. This may be explained by the highly repetitive, rigid antigenic surface array of VLPs which are spaced by 5 nm and displaying RBD domains at a spacing of 5-10 nm. 37 Such array of highly organized epitopes is considered a pathogenassociated structural patterns (PASPs) which are recognized by the immune system. 38 In contrast, naturally induced Abs by SARS-CoV-2 are low in number and wane rapidly. 39 It has been shown that B and T cells are primed by mucosal vaccination or natural infection, express receptors which promote homing of these cells to mucosal sites as Ab-secreting cells or effector or tissue-resident T cells. 40 We have shown in our previous studies that VLP-specific T H cell response mediate specific B cell isotype-switch.

Furthermore, the packaged RNA in VLPs can drive CD8+ T cells as well as T H 1 responses. 41 The role of T cells after i.n. vaccination using VLPs is an area we are currently investigating.

Collectively, we have shown in this study that our COVID-19 vaccine candidate CuMV TT -RBD is highly immunogenic and capable of inducing both mucosal and systemic IgG and IgA response against SARS-CoV-2 upon i.n. administration. The induced Abs could effectively recognize and neutralize wt as well as the emerging VOCs.

The ability of the vaccine candidate to stop nasal viral shedding and transmission is currently under investigation. As COVID-19 pandemic continues to present a global threat to human health, it seems rational to further develop an i.n. vaccine based on conventional platform.

This publication was funded by Saiba AG, the Swiss National Science Foundation (SNF grants 31003_185114 and IZRPZO_194968). Open access funding provided by Inselspital Universitatsspital Bern.

[Correction added on 14-May-2022, after first online publication:

CSAL funding statement has been added.] 

COVID-19 vaccines that reduce symptoms but do not block infection need higher coverage and faster rollout to achieve population impact

Clinical progression and viral load in a community outbreak

Atlas of ACE2 gene expression reveals novel insights into transmission of SARS-CoV-2

ACE2 receptor expression and severe acute respiratory syndrome coronavirus infection depend on differentiation of human airway epithelia

Mechanisms of SARS-CoV-2 Transmission and Pathogenesis

SARS-CoV-2 receptor ace2 is an interferon-stimulated gene in human airway epithelial cells and is detected in specific cell subsets across tissues

SARS-CoV-2 vaccines in development

Role of IgA versus IgG in the control of influenza viral infection in the murine respiratory tract

Role of IgA in the defense against respiratory infections: IgA deficient mice exhibited increased susceptibility to intranasal infection with Mycobacterium bovis BCG

The establishment of resident memory B cells in the lung requires local antigen encounter

Intranasal ChAdOx1 nCoV-19/AZD1222 vaccination reduces shedding of SARS-CoV-2 D614G in rhesus macaques

Immune responses after live attenuated influenza vaccination

Efficient induction of mucosal and systemic immune responses by virus-like particles administered intranasally: Implications for vaccine design

Incorporation of tetanus-epitope into virus-like particles achieves vaccine responses even in older recipients in models of psoriasis, Alzheimer's and Cat Allergy

Immunochemical analyses of serum antibodies from pig herds in a Salmonella non-endemic region

Murine bronchoalveolar lavage

Evaluation of SARS-CoV-2 neutralizing antibodies using a CPE-based colorimetric live virus micro-neutralization assay in human serum samples

The theory and practice of the viral dose in neutralization assay: Insights on SARS-CoV-2 'doublethink' effect

The 3Ds in virus-like particle based-vaccines: "Design, Delivery and Dynamics

A scalable and highly immunogenic virus-like particle-based vaccine against SARS-CoV-2

Neutralization of MERS coronavirus through a scalable nanoparticle vaccine

The potential significance of high avidity immunoglobulin G (IgG) for protective immunity towards SARS-CoV-2

Murine IgG1 complexes trigger immune effector functions predominantly via Fc gamma RIII (CD16)

BNT162b2 mRNA COVID-19 vaccine induces antibodies of broader cross-reactivity than natural infection, but recognition of mutant viruses is up to 10-fold reduced

Mucosal immunity in COVID-19: A neglected but critical aspect of SARS-CoV-2 infection

Mucosal delivery routes for optimal immunization

Major findings and recent advances in virus-like particle (VLP)-based vaccines

RNA and toll-like receptor 7 license the generation of superior secondary plasma cells at multiple levels in a B cell intrinsic fashion

Alveolar macrophages and lung dendritic cells sense RNA and drive mucosal IgA responses

Development of a vaccine against sars-cov-2 based on the receptor-binding domain displayed on virus-like particles

Enhanced SARS-CoV-2 neutralization by dimeric IgA

IgA subclasses have different effector functions associated with distinct glycosylation profiles

Fc receptor-targeted mucosal vaccination as a novel strategy for the generation of enhanced immunity against mucosal and non-mucosal pathogens

Coronavirus RNA proofreading: Molecular basis and therapeutic targeting

Molecular definition of severe acute respiratory syndrome coronavirus 2 receptor-binding domain mutations: Receptor affinity versus neutralization of receptor interaction

SARS-CoV-2 structural features may explain limited neutralizing-antibody responses

Vaccine delivery: A matter of size, geometry, kinetics and molecular patterns

The trinity of COVID-19: immunity, inflammation and intervention

Resident memory CD8+ T cells in the upper respiratory tract prevent pulmonary influenza virus infection

Enhanced Neutralizing Antibody Titers and Th1 Polarization from a Novel Escherichia coli Derived Pandemic Influenza Vaccine

Additional supporting information may be found in the online version of the article at the publisher's website.