INTRODUCTION
Coronavirus disease 2019 (COVID-19) pandemic, caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has led to more than 765 million reported cases and 6.9 million deaths declared by the World Health Organization as of 12 May 2023 (
https://covid19.who.int/), severely disrupting public health and social and economic infrastructures (
1,
2). Apart from the new coronavirus, seasonal influenza causes sustained epidemics in most nontropical countries and approximately leads to 650,000 deaths every year (
3). Of special note, SARS-CoV-2 and influenza viruses share similarities clinical manifestations of the respiratory syndrome, common transmission mechanism, the same infection tissues, and seasonal coincidence (
4–
6). Recent clinical reports revealed concurrent infections with influenza A virus in 22.3% of cases who died of SARS-CoV-2 and in 19.3% of living patients in Northeastern lran (
7). In addition, pre-infection with influenza A virus substantially strengthens the entry of SARS-CoV-2 into host cells, resulting in more severe lung damage and a higher mortality rate (
8). Coinfection of SARS-CoV-2 and flu resulted in 43.1% of dead cases, which SARS-CoV-2 infection alone led to 26.9% of dead cases (
9). Given that the concomitant circulation infections of SARS-CoV-2 with influenza virus, urgency in the development of an available and viable vaccine capable of controlling both kinds of pandemics and preventing next wave pandemic is of vital important.
SARS-CoV-2 is a single-stranded RNA virus that encodes membrane (M), envelope (E), nucleocapsid (N), and surface spike (S) proteins (
10). S protein initiates viral entry through its receptor binding domain (RBD) to engage the host cell receptor, angiotensin-converting enzyme 2 (ACE2) (
11). Recent reports have described that more than 90% of neutralizing humoral responses against SARS-CoV-2 are accountable for RBD-directed antibodies (Abs) (
12–
14). Furthermore, it has been proved that RBD is the predominant target of neutralizing Abs of COVID-19 convalescent plasma, which has been used for rescuing mortality in severe cases (
15,
16). The relationship between RBD-directed Abs and neutralization capability of COVID-19 patient plasma motivates using RBD as a subunit vaccine immunogen for stimulation of a more focused immune response targeting well-conserved domains (
17). To date, multiple SARS-CoV-2 vaccines have been developed, of which live attenuated vaccines are particularly attractive as they activate all branches of the host immune system. However, preexisting cross-reactive immunity originated from natural contact with other human coronaviruses might potentially restrict their protective activity against SARS-CoV-2 (
18). Moreover, live attenuated vaccines have the potential to revert to a wild-type phenotype causing severe diseases. Compared with live-attenuated virus or some viral vectored vaccines, RBD subunit vaccine could eliminate the concerns of preexisting immunity, virulence recovery, and incomplete inactivation due to specific viral antigenic fragments excluding any components of infectious viruses (
19). Nevertheless, RBD subunit vaccine would necessitate a potent adjuvant or multiple doses to elicit adequate immunogenicity mainly arising from their relatively low immunogenicity due to rapid degradation and clearance (
20).
Influenza pandemics occur when a strain of influenza virus that has a viral surface protein hemagglutinin (HA), to which there is little or no existing immunity, transmits from human to human within the population (
21). Among them, pandemic influenza A H1N1 virus circulates in humans and causes annual epidemics around the world since first emerged in April 2009 (
22). Administration of influenza A vaccine is one effective strategy to prevent infections and severe illnesses. Both live attenuated influenza vaccine and inactivated influenza vaccine were widely used for the public against influenza pandemics. Studies have shown that live attenuated influenza vaccine can reduce the rate of infection as much as 94% and have a higher immunogenicity ability compared to inactivated influenza vaccine (
23). However, live-attenuated viruses have potential safety concerns, such as increased risks of asthma or wheezing (
24). Comparatively, inactivated influenza A virus vaccine exhibited a good safety and intrinsic immunogenicity that provides efficient protection for the individual and moderates the impact of an outbreak of influenza on the community (
25). In particular, clinical analysis indicated that COVID-19 patients who inoculated with influenza A virus vaccine in 2020, even if received after the onset of SARS-CoV-2 infection-related symptoms, had a higher chance of survival and had less need of hospital care than those patients without receiving influenza vaccination (
26,
27).
Inspired by both the outstanding characteristics of inactivated influenza A virus and RBD, we conjugated RBD onto the surface of inactivated influenza A virus, creating a SARS-CoV-2 virus-like particle (VLP) vaccine with potential two-hit protective activity that not only induced the generation of RBD-specific immunoglobulin G (IgG) antibody but also produced HA antigen-specific antibody responses (
Fig. 1). Here, inactivated influenza A (H1N1) virus (Flu) was used to conjugate with RBD (designated as Flu-RBD). We demonstrated that Flu could serve as a safe and flexible platform to improve the retention of RBD in draining lymph nodes (LNs) and trigger a stronger humoral and cellular immunity relative to free RBD and the mixture of unconjugated RBD with Flu. This strengthened the functional immunogenicity of RBD antigen. We tested this vaccine using a hamster model of live SARS-CoV-2 challenge, where two doses of Flu-RBD vaccination were verified to be safe and efficacious, providing protective activities against SARS-CoV-2 infection. Furthermore, we demonstrated that Flu-RBD vaccinations exhibited substantial neutralization ability against SARS-CoV-2 Delta pseudovirus variants and wild-type influenza A H1N1 inactivated virus in mice.
DISCUSSION
To date, none of the 14 recommended treatments such as remdesivir, hydroxychloroquine, and IFN showed significant impact on overall mortality or hospital length of stay in COVID-19 hospitalized patients (
45). Therefore, safe, effective, and widely available vaccines are essential to eliminating COVID-19 pandemic (
46). Despite that multiple SARS-CoV-2 vaccine candidates have been approved for emergency usage by using conventional viral, immunogens, and genetic (DNA and mRNA) approaches (
30,
47,
48), none of them is constructed to protect the public against COVID-19/flu coinfection, which is difficult to distinguish regarding their similar clinical presentations (
6,
49). Of special note, influenza A virus pre-infection strongly enhances the infectivity of SARS-CoV-2 by increasing virus entry, resulting in a higher mortality rate. Some clinical reports indicated that the trained immunity induced by influenza vaccines could lead to some degree of protection against COVID-19. Those who got an influenza vaccine that were 30% less likely to be infected by SARS-CoV-2, and 89% less likely to evolve serious COVID-19 relative to those who did not receive the influenza vaccine (
50). Here, we developed a promising vaccine candidate with two-hit protection against both SARS-CoV-2 and flu infections. RBD is the primary target of neutralization potency in COVID-19 human convalescent sera (
13). RBD-based subunit vaccines have shown to be efficient and promising in preclinical studies by producing a SARS-CoV-2–specific antibody response (
51). Inactivated influenza A H1N1 virus (Flu) vaccine against influenza has been developed for decades and demonstrated to be effective and safe even for pregnant women (
52). Inspired by their advantages, we conjugated the RBD of SARS-CoV-2 with Flu as a two-hit vaccine.
We found that Flu-RBD led to a significantly greater shift in RBD distribution from injection site toward LNs by 24 hours than free RBD in mice. Moreover, Flu-RBD prolonged the retention of RBD in the draining LNs and enhanced the internalization of RBD by APCs, revealing that Flu could function as an efficient platform for delivery of RBD into LNs. Accordingly, Flu-RBD vaccinations produced a higher titer of RBD-specific IgG Abs compared to RBD. Furthermore, Flu-RBD vaccine preserved the immunogenicity of Flu vaccine by generating HA-specific IgG Abs and T cell responses, which play key roles against viral invasion. Collectively, our results showed that Flu-RBD VLP vaccine could evoke both RBD and HA specific immune responses.
We used Syrian golden hamsters, an animal model for studying SARS-CoV-2 pathogenesis and transmission (
37), to investigate the safety and efficacy of Flu-RBD vaccine. Vaccinations of hamsters with two doses of Flu-RBD provided effective protective immunities against SARS-CoV-2 infection as demonstrated by reduced viral load and high level of antibody titers. Histopathological analysis of lung tissues from hamsters vaccinated with Flu-RBD exhibited minimal lung pathology and low numbers of SARS-CoV-2 infected cells as opposed to SARS-CoV-2 infected hamsters, showing intensive damage and stronger SARS-CoV-2 signals.
Continued emergence of SARS-CoV-2 variants poses severe threats to present COVID-19 vaccines, particularly as the Delta variant (B.617.2) that causes sharp rises in infections in many countries, even some with relatively high vaccination coverage (
43). Our results suggested that vaccinations with two doses of Flu-RBD could elicit potent neutralization activity against SARS-CoV-2 Delta pseudovirus. Furthermore, the serum from mice vaccinated with Flu-RBD efficiently blocked the Delta pseudovirus entry into host cells, indicative of its strong cross-protective capacity.
With more scientists predicting that SARS-CoV-2 is expected to become endemic and adopt a seasonal pattern of annual epidemics, several recent clinical trials combing SARS-CoV-2 vaccine with the annual influenza vaccine have been conducted to evaluate their safety and potential against COVID-19/influenza infection (
53–
55). In a phase 3 trial conducted by Novavax, coadministration of NVX-CoV2373 with influenza vaccine has proven safe, with a reactogenicity profile similar to that of either vaccine administered alone (
53). In a phase 4 trial, concomitant administration of ChAdOx1 (Oxford-AstraZeneca) or BNT162b2 (BioNTech) with influenza vaccine did not raise any safety concerns and preserved antibody responses to both vaccines (
54). In another phase 2 trial, interim results showed that no safety concerns for concomitant administration of mRNA-1273 vaccine with influenza vaccine in elderly individuals (
55). Although promising, these strategies were simply combo. We presented an alternative strategy with the development of a single vaccine efficiently targeting two respiratory viruses, SARS-CoV-2 and influenza, which is more cost-effective, enhanced immunogenicity, improved patient convenience, and fewer missed opportunities to vaccinate over the reported strategies.
A limitation of our study is that we did not study the protective activity of Flu-RBD against live Flu infection in an animal model, and future studies are planned to evaluate this question. However, it is worth noting that Flu-RBD vaccine maintained the immunogenicity of flu vaccine by generating HA-specific IgG antibody in both mice and hamsters and HA-specific T cell responses, as well as showing great capacity against wild-type influenza A H1N1 inactivated virus infection. In summary, we developed a two-hit protection vaccine against SARS-CoV-2/flu infections based on RBD conjugated with inactivated influenza A virus. This two-in-one vaccine improves immunogenicity of RBD antigen, reduces vaccine hesitancy, as well as eases the public’s fatigue toward vaccinations and reduces the burden on health care services for vaccine delivery. We expect that our two-hit VLP vaccine will provide important insights into developing safe and effective vaccines for targeting coinfection of circulating viral threats.
MATERIALS AND METHODS
Synthesis of Flu-RBD
Before preparing Flu-RBD, RBD were reacted with AZO-PEG-NHS (PG2-AZNS-3k, NANOCS). Briefly, 10 μg of recombinant SARS-CoV-2 RBD protein (40592-V08B, Sino Biological) was reconstituted and dropped into AZO-PEG-NHS solution (2 mg/ml) and reacted for 24 hours at 4°C to form RBD-PEG-AZO. To remove unreacted AZO-PEG-NHS, the above mixture was dialyzed using Slide-A-Lyzer MINI Dialysis Units [10,000 molecular weight cut-off (MWCO)] and concentrated by Amicon Ultra-4 filter (10-kDa cutoff).
Meanwhile, inactivated influenza A (H1N1) virus (23-047-299, Thermo Fisher Scientific) defined as Flu was reacted with DBCO-PEG-NHS (PG2-DBNS-5k, NANOCS) at 4°C for 24 hours to generate DBCO-PEG-Flu. Unreacted DBCO-PEG-NHS removal was performed by centrifugation via Amicon Ultra-4 filter (100 kDa). To synthesize Flu-RBD, 10 μg of RBD-PEG-AZO was added into 1011 of DBCO-PEG-Flu solution and reacted at 4°C for overnight. The resultant Flu-RBD was ultracentrifugated at 100,000g for 70 min at 4°C and washed with PBS twice (10 mM, pH 7.4) and resuspension in PBS buffer. To quantify the number of RBD protein on Flu-RBD, 105 of Flu-RBD particles were lysed by radioimmunoprecipitation assay buffer and quantification via an ELISA assay (EH492RB, Thermo Fisher Scientific).
Synthesis of RBD-RhB and Flu-RBD-RhB
One hundred micrograms of NHS-RhB (186-1425, Thermo Fisher Scientific) was reacted with 10 μg of RBD in PBS buffer at 4°C overnight. Free NHS-RhB removal was performed by centrifugation and washed with PBS buffer three times via Amicon Ultra-0.5 filter (10-kDa cutoff). RBD-RhB was mixed with AZO-PEG-NHS to generate RBD-PEG-AZO-RhB and then reacted with DBCO-PEG-Flu to synthesize Flu-RBD-RhB according to above methods.
Preparation of mRNA of full-length spike protein and its loading in lipids
The S protein plasmid for in vitro transcription was created using the full-length DNA sequence of the trimerized S surface glycoprotein (Gene Bank: QHD43416.1; amino acids 1 to 1273) of the SARS-CoV-2 isolate Wuhan-Hu-1. The mRNA construct of S protein was prepared, extracted, and purified according to our previous report (
56). Purified mRNA was formulated with lipids (ALC-3015, ALC-0159, 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC), and cholesterol) using the ethanol-drop process to obtain RNA-lipid nanoparticles (LNP@mRNA) according to Pfizer’s report (
32). The molar ratios of the ALC-3015:ALC-0159:cholesterol:DSPC used are 46.3:1.6:42.7:9.4, and each dose of LNP@mRNA vaccine consists of 0.43 mg of ALC-3015, 0.05 mg of ALC-0159, 0.09 mg of DSPC, 0.19 mg of cholesterol, and 30 μg of mRNA according to the released document from Pfizer (
34).
Animal procedures
With respect to World Organization for Animal Health recommendation, CD1 mice are strain for the vaccine potency assay in mouse (
57), which have been used for evaluating the immunogenicity of various vaccines including RB51 vaccine, Zika vaccine, PCV15 vaccine, etc. Accordingly, female CD1 mice [Crl:CD1(ICR)] that are 6 to 8 weeks old and purchased from Charles River Laboratory (MA, USA) were used for assessing the immunogenicity of Flu-RBD vaccine. All animal studies were compliant with the Institutional Animal Care and Use Committee (IACUC) at North Carolina State University (protocol # 19-806-B). RBD-RhB and Flu-RBD-RhB were administered into the right hint limb of mice via i.m. injection. After 4 or 24 hours postinjection, the right, left, and cervical draining LNs as well as major organs were harvested and imaged via the Xenogen Live Imager (PerkinElmer, MA, USA).
PBS, Flu (1010/kg), RBD (0.71 μg/kg), Flu-RBD (1010/kg), and the mixture of Flu and RBD (Flu-M-RBD) vaccinations were i.m. injected with two doses 1 week apart. Seven days later, blood, spleen, BALF, and LNs were collected for further analysis. To evaluate the immune responses of booster, CD1 mice administrated with two doses of Flu-RBD (prime) were received a homologous booster dose 4 weeks later. Seven days later, the blood of mice was collected for further analysis.
IgG antibody titer
One hundred microliters of RBD or influenza A H1N1 HA recombinant protein (4 μg/ml; A42579, Thermo Fisher Scientific) was added into micro titer plates (Nunc Cell Culture, Thermo Fisher Scientific) for overnight at 4°C. After that, 200 μl of 5% (w/v) bovine serum albumin (B6717, Sigma-Aldrich) in phosphate-buffered saline with 0.1% Tween 20 (PBS-T) buffer was added to each well to block nonspecific binding and incubated at 37°C for 1 hour. After three times of wash with PBS-T, serum samples with serial dilutions (1:100, 1:1000, 1:3000, 1:5000, 1:10,000, 1:50,000, 1:100,000, and 1:200,000) and control serum samples (1:100 dilution) were added to the wells and incubated at 37°C for 1.5 hours. After washing intensively, 100 μl of horseradish peroxidase–labeled anti-mouse IgG secondary antibody with 1:10,000 dilution was added for 1 hour of incubation at 37°C. After three times of wash, 100 μl of trimethylboron was added into each well and incubated at room temperature for 30 min. Fifty microliters of 2 M H2SO4 (Sigma-Aldrich) was used for stopping the reaction, and the optical absorption was determined at 450 nm via a plate reader. The endpoint IgG titer was determined as the highest reciprocal dilution of serum that exhibits an optical density greater than twofold of the mean control group.
Cytokine measurements in splenocytes
Splenocytes from each vaccinated mouse were seeded in an ELISpot plate (EL485, R&D Systems) (106 cells per well) and stimulated with RBD peptide pool (PP002-A, SinoBiological) and H1N1 HA peptide pool (130-097-285, Miltenyi Biotec). Antigen-specific cells secreting IFN-γ were detected according to the protocol of the manufacturer. The formed SFUs were imaged by an anatomical microscope (Nikon, Minato City, Tokyo, Japan) and counted by ImageJ software. Furthermore, the collected splenocytes were also plated in six-well plates (5 × 106 per well) and restimulated with RBD peptide pool and H1N1 HA peptide pool. After 48 hours of incubation, the culture supernatant was collected. The Mouse IL-6 Kit (RAB0308, Sigma-Aldrich) and the Tumor Necrosis Factor α Kit (RAB0477, Sigma-Aldrich) were used to detect IL-6 and TNF-α levels from collected supernatant.
T cell immunities in splenocytes
At day 7 postprime, mouse splenocytes were collected and seeded to the plate (1 × 106 per well). The RBD peptide pool or influenza A HA peptide pool was added to stimulate splenocytes for 12 hours at 37°C, respectively. Brefeldin A solution was added into the plate with a final concentration of 1 μg/ml and incubated for another 4 hours. Following that, the splenocytes were collected and stained with CD4–fluorescein isothiocyanate (FITC) (100406, BioLegend) antibody or CD8-FITC (ab22504, Abcam) antibody at a dilution of 1:100. After washing with magnetic-activated cell sorting flow buffer, splenocytes were fixed with 4% paraformaldehyde (PFA) and permeabilized using saponin and incubated with IFN-γ–phycoerythrin (PE) (507806, BioLegend), IL-4–PE (504103, BioLegend), or IL-17a–APC (17-7177-81, Invitrogen) Abs for 1 hour. The stained splenocytes were analyzed by a CytoFLEX flow cytometer (Beckman Coulter). Data analyses were carried out by FCS Express V6.
Syrian golden hamster studies with live SARS-CoV-2
Six to 8 weeks old of 20 male and female Syrian golden hamsters (Envigo) were randomly assigned to four groups and housed at Bioqual. Hamsters were i.m. administered with two doses of PBS, Flu (1010/kg) or Flu-M-RBD, Flu-RBD (1010/kg) at week 0 and week 1 (n = 5 per group, 2 females/3 males). At week 2, the hamsters were challenged with 100 μl of 2 × 104 50% tissue culture infective dose SARS-CoV-2 via intranasal routes (50 μl in each nare). BAL, OS, and blood were harvested at the indicated time. All hamster experiments were conducted in accordance with all relevant local, state, and federal regulations that were approved by the Bioqual IACUC (20-091P).
Viral load assay
Following procedures previously described by Duke Human Vaccine Institute (
58), SARS-CoV-2 RNA copies per milliliter were tested by qRT-PCR assay. Briefly, viral RNA was extracted using the QIAsymphony SP sample preparation platform with the DSP Virus/Pathogen Midi Kits (Qiagen). A primer specific to SARS-CoV-2 envelope gene was annealed to extracted RNA, which was reverse-transcribed into cDNA using SuperScript III Reverse Transcriptase and RNaseOut. The cDNA was treated with ribonuclease and then added to the custom 4× TaqMan Gene Expression Master Mix including primers and a fluorescently labeled hydrolysis probe. The qRT-PCR reaction was performed on a QuantStudio 3 Real-Time PCR system. SARS-CoV-2 RNA copies per reaction were interpolated via quantification cycle data and a serial dilution of a highly characterized custom DNA plasmid with SARS-CoV-2 envelope gene sequence. The limit of quantification for the assay was approximately 62 RNA copies/ml of sample.
SARS-CoV-2 Delta pseudovirus neutralization assay
SARS-CoV-2 Delta pseudoviruses were constructed by cotransfecting HEK293T cells with the plasmids of plv-spike-v8 (InvivoGen), pLenti-EF1pluciferase-PGK-RFP-T2A-PURO lentiviral reporter (LR252, ALSTEM), and pspax2 (64586, addgene) via Lipo3000 (L3000015, Thermo Fisher Scientific). After 48 to 72 hours, Delta pseudoviruses were harvested from culture medium through centrifugation (3000 rpm, 10 min), aliquoted, and stored at −80°C before use.
Seven-week-old female CD1 mice were vaccinated with two doses of PBS, RBD, Flu, Flu-M-RBD, or Flu-RBD once a week. Seven days later, mice were inoculated with SARS-CoV-2 Delta pseudovirus carrying both RFP and luciferase reporters. Lung organs were harvested and imaged at 24 hours postchallenge by the Xenogen Live Imager. To study the distribution of Delta pseudovirus in the lung, the collected lung tissues were fixed in 4% PFA, dehydrated using 30% sucrose solution, and then frozen in O.C.T. compound (Tissue-Tek). Cryo-sections were permeabilized and blocked with Dako containing 0.1% saponin solution for 1 hour and then stained with phalloidin antibody (ab176753, Abcam). The ProLong Gold Antifade reagent with DAPI was used to counterstain nuclei and prevent the fade of the fluorophore. Imaging was performed via Olympus FLUOVIEW CLSM.
The sera samples from vaccinated mice were collected to assess their neutralization activity against SARS-CoV-2 Delta pseudovirus. A549 cells expressing human ACE2 and transmembrane serine protease 2 (TMPRSS2) were obtained from InvivoGen (a549-hace2tpsa). Sera samples with the indicated dilutions were incubated with SARS-CoV-2 Delta pseudovirus for 1 hour at 37°C. After that, the mixture was added in the A549 cells and incubated for another 24 hours. The luciferase signals from infected cells were determined by the Dual-Glo Luciferase Assay System (Promega). Furthermore, the A549 cells infected with the mixture of SARS-CoV-2 Delta with serum at a ratio of 1:1 were collected for flow cytometry analysis via the RFP reporter.
Wild-type influenza A H1N1 inactivated virus clearance assay
Wild-type influenza A H1N1 (VNV-019) inactivated virus was purchased from Creative Biogene and labeled with 1,1-Dioctadecyl-3,3,3,3-tetramethylindodicarbocyanine (DID) dye (V22887, Thermo Fisher Scientific). Seven-week-old female CD1 mice were i.m. administered with PBS, Flu-RBD, LNP@mRNA, or LNP@mRNA plus Fluz in two doses once a week. Seven days after two vaccinations, every mouse was challenged with H1N1 inactivated virus with red fluorescence. Lung organs were harvested and imaged at 24 hours postadministration by the Xenogen Live Imager. Immunofluorescence imaging of lung tissues was performed by an Olympus FLUOVIEW CLSM.
Statistical analysis
All experiments were performed at least three times independently. Results are shown as means ± SD. Comparisons between any two groups were performed using the two-tailed, unpaired Student’s t test. Comparisons among multiple groups were performed using one-way analysis of variance (ANOVA), followed by the post hoc Bonferroni test. Group data were analyzed by two-way ANOVA with Tukey’s multiple comparisons test. Single, double, triple, and four asterisks represent P < 0.05, 0.01, 0.001, and 0.0001, respectively; P < 0.05 was considered statistically significant.
Acknowledgments
We thank S. Kar, M. Porto, M. Lok, and B. Spence at BIOQUAL Inc. for performing the hamster studies. We thank N. De Naeyer, C. T. DeMarco, and T. N. Denny at Duke Human Vaccine Institute for performing viral load assay. We are indebted to J. Chung for synthesis of spike mRNA.
Funding: This work was supported by grants from the National Institutes of Health (HL123920, HL144002, HL146153, HL147357, HL149940, and HL154154 to K.C.; HL164998 to P.C.D) and the American Heart Association (19EIA34660286 to K.C.).
Author contributions: K.C., Z.W., and Z.L. conceived the idea. Z.W., K.C., and Z.L. designed the overall experiments. Z.W., Z.L., W.S., and D.Z. performed the experiments. Z.W. and K.C. wrote the article. W.S., S.H., and P.-U.C.D. reviewed and edited the article. K.C. supervised this research. All authors read and approved the final article. All authors provided the corresponding author with written permission to be named in the article.
Competing interests: The authors declare that they have no competing interests.
Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials.