Polar functional group-containing glycolipid CD1d ligands modulate cytokine-biasing responses and prevent experimental colitis

The MHC class I-like molecule CD1d is a nonpolymorphic antigen-presenting glycoprotein, and its ligands include glycolipids, such as α-GalCer. The complexes between CD1d and ligands activate natural killer T cells by T cell receptor recognition, leading to the secretion of various cytokines (IFN-γ, IL-4, IL-17A, etc.). Herein, we report structure–activity relationship studies of α-GalCer derivatives containing various functional groups in their lipid acyl chains. Several derivatives have been identified as potent CD1d ligands displaying higher cytokine induction levels and/or unique cytokine polarization. The studies also indicated that flexibility of the lipid moiety can affect the binding affinity, the total cytokine production level and/or cytokine biasing. Based on our immunological evaluation and investigation of physicochemical properties, we chose bisamide- and Bz amide-containing derivatives 2 and 3, and evaluated their in vivo efficacy in a DSS-induced model of ulcerative colitis. The derivative 3 that exhibits Th2- and Th17-biasing responses, demonstrated significant protective effects against intestinal inflammation in the DSS-induced model, after a single intraperitoneal injection.

Invariant natural killer T (iNKT) cells expressing an invariant α chain T cell receptor (TCR; Vα14-Jα18 in mice and Vα24-Jα18 in humans), play a critical role in regulating various immune responses 1,2 . iNKT cells recognize glycolipid ligands present in the nonpolymorphic MHC class I-like molecule CD1d protein by using TCRs, and then are activated. iNKT cell activation induces the secretion of various cytokines including Th1 (e.g. IFN-γ), Th2 (e.g. IL-4) or Th17 cytokines (e.g. IL-17A). Th1 cytokines are involved in the enhancement of inflammatory responses, and mainly induce cellular immune responses leading to antitumor 3,4 or antiviral effects 5,6 . Thus, controlling Th1 responses has recently attracted significant attention as a therapeutic target for a range of cancers and viral infections. Th2 cytokines promote humoral immunity by participating in the antibody-mediated immunity control of extracellular pathogens. The production of Th2 cytokines is also associated with the amelioration of certain autoimmune diseases (e.g., IBD, type 1 diabetes and multiple sclerosis) [7][8][9] . Th17 cytokines can contribute to host defense against extracellular pathogens such as Streptococcus pneumoniae [10][11][12][13] . Thus, controlling Th17 cytokines can facilitate the development of vaccine adjuvants against pneumococcal infection. However, the details of the role Th17 cytokines in the CD1d-NKT system remain unclear. In addition to each characteristic, these cytokines can positively or negatively interact with each other, leading to the modulation of various immune responses. For instance, the Th1 and Th2 signaling pathways can negatively cross-regulate each other 14 . The Th1 and Th17 cytokines demonstrate synergistic behavior, leading to protection against extracellular pathogens, although in some cases, the disturbance of Th1 and Th17 can result in immunopathology 12  www.nature.com/scientificreports/ cytokine induction levels and/or balance is a key factor in the development of effective immunotherapies [15][16][17] . Therefore, multiple research groups have focused on the selective regulation of cytokine production 14 . Importantly, previous structure-activity relationship studies (SAR) on CD1d ligands have shown that differences in the ligand structure can influence cytokine polarization from NKT cells 14 , although the underlying mechanism is not fully understood. Thus, to date, many researchers have actively explored and developed novel CD1d ligands on the basis of SAR studies of glycolipid ligands. Representative CD1d ligands, namely, α-galactosyl ceramide (α-GalCer, KRN7000) 18 , 7DW8-5 19 , OCH 20 and a phytosphingosine-modified α-GalCer derivative 21 are shown in Fig. S1. α-GalCer, one of the most potent ligands, induces high level production of various cytokines, and has had good therapeutic effects in some models, including cancer 18 or autoimmune models 9 . 7DW8-5 displays Th1 polarization and has been identified as a lead candidate for HIV and malaria vaccine adjuvants 19 . OCH is an α-GalCer variant containing a truncated sphingosine moiety, and it shows IL-4 selectivity. OCH has been effective in autoimmune models such as a murine experimental autoimmune encephalitis (EAE) model 20 or a murine dextran sulfate sodium (DSS) model of ulcerative colitis (UC) 22 . The phytosphingosine-modified α-GalCer derivative induced greater secretion of a Th2 cytokine, IL-4, over those of IFN-γ and IL-17, and was effective in the EAE model 21 .
In this context, we conducted SAR studies based on α-GalCer to obtain the potent and/or cytokine-biasing ligands [23][24][25] . In particular, we focused on a large hydrophobic pocket of CD1d (A' pocket) that recognizes long fatty acyl chains of glycolipid ligands and developed CD1d ligands with polar functional groups such as amides or amines that display potent cytokine induction levels and Th2 cytokine polarization. Our molecular dynamics (MD) simulation studies predicted that amide groups on the ligands can interact with confined polar amino acid residues, Gln14 and Ser28 in the large hydrophobic pocket through direct and/or water-bridged hydrogen bonds 23 . However, the ligand-recognition mechanism of the lipid binding sites is still not understood, and the functional groups introduced onto the fatty acyl chains have not been fully optimized.
In this study, we conducted a detailed SAR study of α-GalCer derivatives containing functional groups in their long fatty acyl chains to obtain new potent ligands and elucidate the ligand-recognition mechanism of lipid binding. Additionally, their cytokine production selectivity (for IFN-γ, IL-4 and IL-17A) was investigated. We also evaluated their physicochemical and pharmacokinetic (PK) profiles and in vivo efficacy in a murine autoimmune model, a DSS model of inflammatory bowel diseases (IBD).

Results and discussion
Ligand design. We have previously reported several α-GalCer derivatives (1, 2, 3, and 6) containing amide and amine groups in their lipid moieties, as shown in Fig. 1 (1a: C26:0 α-GalCer derivative having one amidecontaining acyl chain, 1b: C20:0 α-GalCer derivative having one amide-containing acyl chain, 2: bisamidecontaining α-GalCer derivative having two amide-containing acyl chain, 3: Bz amide-containing α-GalCer derivative having one amide containing terminal Ar group, 6a: C26:0 α-GalCer derivative having one aminecontaining acyl chain, and 6b: C20:0 α-GalCer derivative having one amine-containing acyl chain) 23,24 . The cytokine induction levels of all the ligands were higher or equal to that of C26:0 α-GalCer. Based on these findings, we designed CD1d ligands containing polar functional groups (4a: N-methyl amide, 5a: urea and 7a: carbonyl groups) to obtain highly potent and/or cytokine-biasing ligands and investigate the recognition of their polar residues by the A' pocket in detail (Fig. 1). Furthermore, their corresponding C20:0 analogues (4b, 5b and 7b) were also prepared. Previous computational studies, including MD simulations, indicated that the formation of hydrogen bonds (shielded hydrogen bonds) between the amide group of the ligands and the polar residues in the hydrophobic A' pocket of CD1d might enhance the recognition of the ligands by CD1d 23 . In contrast, the introduction of amide groups to saturated long acyl chains would partially rigidify the flexible lipid chains due to the planar structure of amide bonds, and this would mostly likely affect the interaction of these chains with CD1d. Thus, we also designed trans/cis-alkene-containing ligands 8 and 9 to elucidate the relationship between lipid conformation and CD1d affinity.
Synthesis of the ligands. The synthesis of the α-GalCer derivatives containing polar and alkene groups is shown in Fig. 2. First, we synthesized N-methyl amide derivatives 4a and 4b (Fig. 2a). The selective methylation of Ns amide derivative 10 24 , removal of the Ns group and acylation with tridecanoic acid or heptanoic acid gave protected N-methyl amide derivatives 11a and 11b. Cleavage of the Bn groups with Pd(OH) 2 /C afforded desired N-methyl amide derivatives 4a and 4b. The preparation of urea analogues 5a and 5b starting from known compound 12 24 is shown in Fig. 2b. Removal of the Boc group under acidic conditions provided the corresponding amine, which was treated with the appropriate isocyanates (C 11 H 23 NCO or C 5 H 11 NCO) to give protected urea derivatives 13a and 13b. Then, the protected urea derivatives were converted to desired products 5a and 5b. Next, we examined the synthesis of carbonyl derivatives 7a and 7b (Fig. 2c). The required fatty acids containing carbonyl groups were prepared from known dicarboxylic acid derivative 14 26 . After the activation of monoacid 14 with i-BuOCOCl, treatment with methylenetriphenylphosphorane gave the corresponding phosphorane intermediate, which was reacted with the appropriate aldehydes to provide carbonyl compounds 15a and 15b 27 . Removal of the Bn groups and reduction of the alkene moieties furnished desired carbonyl-containing products 16a and 16b. With the desired fatty acid derivatives in hand, the sphingosine fragments were installed. The reduction of known azide 17 28 followed by condensation with 16a and 16b and removal of the Bn groups provided carbonyl derivatives 7a and 7b. Finally, we focused on the synthesis of the alkene-containing derivatives (Fig. 2d). We planned to selectively synthesize both trans/cis-alkene-containing derivatives 8a, 8b, 9a and 9b to examine the differences in the affinities of these isomers toward CD1d. Propargyl alcohol 19 was converted to alkynes 20a and 20b in moderate yields (4 steps) 29,30 . After investigating the trans-selective reduction of alkyne 20, we found that the treatment of 20 with LiAlH 4 in diglyme and THF at 130 °C gave the alcohol containing www.nature.com/scientificreports/ trans-alkene moieties 30 , and subsequent TEMPO oxidation afforded desired fatty acids 21a and 21b. In contrast, the cis-selective reduction of alkyne 20 was accomplished using nickel boride under a H 2 atmosphere 31 . The subsequent removal of the TBS groups and TEMPO oxidation provided desired fatty acids 22a and 22b. With the alkene-containing fatty acids in hand, we next attempted to convert the glycolipid forms starting from known azide 17. Cleavage of the Bn groups, reduction of the azide group and selective N-acylation with 21 and 22 in the presence of HATU furnished alkene-containing α-GalCer derivatives 8a, 8b, 9a and 9b.

The immunological evaluation of the α-GalCer derivatives in vitro.
With the α-GalCer derivatives in hand, we next performed in vitro immunological evaluations of the CD1d-NKT cell axis based on the following: (1) an antigen-presenting cell (APC)-free assay 32-34 , (2) a binding assay using AlphaScreen 24 , (3) the evaluation of cytokine production by mouse splenocytes. We used the APC-free assay to investigate the agonistic activities and binding characteristics of these compounds to CD1d. In particular, the stability of the ligand-CD1d complexes and the binding potential of the ligand-CD1d complex to NKT-TCR can be evaluated. The binding assay using AlphaScreen can reveal the binding affinities of these compounds for CD1d. The evaluation of cytokine production by mouse splenocytes was conducted to examine cytokine polarization through the CD1d-NKT cell system. The previously reported derivatives, amide-containing derivative 1 and amine-containing derivative 6, were also evaluated in these assays to allow comparisons between different functional groups. Initially, we conducted the APC-free assay using the synthesized α-GalCer derivatives ( Fig. 3a-d). The APCfree assay with derivatives 1a, 1b, 2, and 3 has already been reported, and the results demonstrated that all four derivatives resulted in higher levels of cytokine production relative to that caused by α-GalCer 23 .
The α-GalCer derivatives containing polar functional groups in their long fatty acyl chain (C26:0) were tested as shown in Fig. 3a. The introduction of any of the tested polar functional groups enhanced the cytokine production levels in a dose-dependent manner. Notably, the modification of the amide group to an N-methyl amide group (4a) led to an improvement in the cytokine production activity at 10 nM concentration (exact values of IL-2; 1a: 177 ± 27 vs. 4a: 749 ± 92). Urea-containing derivative 5a showed activity equal to that of amide-containing derivative 1a (1a: 177 ± 27 vs. 5a: 85 ± 26 at 10 nM). The introduction of amine or carbonyl groups, which are considered to be more flexible than amide or urea groups, resulted in potencies slightly higher than those of amide-or urea-containing derivatives at 10 nM concentration (1a: 177 ± 27 vs. 6a: 323 ± 41 and 7a: 424 ± 32). Next, we examined the binding potential of derivatives 1a, 4a, 6a and 7a using AlphaScreen system previously reported by our group 24 . The dose response curves and IC50 values of these compounds are shown in Fig. 3e, f. All the polar functional group-containing derivatives exhibited higher binding affinities than that www.nature.com/scientificreports/ of α-GalCer. N-Methyl amide-containing derivative 4a had a higher binding affinity for mCD1d. These results are consistent with those of the APC-free assay. In terms of the C20:0 α-GalCer derivatives, the introduction of polar functional groups significantly increased the cytokine production activities (Fig. 3b). Unlike the α-GalCer derivatives containing a long fatty acyl chain (C26:0), N-methyl amide-, urea-, and carbonyl-containing derivatives 4b, 5b, and 7b exhibited activities comparable to that of amide-containing derivative 1b (1b: 662 ± 46 vs. 4b: 546 ± 87, 5b: 376 ± 92 and 7b: 600 ± 20 at 10 nM). Interestingly, carbonyl-containing derivative 7b caused higher levels of cytokine production than did amine-containing derivative 6b (6b: 74 ± 26 vs. 7b: 600 ± 20 at 10 nM). These results support our previously reported MD simulation studies that suggested that the carbonyl group of the amide group is mainly associated with the Gln14 or Ser28 residues of mCD1d through hydrogen bond formation. The reduced activity of amine derivative 6b might be due to its inability to form a strong hydrogen bonding network. Taken together, the agonistic activities and binding characteristics are dependent on the introduction of polar functional groups. Furthermore, the effects of each polar functional group on the activities varied depending on the length of the acyl chain. We next investigated the relationship between the lipid conformation and the binding characteristics when using alkene-containing derivatives (Fig. 3c, d). Interestingly, the presence of alkene groups enhanced the levels of IL-2 production (α-GalCer: 274 ± 53 vs. 8a: 1626 ± 371 and 9a: 1556 ± 91 at 100 nM). Although significant differences in activities were not observed between the trans-alkenes and cis-alkenes, the cis-alkene-containing C20:0 derivatives tended to slightly increase cytokine production compared with the corresponding trans-alkenecontaining derivative (8b: 334 ± 19 vs. 9b: 925 ± 57 at 100 nM). The amide-containing derivative had higher levels of cytokine induction than the corresponding alkene-and alkane-containing derivatives (1a: 2,502 ± 114 vs. 8a: 1626 ± 371 and α-GalCer: 274 ± 53 at 100 nM). These results indicated that lipid conformation could be an important factor for enhancing the stability of the ligand-CD1d complexes. Correspondingly, when considering that the introduction of an amide group to the acyl chain enhanced the cytokine production levels, the planarity of the amide bonds (conformational restriction of amides) could affect the CD1d recognition as well as the formation of hydrogen bonds between the amide group and the polar residues of CD1d, as supported by our previous reported computational studies 23 . Taken together, the order of the binding potentials upon introduction of various functional groups was N-methyl amide > amide > alkene > alkane. The binding of lipid acyl chains with amide groups to CD1d might be primarily affected by the rigidity and polarity of the chains.

Evaluation of water solubility and metabolic stability in vitro and phamacokinetics (PK) profiles.
With the in vitro immunological data in hand, we next evaluated the water solubility and metabolic stability of the α-GalCer derivatives (Table S1) to select compounds for further in vivo evaluation. Although many SAR studies focusing on α-GalCer as a CD1d ligand have been reported, little information is available about their ADME (absorption, distribution, metabolism and excretion) properties. Moreover, α-GalCer is known to exhibit limited solubility in aqueous solvents due to the high lipophilicity of its two long lipid chains. Indeed, α-GalCer is extremely insoluble in aqueous buffer, and even DMSO. Thus, it requires specialized techniques, such as sonication or heating, to obtain reproducible results. On the other hand, our newly deigned ligands were expected to offer better water solubility due to the introduction of polar functional groups into the hydrophobic lipid chains. The solubilities of α-GalCer and a series of α-GalCer derivatives were evaluated (Table S1). Some compounds, including α-GalCer, showed poor solubility in this examination (< 5 μM), meaning that their levels were below the detection limits. The C20:0 amide-and Bz amide-containing derivatives (1b and 3) exhibited measurable solubilities (5 and 7.2 μM, respectively). Next, the metabolic stabilities of the derivatives were tested using human and rat liver microsome assays (Table S1). Although the C26:0 amide-, C20:0 amide-and aminecontaining derivatives (1a, 1b and 6a) displayed good metabolic stability for both species, the N-methyl amide derivative (4a) was metabolized more rapidly. The bisamide-and Bz amide-containing derivatives (2 and 3) were resistant to rat microsomal metabolism.
Next, the PK profile of the glycolopid ligand in mice was evaluated as shown in Fig. 6. A limited number of studies have reported on PK profiles of glycolipids including α-GalCer 40 . As shown in Fig. 6a, despite its relatively large size (MW = 781.1) and high lipophilicity, a measurable plasma concentration was achieved after intraperitoneal injection of 100 μg/kg of 3 (AUC = 466 ng•h/mL; T 1/2 = 5.2 h; T max = 1.0 h). 41,42 . NKT cell activation by CD1d ligands prevented the murine DSS model of UC, which is a clinical manifestation observed in chronic inflammatory diseases of the gastrointestinal (GI) tract (IBD) 43 . In particular, the potent CD1d ligands, α-GalCer and OCH (a representative Th2-biasing ligand) were effective in the murine DSS model 22 .   www.nature.com/scientificreports/ using the APC-free assay as shown in our previous report 23 . Furthermore, from a synthetic point of view, the Bz amide-containing derivative 3 was the attractive candidate, because amide-containing derivatives are easier to synthesize than the amine-or carbonyl-containing derivatives (6b or 7b) that also showed Th2 and Th17-biasing responses. Indeed, compared to the Bz amide-containing derivative 3, two or more extra steps were required to synthesize the derivatives 6b or 7b. Additionally, we also selected bisamide-containing derivative 2 that displayed relatively less selective cytokine responses but potent cytokine induction activities. The evaluation of the derivatives using DSS-induced model was carried out according to the procedure reported by Ueno et al. 22 in which α-GalCer and OCH were administered intraperitoneally, and demonstrated their in vivo efficacy in the model. DSS treatment causes progressive weight loss, diarrhea and colon inflammation with a significant reduction in colon length, which leads to increases in the clinical scores of disease symptoms (DAI: disease activity index). C57BL/6 mice (n = 12 per group) were continuously treated with 2.0% DSS in their drinking water from day 1 to day 7. The mice were injected intraperitoneally with 100 μg/kg of the derivatives or PBS (control) on day 4. On day 8, the DAI scores and colon lengths of the mice administered the derivatives were compared with those of the mice administered PBS (control). Two out of twelve mice treated with bisamide-containing derivative 2 (less selective but potent ligand) had died by day 8. As shown in Fig. 7, treatment with the Bz amide-containing derivative 3 (Th2-and Th17-biasing ligand) resulted in a significant improvement in the DAI scores (P < 0.05). In contrast, treatment with the bisamide-containing derivative 2 did not significantly alter the DAI scores to control group. Colon shortening is a feature of colon inflammation in the DSS-induced model. In the model, the colon length of healthy mice is about 8 cm (DSS (−)), whereas that of mice upon DSS treatment is about 4 cm (DSS + PBS) as shown in Fig. 7b. Relative to PBS (control), treatment with both derivatives 2 and 3 significantly prevented the colon shortening caused by DSS (P < 0.05). As such, Th2-and Th17-biasing derivative 3 displayed protective effects against intestinal inflammation after a single intraperitoneal injection. In this study, we selected the intraperitoneal injection as the administration method of lipid ligands, and used DAI scores and colon length to investigate the protective effects of derivatives against colitis. On the other hand, the difference in the administration method of immunomodulators, such as lipid ligands, might have a significant impact on their efficacy 46,47 . In addition, the cytokine profiles in the colon upon ligand treatment could also provide useful information regarding their efficacy. Further research is currently being conducted into the effects of administration methods on efficacy and the examination of colon cytokine profiles.

Conclusion
In this study, we conducted a detailed SAR study of α-GalCer derivatives containing various functional groups in their lipid acyl chains and identified several potent CD1d ligands that effected cytokine induction or showed cytokine polarization. In particular, the N-methyl amide-containing ligand is among the most potent CD1d ligands reported to date. The studies have indicated that the binding affinities and/or total cytokine production levels could depend on the flexibility of lipid part in addition to hydrophilic interactions (e.g. hydrogen bond formation). We also demonstrated the potential use of amide-containing lipids as the isosteres of alkene-containing lipids. Many alkene-containing lipid moieties, such as naturally occurring fatty acids and sphingosines, are known to exhibit activities distinct from those of the corresponding saturated lipids. Our approaches might be applicable to SAR studies to effectively identify chemical probes for elucidating the function of lipid binding proteins, or drug candidates.
In terms of cytokine polarization, most of the ligands containing polar functional groups displayed Th2-and Th17-biasing immune responses. Intriguingly, the ligands containing the Bz-amide group and flexible functional groups such as amine and carbonyl groups tended to show Th2-and Th17-biasing profiles.
Finally, based on our in vitro immunological evaluation and investigation of the physicochemical properties, we selected two ligands, bisamide-and Bz amide-containing derivatives 2 and 3, and evaluated their in vivo efficacies in a DSS-induced model of UC. A single dose of derivative 3 (Th2-and Th17-biasing ligand) significantly www.nature.com/scientificreports/ attenuated the severity of disease in DSS-induced colitis. Therefore, our designed Th2-and Th17-biasing ligands could be used to regulate autoimmune disorders through the CD1d-NKT cell axis.
The amide-containing ligands were synthesized as described in the Supplementary Information. All lipids were dissolved in DMSO and stored as aliquots at − 30 °C.
Cell culture. NKT  Pharmacokinetic profiles in mice. The pharmacokinetic study was approved by The Institutional Animal Care and Use Committee of Ono Pharmaceutical Co., Ltd. and was performed in accordance with the ethics criteria contained in the animal welfare bylaws of the Committee. The pharmacokinetics of 3 were studied following intraperitoneal administration. The 10 μg/mL dosing solution, which contained DMSO/Phosphate buffered saline = 1/9, was administered intraperitoneally to six male C57BL/6 mice (Charles River Laboratories, Japan, Inc.). Blood samples were collected into heparinized syringes at designated times and centrifuged to obtain the plasma samples. The plasma samples were assayed after protein precipitation by acetonitrile/ethanol (70/30, v/v), followed by HPLC/MS/MS analysis. Plasma concentration-time data were analyzed by noncompartmental methods. www.nature.com/scientificreports/ Efficacy of in vivo CD1d ligand treatment in DSS-induced colitis. The animal experiments using murine DSS model were conducted in accordance with the guidelines of Keio university institutional animal care and use committee. Specific pathogen-free C57BL/6 mice were purchased from Japan Clea (Tokyo, Japan). All mice were housed under specific pathogen-free conditions in microisolator cages in an animal facility, and male mice (7 weeks of age) were used. DSS (dextran sulfate sodium) was added to the water supply provided to the animals at a concentration of 2.0% (wt/vol) for days 1-7. Derivatives 2 and 3 were first dissolved in DMSO at 100 μg/mL and then diluted in PBS solution. The derivatives (100 μg/kg in 200 μL of solution) were injected intraperitoneally on day 4 after the induction of colitis. Control animals received 200 μL of PBS solution containing the same concentration of DMSO. Mice were sacrificed on day 8 after DSS administration. Intestinal tissues were removed and opened longitudinally. The length of the colon was measured after exclusion of cecum. The disease activity index (DAI), defined as stool blood, stool form and weight loss, and the length of the colon were used to analyze the therapeutic benefits of derivatives 2 and 3.

Statistical analysis.
All the results were expressed as the mean ± SEM. All statistical analyses were performed with GraphPad Prism 7.0 software package. Statistical analyses of data were performed by using the oneway ANOVA test when more than 2 groups were compared with post-hoc analysis (Tukey's Test) to determine statistical significance. Statistical significance was set at a p value of less than 0.05.

Reporting summary. Further information on the research design is available in the Nature Research
Reporting Summary linked in this article.

Data availability
Scientific RepoRtS | (2020) 10:15766 | https://doi.org/10.1038/s41598-020-72280-4 www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.