|Home | About | Journals | Submit | Contact Us | Français|
QS-21 is a potent immunostimulatory saponin that is currently under clinical investigation as an adjuvant in various vaccines to treat infectious diseases, cancers, and congnitive disorders. Herein we report the design, synthesis, and preclinical evaluation of simplified QS-21 congeners to define key structural features that are critical for adjuvant activity. Truncation of the linear tetrasaccharide domain revealed that a trisaccharide variant is equipotent to QS-21 while the corresponding disaccharide and monosaccharide congeners are more toxic or less potent, respectively. Modification of the acyl domain in the trisaccharide series revealed that a terminal carboxylic acid is well-tolerated while a terminal amine results in reduced adjuvant activity. Acylation of the terminal amine can restore adjuvant activity and enables the synthesis of fluorescently-labeled QS-21 variants. Cellular studies with these probes revealed that, contrary to conventional wisdom, the most highly adjuvant active of these fluorescently-labeled saponins does not simply associate with the plasma membrane, but rather is internalized by dendritic cells.
There is an ever increasing need to develop effective vaccines to combat acute and chronic infections and diseases, in both therapeutic and prophylactic settings. The traditional use of live, attenuated pathogens as immunogens in vaccines has proven successful in addressing a host of maladies, yet there are many pathogens and diseases for which this strategy has proven unsuccessful. As a consequence, many contemporary approaches to vaccine development employ recombinant or synthetic subunit vaccines, which offer improved safety and more precise targeting. However, they are often characterized by poor immunogenicity. As such, subunit antigens must be co-administered with an adjuvant to enhance the immune response.1–8
The discovery of novel adjuvants has emerged as a critical frontline effort in the development of modern vaccine formulations. The adjuvant component enables dose-sparing of rare, expensive, and otherwise impotent antigens. An adjuvant also extends the immunotherapeutic benefits to poor responders, such as elderly or immunocompromised patients. While numerous classes of adjuvants have been explored for vaccine therapies over the past several decades, Alum, a mixture of aluminum salts (hydroxide, phosphate, sulfate),9–12 is still the most popular and is present in approximately half of all human vaccines in the U.S. First introduced for human use in the 1930s, Alum was the only approved adjuvant in man for more than 70 years. It was not until late 2009 that the FDA approved GlaxoSmithKline’s AS04 adjuvant (a proprietary combination of Alum and monophosphoryl lipid A)13 for the Cervarix vaccine to immunize against human papillomavirus (HPV). Thus, there remains a great need for novel adjuvants for use in vaccine therapy.
Currently, one of the most promising new adjuvants undergoing clinical investigation is QS-21, a natural product extract obtained from the bark of the Quillaja saponaria tree, found in the desert regions of Chile, Bolivia, and Southern Peru. This semi-purified extract consists of a mixture of triterpene saponins (Figure 1), whose principal constituents are characterized by complex bis(desmosides) 1 (QS-21-apiose) and 2 (QS-21-xylose). These immunoactive constituents of QS-21 contain four distinct structural domains: the central triterpene (quillaic acid), a branched trisaccharide at the C3 position of the triterpene, a linear tetrasaccharide attached at C28 of the triterpene, and a branched acyl chain linked to the central fucose moiety of the linear tetrasaccharide.
The importance of QS-21 as an investigational adjuvant is evident in its use in more than 100 clinical trials involving >6,000 total patients. QS-21 has exhibited a remarkable ability to augment clinically significant responses to vaccine antigens targeting a wide landscape of diseases and degenerative disorders (e.g., cancer, Alzheimer’s disease, malaria, HIV, hepatitis). Despite these promising indications, further commercial advancement of QS-21 is hampered by low isolation yields, inconsistent composition, and lack of purity. Moreover, the instability of QS-21, characterized by spontaneous hydrolysis of the acyl domain, limits its efficacy, enhances its local reactogenicity, and complicates formulation and storage protocols. Finally, these hurdles in advancing QS-21 are further exacerbated by the fact that the mechanism by which it potentiates the immune response is unknown. In the absence of insights on this front, there exists no rational path forward to design novel saponin adjuvants with improved vaccine efficacy.
Our previous synthetic studies on QS-21 have permitted access to each specific isomeric saponin within the mixture,14–17 thereby obviating the problems associated with inconsistent purity and composition. While the initial synthetic route was rather lengthy, a significantly streamlined semi-synthetic route was subsequently developed.18 This involved isolation of the entire Quillaja prosapogenin half of the adjuvant ([QPS]-OH, 3, Figure 2) from the commercial semipurified Quillaja saponaria extract Quil-A, followed by derivatization into a selectively protected form (protected Quillaja prosapogenin, [PQPS]-OH, 4). Integration of this technology into the synthesis of novel saponins offers a convenient means by which to synthesize and append distinct oligosaccharides and acyl chain variants for evaluation.
Our initial work in this area involved the preparation and evaluation of three non-natural analogues comprising acyl chain domain variants 5–7 (Figure 2).19 These syntheses involved modification of the fucose moiety within the linear tetrasaccharide domain of the saponin to reduce hydrolytic instability, allowing the attachment of three distinct lipophilic acyl chains. This initial study revealed that significant structural modification of the central fucosyl pyranose and of the acyl chain of QS-21 causes little or no impairment of adjuvant activity. Moreover, differing toxicity profiles were observed.
Although these initial findings highlighted the exciting prospect of designing improved saponin adjuvants, these three compounds offered a rather narrow insight into the structure–activity profile of these adjuvants. For example, while all three synthetic Quillaja saponins were competent immunostimulators, the complex acyl chain in 5 required an excessively long synthesis, the simplified acyl chain in 6 imparted unacceptable levels of toxicity, and the non-glycosylated acyl chain in 7 limited its aqueous solubility.
Herein, we report efforts that have defined the specific substructures of the saponin adjuvant that are critical for adjuvant activity. Multiple variations of both the acyl chain and linear tetrasaccharide domains have been explored systematically, leading to the development of highly simplified structures that show excellent immunostimulatory properties with encouraging toxicity profiles. In addition, these efforts led to the development of a synthetic Quillaja saponin scaffold that is amenable to attachment of fluorescent reporter groups while still retaining adjuvant activity. The development of this novel class of biochemical tools allows, for the first time, direct investigations of QS-21-derived molecular probes to unravel the cellular mechanisms of saponin immunostimulation.
The non-optimal chemical and biological properties associated with the first three acyl chain domain variants 5–7 prompted us to investigate two additional variants (Scheme 1). The cholestanyl variant 9 was designed based on the known propensity of QS-21 to associate with cholesterol, which has led to hypotheses that saponin binding to membrane-associated cholesterol may be a key enabling feature in its putative mechanism of action. Indeed, the immunostimulatory complex, ISCOMATRIX,20–22 comprises a proprietary mixture of QS-21, cholesterol, and various phospholipids, and is CSL Behring’s favored adjuvant product in clinical trials with vaccines against influenza, hepatitis C virus (HCV), HPV, and cancer.22 The carboxyacyl variant 10 was designed to overcome the water solubility issues associated with the simple dodecanoyl variant 7, through incorporation of a charged carboxylate moiety to increase polarity of the otherwise lipophilic acyl chain.
Governing criteria in both of these acyl chain domain variants are simplicity and highly efficient synthesis. In this regard, 5α-cholestan-3β-ol (11) is commercially available, and dodecanoic acid mono-benzyl ester (12) is readily prepared in one step from its commercially available diacid precursor.23 The preparation of these analogues began with the protected Quillaja triterpene bis(desmoside) 8 (Scheme 1), previously synthesized as a late stage intermediate in the preparation of the first-generation analogues 5–7. Its amino functionality was acylated with 5α-cholestan-3β-yl chloroformate, generated in situ by treatment of 5α-cholestan-3β-ol with phosgene. The resulting cholestanyl adduct was then exposed to global deprotection conditions, via hydrogenolysis and TFA/H2O hydrolysis, to afford the cholestanyl saponin 9 (SQS-0-0-4-6). In a similar manner, amine 8 was treated with the acyl chloride derivative of dodecanoic acid mono-benzyl ester to afford the corresponding fully protected saponin. Deprotection via hydrogenolysis and TFA hydrolysis afforded the carboxyacyl saponin 10 (SQS-0-0-4-5).
Evaluation of these novel saponins for immunostimulatory activity involved vaccination of mice with a multiple antigen formulation that included both GD3–KLH and MUC1–KLH conjugates. These constructs represent clinically relevant oncology vaccine antigens (GD3: melanoma, sarcoma; MUC1: prostate, breast), conjugated to the immunological carrier KLH, and serve as useful models for comparing adjuvant performance. Monitoring antibody responses to both the antigen and the carrier protein provides a useful assessment of adjuvant performance in antigens of diverse immunogenicity, ranging from a poorly immunogenic glycolipid (GD3), to a moderately immunogenic non-glycosylated peptide (MUC1), and highly immunogenic protein (KLH). Groups of five mice (C57BL/6J, female, 6–8 weeks of age) were vaccinated with GD3–KLH and MUC1–KLH at a 5 µg dose. The antigens were co-administered with the adjuvant of interest in a vaccination protocol involving three subcutaneous injections at days 0, 7 and 14, plus a booster at day 65. As the negative control, mice were vaccinated with the GD3–KLH and MUC1–KLH antigens only. As a positive control, vaccinations were performed with naturally derived QS-21 (NQS-21), obtained by fractionating a mixture of saponins from Quillaja saponaria,24 and also with synthetic QS-21 as its reconstituted 65:35 mixture of apiose and xylose isomers (SQS-21). Antibodies against each antigen in mice sera were detected by ELISA (Figure 3a–c, see Supporting Information for experimental details). As a standard initial assessment of overall toxicity, the weight loss of the mice was monitored at 0 h, 24 h, 48 h and 72 h after the first vaccination (Figure 3d).
The resulting antibody titers indicate that both acyl chain domain variants, cholestanyl saponin 9 (SQS-0-0-4-6) and carboxyacyl saponin 10 (SQS-0-0-4-5), induce anti-GD3 titers comparable to that of NQS-21 and SQS-21 (Figure 3c); however, stark differences were observed in comparisons of antibody levels generated against the MUC1 peptide and KLH protein antigens (Figures 3a, 3b). Cholestanyl saponin 9 was essentially inactive as an adjuvant in mice vaccinated with either of these antigens. In contrast, mice vaccinated in the presence of carboxyacyl saponin 10 exhibited anti-MUC1 and anti-KLH titers comparable to those of both positive controls (NQS-21, SQS-21). Interestingly, both synthetic saponins 9 and 10 exhibited noticeably lower toxicity, as judged by weight loss, in comparison to NQS-21 and SQS-21 (Figure 3d). Based on its superior immunostimulatory performance, carboxyacyl saponin 10 was selected as a lead for further structural modifications.
Having identified in carboxyacyl saponin 10 an acyl chain domain that offers both ease of synthesis and potent immunostimulating effects, we next focused our efforts on modification of the linear tetrasaccharide domain of the QS-21. Although the synthetic route for the preparation of the linear tetrasaccharide in 10 is highly convergent, the synthesis of each of the selectively protected monosaccharide building blocks requires 3–7 steps, with that of the terminal apiose sugar being the most lengthy. Thus, we prepared systematically truncated versions of this linear tetrasaccharide 13–15 (SQS-0-0-5-5, SQS-0-0-6-5, and SQS-0-0-9-5) (Figure 4).
We first sought to assemble the selectively protected linear trisaccharide 22 (Scheme 2), in which the terminal apiose sugar has been omitted. As shown, commencing with allyl 2,3-di-O-isopropylidinerhamnopyranoside (17), the C4-hydroxyl group was glycosylated with tri-O-benzylxylopyranose (16) under dehydrative coupling conditions with Ph2SO and Tf2O. The condensation proceeded with β-anomeric selectivity to afford the disaccharide 18 in 71% yield. Selective removal of the allyl acetal in 18 proceeded in quantitative yield to provide disaccharide 19, which serves as a competent glycosyl donor in a second dehydrative glycosylation with the previously prepared 4-deoxy-4-azidogalactopyranoside acceptor 20,19 to furnish trisaccharide 21 with complete α-selectivity. Subsequent conversion of the silyl acetal in 21 to a latent leaving group was performed in a two-step procedure, involving desilylation followed by treatment with CCl3CN in the presence of DBU to give the α-trichloroacetimidate 22.
The two other truncated carbohydrate fragments, 25 and 26, were prepared in an analogous fashion from similar substrates (Scheme 3). Thus, the C2-hydroxyl group of the 4-deoxy-4-azidogalactopyranoside acceptor 20 underwent dehydrative glycosylation with 4-O-benzyl-2,3-di-O-isopropylidenerhamnopyranoside (23) to afford disaccharide 24, allowing for subsequent conversion of its anomeric functionality from the TIPS acetal to the trichloroacetimidate 25. Finally, preparation of monosaccharide 26 was accomplished through straightforward functional group modifications of 4-deoxy-4-azidogalactopyranoside 20, involving C2-O-acetylation, carefully monitored desilylation to avoid unwanted acyl transfer, and trichloroacetimidate formation.
The protected carbohydrate domains 22, 25, and 26 were then advanced to saponin analogues 13–15, respectively, having truncated linear tetrasaccharide domains (Scheme 4). These late stage assembly steps are exemplified by the synthesis of the saponin 13 (Scheme 4a), in which the previously prepared selectively protected Quillaja prosapogenin 4 ([PQPS]-OH) is glycosylated at its C28-carboxylic acid by the trisaccharide 22 under Schmidt coupling conditions to afford the glycosyl ester 27 with complete β-anomeric selectivity. Reduction of the azido group in 27 to its corresponding amine 28 was accomplished with PhSeH under basic conditions. Based on the promising activity of the carboxyacyl saponin 10 above (Figure 3), the amino group was also acylated with dodecanoic acid mono-benzyl ester 12 to install the acyl chain domain. Subsequent global deprotection via hydrogenolysis (H2, Pd/C) and acid hydrolysis provided the carboxyacyl trisaccharide 13 as the single-sugar truncation analogue of the immunoactive carboxyacyl tetrasaccharide saponin 10. Following similar synthetic protocols, the progressively carbohydrate-truncated saponins carboxyacyl disaccharide 14 (Scheme 4b) and carboxyacyl monosaccharide 15 (Scheme 4c) were prepared from their corresponding protected saccharides 25 and 26, respectively.
Evaluation of these three truncated saponin analogues for immunoadjuvant activity involved vaccination of mice with a three-component vaccine that contained the GD3–KLH and MUC1–KLH conjugates as well as OVA. The resulting antibody responses to GD3 were very similar in the mice vaccinated with carboxyacyl trisaccharide 13, carboxyacyl disaccharide 14, carboxyacyl monosaccharide 15, and the positive controls, NQS-21 and SQS-21 (Figure 5d). However, antibody titers against KLH, OVA and MUC1 peptide decreased with each subsequent saccharide truncation (Figures 5a–c). Although trisaccharide 13 induces greater weight loss (Figure 5d), two of five mice died after vaccination with disaccharide 14, while only one mouse died when treated with trisaccharide 13 or monosaccharide 15. In addition, while monosaccharide 15 was the least toxic among the saponins tested, as assessed by weight loss, it was also the least efficacious. Thus, taking mortality, toxicity, and efficacy into consideration, the carboxyacyl trisaccharide saponin 13 was deemed superior to the disaccharide 14 and the monosaccharide 15. Overall, trisaccharide 13 also exhibits comparable efficacy to SQS-21-apiose (1), but is more readily accessible (25 vs. 76 synthetic steps).
The above results reveal the importance of incorporating at least the trisaccharide moiety of the linear tetrasaccharide domain to maintain optimal adjuvant activity and present an attractive path forward for accessing additional simplified adjuvants, given that the significant efforts required for the synthesis and coupling of the selectively protected apiose sugar can be circumvented. Thus, the simplified carboxyacyl trisaccharide saponin 28 was subsequently used as the workhorse synthetic saponin scaffold for the investigation of an additional property of the acyl chain – ionic charge. Initial studies with the 12-carboxy-dodecanoyl chain in both carboxyacyl tetrasaccharide 10 and carboxyacyl trisaccharide 13 signaled that adjuvant activity and aqueous solubility can be maintained with the introduction of a carboxylate anion (at physiological pH).
To explore whether a cationic moiety would impart the same beneficial properties, two amine-containing saponins 33 and 34 were prepared from selectively protected precursor 28 (Scheme 5). Acylation of the amino group in 28 with (Boc)2O provided the corresponding carbamate. The fully protected intermediate was then subjected to the standard hydrogenolysis/hydrolysis deprotection protocol to provide amino trisaccharide 33 (SQS-0-0-5-10), which lacks the lipophilic acyl chain domain entirely. Preparation of a second amine-containing variant involved acylation of the amino group in 28 with 6-((t-butoxycarbonyl)amino)hexanoic acid (36) followed by hydrogenolyis/hydrolysis deprotection, providing the extended lipophilic aminoacyl trisaccharide 34 (SQS-0-0-5-11).
We recognized that the amine-containing saponin 34 also opened the door to late-stage, chemoselective elaboration with other moieties, such as fluorescent labels that might be useful in studying saponin subcellular localization. Along these lines, treatment of aminoacyl trisaccharide 34 with fluorescein isothiocyanate (FITC) (37) afforded the fluoresceinyl aminoacyl trisaccharide 35 (SQS-0-0-5-12).
Immunological evaluation of this new group of saponins was accomplished through vaccination of mice with the MUC1–KLH conjugate. Antibody responses to KLH and MUC1 in mice vaccinated with carboxyacyl trisaccharide saponin 13 (SQS-0-0-5-5) were again comparable to both NQS-21 and SQS-21 (Figure 6a,b). In contrast, mice vaccinated with either amino trisaccharide 33 (SQS-0-0-5-10) or aminoacyl trisaccharide 34 (SQS-0-0-5-11) generated antibody titers that were similar to no-adjuvant treatment control. This is consistent with the hypothesis that the positive charge imparted by the ammonium group in the acyl chain domain attenuates saponin adjuvant activity. Interestingly, the toxicity (weight loss) of the amine-containing saponins 33 and 34 was also decreased compared to NQS-21, SQS-21, and carboxyacyl saponin 13 (Figure 6c).
Notably, upon masking the free amine of 34 by acylation with FITC, efficacy is restored in the resulting fluoresceinyl aminoacyl trisaccharide 35 (SQS-0-0-5-12) to levels comparable to both positive controls NQS-21 and SQS-21 (Figure 6a,b). Importantly, toxicity (weight loss) also remains low in this saponin, comparable to levels seen with aminoacyl trisaccharide 34 (Figure 6c). To the best of our knowledge, while fluorescently-labeled saponins have been reported25, the fluorescein-labeled saponin 35 represents the first example of a fluorescent saponin with confirmed adjuvant activity.
Building upon the promising adjuvant activity and toxicity profile of the fluorescein-labeled saponin 35 above, we next investigated two additional fluorescently-labeled saponins (Scheme 6) to provide probes with alternative excitation and emission wavelengths and increased stability at acidic pH that may be encountered in the intracellular environment. Chemoselective acylation of the aminoacyl trisaccharide scaffold 34 with Cascade Blue derivative 4026 afforded the corresponding Cascade Blue-labeled saponin 38 (SQS-0-0-5-15). Analogous acylation with BODIPY 581/591 C3 succinimidyl ester (41) afforded BODIPY-labeled saponin 39 (SQS-0-0-5-17).
Preclinical evaluation of these fluorescently-labeled saponins revealed that, although median antibody titers to KLH from mice treated with Cascade Blue-labeled saponin 38 were comparable to SQS-21, antibody responses to both OVA and MUC1 were attenuated (Figure 7a–c). Since Cascade Blue imparts polyanionic character to the saponin, the attenuation in antibody response to OVA and MUC1 indicates that saponin adjuvant activity is highly sensitive to charge. In mice treated with BODIPY-labeled saponin 39, antibody titers generated to KLH, OVA, and MUC1 were all lower compared to NQS-21 and SQS-21. Although introducing either of these two fluorophores attenuates adjuvant activity, both fluorescently-labeled saponins were also less toxic compared to the positive controls (Figure 7d).
Currently, there are only indirect means to investigate the mechanism of action of immunopotentiating saponins. Theories for the saponin mode of action include perturbation of cell membrane components to increase immune cell maturation as well as increased delivery of antigens into cells. Although there is limited knowledge on which cell types are involved in saponin-mediated immunostimulation, we elected to investigate dendritic cells, as there is mounting evidence that highlights their importance in adjuvant activity across a panel of various adjuvants, including MF59,27 α-galactosyceramide,28 and MPL-A.29
Immature dendritic cells were treated with the fluorescently-labeled saponins 35, 38, and 39, and imaged via confocal microscopy. Interestingly, the immunopotentiating fluorescein-labeled saponin 35 (SQS-0-0-5-12) (Figure 8a) localized to a discrete structure within the cell, distinguishing itself from the other two probes 38 (SQS-0-0-15) and 39 (SQS-0-0-5-17), which are poor immunopotentiators. BODIPY-labeled saponin 39 (Figure 8b) localized to the plasma membrane while Cascade Blue-labeled saponin 38 (Figure 8c) did not associate with dendritic cells at all. These findings suggest that, in order for a saponin to be an efficacious adjuvant, it is necessary for it to be internalized into dendritic cells. This mechanism is shared by Alum, wherein internalization results in activation of NALP-3 inflammasome.11 The low efficacy observed in Cascade Blue-labeled saponin 38 may result from its inability to engage the cell membrane due to the repulsion between its three sulfonate anions with the phospholipid anions that adorn the plasma membrane. The importance of the saponin scaffold in the trafficking of these fluorescently-labeled saponins within dendritic cells was established by comparison to the trafficking of the corresponding fluorescently-labeled glycine methyl esters (Figure 8d–f; see Supporting Information for synthesis and characterization).
To determine whether or not the majority of the fluorescein-labeled saponin 35 remains intact intracellularly, peripheral blood mononuclear cells were treated with 35 and the cellular lysate was analyzed by RP-HPLC. A single peak (tr = 18.1 min) corresponding to 35 was observed (Figure S1), supporting the supposition that the majority of the immunoactive fluorescein-labeled saponin 35 exists as an intact molecule within the cells.
The synthesis and preclinical evaluation of the QS-21 derived tailored saponins reported herein have defined key structural features that are critical for adjuvant activity and have enabled the development of the first generation of chemical probes to study saponin mechanisms of action. Through systematic truncation of the linear tetrasaccharide domain, it was discovered that an optimal balance of synthetic accessibility, toxicity, and efficacy is achieved with the corresponding trisaccharide variant 13 (SQS-0-0-5-5) (Figure 4). As a result, many synthetic challenges associated with the previous state-of-the-art, 42-step synthesis of carboxyacyl tetrasaccharide 10 (SQS-0-0-4-5) have been circumvented with the identification of this significantly simplified analog, which is accessed in only 25 steps. This will allow future investigations of other domains of QS-21, such as the branched trisaccharide and triterpene core, to be carried out more efficiently.
The aminoacyl trisaccharide saponin 34 has provided an efficient avenue for late-stage, chemoselective functionalization with a series of fluorophores. This led to the identification of the first adjuvant-active, fluorescently-labeled saponin 35 (SQS-0-0-5-12), which is equipotent to and also less toxic than NQS-21 and SQS-21 (Figure 5). As an example of the utility of these molecules, the panel of fluorescent saponins was tracked with confocal microscopy (Figure 8), leading to the intriguing finding that only the most efficacious saponin adjuvant was internalized into dendritic cells. Further detailed investigations into the trafficking of fluorescently-labeled saponin 35 and other saponin adjuvants in various cell types are warranted to assess the role of dendritic cell internalization in saponin mechanisms-of-action. While we have described herein the ability to decorate aminoacyl trisaccharide saponin 34 with fluorophores, a similar strategy can also be envisioned to attach radiolabels for in vivo biodistribution studies and affinity tags to identify molecular targets of saponin adjuvants in the future. Thus, the development of this new class of saponin adjuvant probes affords unique opportunities to dissect the mechanisms-of-action of QS-21 and other saponin adjuvants.
This work is dedicated to the memory of our friend, mentor, and colleague, Prof. David Y. Gin. This work was supported by the NIH (R01 GM058833, R01 AI085622), and Experimental Therapeutics Center of MSKCC. We thank Prof. Samuel J. Danishefsky (MSKCC) and Prof. Derek S. Tan (MSKCC) for helpful discussions and assistance with the preparation of the manuscript, and Dr. George Sukenick, Dr. Hui Liu, Hui Fang, and Dr. Sylvi Rusli (MSKCC Analytical Core Facility) for expert mass spectral analyses. A.F.-T. thanks the Ministerio de Educación of Spain and Comisión Fulbright for postdoctoral fellowship support.
Supporting Information Available: Experimental details for synthetic procedures and analytical data for isolable synthetic intermediates. This material is available free of charge via the Internet at http://pubs.acs.org.