DNA-Programmed Lipid Nanoreactors for Synthesis of Carbohydrate Mimetics by Fusion of Aqueous Sub-attoliter Compartments

Abstract

Lipid nanoreactors are biomimetic reaction vessels (nanoreactors) that can host aqueous or membrane-associated chemical and enzymatic reactions. Nanoreactors provide ultra-miniaturization from atto- to zeptoliter volumes per reaction vessel with the major challenge of encoding and spatio-temporal control over reactions at the individual nanoreactor or population level, thereby controlling volumes several orders of magnitude below advanced microfluidic devices. We present DNA-programmed lipid nanoreactors (PLNs) functionalized with lipidated oligonucleotides (LiNAs) that allow programming and encoding of nanoreactor interactions by controlled membrane fusion, exemplified for a set of carbohydrate mimetics with mono- to hexasaccharide azide building blocks connected by click-chemistry. Programmed reactions are initiated by fusion of distinct populations of nanoreactors with individually encapsulated building blocks. A focused library of triazole-linked carbohydrate-Cy5 conjugates formed by strain-promoted azide-alkyne cycloadditions demonstrated LiNA-programmed chemistry, including two-step reaction schemes. The PLN method is developed toward a robust platform for synthesis in confined space employing fully programmable nanoreactors, applicable to multistep synthesis for the generation of combinatorial libraries with subsequent analysis of the molecules formed, based on the addressability of the lipid nanoreactors.

Introduction

Leakage-free compartmentalization is a fundamental design aspect in nature to control chemical reaction sequences spatially and temporally. A major obstacle to mimic cellular compartmentalization has been the lack of technology for leakage-free fusion of multiple lipid nanoreactors in sequential or parallel fashion. Lipid membrane anchors incorporated in LiNAs are responsible for persistent anchoring into the lipid bilayer membranes. Complementary pairs of membrane-anchored LiNAs mediate the “zipping” process to force membranes in juxtaposition and close proximity and thereby promote practically leakage-free fusion and content mixing between PLNs as an emerging strategy for carrying out chemical and enzymatic reactions (Figure 1A).¹⁻³ The ability of LiNAs to fuse lipid membranes⁴⁻⁶ has enabled applications in cell-free expression,⁷ sensing,⁸⁻¹⁰ minimal cell systems,⁴⁹ and drug delivery.^11,12,50 However, the potential for LiNA-programmed chemistry and combinatorial fusion designs that is inherent in the sequence-dependent hybridization of LiNA sequences has not been exploited. The LiNA fusion platform employs orthogonally hybridizing duplex sequences on at least three liposome populations for combinatorial or cascade fusions.^2,3 The presented design has recently been implemented in a single-particle microscopy setup with surface immobilized PLNs, with up to seven consecutive LiNA-mediated stochastic fusion events with minute amounts of dsLiNAs (∼5–10 per PLN) necessary for each fusion event.³ These results urged us to develop the PLN fusion concept toward ultra-miniaturized chemical nanoreactors that comprise sub-attoliter volumes, which are orders of magnitudes below currently used advanced microfluidic platforms. Chemistry has been exemplified with a focused library of carbohydrate mimetics using single-step and two-step programmed reactions.

Figure 1.

(A) DNA-programmed liposome nanoreactors (PLNs) that generate products by DNA hybridization.^31,32 Box: illustration of liposome assembly mediated through the hybridization of complementary lipidated DNA (cLiNA) sequences.³⁰ LiNAs consists of a terminal T-nucleotide, a lipid anchor (X), a triethylene glycol linker (P3), and 17 nt DNA sequences; LiNAs anchor spontaneously and stably into liposomes. (B) Synthesis of carbohydrate mimetics in PLNs. A combination of encapsulated sCy5(3)-DBCO (Cy5(3); 0.2 mM) and glycosyl azides (1–10; 1 mM) that react via the strain-promoted alkyne-azide cycloaddition (SPAAC) reaction. cLiNAs-AA′ and ncLiNAs-AA for fusion by complementary and non-complementary LiNAs, respectively.

Carbohydrate mimetics are of interest in the context of sugar–protein and sugar–nucleic acid interactions and play a critical role in a plethora of cellular processes.¹⁴⁻¹⁶ Especially, fluorescently labeled carbohydrate mimetics are potentially useful as target-specific fluorescent markers in the context of metabolic tracing, receptor binding studies, or carbohydrate–membrane interactions via live cell imaging^17,18 or conjugated to proteins and nucleic acids via triazol-linkages with potential applications in drug discovery.^16,19 Here, we present encoded libraries of small molecule carbohydrate reactants within the lumen of LiNA-functionalized lipid nanoreactors (PLNs). Thus, we achieve LiNA-programmed and encoded chemical reactions that are not, as developed for DNA-encoded chemical libraries (DELs), performed via covalent attachment of all building blocks to DNA but are based on the mechanical entrapment of chemical building blocks within LiNA-barcoded PLNs.¹³ Strategies for setting up libraries of carbohydrate mimetics and subsequent screening for bioactive “hit” molecules are important for drug discovery and still an unmet need.²⁰⁻²² Existing strategies such as DELs have been established as a high-throughput (HTP) combinatorial methodology for the discovery of bioactive lead compounds.²³⁻²⁶ However, DELs have a number of intrinsic disadvantages and limitations, e.g., interference of bound reagents and DNA in target binding, general complexity of DELs with the need to synthesize DNA-conjugates of each building block, which chemically blocks at least one position within the respective building block as well as sterically shields one side of the molecule, and the attached negatively charged DNA strand will add significant steric bulk which may affect reaction-speed and regioselectivity of reactions.²⁷⁻²⁹

We demonstrate the synthesis of different individual carbohydrate-dye conjugates (the dye label is used to quantify reaction yields by UPLC-fluorescence analysis with a detection limit in the pM range) during fusion of two PLNs. As an extension to single-step reactions, examples of parallel two-step reactions are investigated. Alongside optimization of the chemistry, we optimized the lipid composition of the PLNs to reduce membrane permeability of carbohydrate building blocks (especially mono- and disaccharides) while preserving the overall fusogenicity and efficient content mixing of PLNs. The strain-promoted azide-alkyne cycloaddition (SPAAC) reactions proceeded with 36–63% yield at the bulk level and with up to 92% among the PLNs that underwent fusion. Thus, we obtained di-, tri-, tetra-, and hexasaccharide conjugates, while monosaccharides exhibited too high lipid membrane permeability for chemistry to result in practically useful yields. The ability to program multiple synthetic steps will presumably allow the preparation of larger libraries with increased chemical diversity.

Results and Discussion

PLN Optimization and Fusion Assays

In the previously reported LiNA-programmed fusion design,^2,33−35 a lipid mixture of DOPC/DOPE/cholesterol (50/25/25, molar ratio) was used but further optimized with an increased amount of DOPE to increase fusion and content mixing.^36,37 Fusion was monitored by a sulforhodamine B (SRB, Figure S2) dequenching assay.^38,39 Liposomes composed of DOPC/DOPE/DOPG/cholesterol (29/45/1/25, molar ratio) showed an increased fusion performance compared to the original composition (Figure 2A). However, membranes with a higher DOPE content also exhibited increased passive leakage of SRB (Figure 2A). The addition of membrane anchored poly(ethylene glycol) (PEG) polymers (DOPE-PEG₂₀₀₀, 2 mol %) was able to counteract the increased permeability (passive leakage),^40,41 and both leakage during fusion and passive leakage were much lower for all temperatures after the addition of DOPE-PEG₂₀₀₀ compared to non-PEGylated liposomes while maintaining excellent fusion performance (Figure 2A). The optimized lipid composition was used for all subsequent PLN reactions to achieve optimal chemical yields with low content leakage. To monitor chemical reactions for PLNs, a fluorophore/quencher ANTS/DPX assay (ANTS-8-aminonaphthalene-1,3,6-trisulfonic acid/DPX-p-xylene-bis-pyridinium bromide, SRB Figure S4)^42,43 was used instead of SRB, which was only used during lipid mixture optimization studies. Upon content mixing between a population encapsulating ANTS (L_{ANTS}) and another encapsulating excess DPX (L_{DPX}), the ANTS signal (I_t) is quenched, which compared with maximal values (I_100%, corresponding to 100% fusion yield) and minimal values (I_0%, corresponding to 0% fusion yield) allowed quenching to be measured reproducibly. Multiple fusion events of a specific L_{ANTS} do not further quench the signal (Figures S7 and S8). After mixing L_{ANTS} and L_{DPX}, the fluorescence intensity dropped by 33% compared to the maximal I_100%, i.e., 33% of the L_{ANTS} population fused with at least one L_{DPX} liposome. Next, we kept the L_{ANTS} concentration unchanged but increased the equivalents of L_{DPX} to 2× and 3×, resulting in fusion yields of 51 and 68%, respectively. For optimal reaction yields, the population ratio was kept at 1:3 in subsequent experiments.

Figure 2.

(A) Comparison of content mixing performance during fusion of liposomes with different lipid compositions. DOPC/DOPE/DOPG/cholesterol in the composition of 49/25/1/25, 29/45/1/25, and 29/45/1/25 with 2% DOPE-PEG₂₀₀₀ modification. Sulforhodamine B (SRB) assay for apparent CM, leakage, and passive leakage (for detailed description, see Section 1.7 in the Supporting Information). (B) ANTS fluorescence intensity versus time was monitored. In system I, a mixture of L_{ANTS} and L_{0} in a complementary fusion manner, this fluorescence level corresponds to 0% fusion. In system II, the 100% fusion level was determined by the fusion of L_{ANTS/DPX} and L_{0} in a complementary fusion manner. In each system, a concentration of 5 mM DPX on the outside is used to prevent unspecific fusion signals from leakage. The fluorescence decline at 0 min was due to the dilution during addition of another liposome population (for detailed explanation, see Section 1.8 in the Supporting Information). Inset: fusion percentages were calculated by mixing L_{ANTS}/L_{DPX} for increasing lipid equivalents of the quencher population. cLiNAs-AA′ for fusion. All bar data are the mean ± s.d.; n = 3.

Carbohydrate Building Blocks and Library Generation by Click-Chemistry

To demonstrate chemistry in PLNs, we chose strain-promoted azide-alkyne cycloadditions (SPAAC). A series of azido-functionalized mono-, di-, tri-, tetra-, and hexasaccharides (compounds 1–10, Figure 1B)⁴⁴ were reacted with dibenzylcyclooctyne-sulfo-cyanine5 (Cy5, as a label for the analytical UPLC fluorescence assay) upon fusion. To optimize the reaction yield using PLNs, two heating modes were performed (heating mode I: 50 °C for 1 h; heating mode II: 50 °C for 5 min followed 37 °C for 6 h) and compared (Figure 3).

Figure 3.

Schematic experiment designs and yields of the SPAAC reaction. (A) SPAAC between glycosyl azides and Cy5(3) resulting in dye-saccharide conjugates in PLNs. (B) Explanatory scheme for SPAAC reactions by cDNA PLNs including control experiments to verify that reactions are triggered by LiNA-mediated fusion (not drawn to scale). (C) Yields of SPAAC reactions in PLNs. Entry numbers 1–10 correspond to the reaction between glycosyl azides 1–10 and Cy5. Note: Data represent the percentages of yield (%), and conversion is based on Cy5. [a] Heating mode I, 50 °C for 1 h; [b] heating mode II: 50 °C for 5 min followed by 37 °C for 6 h. Sugar permeability and lysis tests were carried out under heating mode I. [c] Product retention time. For further details, see Sections 1.10 and 1.11 in the Supporting Information.

Starting from monosaccharides 1 and 2, only low yields were observed for fusion by cLiNAs under both heating methods, whereas the control permeability experiment provided quantitative conversion of Cy5-1 and Cy5-2 (≥99%; Figure 3C), demonstrating that monosaccharide azide building blocks (azide substituents reduce polarity, calculated as clog P, by more than one order of magnitude) readily cross the lipid bilayer. The PLN net yields of disaccharides 3 and 4 for cLiNAs were 40 and 41%, respectively, using heating mode I (1 h, 50 °C). Since 90% fusion is completed within 5 min at 50 °C (Figure S3), mode II (5 min, 50 °C followed by 37 °C for 6 h) was employed, combining high fusion efficiency with minimized saccharide leakage with improved yields of 63 and 62%, respectively. Only 1–3% product was observed for PLNs engrafted with ncLiNAs, particularly for the 1 h readout in mode I. We assume that passive diffusion of highly permeable building blocks into the “target” PLN population may cause the low level of formed byproducts from ncLiNAs. These observations underlined the applicability of sequence-specific DNA hybridization to facilitate fusion-triggered reactions, suggesting that increased reactant retention greatly increases yields during product formation especially for longer reaction times with ncLiNAs. In the lysis experiments, the PLN populations without LiNA were mixed and lysed, leading to an approx. 3000-fold reactants dilution; the yields were all ≤3.2% (see Experimental Section 1.14). The influence of different carbohydrate reactants (compounds Cy5 and 3) on fusion yield was evaluated, but no significant difference was observed for the reactant concentration used (Figure S6). Moreover, the optimized lipid composition and PEG modification resulted in significantly improved product yield (Table S11), demonstrating that both an increase in liposome fusion efficiency and reactant retention within the liposomes is directly reflected in higher reaction yields.

The PLN generated carbohydrate library was further expanded to larger tri-, tetra-, and hexasaccharides (compounds 5–10) and several trisaccharides (5–8), providing Cy5-conjugates in slightly lower yields (36–45%, mode II) than seen for disaccharides. The permeability experiments showed that, counterintuitively, trisaccharides are more prone to leakage. The literature suggests that disaccharides have stronger interactions with the lipid membrane than trisaccharides,^45,46 observed via a decrease in the lateral mobility of lipids in the bilayer, which may be responsible for the lower membrane permeability. Larger tetra- and hexasaccharides all led to good yields (49 and 52%) of the corresponding triazole products despite the reduced reaction speed for larger carbohydrate building blocks supporting a decrease in membrane permeability as a major contributor to improved yields. These results provide a proof of principle for performing reactions in PLNs with practically useful yields with low reactant leakage except for monosaccharides. The high leakage of monosaccharides is attributed both to their small size and also to the greatly reduced polarity by introduction of a nonpolar azide substituent, which decreases polarity by more than one order of magnitude (see clog P values in the SI, Figure S30), which we assume contributes greatly to the leakage observed. However, chemistry with unmodified or more polar monosaccharides (e.g., amino sugars) may proceed in similar yields to larger well-contained building blocks. Yields of dye-disaccharide products correspond to almost full conversion in the fused populations of PLNs. Factoring in a 68% fusion yield (content mixing), calculated yields of 92 and 90% are obtained for 3-Cy5 and 4-Cy5, respectively.

We assume that full conversion of the limiting Cy5 was not achieved due to leakage of building blocks, likely during the collapse of the hemifusion diaphragm after formation of a fusion pore since this is not observed for ncLiNAs, which strongly suggests a fusion-induced leakage process.^47,48 Approx. 2.5% of the Cy5 dye leaked out of the liposomes, as confirmed by size-exclusion chromatography on fused populations. However, leakage of monosaccharide or Cy5 building blocks is not limiting PLN applications since leaked reactants are strongly diluted (>3000×, Figure S12) and very unlikely to react outside the PLNs in relevant quantities.

To realize a versatile platform for chemistry inside sub-attoliter compartments, parallel as well as successive multistep chemistry triggered by fusion is desirable. We employed two pairs, LiNA-AA′ and LiNA-BB′, with similar binding affinities (T_m) and fusion efficiency and without cross hybridization (Table S1). The LiNA-encoding can be used to program both parallel and cascade reactions in a “one-pot” approach.

For parallel reactions, two different glycosyl reactants (3 and 10) and alkynyl-dyes (Cy5 and Cy3) were encapsulated separately. The four PLN populations were encoded with LiNAs (Figure 4a), mixed, incubated under heating mode II, and analyzed. The expected products (3-Cy5 and 10-Cy3) are formed in good yields, similar to the individual control reactions with only traces of observed cross-products (Figure 4b,c, 3-Cy3 or 10-Cy5). Next, a two-stage fusion cascade was employed to achieve a compartmentalized linear multistep synthesis; heating modes II and III (50 °C for 5 min followed by 50 °C for 2 min under MW) were used for first- and second-round fusions, respectively. Three PLN populations were encoded with LiNAs (fusion mode I, Figure 5A): L_{Cy5} engrafted with LiNA-A; L_{11} engrafted with LiNA-A′ and LiNA-B; L_{10} engrafted with LiNA-B′. Only L_{Cy5} liposomes fused with L_{11} are encoded and able to fuse with L_{10}. This strategy exclusively provided product Cy5-11-10, albeit at a lower yield than for a single-step reaction (8.6% over a two-step reaction). This is explained by the fusion of L_{11} with L_{10}, producing 11-10 instead of the targeted product, further enforced by the 1:3 PLN ratio and the resulting excess of L_{11}; however, due to the lack of fluorescence or UV activity, the product 11-10 remains invisible and represents a significant obstacle for byproduct quantification at pM concentration range. By engrafting both LiNA-A and LiNA-B on L_{Cy5}, these liposomes can fuse with L_{11} (LiNA-A′) and L_{10} (LiNA-B′) liposomes independently (fusion mode II).

Figure 4.

(A) Schematic representation of DNA-programmed liposome nanoreactors. This strategy enabled parallel chemical reactions in one pot by utilizing two complementary sets of LiNAs, A/A′ and B/B′, allowing reactions between Cy5 and 3 and between Cy3 and 10. (B) UHPLC chromatograms of starting reactants and products formed inside the parallel PLNs. (C) Retention times of starting reactants Cy5 or Cy3; encoded PLN products, Cy5-3 and Cy3-10; and cross products from promiscuous fusion/reaction, Cy5-10 and Cy33. For details, see Experimental Section 1.12 and Figure S13.

Figure 5.

Two-step PLN fusion cascade encoded by two sets of cLiNAs-AA′ and cLiNAs-BB′. (A) For fusion mode I: the PLN ratio (L_{Cy5}/L_{11}/L_{10}) is 1:3:3, L_{Cy5} encoded with LiNA-A, L_{11} encoded with LiNA-A′ and LiNA-B, L_{10} encoded with LiNA-B′. Isolated yields are shown in the table. (B) For fusion mode II, the PLN ratio (L_{Cy5}/L_{11}/L_{10}) is 1:3:3, L_{Cy5} encoded with LiNAs-A and LiNAs-B, L_{11} encoded with LiNA-A′, L_{10} encoded with LiNA-B′. Isolated yields are shown in the table with yields determined by UHPLC, referencing the ratio of the peak area of each compound to the total peak area (Supporting information, Experimental Section 1.12). Note: (a and c) the difference between fusion modes I and II is in the LiNA encoding method, with both modes providing reaction products in two steps; (b) data represent the percentages of yield (%) and conversion based on Cy5 and Cy3, respectively. (C) Two-step PLN encoded reaction scheme.

Fusion mode II produced the same product at a much higher yield than for fusion mode I (16% vs 8.2%, respectively). Fusion mode II increases the fusion efficiency between L_{Cy5-11} and L_{10} (second stage) while preventing fusion between L_{11} and L_{10}. Instead, a small amount of byproduct Cy5-10 (10%) was formed via fusion of L_{10} with L_{Cy5} remaining from the first stage. In control experiments, when encoding with LiNA (A’ or B′) being absent from the cascade, the (by)product inside LiNA-uncoded liposomes disappeared, thus proving that two sets of LiNAs promote two-step fusion. The two-step fusion performance has independently been monitored by an SRB assay (Figure S14). These results indicate that PLNs allow chemical reaction pathways via LiNA-encoding and illustrate the versatility of the approach.

Conclusions

Model reactions with a library of carbohydrate building blocks of mono-, di-, tri-, tetra-, and hexasaccharides demonstrated the feasibility of encoding chemistry and multistep chemical reactions within sub-attoliter (10^–18 L) lipid compartments with practically useful yields. Micrometer-scale experimental systems, such as microfluidics, parallel micro pipetting, microarrays, or lab-on-a-chip technologies, are very successful tools for performing high-throughput (HTP) combinatorial methodology. However, there is a major difference in the scale of volumes that are controlled; the PLN particle sizes between 50 and 200 nm reprensent sub-attoliter (10^–18 L) volumes which can be addressed and turn PLNs into ultra-miniaturized nanoreactors, utilizing the aqueous lumens as carriers for reagents and selected building blocks for programmed reactions, whereas even the most advanced micro/nanofluidic devices are limited to femtoliter volumes (10^–15 L) but most commonly used systems typically operate at picoliter volumes (10^–12 L). This provides several orders of magnitude advantage in material consumption at the same concentration inside the PLNs, which would matter when a platform is massively upscaled for millions of compounds as reported for DELs. The azido-saccharide building blocks used in this study illustrate that neutral molecules can indeed be used for synthesis in the aqueous lumen of liposomes with unique differences displayed in membrane permeability. We observed acceptable (>15%) to very good conversion yields (>90%) for the effectively fused sub-population with a moderate excess of reagents (1:3). When corrected for fusion yield (effective number of fused PLNs), the conversion inside the nanoreactor was almost quantitative in the case of the disaccharide-sCy5 conjugates (≥90%). Two-step reactions encoded in two different ways provide the encoded products and further underline that reaction products remain encapsulated and are accessible for further reaction steps. Extending a focused library in future applications, the platform using the two-step chemistry shown with 10 different DBCO-dyes, 10 different azido-alkynyl-saccharides, and 10 different azido-saccharides would already enable the generation of a library with 1000 distinct carbohydrate mimetics. Application of different chemistries, increasing the step number to three and four would exponentially increase the size of the corresponding library and the chemical diversity. The current study was limited to two steps by the SPAAC chemistry, and transition to a combination of different chemistries would allow extension to more steps and subsequently allow libraries with more individual compounds and more chemical diversity. Based on a recently published single PLN microscopy study, up to seven consecutive stochastic fusion events of immobilized PLNs showed the potential for new combinatorial approaches, which may allow generation of LiNA-encoded dynamic combinatorial libraries as an extension to current applications of DELs. The strength of the presented LiNA-encoded PLN platform is in the generation of DELs with LiNA encoding of the PLN instead of the individual chemical building blocks. This greatly reduces the complexity inherent in covalent modification of all building blocks and thereby removes a major limitation of currently used DEL platforms for high-throughput compound and lead structure generation as currently applied to accelerated drug discovery. The PLN platform is not limited to carbohydrates but could potentially be applied to build other synthetic (bio)polymers or natural product mimetics using either chemical or enzymatic approaches. Neither LiNA barcoding nor fusion are time-limiting steps in the application of PLNs for combinatorial processes, while using ultralow amounts of chemicals or enzymes for individual reactions is potentially very beneficial for larger combinatorial libraries.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacs.3c04093.

• Detailed description of experimental procedures, saccharide synthesis and characterization, liposome NTA data, SRB data and ANTS/DPX data (PDF)

The authors declare no competing financial interest.