Marrying chemistry with biology by combining on-chip solution-based combinatorial synthesis and cellular screening

Drug development often relies on high-throughput cell-based screening of large compound libraries. However, the lack of miniaturized and parallelized methodologies in chemistry as well as strict separation and incompatibility of the synthesis of bioactive compounds from their biological screenings makes this process expensive and inefficient. Here, we demonstrate an on-chip platform that combines solution-based synthesis of compound libraries with high-throughput biological screenings (chemBIOS). The chemBIOS platform is compatible with both organic solvents required for the synthesis and aqueous solutions necessary for biological screenings. We use the chemBIOS platform to perform 75 parallel, three-component reactions to synthesize a library of lipidoids, followed by characterization via MALDI-MS, on-chip formation of lipoplexes, and on-chip cell screening. The entire process from the library synthesis to cell screening takes only 3 days and about 1 mL of total solutions, demonstrating the potential of the chemBIOS technology to increase efficiency and accelerate screenings and drug development.

This manuscript describes the on-chip platform for synthesis, characterization and biological screening demonstrating the potential for the drug development. The author synthesized and screening lipidoids as transfection agent to proof the concept. Taken together, I recommend publication of a small number of concerns have been addressed: (1) A shown in Figure 4, the zeta-potential of some lipoplexes appeared to be negative charge. Hove the author concern about the N/P value for the lipoplexes formation?
Reviewer #3 (Remarks to the Author): Levkin and coworkers have developed an interesting technology for the rapid synthesis of chemical libraries that can be screened structurally and, in some cases, biologically. By modifying the chemistry of a slide surface, it is possible to control the introduction of specific reagents, both chemical and biological. Overall, this is an interesting study, although there are some concerns that need to be addressed.
1. Figure 1 should only be the right-most figure, as the other parts are already well known.
2. The authors indicate that there is a "methodological gap" between synthesis and screening and that has been a major cause of a lack of progress in drug development efficiency. I do not agree. The problem has typically been downstream -poor lead optimization, limited specificity, potential toxicity, etc.
3. There have been myriad combichem approaches developed over the past 20+ years. Rapid screening was performed, even biological screening. The argument over chemical incompatibility is weak.
4. The lack of academic studies is not so much due to the cost as it is the goal within a noncommercial environment. Fundamental studies do not typically require very large libraries. Moreover, combinatorial libraries are losing favor to much more focused libraries that can interrogate a much broader structurally universe. 5. I like the synthetic scheme of Slide 3 that allows rapid synthesis, minimal cross-talk, and then offline analysis on the chip (e.g., MALDI). But, off-line analysis still required if there is a need to know if mixtures exist and the approximate conversions.
6. A technical issue should be addressed. In the schemes, two chips are contacted through liquidliquid contact. For this to be successful, there needs to be very good alignment of the two chips. Anything off even several microns (depends on spot size) will reduce effectiveness. How is this accounted for by the authors? 7. The ability to include biological reagents (such as plasmid DNA) is quite useful. The authors showed improved transfection, albeit with higher toxicity. This opens up opportunities for downstream applications, e.g., CRISPR. That may need to be addressed in more detail.
Reviewer #4 (Remarks to the Author): The ms describes an essentially novel microarray droplet technology based on patterns with different types of wettabilities and chemical or biological compounds/cells allowing for the synthesis of model chemical compounds, converting them into an aqueous droplet array and finally testing a library of compounds in a cellular assay (here transfection).
The technology builds partly on microarray drop array techniques developed earlier in the KIT group, but the combination is entirely new in my view, addressing the highly important fabrication of chemical/pharmaceutical libraries and its evaluation in cell-based assays. This microarray microdoplet technique is highly innovative and has the potential of massively shortening the time and reducing the amount of compounds needed for library synthesis, microarray preparation, and cell-based highly parallel readouts.
The presentation of the result is very professional including state of the art, clear argumentation, background info, process development and readouts, nicely designed figures, and references.
A topic that is (I believe) not addressed is the issue of byproducts produced during the library synthesis of chemical compounds. Although turnover is quite impressive, small amounts of side products might have important undesirable effects such as cell toxicity. This should be addressed. Also, the issue of what clean room standards would have to be used in a commercial fab lab woulds be useful to know.
Otherwise, I highly recommend publication after the proposed revisions. I believe that this publication will be of great interest to a large community in both academia and industry.

Reviewer #1:
Comment: Drug lead synthesis and screening is an arduous and time-intensive endeavor that is made more challenging by the low success rate in translating a lead into clinical testing and commercialization. Accessing chemical space rapidly in high-throughput screening is a strategy to overcome these limitations, which benefits from miniaturization and automation. Here, Benz, et al. describe a method for rapid on-chip, solution-phase rapid parallel organic synthesis of a lipidoid library, MALDI-MS characterization, on-chip liposome formation, and on-chip cellular screening of transfection efficiency of said liposomes. The authors have designed an on-chip platform that allows for the entire process was completed rapidly, and requires approximately 1 mL of total solvents. This manuscript has many strengths that include: 1) a solution-phase small molecule microarray-like system that represents an advance from previous methods in that "compounds" (here lipidoids) can be screened in solution; 2) a system to modulate droplet size on a microarray slide; 3) a method to quickly characterize the compounds by transferring the droplet to a MALDI-MS plate; and 4) (though not discussed), a potential system to screen membrane proteins in the droplets produced, which would represent a huge advance in the field. This latter point is not proven, but demonstration of this capability would improve the impact of the manuscript significantly.
Unfortunately, the authors claim, yet fail to demonstrate, that these methods are applicable to both synthesis and screening of a broad range of ligands and targets. By focusing on screening lipidoids for transfection efficiency, this technology is likely only applicable to a small group of researchers. In addition, the chemical transformations do not appear to be robust (see below). Finally, there are other scientific concerns that limit enthusiasm for publication in Nature Communications. Therefore, unless more broad applications of this technology can be demonstrated in a new manuscript, and the concerns below are met, this reviewer does not feel that the merit of this manuscript rises to the level of the journal and may be more appropriate for a more specialized journal.
While design of the LSTL and HSTL chips features an elegant design and the supplementary video is impressive, a discussion of the scope of chemical reactions that can preformed is lacking. In order for this approach to have true potential as a drug discovery platform as claimed, the authors should demonstrate other chemical transformations that are possible on the chemBOIS chip and address other compatible solvents (not just DMSO).

Response:
We thank the reviewer for highlighting the strength of our work. The point raised by the reviewer regarding the screening of membrane protein is very interesting. We plan to do the experiment in the extension work of this project. Here we describe a method to miniaturize and parallelize solution-based combinatorial synthesis and demonstrate how to combine it with biological screening. As a proof of concept, we performed only one type of reaction and we plan to do many more in future work.
In our model reaction (synthesis of lipidoids) we indeed utilized just one organic solvent (DMSO). As correctly stated by the reviewer, other chemical reactions might need other solvents. In order to demonstrate the applicability of the chemBIOS platform to other types of solvents and, thus, chemical transformations, we performed an additional experiment to test the possibility to form droplets confined to the omniphilic spots using various commonly used organic solvents such as n-hexane (γlv = 18.4 mN m -1 ), ethanol (γlv = 22.1 mN m -1 ), 1-decanol (γlv = 28.5 mN m -1 ) and DMF (γlv = 37.1 mN m -1 ):
Thus, the platform can be readily utilized for different types of solution-based reactions using various organic solvents. Nevertheless, we are aware of the challenges to adapt some other typical procedures utilized in organic synthesis. For example, reactions at elevated temperature, utilizing solid reagents, highly exothermic reactions will be still challenging using the chemBIOS platform. In our future and ongoing research, we will evaluate the possibility to perform various types of reactions, as well as various types of purification procedures common for organic chemistry (e.g. high-throughput crystallization, extraction and even sublimation).
We added the following text to the discussion part: "Despite the fact, that the platform can be readily utilized for many types of solution-based reactions using various organic or aqueous solvents, there are still some challenges in adapting other chemical techniques. Reactions at elevated temperatures and highly exothermic reactions as well as adding solid reagents or purging gaseous educts might be difficult to realize using the chemBIOS platform in its current state. Chemical reactions requiring a protective atmosphere or controlled pressure can be performed but would require a closed chamber with controlled pressure and atmosphere." Comment: It should also be mentioned that this method may suffer from inconclusive screening results as there are no purification steps. Therefore, it is presumed by the authors that high fidelity reactions are a requirement. While the 2-thiopyridone and other impurities produced in this reaction may not have impacted the GFP transfection assay, that may not be true for the synthesis of other scaffolds, screening of other targets, or other phenotypic assays.

Response:
We agree with the reviewer that byproducts might have an impact on the biological assays used in the screening and, although we did not see such an effect in our transfection screening, on-chip purification will be essential for the future development of the technology. We are currently working on this issue and the advantage of the open system of the droplet microarrays is that it is compatible with various purification methods commonly used in solution based organic chemistry, such as high-throughput parallel crystallization, extraction, sublimation and even centrifugation. In order to demonstrate the feasibility of such high-throughput parallel purification, we performed an additional experiment showing a proof-of-principle on-chip twophase extraction. We added the following part to the result part of the manuscript: "The open format of the chemBIOS system is compatible with further on-chip steps, such as investigation, treatment or conversion of the chemical compound library, to achieve highthroughput parallel on-chip purification. In a proof-of-principle experiments, we showed how crude products could be purified by an on-chip two-phase liquid extraction (Fig. 4a,b). Therefore, we dissolved Nile red and methylene blue in 1-octanol which resulted in a dark blue solution. Next, we applied this solution on several spots on an LSTL slide D and sandwiched this slide with an HSTL slide E containing water droplets. Sandwiching of both solvents resulted in the formation of an interface between the solvents since they are not miscible (Fig. 4a). Methylene blue is highly soluble in water and therefore was extracted from the organic phase into the aqueous phase. Nile red is insoluble in water and, therefore, remained in the organic phase. After 10 min the color of organic phase turned red, while the color of aqueous phase turned blue, indicating a successful separation of both dye compounds. The UV-Vis spectrum of the mixture shows two local absorbance maxima, one at 541 nm which corresponds to Nile red and another at 654 nm corresponding to methylene blue. After the extraction, the intensity of the absorbance maximum of methylene blue in the organic phase decreased while the intensity of Nile red remained constant. In the aqueous phase, no Nile red could be detected. Source data are provided as a Source Data file.
We analyzed the mixture before purification, as well as both the organic and aqueous phase after the purification by UV-Vis spectroscopy (Fig. 4b) and liquid chromatography mass spectrometry (LC-MS) (Supplementary Fig. 9 and 10) to estimate the success of purification. Both analysis methods showed comparable results. While Nile red remains dissolved only in the organic phase, a significant amount of methylene blue was extracted into the aqueous phase. Analysis by UV-Vis spectroscopy enables quantification of this result by the use of Beer-Lambert law. Therefore, the extinction coefficients of Nile red and methylene blue in acetonitrile were calculated to be 43,506 M −1 cm −1 at 541 nm and 72,653 M −1 cm −1 at 654 nm, respectively ( Supplementary   Fig. 11). We estimated the concentration of methylene blue to be 11.9±0.5 mM in the original organic phase mixture, while 3.3±0.6 mM and 4.1±0.4mM in the organic and aqueous phase, respectively, after extraction. The concentration of Nile red was estimated to be 6.0 mM in the original mixture, while 4.3±0.3 mM and 0.3±0.04 mM in the organic and aqueous phase after the extraction. This corresponds to the purification of the methylene blue from 66% in the mixture to 91% in the aqueous phase and that of Nile red from 34% in the mixture to 57% in the organic phase by a single purification step." Furthermore, we complemented the discussion part by adding the following: "The on-chip parallel high-throughput purification is another very important challenge on that we have to focus in further studies. The advantages of the chemBIOS platform are that it possesses open and flat configuration making all microcompartments accessible and potentially compatible with parallel high-throughput purification methods, e.g. on-chip liquid-liquid extraction (Fig. 4a,b)." The peculiarity of the lipidoid reaction we presented here is that it produces the same byproduct for all the lipids. Once the best lipids are identified, they can be synthesized in the bulk, purified accordingly and then can be used for more focused secondary screenings, where purity of the product is an important issue. The aim of a primary screeningwhich was demonstrated by our chemBIOS systemis to accelerate and facilitate the primary synthesis and screening of large compound libraries in short time.

Comment: This platform's open droplet microarray displays clear advantages in chip-to-chip
transfer. Single step transfer to MADLI is important for rapid characterization, however it should be stated that MALDI may not provide an accurate assessment of reaction yield due to variation in ionization and that NMR and/or liquid chromatography should be performed on hit compounds.

Response:
We agree with the reviewer that MALDI-MS is not always quantitative due to the difference in ionization. However, we did not use the MALDI-MS to assess reaction yield. We used UV-Vis spectroscopy to estimate the reaction yield by measuring the absorption of the by-

Response:
We agree with the reviewer that a lot of research was performed in this field. However, what we wrote in our manuscript is that most of the organic chemistry methods developed so far, including many combichem methods, are incompatible with the biological screenings that are usually performed in polystyrene or COC microtiter plates (see pages 2 and 3). Most of the existing combihem methods designed for biological assays (e.g. solid-phase peptide or oligosaccharide arrays, DNA-encoded libraries, one-bead-one-compound, etc.) are not compatible with more physiologically relevant cellular assays (2D cell culture, 3D spheroid or even in vivo screenings) where freely diffusing compounds in an array format are needed. These methods require additional steps for releasing the compounds into the cell suspension whereby most releasing methods are not bioorthogonal (e.g. cleavage of the compounds under strong acidic or strong alkaline conditions or requiring other cell toxic treatments). Furthermore, the above-mentioned methods suffer from the lack of compartmentalization. These are the main challenges that we solve with our chemBIOS platform based on compartmentalized microdroplet reactions. Each droplet acts as a separated reaction space, where we can do miniaturized and parallelized solution-based synthesis. We revised the introduction part to make it clearer regarding this point: "Moreover, most of the existing SPS methods are not compatible with more physiologically relevant cellular assays (2D or 3D cell culture) where freely diffusing compounds, and, therefore, compartmentalization of individual cell experiments, are required."

Comment:
The lack of academic studies is not so much due to the cost as it is the goal within a non-commercial environment. Fundamental studies do not typically require very large libraries. Moreover, combinatorial libraries are losing favor to much more focused libraries that can interrogate a much broader structurally universe.

Response:
The method is in general applicable for both random combinatorial libraries, as well as for more focused libraries. We disagree that fundamental studies do not typically require very large libraries. The main reason why academic labs do not perform such screenings is the unaffordable costs of such libraries. The goal of this project is to develop a platform that is potentially able to make large compound screenings affordable to every academic lab in the future. Such "democratization" of screenings can have long standing important positive consequences on drug discovery and related fields.
Comment: I like the synthetic scheme of Slide 3 that allows rapid synthesis, minimal cross-talk, and then off-line analysis on the chip (e.g., MALDI). But, off-line analysis still required if there is a need to know if mixtures exist and the approximate conversions.

Response:
We thank the reviewer for this comment. Off-line analysis is required at the reaction optimization stage and, of course, would be useful at the step of combinatorial screening of large libraries. However, the ultimate goal of this technology will be the synthesis of thousands or hundreds of thousands compounds in parallel (one microtiter plate-sized glass slide can accommodate several thousand droplets at the moment). At such a scale, it will be either impossible to analyze every compound by an off-chip HPLC or NMR not only due to the time constricts (10.000 compounds * 5 min analysis time = 35 days) but also due to the sensitivity limitations. In this respect, the method can be compared with peptide microarrays or DNA microarrays where only on-chip high-throughput analytical methods are compatible. In our case, we have an advantage that compounds are not bound to the surface and, therefore, MALDI-MS (very sensitive, fast and informative method) can be applied for the on-chip characterization. In the initial stages, we will of course utilize off-chip methods to characterize the synthesis efficiency, yields, limitations, etc.
Comment: A technical issue should be addressed. In the schemes, two chips are contacted through liquid-liquid contact. For this to be successful, there needs to be very good alignment of the two chips. Anything off even several microns (depends on spot size) will reduce effectiveness. How is this accounted for by the authors?
Response: Thanks for this comment. For the perfect aligning we have used an in-house developed aligner device. The procedure is discussed, and photographs of the device are added to the supplementary information. (d) Schematic describing the process of precise sandwiching of two droplet arrays. The alignment device consists of a lower and upper frame. Slide A is set into the lower frame of the alignment device and fixed with a spring holder. Slide B is set into the upper frame of the alignment device and fixed, too. The lower frame shows round rods in each edge of the frame, whereas the upper frame shows round recesses on the same positions. The upper frame is put on the lower frame so that the rods from the lower frame slide into the recesses of the upper frame and therefore fix both frames in the lateral position. The distance in height of both frames can be precisely controlled by four screws until the droplets of slide B merge with the droplets of slide A.

Comment:
The ability to include biological reagents (such as plasmid DNA) is quite useful. The authors showed improved transfection, albeit with higher toxicity. This opens up opportunities for downstream applications, e.g., CRISPR. That may need to be addressed in more detail.

Response:
We added a discussion of other potential biological applications (such as screening of membrane proteins and CRISPR) to the revised manuscript.
"The screening of lipidoids for their cell transfection efficiency using plasmid DNAs is an example of a biological assay that can be combined with synthesized libraries. Other possible biological experiments include, inter alia, CRISPR/Cas9-based gene editing, screening of membrane proteins, gene-knockdown, screening of cell spheroids or organoids."

Reviewer 4:
Comment: The ms describes an essentially novel microarray droplet technology based on patterns with different types of wettabilities and chemical or biological compounds/cells allowing for the synthesis of model chemical compounds, converting them into an aqueous droplet array and finally testing a library of compounds in a cellular assay (here transfection).
The technology builds partly on microarray drop array techniques developed earlier in the KIT group, but the combination is entirely new in my view, addressing the highly important fabrication of chemical/pharmaceutical libraries and its evaluation in cell-based assays. This microarray microdoplet technique is highly innovative and has the potential of massively shortening the time and reducing the amount of compounds needed for library synthesis, microarray preparation, and cell-based highly parallel readouts.
The presentation of the result is very professional including state of the art, clear argumentation, background info, process development and readouts, nicely designed figures, and references.
A topic that is (I believe) not addressed is the issue of byproducts produced during the library synthesis of chemical compounds. Although turnover is quite impressive, small amounts of side products might have important undesirable effects such as cell toxicity. This should be addressed.

Response:
We thank the reviewer for the positive response and appreciation. We agree with the reviewer that the on-chip parallel high-throughput purification is a very important but also interesting challenge that need to be better discussed in the manuscript. We are currently working on the implementation and optimization various strategies to achieve parallel high-throughput purification essential for such screenings. The advantages of our droplet microarray platform are that it is an open platform where all microcompartments are accessible externally, which potentially makes it compatible with parallel miniaturized and high-throughput extraction, crystallization, or even sublimation. We would like to emphasize that such purification methods are usually incompatible with most of the high-throughput miniaturized combinatorial platforms (droplet microfluidics, microfluidics in general, slipchip, etc.).
In order to address this point, we performed a proof-of-principle experiment showing the possibility to do on-chip purification by two-phase liquid-liquid extraction.