Continuous collective analysis of chemical reactions

Abstract

The automated synthesis of small organic molecules from modular building blocks has the potential to transform our capacity to create medicines and materials^1-3. Disruptive acceleration of this molecule-building strategy broadly unlocks its functional potential and requires the integration of many new assembly chemistries. Although recent advances in high-throughput chemistry^4-6 can speed up the development of appropriate synthetic methods, for example, in selecting appropriate chemical reaction conditions from the vast range of potential options, equivalent high-throughput analytical methods are needed. Here we report a streamlined approach for the rapid, quantitative analysis of chemical reactions by mass spectrometry. The intrinsic fragmentation features of chemical building blocks generalize the analyses of chemical reactions, allowing sub-second readouts of reaction outcomes. Central to this advance was identifying that starting material fragmentation patterns function as universal barcodes for downstream product analysis by mass spectrometry. Combining these features with acoustic droplet ejection mass spectrometry^7,8 we could eliminate slow chromatographic steps and continuously evaluate chemical reactions in multiplexed formats. This enabled the assignment of reaction conditions to molecules derived from ultrahigh-throughput chemical synthesis experiments. More generally, these results indicate that fragmentation features inherent to chemical synthesis can empower rapid data-rich experimentation.

Extended Data Fig. 1 |. Synthesis and fragmentation analysis of TIDA boronates.

a, A set of 60 TIDA boronates (S6-S65) were prepared from halo-TIDA boronates (1, S1-S5) and a series of amines via Buchwald-Hartwig coupling reactions. b, Product ion scanning tandem mass spectrometry analysis of these 60 TIDA boronates (S6-S65) revealed consistent loss of 86 Da. Reported intensity is the peak product ion intensity for the [M+H]⁺ to [M+H-86]⁺ transition. Data were not normalized relative to parent ion intensity. c, Collision energies associated with peak product ion intensity for the [M+H]⁺ to [M+H-86]⁺ transition for TIDA boronates (S6-S65) were found to be consistently within the 30–40 V range. Full spectra for these data can be found in the supplementary materials and Extended Tabular Data Sheet 2.

Extended Data Fig. 2 |. Neutral loss analysis of TIDA boronates.

a, Schematic representation of neutral loss mass spectrometry, analyte MS data are collected only when parent Q1 ions and daughter Q3 ions are separated by the mass of the desired neutral lost fragment. b, Schematic representation of acoustic droplet ejection mass spectrometry. Nanoliter droplets are directly introduced into a mass spectrometry via acoustic ejection into an open port interface. c, Neutral loss acoustic droplet ejection mass spectrometry (NL-ADE-MS) data for 60 TIDA boronates (S6-S65) using loss of 86 Da fragments shows a linear response R²(avg) = 0.99. d, Relative signal intensity for NL-ADE-MS data collected for 60 TIDA boronates (S6-S65) at equimolar concentrations (10 uM) revealed all signals to be within approximately 1 order of magnitude when data are collected using a fixed molecular weight dependent collision energy 36 V for <460 Da and 42 V for >460 Da averaged across 24 replicates. Data can be found in Extended Tabular Data Sheets 3 and 4.

Extended Data Fig. 3 |. Neutral loss acoustic droplet ejection mass spectrometry method optimization.

100 diverse compounds (see: Fig. S9) spanning a, mW, b, LogP, and c, polar surface area were selected to validate the specificity of TIDA boronate fragmentation. d, NL-ADE-MS analysis of these 100 compounds revealed only two weak hits for the neutral loss of 86 Da demonstrative of high specificity of TIDA boronate fragmentation. e, Collision energy scales with molecular weight across our 60 TIDA boronate test set, heavier molecules will require more kinetic energy to affect a similar fragmentation efficiency. f, To test the influence of non-idealized collision energies we grouped TIDA boronates in narrow and wide molecular weight ranges. NL-ADE-MS data were collected for scan ranges covering each grouping using the averaged collision energy for each group. Peak product signal intensities were tested for a wide range at 36 V and 42 V collision energy. Comparison against narrower scan range data with range optimized collision energies showed that non-idealized collision energies minimally impacted the average signal intensity across our TIDA boronate test set. g, Influence of mass scan range on sample-to-sample NL-ADE-MS data collection. h, Alternating injections of either TIDA boronate 1, Boc 7, or THP 9, and blank wells allowed assessment of carry-over. Peaks separated by 1.2 s demonstrated no significant residual signal. i, Authentic samples of TIDA boronate S5 were mixed with reaction mixture components derived from Buchwald-Hartwig coupling reactions and analyzed by NL-ADE-MS. Signal intensities were essentially identical in the presence or absence of these reaction components, demonstrating that matrix effect have limited influence on NL-ADE-MS. j, Assessment of the optimal injection volume for NL-ADE-MS analysis. Using boronate S2 ejection of a single 2.5 nL droplet was the most reproducible, whereas ejection of 10 nL (4 droplets) was more variable. Data can be found in Extended Tabular Data Sheets 5–11.

Extended Data Fig. 4 |. Head-to-head comparison of NL-ADE-MS and LC-MS.

Whole 384-well plate reactions were performed spanning 64 reaction conditions for a range of miniaturized chemical transformations to access 6 chemical products in each case (a, 1a-1f, b, 1g-1l, c, 1m-1r, d, 2a-f, e, 3a-f, f, 4a-f). Relative product % was normalized to the highest output by both NL-ADE-MS and LC-MS and then averaged across the 6 expected products to provide plots featuring rank ordering of the 64 possible reaction conditions. These data show excellent agreement between NL-ADE-MS and LC-MS affirming the ability of NL-ADE-MS to rapidly acquire accurate reaction outcome data. Each individual 384-well plate required only 7.68 min of data collection by NL-ADE-MS, whereas the equivalent LC-MS data set needed 19.2 h. Inset tables show the top 3 conditions selected by NL-ADE-MS. Starting material fragmentation properties were used to define NL-ADE-MS analysis for the subsequent products per Extended Data Fig. 5: 1 (CE = 31), 2 (CE = 30), 3 (CE = 32), 4 (CE = 23). Protocols can be found in Supplementary Information and data can be found in Extended Tabular Data Sheets 12–17.

Extended Data Fig. 5 |. Starting materials define fragmentation barcodes for downstream products.

a, Despite being of lower molecular weight starting materials should closely approximate the collision induced dissociation parameters for the resulting products. In this manifold an underestimation of required collision energy for the product minimizes potential overlap with other fragmentation channels. NL-ADE-MS analysis of products arising from 1 (b, S6, c, S7, d, S8) and S1 (e, S17, f, S18, g, S19) using both idealized (gray) and starting material (black) defined fragmentation parameters show that linear concentration responses are followed with the starting material defined collision energy over a wide concentration range. h, In contexts where the desired fragmentation is obfuscated by another fragmentation channel, such as −56 Da associated with loss of a tertiary butyl group (gray traces in panels i and j), in-source fragmentation allows a “pre-fragmented” parent ion to be employed in the desired fragmentation analysis which is the loss of −86 Da associated with TIDA boronates (shown in orange). In two competitive examples weak easily fragmented tertiary butyl groups can be in-source fragmented to allow the desired TIDA boronate fragmentation channel to predominate (i, S96, and j, S21). Collectively these data validate that starting material fragmentation behavior is a fundamental feature which translates into resulting chemical products. Protocols can be found in Supplementary Information and data can be found in Extended Tabular Data Sheets 18 and 19.

Extended Data Fig. 6 |. Fragmentation-first experimentation.

Starting material fragmentation behavior defines high-throughput analytical readouts by NL-ADE-MS for a wide range of substrates including clarithromycin derivative 5 (CE = 32, 158 Da fragment), amphotericin B 6 (CE = 29, 181 Da fragment), C5-lenalidomide 12 (CE = 25, 111 Da fragment), and C5-lenalidomide-Br 13 (CE = 26, 111 Da fragment). Whole 384-well plate reactions were performed spanning 64 reaction conditions for a range of miniaturized chemical transformations to access 6 chemical products in each case (a, 5a-5f, b, 5g-5l, c, 6a-f, d, 12a-f, e, 12g-l, f, 13a-f). Relative product % was normalized to the highest output by both NL-ADE-MS and LC-MS and then averaged across the 6 expected products to provide plots featuring rank ordering of the 64 possible reaction conditions. Inset tables show the top 3 conditions selected by NL-ADE-MS. These data show excellent agreement between NL-ADE-MS and LC-MS affirming the ability of starting material defined parametrization of NL-ADE-MS to rapidly acquire accurate reaction outcome data. Each individual 384-well plate required only 7.68 min of data collection by NL-ADE-MS, whereas the equivalent LC-MS data set needed 19.2 h. Protocols can be found in Supplementary Information and data can be found in Extended Tabular Data Sheets 20–22.

Extended Data Fig. 7 |. Further examples of fragmentation-first experimentation.

Starting material fragmentation behavior defines high-throughput analytical readouts by NL-ADE-MS for a wide range of substrates including boc protected amine 7 (CE = 15, 56 Da fragment), boc proline 8 (CE = 11, 56 Da fragment), THP protected alcohol 9 (CE = 22, 84 Da fragment), (+)-JQ-1 10 (CE = 23, 56 Da fragment), and indomethacin 11 (CE = 29, 219 Da fragment). Whole 384-well plate reactions were performed spanning 64 reaction conditions for a range of miniaturized chemical transformations to access 6 chemical products in each case (a, 7a-7f, b, 7g-7l, c, 8a-f, d, 9a-f, e, 10a-f, f, 10g-l, g, 9g-l, h, 11a-f, i, 11g-l). Relative % were calculated as described in Extended Data Fig. 6 and were uniformly consistent with supporting LC-MS data. Protocols can be found in Supplementary Information and data can be found in Extended Tabular Data Sheets 23–31.

Extended Data Fig. 8 |. Eight-plex NL-ADE-MS analysis.

Direct comparison individual and multiplexed data were collected for each component of the 8-plex experiment described in Fig. 4. Head-to-head comparison showed that multiplexed data were tightly correlated to individual data for: a, Boronate 1, b, Boronate 14, c, Boronate 15. d, Boronate 16, e, Boronate 17, f, Boronate 18, g, Boronate 19, and h, Boronate 20. Colors reflect individual specific reaction conditions. Using 4 TIDA boronates i, 1, j, 14, k, 15, and l, 16, and all 96 amines (21-116) as representative examples, NL-ADE-MS allows reaction conditions to be prioritized for specific boronates directly from ultra-high throughput experiments. These data show excellent correlation against LC-MS data (R² (avg) = 0.98). In all cases the best condition was selected or was within 5% where selection of either condition would provide a similar outcome. This subset of the NL-ADE-MS data required only 7.68 min whereas the equivalent LC-MS dataset required 6.4 days of continuous data collection a 1,200-fold increase in pace. Inset tables show the top 3 conditions selected by NL-ADE-MS. From an amine centered standpoint, using all 8 TIDA boronates (1, 14–20) and amines (45–68) as a representative example, NL-ADE-MS allowed reaction conditions to be prioritized for specific amines directly from ultra-high throughput experimentation (representative example shown in panel m, for other examples see Extended Tabular Data Sheet 37). These data show excellent correlation against LC-MS data (R²(avg)= 0.91, see supplementary tabular data sheet 37). n, In 8 of 24 cases the best condition was directly selected, however in many cases condition data were close to one another (i.e. panel m). With the goal of assessing the utility of these data for selecting the highest output chemical reaction conditions we applied a nearest neighbor constraint of “within 5%” (panel c). This allowed much more effective determination of reaction data, and by using these constrains we found that in 22 of 24 cases (92%) the best conditions selected by NL-ADE-MS were within 5% of the top 3 conditions identified by LC-MS. Protocols can be found in Supplementary Information and data can be found in Extended Tabular Data Sheets 40 and 41.

Extended Data Fig. 9 |. Six-plex NL-ADE-MS analysis for diverse transformation.

A suite of transformations were investigated for their suitability for multiplexing in a 6-plex format spanning products derived from a, C-O coupling of TIDA boronate 1; b, reductive amination of TIDA boronate 2; c, Buchwald-Hartwig coupling of clarithromycin derivative 5; d, Suzuki coupling of clarithromycin derivative 5; e, Buchwald-Hartwig coupling of boc protected amine 7; f, Buchwald-Hartwig coupling of THP protected alcohol 9; g, Buchwald-Hartwig coupling of (+)-JQ1 10; h, Amidation of C5-lenalidomide 12; and i, Buchwald-Hartwig coupling of C5-lenalidomide Br 13. Multiplexing reduced the data acquisition time from 7.68 min to 1.28 min for each set of 384 reactions. Relative product % was normalized to the highest output by both individual and multiplexed data, then averaged across the 6 expected products to provide plots featuring rank ordering of the 64 possible reaction conditions. Tight correlations between individual and multiplexed data were observed in all cases, even in data sparse scenarios where ~80% of data were non-productive reactions as seen in the derivatization of clarithromycin building block 5 (panels c and d). Fragmentation parameters were defined by the respective starting materials clarithromycin derivative 5 (CE = 32, 158 Da fragment), boc protected amine 7 (CE = 15, 56 Da fragment), THP protected alcohol 9 (CE = 22, 84 Da fragment), (+)-JQ-1 10 (CE = 23, 56 Da fragment), C5-lenalidomide 12 (CE = 25, 111 Da fragment), and C5-lenalidomide-Br 13 (CE = 26, 111 Da fragment). Protocols can be found in Supplementary Information and data can be found in Extended Tabular Data Sheets 42–50.

Extended Data Fig. 10 |. Further applications of fragmentation first experimentation.

Rapid quantitative readouts can be collected using NL-ADE-MS where appropriate standards are available allowing direct assessment of conversion for a, Buchwald-Hartwig coupling of TIDA boronate 1; b, Suzuki coupling of TIDA boronate 1; c, C-O coupling of TIDA boronate 1; d, Suzuki coupling with clarithromycin derivative 5; e, Buchwald-Hartwig coupling of (+)-JQ1 10; and f, amidation of C5-lenalidomide 12. These quantitative starting material conversion data align well with relative product outputs providing an orthogonal proxy for reaction progress in high-throughput formats. g, The unifying fragmentation features which underlie NL-ADE-MS allow reaction monitoring in native contexts using a single method to collect multiple data points per time point in parallel for the hydrolysis of TIDA boronates 1, S2, and S5 under aqueous basic conditions. h, Beyond NL-ADE-MS common fragmentation features can drive selective extraction and analysis of reaction data using conventional LC-MS/MS. Relative % were calculated as described in Extended Data Fig. 4. i, A central aspect of reaction development is tracking starting material fate, fragmentation-first thinking simplifies this process because starting material associated entities can be directly extracted by their unique fragmentation signatures. In a representative example, Buchwald-Hartwig coupling shows the expected starting material and product signals alongside side products including proto-dehalogenation and hydroxylation of the activated 2-pyridyl halide, alongside other ions at m/z 404 and 472 which would warrant further study. Protocols can be found in Supplementary Information and data can be found in Extended Tabular Data Sheets 51–58.

Fig. 1 |. Commonality outlines a roadmap to rapid analysis.

a, Common chemical features are found throughout chemical synthesis from modular routes to targeted approaches. b, Standard methods for analysing chemical reactions typically follow one-at-a-time asynchronous workflows to chromatographically separate desired analytes from other components. c, Commonality of synthetic origin generalizes reaction analysis by embedding starting material-specific fragmentation fingerprints into the resulting products. These fingerprints allow direct, simple and sensitive extraction of product signals by tandem mass spectrometry, which eliminates the need for chromatography enabling analysis to operate in a continuous and multiplexed manner. MS, mass spectrometry.

Fig. 2 |. Loss of common fragments drives single-second reaction analysis.

a, TIDA boronates enable rapid identification by tandem mass spectrometry through the loss of neutral fragments at a pace of 1.2 s per sample. b, A wide range of miniaturized chemical reactions were readily applicable to NL-ADE-MS analysis (see Extended Data Fig. 4). c, Data derived from NL-ADE-MS allow optimal solvents, catalysts and bases to be assigned to chemical reaction products. %LC-MS and %NL-ADE-MS are scaled relative to the maximum MS intensity for a specific product and then averaged across all target products for each reaction condition. Bpin, pinacol boronic ester; BTMG, 2-tert-butyl-1,1,3,3-tetramethylguanidine; MS, mass spectrometry; LC-MS, liquid chromatography-mass spectrometry, DMSO, dimethylsulfoxide; DMF, dimethylformamide; DMAc, dimethylacetamide; NMP, N-Methyl-2-pyrrolidone; Ph, Phenyl; Ms, Methane sulfonyl.

Fig. 3 |. Fragmentation-first design of high-throughput experimentation.

a, Fragmentation pattern analysis of chemical building blocks identifies parameters that inform rapid analysis of chemical reaction products by NL-ADE-MS. b–d, Many classes of chemical targets are amenable to this approach, including natural product derivatives (b), protecting groups (c) and drug-like building blocks (d). Optimal reaction conditions and solvents could be rank ordered by NL-ADE-MS for a wide range of reaction types (see Extended Data Figs. 6 and 7).

Fig. 4 |. Ultrahigh-throughput multiplexed analysis of chemical reaction mixtures.

a, Automated liquid handling enabled the creation of 12,288 reaction mixtures spanning 738 potential chemical products. Acoustic liquid handling enabled analytical samples to be reformatted into a single 1,536-well plate for analysis. Multiplexed NL-ADE-MS determined the relative outcomes of these 12,288 reactions in 32 min. Comparative LC-MS data showed that multiplexed NL-ADE-MS selects the best reaction conditions with comparable efficiency for specific halides (b), amines (c), or across substrates (d). The data point colour scheme represents each of the 16 separate reaction conditions, see Supplementary Information for further details. %LC-MS and %NL-ADE-MS are scaled relative to the maximum MS intensity for a specific product and then averaged across all target products for each reaction condition.