Generalization of High-Throughput Experimentation in Organic Chemistry: Case Study on the Flortaucipir Synthesis

Abstract

High-Throughput Experimentation has undergone an outstanding evolution in the past two decades and has proven to be a game-changer in the acceleration of reaction discovery and optimization. Despite a good implementation in the pharmaceutical industry and a demonstrated accessibility to the technology, the generalization of High-Throughput Experimentation as a standard method for optimizing reactions is not yet observed. The perspective aims at discussing the necessity of generalizing such technologies, supported by the case study: the optimization by High-Throughput Experimentation of a key step in the synthesis of Flortaucipir, an FDA-approved imaging agent for Alzheimer’s diagnosis.

1. Introduction

Organic synthesis plays a key role in many crucial industrial sectors such as materials, energy, agrochemicals, cosmetics, and Drug Discovery and Development (DDD) [1]. Devising specific, selective, and robust new chemical reactions giving access to more complex molecules is therefore critical for efficiently managing public health, societal problems, and human welfare. DDD is particularly demanding in novel chemical solutions to help solve the problems posed by the diversity of pathologies, the emergence of personalized medicine, and the high rate of drug attrition during clinical trials. However, DDD is a long, expensive, and risky business that takes 10–15 years. Up to now, one of the most credible solutions for derisking DDD is to test a maximal number of relevant molecules. In the last two decades, High-Throughput Experimentation (HTE) has popped up as one of the best solutions to accelerate and improve reaction discovery and optimization [2,3,4,5,6,7,8,9,10]. It brought a drastic shift in how researchers design and execute chemical experiments (Figure 1). This approach consists of the miniaturization and parallelization of reaction conditions, enabling a large number of experiments to be run simultaneously, with a significant material/time/cost/waste reduction and high reproducibility. This is especially true in catalysis, where the number of conditions to be evaluated increases exponentially.

Over the past decade, HTE platforms developed for synthetic organic chemistry have been created, from standard screening protocols at the micromole scale in a 96-well plate format to HTE campaigns currently conducted at the nanomole scale in 1536-well plates. HTE plays a central role in expediting reaction discovery and optimization, emerging as the method of choice for generating reliable and standardized large experimental datasets, usable to feed predictive algorithms [11,12,13]. This technique has been harnessed and strengthened, leading to robust and consistent methodologies across a wide spectrum of chemical reactions commonly employed by organic chemists. This is especially true in catalysis, where the selection of catalysts and ligands is crucial. Although it has been well implemented in big pharma industries, HTE remains accessible to a very restricted number of chemists, particularly in academia. This is partially due to a lack of systematic training of young generations of chemists on new technologies and also to the false feeling that HTE is a dramatically expensive technology, requiring robotization and automation, despite several attempts from academic HTE centers to prove the reverse [14]. With this perspective, we aim to highlight the incontestable benefits of moving from standard rationally designed batch optimization to HTE methodology. To support our discussion, we applied our strategy to the re-optimization of a key step of the Flortaucipir synthesis.

2. Materials and Methods

HTE campaign: The screening of reaction conditions by High-Throughput Experimentation was performed in a 96-well plate format using 1 mL vials (8 × 30 mm vials number 884001 from Analytical Sales and Services, Flanders, NJ, USA, Figure S1b) in a Paradox reactor (96973 from Analytical Sales and Services, Figure S1c of the Supporting Information). The homogeneous stirring was controlled with stainless steel, Parylene C-coated stirring elements, and a tumble stirrer (VP 711D-1 and VP 710 Series from V&P Scientific, San Diego, CA, USA, Figure S2). The liquid dispensing was performed using calibrated manual pipettes and multipipettes (Thermo Fisher Scientific, Waltham, MA, USA/Eppendorf, Hamburg, Germany). The HTE experiments were designed using an in-house software called HTDesign^® and developed by the GIPSI team at CEA Paris-Saclay (DRF/Joliot) (see Supplementary Information). At the end of the reaction, each sample was diluted with a solution containing 1 µmol of biphenyl as an internal standard (500 µL, 0.002 M) in MeCN. 50 µL aliquots were sampled into a 1 mL deep 96-well plate containing 600 µL MeCN. Ratios of Area Under Curve (AUC) of starting material, products, and side products were tabulated, as well as areas for peaks of interest. The analytical method was described in the materials and equipment.

Analysis: LC-MS spectra were recorded on a Waters Acquity UPLC^® (Waters Corporation, Milford, MA, USA) equipped PDA eλ Detector and SQ Detector 2, mobile phase A: H₂O + 0.1% formic acid, mobile phase B: acetonitrile + 0.1% formic acid.

All additional information regarding the materials and methods used in this study is reported in the Supporting Information.

3. Discussion and Results

To date, many drawbacks remain unanswered when talking about the way chemists think and perform their reaction optimization. The most important one might be the trustworthiness of the results. A non-negligible number of results published in the literature are not reliable/reproducible [15]. This is partially because duplicates/triplicates are not performed, as is systematically the case in biology. Another reason is the description of protocols, mostly under-described: exact temperature of the room, a stirring system employed, including rate per minute (rpm), batches of reagents, light power in photoredox synthesis, etc. [16,17,18,19]. This non-reproducibility might create bias in the comparison of experiments in an optimization reaction, leading to inappropriate decisions. Another bias remains in the choice of parameters screened in the rational approach. As this approach is time-consuming, a restricted number of parameters is selected in many cases. The chance of missing the good parameter to screen, but even more so the most efficient combination of parameters, is non-negligible. The main consequence is the lack of transposability of such methodologies to analogous compounds, a problem that life science industries are facing daily. Last, but not least, negative results remain mostly under-discussed/reported. Those non-reported results are a significant loss for the entire community, as many failed reactions have likely been reproduced by other laboratories. In the meantime, negative results are highly valuable and can provide key insights when studying mechanistic questions, as well as for the development of reliable predictive models [20,21,22,23,24].

When chemists in life science industries tackled the question of acceleration of reaction optimization with HTE, originally to accelerate the Active Pharmaceutical Ingredients (API) production, a fantastic opportunity appeared to answer all the above-mentioned drawbacks. HTE is the method that allows miniaturized reactions to be tested in parallel, enabling faster and more efficient exploration of multiple variables at once. Unlike the traditional one-variable-at-a-time approach (OVAT), HTE accelerates data generation for reaction optimization, discovery, and target molecule access. It also improves cost and material efficiency while providing rich and reliable datasets that can also enhance the performance of machine learning algorithms. It is important to mention that HTE can be implemented in both automated and semi-manual formats. Automated systems offer a high degree of precision and reproducibility by using robotics and software to handle tasks with minimal human action [25,26,27,28,29]. Semi-manual setups, while less automated, still allow for efficient parallel experimentation by combining manual input with some automated processes, making HTE accessible even in laboratories without full automation capabilities [14,30]. Regardless of the setup, experimental design and reaction plate layout are critical to the success of the workflow and must be carefully planned in advance. This includes considerations such as the choice of stirring system, typically tumble stirrers, and the appropriate analytical methodologies to ensure reliable and consistent results. To illustrate the advantages of HTE in reaction optimization, compared to the classical rational trial-error approach, Radar graphs representing the evaluation of both HTE and traditional optimization approaches on 8 different aspects have been drawn, evaluation performed by six chemists, from various horizons (academia or pharma industry, synthesis methodology or drug design). The various evaluated aspects will be discussed in an attempt to convince the community that accelerating their processes is only one of the benefits from the adoption of this technology (Figure 2).

Accuracy: HTE allows researchers to run a large number of experiments under tightly controlled conditions, which is key to accurately identifying the best reaction parameters for a given (catalytic) process. This accuracy comes from several factors: (i) Precise control of variables [31,32]: using parallelized systems and robotics, parameters such as temperature, pressure, solvent, catalyst type, concentration, and time can be kept consistent across experiments, reducing the chance of human error. (ii) Minimization of bias: running experiments in parallel under identical conditions removes many sources of variability, resulting in more reliable and reproducible data. (iii) Real-time monitoring and analysis: Many HTE platforms are paired with analytical tools such as spectrometry or chromatography, allowing researchers to track reaction progress and product distribution as they happen. This provides more accurate measurements of both reaction kinetics and product yields [33].

Reproducibility: Reproducibility is a critical parameter in catalysis to ensure that the results from experimental studies can be reliable and translated to industrial processes, for instance. HTE offers several advantages in this sense. HTE reduces the variation attributed to the operator, which often can lead to inconsistent results. This is even more the case in automated workflows, where the traceability of potential errors in dispensing, for instance, is possible. The same experimental setup can be used across multiple runs, improving the consistency and reproducibility of outcomes, or to discard an erroneous experiment. Also, while performing hundreds/thousands of experiments in parallel, HTE allows for the collection of large amounts of data. This provides a more robust statistical analysis and also reduces the impact of outliers, ensuring that the results are reliable and reproducible across an entire project. Finally, HTE makes it easy to repeat experiments under the same conditions and obtain consistent catalytic performance, which is especially important for industrial scaling. It is important to note that while HTE provides insights, trends, and hits, the yields observed may differ when reactions are scaled up to batch or industrial levels, an inherent limitation of any scale-up process.

Deep investigation: HTE provides the ability to conduct a deeper, more comprehensive investigation into reaction mechanisms, catalyst behavior, and the effects of multiple variables [34]. Exploration of a wide parameter space is one of the most important aspects of HTE. Traditional methods may be limited by the number of variables that can be tested due to time and resource constraints. HTE allows researchers to simultaneously explore a vast range of reaction parameters, enabling a more thorough understanding of how variables like catalytic combination, catalyst concentration, steric effects in bases or ligands, solvent, or even temperature affect the reaction outcome. HTE also allows for the discovery of non-obvious trends. With a higher number of data points, researchers can identify subtle correlations and trends that may not be apparent in smaller experimental setups. This deeper insight can lead to the discovery of more effective catalysts, novel reaction pathways, or even previously overlooked reaction conditions. This is crucial to designing catalysts that maintain high activity and selectivity over time in industrial applications. Finally, larger investigations give rise to serendipity, where non-intuitive hits can be observed.

Negative results: As mentioned in a recent report by Marisa Koslowski [35], the lack of negative results in publications is a highly detrimental aspect. When omitting to present the negative results obtained during reaction optimization of scope and limitation investigation, even in Supporting Information, researchers prevent the scientific community from gaining critical knowledge. Knowing that a reaction or a substrate is not working under specific conditions is highly valuable information that can lead to further reaction developments or mechanistic understanding. Those omissions might be partially due to the pressure of competition. If this phenomenon tends to change slowly, many negative results are still omitted. When performing HTE optimization, the negative results are already well estimated, as their presence justifies the good results.

Machine Learning: Compared to batch experiment data found in the literature, HTE can provide standardized, high-quality datasets with minimized bias and improved reproducibility. Literature data often suffer from inconsistencies, underreported negative results, and incomplete parameter documentation, limiting their predictive power. In contrast, HTE explores broader chemical spaces under well-controlled conditions, generating diverse and reliable datasets ideal for machine learning models. By definition, it can allow for more accurate predictions, better generalization, and efficient reaction optimization. Utilizing HTE datasets will not ensure predictive models’ success but will help machine learning move beyond noisy and fragmented literature data while ensuring a more robust and trustworthy feedstock with limited bias. This initiative is already encouraged by repository databases such as the Open Reaction Database [36].

Case study: With these different points highlighted, we are convinced that the generalization of HTE would be of great benefit to the entire community as a technology to improve and accelerate our current research but also to capitalize on current research for future discovery via predictive approaches. To highlight our thoughts with a case study, we have studied the patented and published syntheses of Flortaucipir (Figure 3). Flortaucipir (formerly known as AV-1451 or T807) is a radiotracer used in positron emission tomography (PET) imaging to detect tau protein deposits in the brain [37,38]. It is primarily utilized in the diagnosis and study of neurodegenerative diseases such as Alzheimer’s disease, where tau tangles play a key role in disease progression. By binding to aggregated tau, Flortaucipir helps to visualize tau pathology in living patients, aiding in research and potential early diagnosis. Approved by the FDA in 2020 under the name Tauvid, it represents a significant advancement in understanding and monitoring tau-related brain disorders.

In the synthesis of Flortaucipir, two key steps have attracted our attention: the Pd-catalyzed Suzuki-Miyaura reactions (steps I and III). Pd-catalyzed reactions represent one of the most screened classes of reactions in HTE [39]. This is explained by the robustness of such reactions and their common use in pharma industries, with C-C coupling being one of the most important synthetic tools for chemists in medicinal chemistry [40,41,42,43,44,45,46,47,48]. After analysis of the literature, we noticed that the reported yields for reaction I seemed quite low regarding the literature-reported yields for such starting building blocks. Reaction conditions reported for the step I have been identified without much optimization, leading to 72% isolated yield using Pd(PPh₃)₄ as catalyst, Na₂CO₃ as base, and 1,4-dioxane/H₂O as solvent system at reflux for 24 h, or 65% isolated yield using Pd(PPh₃)₄ as catalyst, K₂CO₃ as base, and toluene/H₂O as solvent system at 50 °C overnight, as, respectively, published by Zheng et al. [49] and patented by Crew et al. [50].

Figure 3. (a) General synthesis of Flortaucipir: I. Pd-catalyzed Suzuki-Miyaura cross-coupling reaction; II. Intramolecular reductive cyclization; III. Pd-catalyzed Suzuki-Miyaura cross-coupling reaction; (b) reaction conditions and associated isolated yields published/patented for steps I. Adapted from [49,50].

Figure 3. (a) General synthesis of Flortaucipir: I. Pd-catalyzed Suzuki-Miyaura cross-coupling reaction; II. Intramolecular reductive cyclization; III. Pd-catalyzed Suzuki-Miyaura cross-coupling reaction; (b) reaction conditions and associated isolated yields published/patented for steps I. Adapted from [49,50].

Such moderate yields, published or patented, with what we suspect are non-optimal reaction conditions, have been considered by other chemists as “ready-to-use” conditions, leading to repeated experiments with a lack of efficiency. We decided to optimize this specific step by a simple HTE campaign, under semi-manual workflow [14], as proof of concept of this perspective.

Optimization of the reaction condition in step I: the optimization of step I involves the coupling of (4-Bromophenyl)-boronic acid and 3-bromo-4-nitropyridine (Figure 4a). Based on the reported literature on these specific building blocks and general knowledge of Suzuki-Miyaura cross couplings, we have identified the most relevant conditions to be screened (Figure 4b). The catalytic systems screened were the following: Pd(dppf)Cl₂, Pd(PPh₃)₄, Pd₂dba₃/PPh₃, Pd₂dba₃/BINAP, Pd₂dba₃/XantPhos, Pd₂dba₃/SPhos, Pd₂dba₃/PCy₃ and Pd₂dba₃/dppf. Bases screened were Na₂CO₃ in the absence of water, Na₂CO₃ in the presence of water, and Cs₂CO₃ in the presence of water. Finally, the solvents selected were dioxane, DMF, THF, and toluene. The HTE campaign has been performed in a 96-well plate format at a 10 µmole scale with a volume of 100 µL per vial (concentration of 0.1 M). Results are reported as a heat map representing the ratio of the Area Under Curve (AUC) of the desired product to the AUC of the internal standard (4-tert-butyl biphenyl), and the performances of the variables (bases, catalytic system, solvent) are represented as pie charts (Figure 4c,d). The conditions reported in the literature are mentioned in the plate design as well as the best hit conditions (purple squares and yellow squares, respectively, in Figure 4c).

The hit identified as the higher ratio AUCProduct/IS over the HTE campaign (vial E12) highlights Pd₂dba₃ as the Palladium source, XantPhos as the ligand, toluene as the most efficient solvent, and Cs₂CO₃ as the base. This is in accordance with the focused parameter performances in Figure 4d and shows a clear trend in favor of toluene. However, the presence or absence of water seems not to have any influence. Cs₂CO₃ shows a slightly better efficiency over Na₂CO₃, when, surprisingly, the catalytic system does not seem to be an important parameter. The reaction has been reproduced in batch, affording 84% yield, using a calibration curve and LC-MS analysis, which is higher than reported yields, as we have reproduced both vial A5 and vial A8 reaction conditions, both corresponding to the reported conditions, and affording up to 50% yield. This result comforts us in our approach, highlighting the interest of HTE-based optimization as a more routinely used methodology, and is proof of the necessity for a systematic screening of all combinations, in complement to human intuition.

In summary, we have illustrated our above-mentioned discussion, arguing the interest in systematic, even when basic, optimization of reaction conditions by reoptimizing a key step of the Flortaucipir synthesis. With those new conditions, the community is now equipped with a more efficient synthesis strategy, which is extremely important when it comes to radiolabeled synthesis [51]. Importantly, further improvements could be achieved by optimizing steps II and III to significantly enhance the overall yield efficiency.

Our case study on the reoptimization of a key step in the synthesis of Flortaucipir illustrates the previous discussion on the significant advantages of a generalization of HTE. By screening a broader parameter space, we identified optimized reaction conditions that yielded higher efficiencies compared to previously reported methods. This not only improved the overall yield of the synthesis but also provided a more robust and reproducible methodology for the community. Despite its success in the pharmaceutical industry, the widespread adoption of HTE remains limited, particularly in academia. Barriers such as cost misconceptions and a lack of systematic training hinder its integration into standard practice. However, as demonstrated in our study, even simple HTE campaigns can lead to substantial improvements, challenging the notion that HTE is an inaccessible or excessively complex technology. To fully leverage the benefits of HTE, the scientific community must embrace its implementation beyond industrial settings. Training future chemists in HTE methodologies and encouraging the systematic reporting of experimental data, including negative results, will foster a more efficient and transparent research landscape.

4. Conclusions

Despite its success in the pharmaceutical industry, the adoption of HTE remains limited, particularly in academia. This perspective aimed to highlight and discuss the numerous advantages of this technology that would benefit the scientific community. Through the reoptimization of Flortaucipir synthesis, we demonstrated HTE’s ability to identify better reaction conditions efficiently. This case study underscores the necessity of systematically integrating HTE into research workflows. By embracing this technology, the scientific community can accelerate innovation, reduce inefficiencies, and enhance data reliability, ultimately advancing both fundamental and applied chemistry. Wider implementation of HTE will undoubtedly shape the future of organic synthesis and high-quality data collection.