MTCiting: Process Chemistry and Pharmaceutical Manufacturing Optimization

Syngene Amgen Research and Development Center; Amgen Inc.   

The authors report the development of AMG 193, a small-molecule therapeutic exhibiting promising antitumor activity across various cancers. This study details the process development efforts to synthesize multi-kilogram quantities of AMG 193 for clinical evaluation.

A robust manufacturing process was established within an accelerated timeframe, delivering high-purity material through the integration of multiple biocatalytic and chemocatalytic transformations. Key chiral building blocks, including a morpholine fragment, were synthesized via enzymatic catalysis. Additionally, a one-pot synthesis was developed to access the naphthyridine core through an Ir-catalyzed C–H borylation, followed by a Pd-catalyzed Suzuki-Miyaura coupling and subsequent cyclization. While this telescoped sequence was efficient at the bench scale, scalability issues led to irreproducible yields during pilot-scale campaigns.

To address these challenges, the authors performed a systematic investigation of the reaction kinetics. Although previous literature suggested a mandatory hold period for Iridium-catalyzed C–B bond formation, real-time monitoring via in-situ ReactIR and NMR revealed that the initial N-borylation of the aminopyridine intermediate with pinacolborane (HBpin) required an unexpected 12 hours. Furthermore, extended preincubation (up to 20 hours) paradoxically decreased the yield of the subsequent C4 borylation, indicating that excess HBpin inhibited the catalytic cycle.

By replacing HBpin with bis(pinacolato)diboron (B₂pin₂), the authors eliminated the preincubation step and achieved a five-fold reduction in catalyst loading. This optimized site-selective Ir-catalyzed borylation, combined with a highly enantioselective biocatalytic imine reduction, successfully enabled the manufacture of hundreds of kilograms of the naphthyridine and chiral morpholine intermediates.

Arunachalampillai, A., Chandrappa, P., Crockett, R. D., Gaines, C. S., Hu, K., Judd, T., Kommuri, V. C., Murray, J. I., Nidhiry, J., Ortiz, A., Robinson, J. A., Rötheli, A. R., Vernon, R. M., Wei, C. S., Wells, S., Wilson, C. V., Xu, S., Yamamoto, K., & Zetzsche, L. E. (2025). Development of a Manufacturing Route toward AMG 193, an MTA-Cooperative PRMT5 Inhibitor. Organic Process Research & Development, 29(11), 2934–2941. https://doi.org/10.1021/acs.oprd.5c00310

Georgia Institute of Technology; Universidade Federal do Rio de Janeiro; Rio de Janeiro State University 

The authors highlight a critical gap in pharmaceutical process control: while advanced control systems for crystal size distribution (CSD) are widely discussed in literature, they are seldom validated beyond numerical simulations. This study addresses this by developing a robust framework for soft sensors that process Attenuated Total Reflection Fourier-Transform Infrared (ATR-FTIR) spectra and Focused Beam Reflectance Measurement (FBRM) data for integration into a Population Balance Model (PBM)-based Nonlinear Model Predictive Controller (NMPC).

The NMPC strategy uses a nonlinear internal model to regulate the process by predicting future outputs based on historical process data and optimized future inputs. This approach was evaluated using the unseeded crystallization of paracetamol in ethanol to control both crystal size and yield via temperature modulation.

Experiments were conducted in an Optimax synthesis workstation integrated with a comprehensive process analytical technology (PAT) suite:

• ReactIR (ATR-FTIR): For real-time monitoring of solution concentration.
• ParticleTrack (FBRM): To capture chord length distributions (CLD).
• EasyViewer: For in-situ crystal imaging.

Machine learning architectures were utilized to translate raw PAT data into state variables for the controller. 

• Symbolic Regression: This model transformed ATR-FTIR spectra into concentration measurements, which were subsequently used to estimate the third-order moment (_u3) of the CSD.
• Artificial Neural Networks (ANNs): The ANNs provided accurate predictions of the zeroth, first, and second moments (_{u0, u1, u2}) across training and validation datasets.

These soft sensors enabled the NMPC to operate online, utilizing concentration data and the first four moments of the CSD (_{u0 - u3}) to compute optimal control actions in real time.

The controller was evaluated across five experimental cases with varying initial conditions and set-points. The NMPC successfully achieved the control objectives in three experiments. However, performance deviations in the remaining two cases were attributed to crystal disappearance (dissolution) and inaccuracies in supersaturation measurement.

The authors conclude that for high-fidelity NMPC performance, simulations must explicitly account for crystal disappearance mechanisms. Furthermore, they emphasize that a PBM-based NMPC requires a highly accurate solubility model and comprehensive consideration of the thermodynamic and kinetic complexities inherent in the crystallization process.

Lima, F. a. R. D., De Moraes, M. G. F., Secchi, A. R., De Souza, M. B., & Grover, M. A. (2025). Experimental nonlinear model predictive control of crystal size and yield in batch cooling crystallization enabled by soft sensor and Symbolic-Based calibration model. Industrial & Engineering Chemistry Research, 64(49), 23582–23600. https://doi.org/10.1021/acs.iecr.5c03894