42.2.1 Mid‐infrared Spectroscopy

IR spectroscopy examines molecular vibrations based on the absorption of light in the mid‐IR region of the electromagnetic spectrum (~400–4000 cm⁻¹). As the IR absorption bands are often unique to the vibrational modes of chemical functional groups, IR spectroscopy provides a high degree of structural specificity in the analysis of liquids, solids, and gases and has long been used by analytical chemists in structural determination of unknown materials. The enabling technology underlying most IR spectrometers is the Michaelson interferometer, a common optical assembly that uses light interference to measure distances in wavelength units. IR light is transmitted from a source through a beam splitter and movable mirror; after interaction with the sample, an interferogram is detected and then converted to a true absorption spectrum by Fourier transform. Hence the technique is commonly known as Fourier transform infrared spectroscopy (FTIR).

A key difference in FTIR spectrometers used for real‐time analysis relative to standard benchtop units is the configuration of the optical element that transmits light from the spectrometer to the interface with the reaction stream. Traditional benchtop instruments measure IR absorption through a fixed length of sample in a transmission mode. In contrast, online measurements primarily employ a sampling element based on attenuated total reflectance (ATR) (Figure 42.4). In ATR measurements, light is transmitted through a solid material and reflected on the surface in contact with the sample before being sent back to the detector. This reflection establishes an evanescent wave that penetrates less than 2 μm into the sampled medium. Thus the ATR element enables in situ monitoring of the reaction being analyzed with zero sample preparation. ATR materials are chosen for low absorbance and high index of refraction, but as they are directly immersed in reaction streams, chemical inertness is a requirement. Diamond is frequently the material of choice, as it exhibits good IR transmission and is chemically inert to the wide range of chemistries employed in chemical development. Moreover, diamond is very hard, resists abrasion by solid particles, and is stable across a wide range of temperatures and pressures. One drawback is that diamond strongly absorbs IR radiation in the 1900–2300 cm⁻¹ region, which overlaps with vibrational modes corresponding to triple and cumulated double bonds. Silicon is an alternative ATR element that provides high sensitivity across the IR spectrum and is thus ideal for monitoring reactions involving azides, acetylenes, carbon dioxide, etc. However, silicon is less chemically resistant, particularly to strong aqueous bases. Table 42.2 summarizes the physical properties of the materials generally used in ATR elements.

An important challenge of PAT tools that incorporate ATR elements is the difficulty of transmitting light from the spectrometer to the sampled medium. Housing the ATR element in a metal probe, usually Hastelloy for its chemical inertness, facilitates light transmission to the reaction medium as well as the adequate ruggedness to withstand the breadth of chemical environments used in chemical development (strong acids and bases, very high or low temperatures, high pressure, etc.). In early FTIR instruments, the probe and the spectrometer were connected by a hollow light path with a series of adjustable mirrors to focus the IR laser beam to the probe. This system provides high performance and low attenuation across the spectrum, but achieving proper alignment of the probe requires a good level of expertise. A more robust, user‐friendly technology is the use of polycrystalline silver halide fibers, which provide a flexible connection between the probe and the spectrometer. Compared with probes with mirrored ends, fiber‐optic conduits can be aligned in a straightforward manner and display a lower sensitivity to movement and vibration. Since light attenuation through fiber optics increases with length and can be affected by conduit curvature, limitations exist on the physical size and configuration of the fibers, with typical fiber lengths being less than 3 m. One other consideration for the use of FTIR spectroscopy is that water strongly absorbs in the IR region. While this does not totally preclude the technique’s used in aqueous systems [4], it can make analysis difficult as many important regions of the spectrum may be obscured. In general, for solvents with absorptions that overlap with the analyte bands, adequately scaled solvent subtraction techniques [5] may facilitate reaction monitoring. When solvent subtraction is not suitable, other chemometric treatments such as principal component analysis (PCA), partial least squares (PLS) regression, principal component regression (PCR), or multiple linear regression (MLR) may be used to deconvolute reaction mixtures [6]. Figure 42.5 displays the IR transmission bands for solvents commonly used in chemical reactions. In addition to monitoring reactions, FTIR probes are routinely used to monitor crystallizations as the short penetration depth into the sample allows for supersaturation studies [7].

42.2.2 Raman Spectroscopy

Raman spectroscopy probes molecular vibrations and rotations in the 100–3000 cm⁻¹ frequency range. It is considered a complimentary technique to mid‐IR spectroscopy that analyzes the wavelength shifts associated to Raman scattering. When a sample is excited with monochromatic light, a small portion of the scattered radiation undergoes quantized changes in frequency that result in vibrational and rotational transitions. The Raman effect accounts for the new frequencies in the spectrum of the monochromatic light and is reported in “Raman shift” away from the monochromatic light source: Δν = ν_± − (ν_± ± ν₁) = ±ν₁, which is independent on the frequency of the monochromatic light, ν_±.

Whereas IR absorption only occurs when the excited vibrational mode induces a change in dipole moment of the molecule, Raman scattering relies on a change in the molecule’s dipole polarizability – the temporary distortion of the electron density around bonds. Hence, a given vibrational mode may be “Raman active,” “IR active,” or both, and Raman spectra and IR spectra can be very different for a given molecule. For example, Raman spectroscopy can detect symmetric vibrations in diatomic molecules such as chlorine, oxygen, etc., whereas IR spectroscopy cannot. In a standard Raman spectrometer, monochromatic light from a laser is directed to a probe immersed in a reaction stream, where it interacts with the analyte. The scattered photons are then returned to the spectrometer and passed through a series of optical filters selected to exclude the wavelength of the laser itself – the majority of the light returning to the spectrometer will be Rayleigh scattered and thus be the same as the excitation wavelength. The strength of the Raman effect depends on the excitation wavelength and is dramatically increased at high excitation frequencies. Ideally, a lower wavelength light source will provide a stronger Raman spectrum. In practice, however, many samples will fluoresce when irradiated with low wavelength light. This fluorescence overwhelms the Raman scattered photons, making spectral interpretation difficult or impossible. Most commercial laboratory Raman instruments use one of three readily available laser lines, 532, 785, or 1000 nm. Among these, the 785 nm wavelength is the most common, as it provides a good balance between Raman intensity and low fluorescence in reaction systems relevant to pharmaceutical development. It should be noted that ambient light collected by the probe will also be detected and interferes with Raman spectra, so quantitative analysis requires that samples be protected from stray light sources.

Like FTIR probes, Raman probes generally consist of an optical window at the end of a Hastelloy pipe. The optical elements of a Raman probe are not ATR based, but instead rather focused into the reaction sample a few millimeters away from the probe tip. Many materials can be used for the window, including diamond, quartz, fused silica, and sapphire. The latter is the most common material because it offers a suitable arrangement of chemical inertness, strength, and availability. Since glass is Raman inactive, Raman probes can be fit with long working distance optics that enable indirect or noncontact measurement; samples can be analyzed through the glass wall of a vial, instead of immersed directly in the reaction stream. One key advantage of Raman spectroscopy over IR is that light is sent from the laser source to the immersed probe via standard optical fiber. The low attenuation along these fiber runs allows for physical separation between probe and spectrometer of up to hundreds of meters, enabling remote analysis and easing implementation in crowded laboratories and processing facilities. Since water is Raman inactive, it is an excellent technique for analyzing biotransformation streams such as fermentations.

42.2.3 Near‐infrared (NIR) Spectroscopy

NIR spectroscopy is an effective PAT tool that has found wide application in process development, quality control, and raw material testing [8]. The NIR region of the spectrum extends from the upper wavelength end of the visible region at 770–2500 nm (4000–13 000 cm⁻¹) [9]. NIR absorption bands represent overtones or combinations of fundamental vibrational bands corresponding to the 1700–3000 cm⁻¹ region. Chemical bonds that link a hydrogen atom to a heavy atom display vibrational transitions in this region and constitute the foundation of the widespread use of NIR to monitor molecules containing C─H, N─H, O─H, or S─H moieties. Due to the overtone nature of the vibrations, NIR spectroscopy is not suited for structural elucidation or identification of compounds, but is well qualified to perform quantitative analysis aided by chemometric techniques. Nevertheless, despite the complexity of NIR spectra, wavelength–structure correlations that can be used for guidance in qualitative analysis have been tabulated (Table 42.3) [10]. In addition to in situ monitoring of chemical processes, NIR spectroscopy has been used in a variety of fields including medical diagnostics, food quality control, and atmospheric chemistry. Typical instruments available for reaction monitoring are both benchtop off‐line units, capable of analyzing solids or liquids, and in‐line fiber‐coupled probes for direct analysis in a reactor setting.

42.2.4 Ultraviolet–Visible Spectroscopy

Ultraviolet–visible (UV‐Vis) spectroscopy analyzers measure the absorption of a sample in the mid‐ultraviolet to visible region of the electromagnetic spectrum that arise from electronic transitions (~200–700 nm, 50 000–14 000 cm⁻¹). While lacking the chemical specificity of the techniques discussed above, it is commonly used to monitor liquid phase reactions due to its simplicity, sensitivity, and time resolution [11]. Despite spectral overlap, absorbance changes occurring during the reaction can be tracked and assigned to specific components with the aid of calibrated standards, chemometric techniques, or both. Moreover, analysis of the isosbestic points can help elucidate the number of absorbing species and lend support to mechanistic hypotheses [12]. A variety of sampling technologies are available for online and off‐line analysis, with both transmission and ATR‐based interfaces. UV‐Vis is commonly used for analysis of “colorful” compounds such as transition metal complexes [13] and highly conjugated molecules [14]. In the same way as other spectroscopic methods, the absorbance of the reaction solvents can interfere with the absorbance of the analytes, and lower wavelength cutoff limits below which solvent absorbance is excessive must be considered (Table 42.4). Since pharmaceutical processes are typically run at high concentrations in solvents that absorb UV light, care must be taken to avoid quantitative analysis in high‐absorbing regions of the spectrum by reducing path length in transmission mode or focusing on longer wavelengths.

42.2.5 Spectroscopy Interpretation (IR, Raman)

Complementing the observation of spectral differences detected during the course of a reaction with sound structural characterization can provide insight into reaction mechanisms. Spectral interpretation handbooks and databases are available in a number of online and printed resources [15],¹ and most contemporary software packages provide approximate functional group assignments for the data collected.² In general, this information should be helpful for a scientist to be able to assign characteristic peaks to a species one expects to observe in a particular reaction. For more accurate predictions of vibrational modes for a compound of interest, quantum mechanical packages such as Gaussian can be utilized [16].

42.3.1 General Reaction Types

Table 42.5 provides a summary of literature reports for common reaction types with their associated functional group transformations in the mid‐IR region. The abundance and diversity of the reactions explored with in situ spectroscopy methods speaks to the value of PAT to advance the design, implementation, and characterization of chemical processes.

42.3.2 PAT Application in Complex Reaction Systems

The analysis of transformations involving reactive intermediates or multiphase media often presents a challenge for conventional off‐line analytical techniques (e.g. HPLC, GC, NMR). During process development and plant execution of such processes, utilization of in situ spectroscopic methods offers a significant advantage from accuracy, time of analysis, and progress monitoring perspectives.

Off‐line techniques are often destructive for reactive intermediates and may not be able to detect the high‐energy species due to their thermal or chemical instability (Figure 42.6a). While the addition of a derivatization step may facilitate conventional analysis, reproduction of the intermediate concentrations demands an innocuous sampling and a derivatization reaction that is viable, quantitative, and significantly faster than the reaction being monitored. Similarly, representative sampling can be difficult for heterogeneous reactions where multiphasic mixtures (e.g. solid–liquid) are present (Figure 42.6b). A common off‐line approach to measure analyte concentration in slurries involves sampling the slurry, filtering out the solids, and then performing the analysis using chromatographic or spectroscopic methods. Immersion probes can simplify this measurement by using direct detection since the ATR element is in close contact with the liquid phase and the interference from the solids is generally inconsequential [44].

For pressurized reactions the use of in‐line spectroscopy can be crucial, as it eliminates the disturbance to the system associated with conventional sampling (Figure 42.6c). For example, sampling a catalytic hydrogenation under pressure often leads to modification of the reaction conditions due to exposure of the reaction mixture to non‐inert conditions, changes in composition, or variations in the partial pressure of hydrogen upon reaction restart. The ability to perform real‐time analysis using in‐line FTIR or at‐line NIR has been exploited to control selectivities in hydrogenations that afford a mixture of products [45]. Table 42.6 provides a brief summary of the benefits offered by PAT tools relative to conventional off‐line analysis methods.

42.4.1 Case Study 1: Control of Ammonia Content in Solution Using NIR Spectroscopy

Consider the chemical transformation depicted in Figure 42.7, where a pharmaceutical intermediate is synthesized via a reductive amination that involves the use of gaseous reagents under pressure [46]. The reaction is performed in two steps: (i) ketone ammonolysis by reaction with ammonia to form an intermediate ketimine and (ii) heterogeneous catalytic hydrogenation of the ketimine intermediate to give primary amine.

In general, for reactions conducted under pressure where dissolved gas is present in significant amounts, it is important to design a robust purging protocol. Ensuring the solution is free of gas in downstream processing is necessary to guarantee safe handling and emissions controls. In this particular case, residual ammonia from the ammonolysis part of the process negatively impacted the hydrogenation step due to its correlation with increased levels of impurity and ammonia off‐gassing that complicated the charge of Pd/Al₂O₃ catalyst. During the first kilo‐scale implementation of the process, ammonia was removed from solution through a sequence of nitrogen purges. Due to inability to continuously subsurface sparge with nitrogen in the reaction vessel and high ammonia solubility in MeTHF (>1 mol/L), a large number of purge cycles needed to be implemented. Each purge cycle consisted of the following steps: venting of headspace to atmospheric pressure with agitation turned off, nitrogen introduction into the headspace to 1–2 psig, and agitation of solution for several minutes until pressure stabilization (i.e. a new ammonia liquid–vapor equilibrium was established), which was then followed by stopping of the agitator and venting of the headspace gas to a scrubber. This operation helped reduce headspace pressure close to atmospheric and enabled the charge of solid Pd/Al₂O₃ catalyst for the subsequent hydrogenation. Figure 42.8 depicts the pressure profile during the purge of residual ammonia for a representative reaction executed in a 20 L reaction vessel. Table 42.7 summarizes the impurity levels observed at different process stages in the three batches executed during the scale‐up campaign.

The impurity formed during filtration of the Pd/Al₂O₃ catalyst, following hydrogenation completion. Moreover, further investigations demonstrated that impurity formation rates were accelerated by increasing levels of residual ammonia. Interestingly, such high levels of impurity were never observed in laboratory experiments prior to scale‐up. Whereas filtration times are on the order of minutes in the laboratory, typical filtration times for plant‐scale hydrogenations (~1000 L) are on the order of hours. Therefore, implementing a control strategy that established acceptable ammonia levels in post‐ammonolysis stream to prolong post‐hydrogenation stream stability had to be considered. Evaluation of multiple spectroscopic techniques for detection of key reaction components is a good general practice when facing a new chemical transformation in the context of PAT development. In‐line or at‐line testing of reaction components in the solvent system of interest or reaction matrix allows to quickly assess if a particular technique has specificity for the compounds of interest. Figures 42.9 and 42.10 show, respectively, FTIR and NIR spectra of the reaction components. Inspection of the FTIR spectra in the 1500–1800 cm⁻¹ region indicated that the carbonyl band of the starting material at approximately 1700 cm⁻¹ is well separated from bands of other species and suggested that FTIR was well suited for monitoring the ammonolysis. Indeed, in situ FTIR monitoring enabled the rapid collection of kinetic information and characterization of the reaction parameter space with respect to conversion in the early stages of reaction development (Figures 42.11 and 42.12). For reaction times on the order days, the reaction required an ammonia pressure greater than 30 psig and excess of titanium tetra‐isopropoxide. Alternative methods to monitor the pressurized reaction were limited by frequency of sampling, safety, and moisture sensitivity. Unfortunately, due to overlap of amine‐containing functional groups in the product and ammonia in solution, quantitative determination of ammonia content with FTIR at low levels proved challenging.

As presented in Figure 42.10, examination of the overlaid NIR spectra of starting material, imine intermediate with and without ammonia, and product indicated that NIR offers excellent specificity with respect to ammonia content, which appears as a strong peak at approximately 5000 cm⁻¹. Having established that the characteristic peak of ammonia is well separated from other reaction components, the next step in method development was to construct the calibration curve. Although the off‐line measurement of a standard ammonia solution could potentially be employed with success, several drawbacks of such approach became apparent. Depending on a particular set of reaction conditions, the range of ammonia concentrations that needed to be spectroscopically characterized could only be achieved at pressures above atmospheric. Conversely, the need to detect low levels of ammonia made availability of standard solutions impracticable. In addition, due to the gas–liquid equilibrium nature of the dissolution process, the dependence of ammonia concentrations on headspace pressure and solution handling could introduce significant error during the calibration. Using in‐line spectroscopic measurements in the calibration eliminated these uncertainties. Figure 42.13 shows a schematic laboratory setup that utilizes a pressure vessel equipped with a supply line of gaseous ammonia, an NIR probe, and a pressure gauge. Using this setup, a simple protocol was devised to dissolve a known amount of gas in the reaction solvent. First, the reactor headspace was pressurized to a known value P_o while the agitator was turned off. Second, the gas supply was turned off, and the agitator activated until an equilibrium pressure P_f at which a constant gauge reading was achieved. Using ideal gas law, namely, , allows to determine the number of moles of ammonia dissolved in solution for each such operation. Repeating this sequence several times (Figure 42.14) affords the calibration curve shown in Figure 42.15, where ammonia concentration is proportional to ammonia NIR peak height. In this particular case, the ammonia peak at approximately 5000 cm⁻¹ was normalized relative to the MeTHF solvent peak. Development of the analytical method for in situ ammonia detection enabled conduction of laboratory stability studies for post‐hydrogenation streams of variable ammonia content. It was established that ammonia concentrations lower than 0.1 M afforded extended multi‐day stream stability. NIR IPC for residual ammonia concentrations lower than 0.1 M was set in place at the end of the ammonolysis reaction. Implementation of this spectroscopy IPC during continuous subsurface nitrogen sparge in scale‐up batches afforded significant reduction of the impurity levels as shown in Table 42.8.

42.4.2 Case Study 2: FTIR Spectroscopy Enabled Control Strategy in Manufacturing of Brivanib Alaninate

The last step of the chemical sequence for the preparation of anticancer drug brivanib alaninate is shown in Figure 42.16 [47]. The overall transformation involves conversion of the parent drug into a prodrug by formation of its alaninate ester. Two reactions are telescoped in a single step without isolation of the intermediate species: (i) esterification of the parent drug with Cbz‐protected alanine and (ii) removal of the alanine Cbz protecting group via hydrogenolysis. Prolonged aging of the reaction mixture during the hydrogenolysis could result in elevated levels of the impurities shown in Figure 42.17.

During laboratory development, it was established that if hydrogen supply is insufficient or interrupted at early stages of the reaction, catalyst poisoning by CO₂ formed during the hydrogenolysis could occur and result in significant reaction rate inhibition. To address this risk and assure manufacturing robustness, a control strategy that implemented two PAT‐based IPCs was developed: (i) reaction progress IPC‐1 at 20 minutes to protect against stalling associated with catalyst poisoning and (ii) end‐of‐reaction IPC‐2 to protect against over‐aging. An additional IPC‐3 was established to measure residual levels of CO₂ at hydrogenation completion due to the accelerating effect of CO₂ on the hydrolysis reaction rate that affords parent drug as an undesired impurity.

In order to establish suitability of FTIR spectroscopy for both kinetics tracking and CO₂ concentration, reaction components were evaluated with in situ measurements. Figure 42.18 shows spectra of the THF solvent, starting material (BMS‐582656 in THF), and end‐of‐reaction mixture containing brivanib alaninate product in the 1100–1900 cm⁻¹ region. Three peaks corresponding to starting material at ~1710 cm⁻¹, ~1200 cm⁻¹, and ~1120 cm⁻¹ could be monitored during reaction progress. A PLS regression model fit to a series of calibration solutions spectra correlates integration of the above IR peaks with concentrations of starting material during the reaction. In order to improve model accuracy, the training sets covered a range of reaction parameters such as temperature (20–35 °C) and catalyst loading (5–17 wt %). Figure 42.19 illustrates the predictive power of the model obtained. Spectroscopic sensitivity is a key issue worth highlighting in the context of starting material concentration determination. Due to stringent conversion requirements encountered in pharmaceutically relevant reactions, most spectroscopic methods will not be able to detect residual starting material at levels of less than or equal to 1 HPLC %. The FTIR detection limit for the brivanib alaninate hydrogenolysis was approximately 2.5 wt %, which is significantly higher than the desired conversion of 99%. In order to overcome this limitation, a prediction‐based end‐of‐reaction IPC was developed. The idea is to evaluate the reaction rate at conversion levels close to the limit of sensitivity and predict reaction completion time in accordance with a kinetic law. A first‐order rate law

was tested and proved effective at predicting conversion in the concentration regime below FTIR sensitivity. In terms of algorithm design, several data points were measured before the starting material concentration reaches 3–4 wt %, and the first‐order equation was fit to them, thus providing as an output a time when greater than 99% conversion would be achieved.

Figure 42.20 shows the IR spectra of reaction components in the expanded approximately 700–4000 cm⁻¹ region. The CO₂ peak at approximately 2300 cm⁻¹ was well resolved from the rest of reaction components, and a calibration curve could be easily established in a fashion similar to the one described in Case Study 42.1 (Figure 42.21). In situ FTIR was highly sensitive with respect to CO₂ in the reaction matrix and enabled CO₂ detection down to ppm levels. Having demonstrated the suitability of in situ FTIR, stability studies at variable CO₂ levels were conducted to determine the IPC level that made possible extended storage of the post‐hydrogenation reaction stream. It was established that CO₂ levels of less than or equal to 250 ppm were acceptable and appropriate as the IPC‐3 limit.

Following extensive laboratory studies, FTIR‐based IPCs were implemented during a pilot plant campaign in 1000 L hydrogenator. For this first phase of IPC implementation in a scale‐up setting, calculations were performed manually on plant floor using a computer program and logged into quality assurance data systems, following standard GMP practices. Table 42.9 summarizes results from three batches executed during the campaign. All of the IPCs performed well and proved to be an effective element of the overall control strategy to assure API quality.

Following successful implementation of spectroscopy‐based IPCs in the pilot plant, the process transitioned into the manufacturing validation campaign. To eliminate the need to perform manual calculations and automate data acquisition and processing in the manufacturing setting, a graphic operator interface was developed. Figure 42.22 displays a screenshot of the graphic interface during the execution of a batch showing trends for BMS‐582656 wt % (measured and extrapolated), CO₂ levels, and the status of each IPC. Both analytical method and instrumentation demonstrated robustness in all of twenty validation campaign hydrogenation loads. Minor probe realignment and algorithm parameter adjustments to accommodate lower signal‐to‐noise ratios were required only once.

A summary of hydrogenation reaction times to reach greater than 99% conversion as determined by IPC‐2 is shown in Figure 42.23. Reaction time spread is evident, confirming inherent variability associated with the process. Moreover, IPC‐1 failure in one hydrogenation load allowed identification of an incipient leak in the agitator nitrogen seal as the hydrogen partial pressure was not preserved (Figure 42.24). Addition of catalyst kicker charge as prescribed by IPC‐1 logic allowed to drive reaction to completion despite the leak and ensured successful recovery of product. It also informed plant maintenance personnel to investigate the agitator integrity and replace the seal prior to initiation of another batch. Overall, implementation of PAT offered significant advantages over standard at‐line methods to mitigate risks associated with reaction robustness and ensure impurity control.

42.4.3 Case Study 3: Raman Spectroscopy Enabled Control of Reaction Time in Pilot Plant

Consider a chemical process as depicted in Figure 42.25, wherein an elimination reaction is carried out at elevated temperature to form the desired product. Due to low solubility of the starting material, high temperature is required to drive the reaction to completion, but extended holds at elevated temperature lead to product degradation and the formation of quality‐impacting impurities. Thus, control of reaction age time is critical to successful implementation.

To understand the risk of over‐aging the reaction at elevated temperature, a series of kinetic experiments were conducted using HPLC analysis under varying temperatures and concentrations. Modeling the reaction system in DynoChem³ showed that both desired and undesired reactions were well fit by unimolecular, first‐order rate expressions with Arrhenius‐style temperature dependence. Figure 42.26 shows an example data set for conversion as a function of time – the reaction rate rapidly accelerates as the temperature increases. Both desired and undesired reactions exhibited high activation energies, and thus variability in the heating rate had a strong influence on the optimal age time once the desired temperature of 105 °C is reached. Furthermore, due to upstream solvent exchange, the reaction does not begin at 0% conversion, as the thermal elimination reaction occurs slowly during distillation, introducing additional variability. This grew particularly concerning as the process scale increased and the reduced heat transfer rate in larger equipment would introduce variability to the optimal reaction age time and increase the risk of impurity formation due to over‐aging. Furthermore, off‐line HPLC analysis is sufficient to monitor the reaction in small‐scale experiments, but is not viable as a control strategy on scale due to the long analysis time and the processing delays associated with cooling the batch to a safe sampling temperature.

Reaction endpoint determination via in‐line Raman analysis provides an elegant solution to these problems. Using spectroscopy to monitor reaction conversion in real time ensures that each batch is held for the optimal age time, regardless of the thermal “history” during heat‐up. Figure 42.27 shows the Raman spectra over the course of a reaction. A chemometric model was built using distinct peaks for starting material, product, and solvent (which should remain constant throughout) and was demonstrated to be sufficiently quantitative throughout the reaction and within the temperature range of interest. An example trace of reaction rate as monitored by PAT is shown in Figure 42.28, in which reaction conversion via Raman tracks with reactor temperature.

This in‐line Raman method allowed for facile technology transfer, as the reaction hold time was determined independently of equipment parameters. Figure 42.29 shows reaction progression in two different, but similarly sized, pilot‐scale reactors. The Raman method reduced risk to quality by ensuring timely calling of the reaction endpoint, even in the case with slower heat‐up and overshoot, and was successfully demonstrated in twelve pilot‐scale batches to convert up to 80 kg of starting material.