Real-Time Concentration Monitoring Using a Compact Composite Sensor Array for in situ Quality Control of Aqueous Formulations

Abstract

Recent advancements have demonstrated the feasibility of refrigerator-sized pharmaceutical manufacturing platforms (PMPs) for integrated end-to-end manufacturing of active pharmaceutical ingredients (APIs) into formulated drug products. Unlike typical laboratory- or industrial-scale setups, PMPs present unique requirements for process analytical technology (PAT) with respect to versatility, flexibility, and physical size to fit into the PMP space constraints. In this proof of principle study, a novel compact composite sensor array (CCSA) combining ultraviolet (UV) and near infrared (NIR) features at four different wavelengths (280, 340, 600, 860 nm) with temperature measuring capability in a 380 × 30 mm housing (length × diameter, 7 mm diameter at the probe head), were evaluated. The results indicate that the CCSA prototype is capable of measuring the solution and suspension concentrations in aqueous formulations of four model APIs (warfarin sodium isopropanol solvate, lidocaine hydrochloride monohydrate, 6-mercaptopurine monohydrate, acetaminophen) in situ and in real-time with similar accuracy as an established Raman spectrometer commonly applied for method development.

Graphical Abstract

1. INTRODUCTION

In the past decade, the use of continuous-flow technologies for pharmaceuticals have expanded.[1–3] The most recent advancements have demonstrated the feasibility of compact systems for the integrated end-to-end manufacturing (synthesis, purification, formulation) of active pharmaceutical ingredients (APIs).[4–13] Particularly, flexible pharmaceutical manufacturing platforms (PMPs)[5–9] offer potential solutions to (1) minimize drug shortages, (2) facilitate drug manufacturing in remote areas and in response to catastrophic events and natural disasters, (3) address shortcomings in the cost-effective production of orphan drugs, and (4) enable the flexible manufacturing needs for the growing paradigm shift towards personalized medicines, also known as precision medicine.[5–9,14–19]

A vital aspect for the success of the PMP concept is the implementation of the process analytical technology (PAT) guidance and the quality by design (QbD) approach to ensure effective design, analysis, and control strategies that lead to predefined quality attributes of the final products.[20–25] Thus, real-time process monitoring by PAT permits not just process understanding but also quality assurance needed throughout the life cycle of pharmaceutical products.[8–10,26–28] For instance, to control the desired strength of formulated APIs prior to the release from PMPs.[29]

In previous work, we reported that PMPs present unique requirements for PATs with respect to versatility and flexibility while being small in physical size to fit into the PMP space constraints, which are unlike typical laboratory- or industrial-scale needs.[25,29] Ideally, a compact PAT is preferred to maintain a small footprint of the PMPs. In recent years, this requirement has driven research to miniaturize and explore cost-effective PATs.[25,29–40]

The novel composite sensor array employed in this study (Incentive™ Vision IR/UV probe, ITA Instruments Ltd) fulfills these needs. The prototype combines ultraviolet (UV) and near infrared (NIR) features at four different wavelengths (280, 340, 600, 860 nm) with temperature measuring capability in a 380 × 30 mm housing (length x diameter, 7 mm diameter at the probe head). The housing includes the electronic unit and can be connected to any computer via USB (Figure 1).[41] The aim of this study is to investigate a compact composite sensor array (CCSA) as a versatile, flexible, reliable, and small-scale PAT that can monitor in situ the concentration of analytes in solutions and suspensions.[29] The ability to monitor solution and suspension concentrations is crucial for the understanding and control of liquid formulated drug products prior to the release from PMPs.[5,6,29] This ability would also broaden the CCSA application into other areas, for instance, to monitor crystallization processes and to calculate the mass of solids produced or to calculate crystallization kinetic parameters based on desupersaturation curves.[42–46]

Figure 1.

(A) Experimental setup with CCSA and Raman probe in a five-neck jacketed beaker (50 mL) and (B) compact composite sensor array (CCSA) Incentive™ Vision IR/UV probe (all-in-one) with a close up on the probe window.

To the best of our knowledge, no prior study has attempted to assess the capability of an all-in-one IR/UV probe (Incentive™ Vision) as a CCSA. In this work, the solution and suspension concentrations of four model APIs were monitored (Figure 2). Specifically, warfarin sodium isopropanol solvate and lidocaine hydrochloride monohydrate (solution formulated) as well as 6-mercaptopurine monohydrate and acetaminophen (suspension formulated) were measured in their commercial aqueous forms.[47,48] Investigating simplified aqueous formulations in this study is consistent with the ability of PMPs to serve a niche of producing formulations just-in-time and at the point-of-use without the need for extended stabilities and thus complex formulation matrixes.[5–7,29] The concentrations measured with the CCSA were compared to concentrations obtained using an in situ Raman spectrometer. The latter is typically applied in the laboratory and industry[49–56] and used here as a reference PAT for real-time monitoring using partial least squares (PLS) regression. The concentration were validated and correlated with offline UV-vis spectroscopy measurements.

Figure 2.

Molecular structures of warfarin sodium isopropanol solvate (WS) and lidocaine hydrochloride monohydrate (LID) formulated in aqueous solutions as well as 6-mercaptopurine monohydrate (6-MP) and acetaminophen (ACM) formulated in aqueous suspensions.

Finally, the CCSA used in this study is also an example of a process analyzer.[57] The main purpose of analyzers is rapid process monitoring/control in a rather “hostile” production environment away from the well-protected laboratory where research and analytical methods are typically developed employing sophisticated PATs such as Raman spectrometer.[57] Unlike in research, the personnel at production sites often have not undergone extensive training in analytical chemistry and spectroscopy associated with laboratory analytical applications.[57] Yet, they rely on similar end results with respect to quantitative values. In this regard, for instance, an in situ Raman spectrometer may be configured as an analyzer instead of a highly valuable asset in research efforts to develop new analytical methods and troubleshoot processes. However, the use of a sophisticated spectrometer as an analyzer in an operational environment would require a significant capital investment to then utilize only a fraction of its features. All of this supports the relevance of the proposed study to identify new versatility and flexibility while being small in physical size.

2. EXPERIMENTAL SECTION

2.1. Materials

Acetaminophen (ACM, 100%), lidocaine hydrochloride monohydrate (LID, 100%), and glycerol (≥ 99.5%, used as a wetting agent) were purchased from Sigma-Aldrich, USA. Warfarin sodium isopropanol solvate (WS, ≥ 97.0%) and 6-mercaptopurine monohydrate (6-MP, ≥ 98%) were acquired from Ningbo Samreal Chemical Co. LTD, China. Nanopure water (18.23 MOhm/cm, pH = 5.29) was obtained from a water purification system (Gemini, Aries Filter). All materials were used “as received” without further purification.

2.2. Formulation Preparation

All API’s were formulated in their commercial aqueous solution or suspension as defined in the US National Library of Medicine.[47] The details are summarized in Table 1 including the limits defined in the US Pharmacopeia (USP).[48] Only simplified formulation matrixes are investigated within this study as the concept of PMPs allows to deliver drug products at the point-of-use for immediate consumption or to be consumed within few days after the formulation.[5,6,11,29]

Table 1.

Details of active pharmaceutical ingredient (API) formulations explored.

Type of Formulation	API	Excipients	Strength (mg/mL)[47]	USP Limits (%)[48]
Solution	WS	water	2	5
	LID	water	20	5
Suspension	6-MP	water	20	−7 to +10a
	ACM	10 % (v/v) glycerol in water[58]	33	10

^aUSP limits for 6-MP are based on monograph for 6-MP tablets because a USP monograph for 6-MP oral suspension (Purixan®)[47] has not been documented.[48]

2.3. Experimental Setup

A five-neck jacketed beaker (50 mL, Ace Glass Inc.), temperature controlled at 20 °C (recirculating bath, Julabo, F32-ME) and equipped with an overhead stirrer (J-Kem Scientific, OHS-1M) at 300 rpm was employed to evaluate the CCSA. The PAT probes of the CCSA (Incentive™ Vision IR/UV probe, ITA-Instruments Ltd) and the Raman spectrometer (Kaiser Optical Systems) were inserted into a jacketed beaker (Figure 1). The signals of the CCSA at 280, 340, 600, and 860 nm were recorded in 0.5 s intervals over 5 min utilizing the Incentive Vision Monitoring software (v. 1.1). The Raman spectra were obtained employing a R m nRxn2™ multi-channel spectrometer equipped with an immersion probe and a 785 nm laser using automatic cosmic ray filter and intensity correction (15 s exposure time, averaging 3 scans per measurement). The spectra were recorded in 1 min intervals and averaged over 5 min using the iC Raman software (v. 4.1.917). To prevent external light interference, the entire jacketed beaker was covered with aluminum foil (not shown in Figure 1).

2.4. Procedure

Two sets of calibration and validation experiments were conducted to assess the CCSA. The calibration experiments started with blank liquid excipient (water or 10 % (v/v) glycerol in water) as indicated in Table 1. Once the signals recorded were stable, the desired API amounts were added into the jacketed beaker, resulting in incremental changes of the concentration monitored by both PATs throughout the experiments. To overcome signal interference between the CCSA and Raman spectrometer, the measurements for each concentration were performed alternating between both techniques starting with the CCSA. The signal magnitudes for both PATs corresponded to the total API amount present in the aqueous formulations.

Prior to each incremental concentration increase, 1 mL of sample was taken. The samples for the solution formulations of WS and LID were filtered (0.2 μm syringe filter, Fisher Scientific), and diluted to a target concentration, before being analyzed offline with UV-Vis spectroscopy (Agilent, UV Cary 100, temperature controlled at 20°C). The samples for the suspension formulations of 6-MP and ACM were also diluted to a target concentration before being analyzed offline with UV-Vis spectroscopy.[59,60] All measurements were performed with a 200–400 nm scan using the UV Cary Scan software version v. 20.0.470. Offline UV-Vis calibrations of all model APIs are available in the Supporting Information. Each set of calibration experiments consisted of at least four different concentrations representing 0 to 5 times the desired strength of the API formulations.[47] The validation experiments were conducted following the same procedure as described above but with API concentrations (three concentrations per API). These concentrations were different than the concentrations used for the calibration sets.

2.5. Calibration Model Development

For the CCSA the data were first pretreated for each concentration and signal by employing a Fast Fourier Transform (FFT) filter with a 30-point window to reduce the burden on the model development.[57,61] From the data recorded in 0.5 s intervals (preset by Incentive Vision Monitoring software, v. 1.1) over 5 min (≥ 600 data points), the average of the last 30 s (≥ 60 data points) were used to construct the models for each signal versus concentration utilizing mathematical expressions with the best possible fit. Origin (OriginLab Corporation, v. 9.7.0.188) was utilized for the data pretreatment and to solve the nonlinear curve-fitting problems employing the Levenberg−Marquardt algorithm.

All Raman spectroscopic data were analyzed using the multivariate data analysis software SIMCA (Umetrics, v. 16.0) employing the PLS regression model. The standard normal variate (SNV) method was used for data pretreatment, which is widely accepted for eliminating the differences caused by variations of the baseline and scattered data generated by laser intensity fluctuations.[61–63] Details of the data pretreatment are given in the Supplementary Information.

The correlation coefficient (R²), the percent average relative deviation (ARD%), and the root mean squared error of prediction (RMSEP) were determined to assess the correlation between the experimental data and the fitted models as well as the difference between the experimental and calculated concentrations based on the PAT signal employed. ARD% and RMSEP were calculated using equations (1) and (2):

$$ARD\%=\frac{100}{N}{\sum }_{i=1}^N\left|\frac{c_i^{exp}-c_i^{cal}}{c_i^{exp}}\right|$$ (1)

$$RMSEP=\sqrt{\frac{1}{N}{\sum }_{i=1}^N{\left(c_i^{cal}-c_i^{exp}\right)}^2}$$ (2)

where $c_i^{exp}$ and $c_i^{cal}$ are the ith experimental and calculated concentration, respectively, while N is the total number of samples.

3. RESULTS AND DISCUSSION.

The results of both PATs for all four APIs are summarized in Table 2. The plotted data utilized to extract the information needed for the calculations not shown in the Results and Discussion section are given in the Supplementary Information. The high R^² values (≥ 0.99) for all APIs and PATs demonstrate that the chosen calibration models fit the experimental data very well (Table 2).

Table 2.

Summary of the calibration and validation experiments when employing the CCSA and Raman spectrometer to measure the solution and suspension concentrations for all APIs.^a

API	CCSA (IncentiveTM Vision IR/UV probe)				Raman Spectrometer
	Signal	R2	ARD%	RMSEP (mg/mL)	R2	ARD%	RMSEP (mg/mL)
WS	860 nm	0.997	1.16	0.18	0.999	0.47	0.11
LID	280 nm	0.999	0.49	1.24	0.999	0.56	1.33
6-MP	280 nm	0.997	0.34	2.28	0.997	2.44	1.55
ACM	280 nm	0.994	10.12	5.67	0.999	11.48	4.48

^aR² represents the correlation coefficient of the calibration data, ARD% the percentage average relative deviation of the validation data, and RMSEP the root-mean-square error of prediction of the validation data.

The CCSA signals chosen for each API in Table 2 are based on the highest R² achieved for each univariant calibration model and strong intensity-concentration dependency. For instance, while the 280 and 340 nm signals depict very good exponential relationship for LID, the 600 and 860 nm signals show an inconsistent correlation and intensity saturation (plateau), respectively. The detailed CCSA signal comparisons for each API can be found in the Supplementary Information. Figure 3 illustrates the calibration and validation data of the signals chosen for WS, LID, 6-MP, and ACM.

Figure 3.

CCSA calibration models for (A) WS at 860 nm, (B) LID at 280 nm, (C) 6-MP at 280 nm, and (D) ACM at 280 nm. The red squares, blue diamonds, and black trend lines represent the calibration data, validation data and the best possible fit (highest R²) for the calibration data, respectively. The concentrations for all APIs were measured offline with UV-vis spectroscopy.

In Figure 3 it can be seen that the calibration models for the suspension formulated 6-MP and ACM depict almost linear data correlation compared to the solution formulated WS and LID. On the other hand, the calibration data for WS, LID, and 6-MP demonstrate less data scattering compared to ACM, which can be assessed using the ARD% values (Table 2). The scattering can be attributed to the well-recognized difficulties in suspending particles homogeneously within stirred tanks and/or crystallizers, a phenomena referred to in the literature as spatial heterogeneity.[64–67] Recent work has demonstrated that sampling with PAT are governed by the same principles as those defined by the theory of sampling.[57,61,68] However, these sampling errors are typically one or two orders of magnitude larger than analytical errors.[57] Thus, the quality of the CCSA data depends on the sampling errors caused most likely by spatial heterogeneity and not necessarily due to the application of the CCSA. This can be supported by the ARD% values for both CCSA and the Raman spectrometer, utilized here as a well-established reference PAT.[49–56] The ARD% of both PATs present similar values for ACM with 10.12 (CCSA) and 11.48 (Raman spectrometer) as shown in Table 2.

Figure 4 shows the results of the validation experiments for WS, LID, 6-MP, and ACM. It can be observed that the calculated concentrations for both PATs lie within the established USP limits (red broken lines). The quality of the calibration models can be assessed by the low RMSEP values (Table 2. Summary of the calibration and validation experiments when employing the CCSA and Raman spectrometer to measure the solution and suspension concentrations for all APIs.^a) with the exception of ACM. However, both the CCSA and the Raman spectrometer depict similar RMSEP values for ACM with 5.67 and 4.48, respectively, which supports the previous discussion hypothesizing that the accuracy issues are likely related to PAT sampling errors and not to the CCSA.[57,61,68] Consequently, it can be stated that the CCSA can monitor the concentration of the solution and suspension formulated APIs with similar accuracy compared to the in situ Raman spectrometer. Though it has to be mentioned that the concentration range for this early stage comparative (feasibility) study was limited to focus on the target dosage of the APIs (Table 1). On this account, the RMSEP values might not be fully representative for the entire concentration range. This aspect needs to be addressed in future studies of the CCSA.

Figure 4.

Results of validation experiments employing the calibration models for CCSA (blue diamonds) and Raman spectrometer (green circles) for (A) WS, (B) LID, (C) 6-MP, and (D) ACM. The red broken lines illustrate the USP limits. The concentrations for all APIs were measured offline with UV-vis spectroscopy.

4. CONCLUSIONS.

This proof of principle study addresses the important need for miniaturized PATs in PMP applications.[25,29] Here, it could be demonstrated that the CCSA prototype is a versatile PAT capable of measuring solution and suspension concentrations in separately conducted aqueous formulations in situ and in real-time with similar accuracy as an established Raman spectrometer. Additionally, this study also highlights how a more cost-effective system may be used as a robust process analyzer. Moreover, the results also hint at further applications where solution and suspension concentrations are important process parameters such as solution crystallization.[43,69] The CCSA might be particularly interesting for mL-scale crystallizers (≤ 100 mL), typically employed in PMPs, which possess similar needs for compact and versatile PATs.[25,64,70] Consequently, this study might be used as a rationale for necessary further analytical and regulatory studies of the CCSA as a potential compact PAT for real-time in situ concentration monitoring in PMPs.[5–7]

Highlights.

• Integrated end-to-end refrigerator-sized pharmaceutical manufacturing platform, PMP

• PMPs present unique requirements for process analytical technology (PAT)

• Novel compact composite sensor array with ultraviolet & near infrared features

• Sensor measures solution & suspension concentrations of four model drugs

• In situ & in real-time with similar accuracy as an established Raman spectrometer

APPENDIX A.

Supplementary Information related to this article is available in the online version.Details regarding calibration and validation experiments as well as calibration models for warfarin sodium isopropanol solvate, lidocaine hydrochloride monohydrate, 6-mercaptopurine monohydrate, and acetaminophen using the Compact composite sensor array (Incentive™ Vision), RamanRxn2™ multi-channel spectrometer (Kaiser Optical Systems), and UV-Vis spectroscopy (Agilent, UV Cary 100).