Resolving the Speed-Specificity Trade-Off in Biopharma High-Throughput Screening

Abstract

High-throughput screening (HTS) in biopharma often requires a crucial trade-off between speed and chemical specificity. By integrating Raman spectroscopy into automated multiwell plate workflows, researchers can achieve label-free, parallel measurement of multiple samples for multivariate analysis and condition optimization.

Using multilaser excitation options alongside liquid handler and robotic-arm integration, this automated approach provides non-destructive chemical analysis of complex samples. This article explores the performance of high-throughput Raman analysis through three biopharma applications: the quantification of nucleoside triphosphates (NTP) in mixtures, the determination of protein detection limits using lysozyme and bovine serum albumin (BSA), and the monitoring of glucose levels in diverse media.

By delivering reproducible and rapid results across these tasks, automated Raman plate reading serves as a high-speed complementary tool to traditional chromatography, facilitating more efficient drug discovery and process development.

Introduction

High-throughput screening (HTS) is a central component of pharmaceutical research, enabling the rapid evaluation of thousands to millions of compounds during drug discovery and bioprocess development to identify promising candidates.^1,2 However, traditional HTS inherently involves a trade-off between analytical speed and chemical specificity.^3,4

Fluorescence and absorbance assays provide rapid readouts but typically require costly dyes and labels and lack the molecular-level information needed to avoid false positives.⁵ Conversely, chromatography techniques such as high-performance liquid chromatography (HPLC) and liquid chromatography–mass spectrometry (LC-MS) offer high selectivity and precision. However, they are slow, labor-intensive, and reagent-heavy, making them difficult to scale for high-throughput workflows.⁶

Raman spectroscopy is a vibrational technique that provides a molecular fingerprint, enabling label-free, chemically specific analysis with minimal sample preparation.^7,8 Manual Raman workflows require individual sample handling and therefore can hinder measurement speed and throughput.

Automated multiwell Raman workflows can reduce manual handling, allowing measurement of samples prepared in standard 96- and 384-well plate formats. This study evaluates the performance and potential of a Raman plate reader (RPR) in three biopharmaceutical applications: nucleoside triphosphate (NTP) quantification in mixtures, protein limit of detection (LoD) evaluation, and glucose quantification in culture media.

Different laser excitation (532 nm or 785 nm) and acquisition settings were used depending on the application and sample type. Collectively, these representative examples showcase how high-throughput Raman analysis can benefit a wide range of biopharma research interests and complement conventional analytical techniques.

Materials and Methods

All experiments were conducted using standard flat-bottom polystyrene 96-well plates. Raman spectra were collected using an automated RPR equipped with both 532 nm and 785 nm excitation probes. Before measurements, the instrument was automatically calibrated using a diamond reference.

Nucleoside Triphosphate (NTP) Quantification

For the NTP study, random mixtures of adenosine triphosphate (ATP), cytidine triphosphate (CTP), guanosine triphosphate (GTP), and uridine triphosphate (UTP) were prepared in phosphate-buffered saline (PBS) with concentrations ranging from 0 to 2.5 mM. Calibration samples were prepared in triplicate and measured using 532 nm excitation at 150 mW with an integration time of 40 s per well.

A separate batch of mixtures with concentrations ranging from 0 to 2 mM was prepared and measured under identical acquisition conditions for validation. A partial least squares (PLS) regression model was constructed using spectral preprocessing consisting of Savitzky–Golay first-derivative transformation followed by mean centering. The final model used four latent variables (LVs).

Protein Detection Limit Evaluation

Dilution series of bovine serum albumin (BSA, ≥96%) and lysozyme (≥39,000 units/mg protein) were prepared in PBS in 96-well plates. Raman spectra were collected using 785 nm excitation at 300 mW.

For BSA samples, spectra were acquired using a 2.5 s integration time with two accumulations. For lysozyme samples, spectra were acquired using a 30 s integration time.

Separate PLS regression models were constructed for each protein using mean-centered spectral data with two LVs.

Glucose Quantification in Culture Media

For glucose concentration prediction in culture and spent media, all media samples were pipetted into a 96-well plate. Raman spectra were acquired using 785 nm excitation at 300 mW with an integration time of 250 ms and 50 accumulations per well.

Spectra were preprocessed using sixth-order detrending followed by mean centering before PLS model calibration. Two LVs were selected for the final model.

Glucose Detection Limit Evaluation

To evaluate the detection limit of glucose, a dilution series of glucose (≥99.5%) was prepared in PBS with concentrations ranging from 0 to 10 g/L. Raman spectra were collected using a 2.5 s integration time with 12 accumulations.

A PLS regression model was constructed using first-order detrending followed by mean centering. One LV was used in the final model.

Results and Discussion

Nucleoside Triphosphate Quantification

NTPs — ATP, CTP, GTP, and UTP — are fundamental building blocks for nucleic acids and play essential roles in cellular energy transfer and biochemical signaling.⁹ Accurate and timely monitoring of NTP concentrations in mixtures is therefore important across a range of biochemical and analytical applications.

However, conventional NTP quantification typically relies on chromatographic techniques such as HPLC coupled with UV absorbance, fluorescence detection, or LC-MS.^10,11 Although these methods provide high selectivity, they are poorly suited for rapid screening because of limited throughput, extensive sample preparation requirements, and high operating and maintenance costs.

These constraints substantially limit the number of conditions, formulations, or reaction states that can be practically evaluated.

Raman spectroscopy is well suited for addressing these limitations. As demonstrated in Figure 1A, each of the four NTPs exhibits distinct Raman features, enabling direct identification and quantification in mixtures without any additional labeling or chromatographic separation.

To evaluate quantification performance and detection limits with Raman, mixtures of ATP, CTP, GTP, and UTP were prepared in PBS across a concentration range of 0 to 2.5 mM in triplicate. Raman spectra were acquired with 532 nm excitation, with an acquisition time of 40 s per well.

A PLS regression model was constructed to resolve the overlapping spectral contributions, with preprocessing using Savitzky–Golay first derivative and mean centering. Four LVs were selected for the final model. To validate model robustness, a separate batch of NTP mixtures was prepared, and the spectra were acquired on a different day.

Strong agreement was observed between predicted and reference concentrations, with regression slopes greater than 0.997 for all four NTPs, as shown in Figure 1B. The model achieved a root mean square error of cross-validation below 0.03 and enabled accurate prediction of NTP concentrations down to 0.1 mM. These results demonstrate the capability of the RPR to provide rapid, label-free quantification of NTPs in complex mixtures, supporting high-throughput screening and reaction analysis workflows in biopharmaceutical environments.

Protein Detection Limit Evaluation

Proteins are central targets in biopharmaceutical development; however, high-throughput protein detection and quantification are generally challenging. Common techniques such as UV absorbance, Bradford and bicinchoninic acid assays, and enzyme-linked immunosorbent assays suffer from limited specificity, interference from formulation components, and dependence on calibration standards.

HPLC and LC-MS offer higher accuracy but are slow and low-throughput, and they require extensive sample preparation. Fluorescence or label-based assays improve sensitivity but introduce labeling artifacts and high reagent costs.^12,13

Raman-based protein analysis, despite its advantages in sample preparation simplicity, is challenging because of relatively weak Raman scattering and interference from fluorescence. To evaluate the detection capabilities of the RPR for protein analysis, two representative proteins were selected: BSA (66.5 kDa) as a large protein and lysozyme (14.3 kDa) as a smaller protein. A dilution series of each was prepared in PBS in 96-well plates for RPR analysis.

Raman spectra, shown in the left panel of Figure 2 (A, BSA; B, lysozyme), were acquired using 785 nm excitation to minimize fluorescence interference. Two PLS regression models, one for each protein, were constructed with mean-centered spectra across a range of concentrations.

As shown in the right panel of Figure 2, good linearity was observed for both proteins. The RPR successfully detected protein concentrations down to ~3 mg/mL. Specifically, the LoD was determined to be 2.6 mg/mL for BSA and ~5 mg/mL for lysozyme using the following equation.¹⁴

These results establish a baseline detection limit for Raman-based protein analysis using the RPR and demonstrate the feasibility of applying the RPR to protein formulation, stability studies, and drug-protein interaction screening without consuming or destroying samples.

Glucose Quantification in Media

Glucose concentration is a key parameter in cell and yeast culture and fermentation processes, directly affecting cell viability and productivity.^15–17 Although HPLC is commonly used for glucose quantification, its limited throughput and operational cost and complexity restrict its use for rapid feedback during media optimization and process development.^18,19 The ability of the RPR to quantify glucose directly in culture and spent media supernatants was evaluated using 785 nm excitation.

Five types of culture media and yeast spent media containing different glucose concentrations were analyzed in a single 96-well plate. As shown in Figure 3A, background fluorescence levels vary substantially across media types.

To explore the feasibility of predicting glucose concentration using a single model, the Raman spectra were preprocessed using sixth-order detrending followed by mean centering to isolate the glucose signal from the complex background. Predicted glucose concentrations were compared with HPLC-measured concentrations.

Glucose concentrations were accurately predicted in most media types, as shown in the right plot of Figure 3A. However, one medium type (Medium D, cyan) exhibited a different prediction slope, indicating complex matrix effects likely because of its substantially stronger fluorescent background. To address this, a separate PLS model could be constructed specifically for this medium during production or quality control workflows.

Alternatively, samples exhibiting strong background interference can be flagged for targeted HPLC analysis. This hybrid strategy enables efficient resource allocation while maintaining analytical rigor.

By establishing high-throughput Raman spectroscopy as the foundational screening technique, biopharmaceutical facilities can streamline their analytical workflows. Rather than routing every sample through time-consuming and reagent-heavy chromatographic evaluations, Raman acts as a quick initial evaluation tool.

The technique accurately resolves and quantifies most samples falling within standard parameters with no or low fluorescence background in a fraction of the time with minimum labor and consumables. Only the highly complex or strongly fluorescing outliers — such as Medium D in this case study — are selectively diverted to HPLC.

From a methodological standpoint, this hybrid analytical strategy substantially reduces the overall chromatographic and labor burden, minimizing reagent consumption, and alleviating column degradation. In high-throughput screening environments dealing with bioprocess condition optimization, complementing HPLC analysis with front-line Raman screening can reduce analysis time and cost and accelerate development cycles.

This case study highlights the utility of the RPR as a rapid screening and down-selection tool before chromatographic analysis. To set a baseline for glucose quantification capability, a separate study evaluated the LoD of glucose in PBS (Figure 3B). With a simple PLS regression model (first-order detrending, mean centering, and one LV), the glucose LoD was determined to be as low as 0.29 g/L.

Conclusions

Across three biopharmaceutical-relevant case studies, from NTP monitoring to protein detection and glucose quantification, the RPR demonstrates its ability to deliver chemical specificity and label-free quantification rapidly. By bridging the gap between research needs and industrial scale, the RPR offers an automated and practical high-throughput screening platform for accelerating drug discovery and bioprocess development.

Conflict of Interest: The author is an employee of HORIBA, which manufactures the Raman plate reader used in this study.

About the Author

Fiona Xi Xu received her PhD in chemistry from the University of Washington under the advisement of Dan Fu. Her graduate research focused on quantitative analysis of protein and lipid metabolism in cells and tissues using advanced vibrational spectroscopy techniques, particularly stimulated Raman scattering microscopy. She is currently a life science applications scientist at Horiba, where she specializes in Raman spectroscopy for life science and biopharmaceutical applications, using platforms ranging from confocal Raman microscopes to process and high-throughput Raman instruments.

References

1. Beggs M, Blok H, Diels A. The high throughput screening infrastructure: the right tools for the task. SLAS Discov. 1999;4(3):143-149. doi:10.1177/108705719900400308

2. Schnecke V, Boström J. Computational chemistry-driven decision making in lead generation. Drug Discov Today. 2006;11(1):43-50. doi:10.1016/S1359-6446(05)03703-7

3. Vervoort N, et al. Recent advances in analytical techniques for high throughput experimentation. Anal Sci Adv. 2021;2(3-4):109-127. doi:10.1002/ansa.202000155

4. Malo N, et al. Statistical practice in high-throughput screening data analysis. Nat Biotechnol. 2006;24(2):167-175. doi:10.1038/nbt1186

5. Simeonov A, Davis MI. Interference with fluorescence and absorbance. In: Markossian S, et al, eds. Assay Guidance Manual. Eli Lilly & Company and the National Center for Advancing Translational Sciences; 2004.

6. Grebe SK, Singh RJ. LC-MS/MS in the clinical laboratory—where to from here? Clin Biochem Rev. 2011;32(1):5-31.

7. Raman CV, Krishnan KS. A new type of secondary radiation. Nature. 1928;121(3048):501-502. doi:10.1038/121501c0

8. Fernández-Galiana Á, et al. Fundamentals and applications of Raman-based techniques for the design and development of active biomedical materials. Adv Mater. 2024;36(43):2210807. doi:10.1002/adma.202210807

9. Hollenstein M. Nucleoside triphosphates—building blocks for the modification of nucleic acids. Molecules. 2012;17(11):13569-13591. doi:10.3390/molecules171113569

10. Pregeljc D, et al. Increasing yield of in vitro transcription reaction with at-line high pressure liquid chromatography monitoring. Biotechnol Bioeng. 2023;120(3):737-747. doi:10.1002/bit.28299

11. Kellner S, et al. Absolute and relative quantification of RNA modifications via biosynthetic isotopomers. Nucleic Acids Res. 2014;42(18):e142. doi:10.1093/nar/gku733

12. Song JG, et al. Quantitative analysis of therapeutic proteins in biological fluids: recent advancement in analytical techniques. Drug Deliv. 2023;30(1):2183816. doi:10.1080/10717544.2023.2183816

13. Protein quantification: evolution, current technologies, and future directions. Biocompare. Accessed March 10, 2026. https://www.biocompare.com/Editorial-Articles/620081-Protein-Quantification-Evolution-Current-Technologies-and-Future-Directions/

14. Kichou H, et al. Estimating the analytical performance of Raman spectroscopy for quantification of active ingredients in human stratum corneum. Molecules. 2022;27(9):2843. doi:10.3390/molecules27092843

15. Li X, de Assis Souza R, Heinemann M. The rate of glucose metabolism sets the cell morphology across yeast strains and species. Curr Biol. 2025;35(4):788-798.e4. doi:10.1016/j.cub.2024.12.039

16. Maicas S. The role of yeasts in fermentation processes. Microorganisms. 2020;8(8):1142. doi:10.3390/microorganisms8081142

17. Mallick S, et al. Yeast cells experience chronological life span extension under prolonged glucose starvation. Heliyon. 2025;11(4):e42898. doi:10.1016/j.heliyon.2025.e42898

18. Plaga A, Stümpfel J, Fiedler HP. Determination of carbohydrates in fermentation processes by high-performance liquid chromatography. Appl Microbiol Biotechnol. 1989;32:45-49. doi:10.1007/BF00164821

19. Chen Y, et al. HPLC for simultaneous quantification of free mannose and glucose concentrations in serum: use in detection of ovarian cancer. Front Chem. 2023;11:1289211. doi:10.3389/fchem.2023.1289211