Optical Screening and Classification of Drug Binding to Proteins in Human Blood Serum

Abstract

Protein−drug interactions in the human bloodstream are important factors in applications ranging from drug design, where protein binding influences efficacy and dose delivery, to biomedical diagnostics, where rapid, quantitative measurements could guide optimized treatment regimes. Current measurement approaches use multistep assays, which probe the protein-bound drug fraction indirectly and do not provide fundamental structural or dynamic information about the in vivo protein− drug interaction. We demonstrate that ultrafast 2D-IR spectroscopy can overcome these issues by providing a direct, label-free optical measurement of protein−drug binding in blood serum samples. Four commonly prescribed drugs, known to bind to human serum albumin (HSA), were added to pooled human serum at physiologically relevant concentrations. In each case, spectral changes to the amide I band of the serum sample were observed, consistent with binding to HSA, but were distinct for each of the four drugs. A machine-learning-based classification of the serum samples achieved a total cross-validation prediction accuracy of 92% when differentiating serum-only samples from those with a drug present. Identification on a per-drug basis achieved correct drug identification in 75% of cases. These unique spectroscopic signatures of the drug−protein interaction thus enable the detection and differentiation of drug containing samples and give structural insight into the binding process as well as quantitative information on protein−drug binding. Using currently available instrumentation, the 2D-IR data acquisition required just 1 min and 10 μL of serum per sample, and so these results pave the way to fast, specific, and quantitative measurements of protein−drug binding in vivo with potentially invaluable applications for the development of novel therapies and personalized medicine.

Using 2D-IR Spectroscopy to Measure the Structure, Dynamics, and Intermolecular Interactions of Proteins in H2O

“…Difficulties in spectral interpretation are exacerbated in label-free applications of 2D-IR because the link between structural change and the amide I band is yet to be fully elucidated. The potential now exists, however, for the interrogation of large 2D-IR data sets with advanced data processing methods, such as the machine learning tools that have been recently demonstrated . The exciting possibility arises of a predictive tool that is able to mine libraries of protein 2D-IR spectra alongside structural biology databases or artificial intelligence (AI)-based predictions of protein folding to establish spectral signatures of specific structural motifs before offering predictions of spectra of unknown proteins or interpretations of spectral change based on structure.…”

“…Hunt, N. T. 2023951703717045 . Reports the combination of machine learning with 2D-IR spectroscopy to classify drugs binding to proteins in blood serum samples.…”

“…The dependence of albumin binding on the presence of fatty acids means that studies of albumin–drug binding in buffer solution represent a poor model for behavior in vivo because it is virtually impossible to mimic the molecular composition of serum in the laboratory, irrespective of the fact that serum composition also changes dramatically in response to metabolic processes and other bodily functions . With this in mind, it is exciting that a recent study has shown that the interaction signature of four different drugs binding to serum albumin in human serum is drug-specific, with machine learning able to accurately identify drug-free and drug-containing serum samples from a small training data set . As a result, the potential now exists for 2D-IR identification, classification, and quantification of bound drugs in a biomedical sample.…”

“…Results of 2D-IR study using machine learning to identify drugs bound to blood serum proteins. (a–d) The spectral changes attributed to each drug by the machine learning algorithm following addition to neat serum: (a) warfarin, (b) paracetamol, (c) cefazolin, and (d) ibuprofen . Figure adapted from ref .…”

“… 51 With this in mind, it is exciting that a recent study has shown that the interaction signature of four different drugs binding to serum albumin in human serum is drug-specific, with machine learning able to accurately identify drug-free and drug-containing serum samples from a small training data set. 4 As a result, the potential now exists for 2D-IR identification, classification, and quantification of bound drugs in a biomedical sample. From a purely biophysical perspective, it also shows that different drugs binding to the same protein produce different changes to the amide I band.…”