Data mining Raman microspectroscopic responses of cells to drugs in vitro using multivariate curve resolution-alternating least squares

Abstract

Raman microspectroscopy is gaining popularity for the analysis of time-dependent biological processes such as drug uptake and cellular response. It is a label-free technique which acquires signals from a large variety of components, including cell biomolecules and exogenous compounds such as drugs and nanoparticles, and is commonly employed for in vitro analysis of cells and cell populations with no labelling or staining required. By monitoring the changes to the Raman spectra of the cell as a result of a perturbing agent (e.g. inoculation of a drug or toxic agent), one can study the associated changes in cell biochemistry involved in both, the disruption and the subsequent cellular response. The main challenge is that the Raman spectra should be data mined in order to extract the information corresponding to the different actors involved on the process. Here, we study the application of multivariate curve resolution-alternating least squares (MCR-ALS) for extracting kinetic and biochemical information of time-dependent cellular processes. The technique allows the elucidation of the concentration profiles as well as the pure spectra of the components involved. Initially, we used Ordinary Differential Equations (ODE) to simulate drug uptake and 2 responses, which were employed to simulate perturbations to experimental control spectra, creating a dataset containing 36 simulated Raman spectra. Four different scenarios governing the drug exposure-response were evaluated: an undetectable disruption (e.g. radiation), a detectable disruption (e.g. a drug) and disruption with a signal significantly larger than the biological changes induced (e.g. a resonant drug), as well as simultaneous and asynchronous responses. Subsequently, data acquired from the exposure of a pulmonary adenocarcinoma cell line (A549) to Doxorubicin was analysed. The results indicate that MCR-ALS can independently identify and isolate both the spectra of the drug and the cell 2 responses under the different scenarios. The predicted concentrations map out the drug uptake and cellular response curves. The technique shows great potential to investigate non-linear kinetics and modes of action. Advantages and limitations of the technique are discussed, providing guidelines for future analysis strategies.

Multimodal vibrational studies of drug uptake in vitro: Is the whole greater than the sum of their parts?

“…The absence of this band in oPLSR and 2D correlation can be caused by negative contributions of the band at 1660 cm −1 , which could be associated to changes in protein conformation. In contrast, negative correlations of 1660 cm −1 were not identified by MCR-ALS, consistent with previous studies which evidenced some limitations in reporting synchronous negative and positive contributions [9]..…”

“…This can be observed in the poor correlation between the cross-validated (CV) predicted and actual time values (See Figure 1F) plot. This behaviour has been previously discussed with simulated data [9] and evidences the problems of using a linear model in a time dependent data unless the kinetic process is separated in linear trams. Nevertheless, the Root mean square error of cross validation (RMSECV) error obtained (10.39 hours) using four latent variables (LV) was deemed significant by a permutation test (P < .005) [36], showing some correlation between the incubation time and the Raman spectra.…”

“…The simulated Raman data set from ref. [9] was used, but it was extended to simulate also IR spectra. The Raman spectra contained two artificial features, which represent the initial rapid uptake and binding of the drug to RNA/DNA in the nucleus, and a slower, secondary response to represent the subsequent cellular response (See SI3).…”

“…Further information about 2D correlation analysis can be found here [26,27,46]. To aid the interpretation of the synchronous coefficients, a simulated data set was created, in a similar way to previous studies [9,12]. In this way, one can study the expected results and then compare them with the real data.…”

“…These techniques, when combined with imaging microscopic systems, are used to study the distribution of chemical substances in cells and tissues [6,7]. Raman microspectrocopy has been demonstrated to be capable of monitoring sub-cellular drug accumulation and distribution of doxorubicin (DOX) in lung cancer cell lines and to identify the mechanism of the DOX action and cellular resistance [8,9]. With a spatial resolution similar to that of confocal laser scanning fluorescence microscopy [10,11], confocal Raman microspectroscopy indicated an initial accumulation and saturation of the drug in the nucleoli (~6 hours), followed by nucleus of the cells (~12 hours), before it began to accumulate in the cytoplasm (>24 hours) [10,11].…”