Multisource Correlation Analysis (MuSCA) Applied to Raman Spectroscopy for Biochemical Analysis

James Piret, and Robin Turner, of Michael Smith Laboratories (Vancouver, BC Canada) and the University of British Columbia (UBC), have been exploring the benefits of extracting and displaying correlated spectrometric and non-spectrometric variables with a proposed method called multisource correlation analysis (MuSCA). Their work has uncovered several advantages of using Raman spectroscopy for these applications. Here, they discuss their efforts to develop an approach that permitted the integration of diverse biochemical information with measured spectra for co-analysis to characterize the spectra and take advantage of the available spectral information.

Piret and Turner are recipients of the 2022 Society for Applied Spectroscopy (SAS) William F. Meggers Award, presented annually at FACSS SciX. This award is given to the author(s) of the outstanding paper appearing in Applied Spectroscopy. More details on the award are available from the Society for Applied Spectroscopy

In a recent paper that was selected for the Society for Applied Spectroscopy 2022 William F. Meggers Award (1), you examined the benefits of extracting and displaying correlated spectrometric and non-spectrometric variables. Can you briefly describe this concept, the benefits, and how this method compares to the well-known two-dimensional correlation spectroscopy (2D-COS)?

Conventional 2D-COS analysis can be very useful for evaluating correlated and anti-correlated changes in different parts of a spectrum as a function of some perturbation to the system being investigated. It can thus provide useful insights into how the system responds to that perturbation. However, the interpretation of 2D-COS data can be quite challenging in the case of chemically complex systems such as living cells for which the spectra exhibit overlapping features that may have contributions from multiple components at all wavenumbers. For example, nucleic acids have several vibrational bands in the fingerprint region and some overlap with protein bands, some with lipid bands, some with carbohydrate bands, and many overlap with abundant small molecules that have chemical moieties in common with nucleic acids.

Independent (non-spectroscopic) measurements that yield relevant monovariate data can often aid the interpretation of spectral data by providing a direct measurement of some components represented in the spectra. For example, total protein or total nucleic acid biochemical assays can help gauge the amount of overlapping signal at bands where they both contribute. Likewise, highly selective assays for specific proteins or nucleic acids can help interpret spectral data. In general, any independent measurements can aid interpretation and link biological variables with observed changes in overlapped spectral peaks. However, these inferred links can normally be made only between the measured data and features in the spectra that are known to correspond to the measurand, typically specific peaks. It would be even more useful if one could identify and evaluate correlated spectral features that may be distributed throughout the spectrum. This would be feasible if the non-spectrometric data could be co-analyzed with the spectral data, perhaps in an appropriately modified form of 2D-COS analysis.

Our lead author, Dr. Georg Schulze, proposed a method that he dubbed "multisource correlation analysis" (MuSCA) whereby any available non-spectroscopic data from the same sample could be encoded along with the measured spectra for each perturbation point. The non-spectrometric data is encoded using evenly spaced artificial peaks with uniform widths and with amplitudes appropriately scaled relative to the value of the non-spectrometric (for example, biochemical assay) amplitudes at each perturbation point. These artificial peaks are appended to the measured spectra at each perturbation point. The resulting augmented hyperspectral data set is then analyzed using 2D-COS to yield maps (such as covariance and correlation coefficient maps) that directly provide correlations between the spectral and non-spectral data throughout the entire spectrum. It is essentially an adaptation or extension of the concept used in perturbation domain decomposition (PDD).

What inspired you to work on this method and approach? Was there a need for an improvement over the currently used methods?

We are exploring the ability of Raman spectroscopy to monitor cellular changes during the manufacturing of therapeutic cells. Cell changes during these processes are typically manifested by changes in their biochemical composition which can be very informative in terms of the state and quality of the cells (for example,increased or decreased levels of specific proteins as well as other changes in macromolecular composition). These changes can be directly measured by biochemical assays, but those measurements normally depend on costly and destructive assays that require samples of the cells to be removed from the system. Frequent sampling can significantly decrease the number of valuable cells to treat patients or even cause a microbial contamination, putting at risk the ability to perform life-saving therapies. Furthermore, the time delay before cell assay results are available can compromise the ability to respond effectively. So, there is considerable incentive to reduce the need for sampling and offline analyses.

Spectroscopic measurements offer the potential for non-destructive, in situ measurements that obviate the need for sampling and return results much more rapidly (even nearly in real time). Infrared (IR) and Raman spectroscopy have been shown to provide information-rich data about cells and tissues, but the data can be difficult to interpret as outlined above. This is especially true if the biochemical measurands have no clear relationship to the spectral data. For example, measurements of transcription factors in cells can signal the expression of developmentally important genes. Those measurements may correlate with certain features in measured spectra, though it may not be clear a priori what spectral features are relevant or why.

Multivariate analytical methods such as principal component analysis (PCA) can often reveal qualitative differences in spectral data that are useful, but it is usually difficult to infer the biological origin for the observed separation in PCs. We were drawn to the prospect of using analytical approaches like 2D-COS that give quantifiable statistics relating spectral variations. We hoped to be able to characterize the spectroscopic data sufficiently well during method development that product sampling during manufacturing could be significantly reduced. Thus, we sought to develop an approach that permitted the integration of diverse biochemical information with measured spectra for co-analysis in order to characterize the spectra and make use of all of the spectral information available. This supplements our use of specific peak amplitudes that are known to represent relevant chemical moieties in some cellular components, though does not consider unassigned peaks, peak shapes, or features between peaks. MuSCA permits co-analysis of the spectral data and biochemical data at every wavenumber, and thus emphasizes spectral changes that can be linked to their biological origin. Thus, the benefit of understanding how spectra relate to information that is obtained from sampling and offline analyses is that spectral measurements can then substantially reduce or altogether replace such procedures.

One might also want to include an array of non-spectrometric measurements to examine all the relationships between spectrometric and non-spectrometric ones and even mutually between non-spectrometric ones. A benefit that would derive from such an approach is that one might uncover unexpected and potentially useful relationships.

Another recent paper details applications of Raman spectroscopy in the development of cell human therapies (3), offering insight into the current state of emerging therapies. What advantage does the use of Raman spectroscopy provide for the analysis of single cells or tissues?

Many different types of therapeutic cells being developed with the hope of offering much improved treatments or even cures for a number of human diseases. Some cell therapies are already approved and many more are currently being tested in clinical trials. The specific role and value of Raman spectroscopy depends completely on the specifics of the cells and the manufacturing process. Raman can be useful during development of therapeutic cells by providing information about the metabolic or developmental state of the cells under particular experimental conditions, or to assess the quality or functionality of the cells as a final product. For example, we are collaborating with our co-author Prof. Timothy Kieffer, who is working on an improved protocol for directing the differentiation of human pluripotent stem cells into pancreatic cells that produce insulin for treating Type I diabetes. The current protocol involves seven stages of culture in defined media where cells are stepwise induced to develop toward the desired pancreatic cells. Each stage of the process involves the costly culture medium, and for patient safety it is extremely important to be able to monitor the cells and determine early if there are signs of abnormal (off-target) cells developing. It is also important to determine what type of off-target cells may be emerging. One aspect of our work in this collaborative project is to develop Raman spectroscopic methods to address both of these objectives.

Post-development, Raman may also be useful as a process analytical technology (PAT) for quality assurance/quality control (QA/QC) monitoring the manufacturing process. Raman spectroscopy is already used as a PAT for manufacturing other biologics (such as therapeutic antibodies) and small-molecule therapeutics. The complexity of therapeutic cells makes the development of PAT for their manufacture far more challenging such that we (and others) are working to develop innovative methods based on Raman spectroscopy for these processes. Dr. Piret will speak about this in his plenary lecture at SciX in October.

In general, the main advantage of Raman spectroscopy for these applications is, as mentioned above, that it offers a rapid, information-rich, non-perturbing, method of assessing changes in the chemical composition of the cells, which in turn signal the expected (or unexpected) response of the cells. The challenge of course is extracting the relevant information from the spectra that is convolved with much irrelevant information. Another advantage is that Raman spectroscopy can be implemented in a variety of ways depending on where in the process or how the measurements need to be made or what analytes are of most interest. For example, we rely heavily on the implementation of near-infrared Raman spectroscopy interfaced with an optical microscopy platform (Raman microspectroscopy); other applications may be able to exploit fiber-optic probes that can be inserted within a bioreactor to, for example, monitor changes in the composition of the culture medium. Another related advantage is that Raman measurements can be carried out using any optical frequency from infrared to ultra-violet, and there can be advantages to choosing a particular wavelength. There are many possibilities discussed in our review (3).

Please describe some of the challenges you faced in working with collecting and analyzing Raman spectra of therapeutic cells and how these challenges may have affected useful clinical applications.

The main challenge encountered in collecting and analyzing Raman spectra for any application involving cells, not only therapeutic cells, is that the sensitivity of spontaneous Raman is quite low. This requires collecting the Raman scattered light for appreciable times (minutes) for each spectrum in order to achieve satisfactory signal-to-noise performance. This can be problematic for measuring live cells, especially mobile live cells. It is a serious limitation if spatially resolved measurements are needed, for example to produce so-called "chemical images" where spectra need to be measured at each pixel of a chemical image. As a result, we have mainly been working with fixed cells for method development using Raman micro-spectroscopy. This allows cells to be isolated and immobilized in the absence of interfering medium components, and also allows cell samples to be preserved at low temperature for future re-examination, but it is clearly not the conditions that would be needed in a PAT application.

Another challenge is that true in situ measurements are difficult to implement. We have explored the use of spatially offset Raman spectroscopy (SORS) that permits measurements to be made through interfering barriers such as the plastic or glass walls of culture vessels, and fiber-optic probes that can be inserted into the vessels. However, the development of improved exposure systems remains a technical challenge that needs to be overcome to advance clinical applications.

Of great promise are emerging methods based on coherent (cf. spontaneous) Raman spectroscopy such as coherent anti-stokes Raman spectroscopy (CARS) and stimulated Raman spectroscopy (SRS). These techniques can potentially address the sensitivity limitations, allowing near real-time spectral acquisition, and facilitate in situ measurement capability due to their high confocality. These techniques are now commercially available in turn-key instruments that enable measurements within flow-cells or micro-fluidic systems for rapid atline analyses. The analytical methods that we are developing should be transferable or at least adaptable to SRS data acquisition.

In what ways may other researchers become involved in cell therapy research and what other analytical techniques show promise?

There are many dimensions to cell therapy research and analytical measurements are critically important in all of them. There are open questions that could be addressable by almost any analytical science researcher. To suggest just a few areas, these include engineering of appropriate exposure systems for effective data acquisition, development of online-compatible biochemical analysis systems, or advanced imaging systems with analytical capabilities. That said, for analytical scientists to work effectively in this area it is important to have a suitable biomedical research collaborator to elucidate the specific problems that need to be addressed and, importantly, provide access to appropriate cells and biochemical data. Fortunately, there are now biomedical researchers at most institutions that are working on some aspect of therapeutic cell research or development, and they are often networked with biotechnology industrial collaborators. However, those experts rarely approach the analytical scientists, so analytical scientists usually would need to approach them.

In addition to the spontaneous Raman that is our specialty, we mentioned above some other non-linear techniques based on Raman scattering (CARS, SRS). There are also some multi-modal techniques that combine Raman with other analytical modalities such as fluorescence, elastic light scattering, optical coherence tomography, phase-contrast microscopy, and mass spectrometry. Other promising non-spectroscopic methods include new non-linear imaging techniques such as two-photon fluorescence lifetime imaging (FLIM) for cell cluster and organoid analyses, second and third harmonic generation (SHG, THG) for high-order structure analysis. In terms of data analytics, there are also many efforts to implement machine learning techniques for all kinds of data analysis as well as bioprocess monitoring and control. We have worked on some new multivariate data analysis techniques as discussed here, but there are many established methods that are also applicable to cellular data analysis that we and others have used in analytical method development such as PCA (mentioned above, cluster analysis, partial least-squares discriminant analysis, and non-negative matrix factorization. There are also some promising new transducers based on biosensors and electrochemical methods. Basically, almost any new chemical analysis or data analysis technique may be useful in some aspect of therapeutic cell development or manufacturing research. There is a lot of room for analytical scientists and engineers to contribute to this relatively new field and new innovations could have a huge impact.

What are your next steps in working with using Raman spectroscopy of cells for therapeutic purposes?

We are continuing our Raman spectroscopy and other analyses of stem cell processes that produce cells for treating diabetes, as well as investigating the application of our methods to T cells. This includes laboratory benchtop work by examining the many still remaining cellular variations that can be discerned using Raman spectroscopy. We are also pursuing a number of interests in developing improved data analytic methods to further exploit the rich information content of Raman spectra. Overall, we are working towards the ultimate goal to translate our knowledge and experience from the benchtop to manufacturing settings, thus implementing Raman spectroscopy with the associated analytical techniques as a fully-fledged PAT.

References

  1. H.G. Schulze, S. Rangan, M.Z. Vardaki, D. G. Iworima, T.J. Kieffer, M.W. Blades, R.F.B. Turner, and J.M. Piret, Appl Spectrosc 75(5), 520–530 (2021).
    DOI: https://doi.org/10.1177/0003702820979331
  2. (2) H.G. Schulze, A. Jirasek, M.W. Blades, and R.F.B. Turner, Appl Spectrosc 57(12), 1561–1574 (2003).
    DOI: https://doi.org/10.1177%2F0003702820979331
  3. S. Rangan, H.G. Schulze, M.Z. Vardaki, M.W. Blades, J. M. Piret, and R.F.B Turner, Analyst 145, 2770–2105(2020).
    DOI: https://doi.org/10.1039/c9an01811e

James Piret has a Bachelor’s degree from Harvard in Applied Mathematics to Biochemistry, and a Chemical Engineering doctoral degree from MIT in 1989. He is a professor at the University of British Columbia in the Department of Chemical & Biological Engineering and the Michael Smith Laboratories; he is also an associate member of the School of Biomedical Engineering. His research focus is on innovative process and device technology development for mammalian cell culture therapeutic protein or cell manufacturing.

Robin Turner earned a PhD degree in electrical engineering from the University of Alberta in 1990. He is now a Professor at The University of British Columbia with joint appointments in the Michael Smith Laboratories and Department of Electrical & Computer Engineering; he is also an Associate Member of the Department of Chemistry. His current research activities focus on applications of Raman spectroscopy to analytical problems in biochemistry, biotechnology, and biomedical engineering.