Improving Raman Spectral Quality in Autofluorescent Biological Samples

Raman spectroscopy offers specific, non-destructive, analysis of biological samples, enabling both identification and quantitative measurement without the need for extensive sample preparation. Advances in sensors, spectrometer design, and laser technology have expanded its use beyond specialized laboratories into a broader range of biological and medical applications. However, a persistent challenge with biological samples is autofluorescence, which can obscure or overwhelm the Raman signal. A variety of strategies have been developed to mitigate this effect and improve spectral quality. This article examines the impact of autofluorescence on Raman measurements and evaluates several approaches for reducing its influence, including the use of longer excitation wavelengths, minimizing excitation and detection volumes, and applying advanced post-processing techniques.

Because of its molecular specificity and minimal sample preparation requirements, Raman spectroscopy has become widely used for identification of potentially dangerous substances for over a decade. The development of compact diode lasers and sensitive charge-coupled device and complementary metal-oxide-semiconductor (CMOS) detectors has enabled the deployment of Raman systems for security applications in the field.

Raman in Biological and Biomedical Applications

Over the past decade, Raman spectroscopy has increasingly been applied not just as a substance identification tool in security applications, but also as a valuable analytical tool in the biological and medical fields. Raman specificity not only enables the identification of single compounds but also the characterization of complex mixtures, such as tissues or cell cultures. Differences in the spectrum correlate with the cell type, bacteria identity, or antibiotic response.

Compositional information in biotech production is key to controlling the process tightly, thus optimizing yield, quality, and overall productivity. Raman spectra are recorded without breaking the sterile seal; the measurements are taken through a viewport or a dip probe that is part of the sterilized reactor. This allows simultaneous inline real-time measurement of multiple compounds, whereby monitoring the feedstock level as a critical process parameter steers the microorganisms from proliferation to production of the target molecule.

Strategies for Autofluorescence Reduction

Unfortunately, in these application spaces, the often very strong autofluorescence of biological samples can overwhelm the generally weak Raman signal. Traditional approaches to dealing with fluorescence can be limited in their effectiveness or require increased complexity and expense.

In the most common approach, a baseline correction algorithm is applied to fit a smooth function below the Raman spectrum, which is subtracted in a second step.^1,2 Multiple algorithms have been published,^3-7 but many require manual adjustment of some fit parameters, resulting in inconsistent fitting performance. The most significant drawback is the fact that the significant shot noise of the high fluorescence signal is carried over into the baseline-corrected result. Although the fluorescence signal is removed, its associated noise can still hide the Raman peaks.

A second traditional approach is a switch to a longer laser excitation wavelength, which has shown to generate lower levels of autofluorescence.⁸ Figure 1 shows the decrease of the background fluorescence relative to the Raman signal for increasingly longer excitation wavelengths for a sample of glucose. The greatest fluorescence reduction is achieved with 1064-nm excitation, the near-infrared (NIR) wavelengths, however, require the switch to the more costly, and less sensitive, InGaAs detector technology. For this reason, 785-nm has thus emerged as the most common Raman excitation wavelength.

Other approaches have been developed, such as shifted excitation difference Raman spectroscopy (SERDS) and time-resolved Raman spectroscopy. In SERDS, the laser excitation is switched between two neighboring wavelengths, about a typical Raman width apart. As only the Raman peaks shift with this wavelength change, but not the fluorescence, the difference spectrum cancels the fluorescence. Time-resolved Raman utilizes the fact that Raman is an instantaneous process, while fluorescence emerges after a short delay. The instantaneous signal therefore discriminates against fluorescence. Both methods, SERDS and time-resolved Raman, however, require more sophisticated and higher-cost equipment.

In this article, we evaluate two novel strategies for the autofluorescence problem, reducing some of those limitations: advanced post-processing and reduction of excitation and detection volume – and the combination of the two. Rather than leading to a more complex instrument, these methods are realized on a new, pocket-size, Raman system.

Materials and Methods

Two setups are employed in this article. In the fermentation demonstration, we use a Wasatch Photonics compact Raman spectrometer with 785-nm laser excitation, fiber-coupled to a probe inserted into a stirred solution of 50 g/L glucose and 2 g/L yeast. The fermentation is observed over 14 hours, with spectra collected in “batch mode” every two minutes at 5 seconds integration time averaged over 20 scans.

For the cell culture experiments, we use a handheld Wasatch Photonics 785 XS Raman system with integrated 100-mW single-mode laser and probe optics, both with the standard 19-mm focal-length front optics as well as with a 50x microscope objective for the tighter focus comparison. The system is mounted pointing down and positioned with an x/y/z stage to focus onto various cell cultures grown on standard nutritional agar plates. Individual, non-averaged spectra are collected with 2-second integration time only from a narrow vertical band in the center of the 2D CMOS detector of the Raman system.

In both experiments, the resulting Raman spectra are processed using a custom neural network of the U-Net type (with the internal project name “DALAI”), which was trained to remove fluorescence and noise from input spectra, while preserving Raman peak intensities.

Results and Discussion

Fermentation Experiment

The original Raman spectra of the fermentation process (Figure 2) show the growing overall background due to the increasing yeast cell count in the broth. They also reflect the appearance of the ethanol product with its Raman peak at 890 cm^-1. The finer details are hidden under the strong fluorescence background.

DALAI post-processing extracts background- and noise-free Raman peaks from the original spectra, as shown in the middle panel of Figure 2. The comparison to pure Raman spectra for glucose (feed) and ethanol (main product) in the bottom panel of Figure 2 illustrates the good agreement between the DALAI-processed and the expected peaks.

In addition to glucose and ethanol, we also see peaks for side products of the fermentation, at around 700 and 1800 cm^-1. Additional comparison measurements show that these peaks are associated with the production of carbonic acid.

The DALAI-processed spectra reflect the expected evolution of the fermentation: decrease of glucose peak intensity and increase in the ethanol peak height, illustrating the quantitative nature of the post-processing step.

Cell Culture Experiment

Cell culture Raman spectra are traditionally recorded on larger, research-grade Raman microscopes. These instruments yield spectra with a good ratio of Raman signal to fluorescence background signal, better than what is achievable with standard, compact Raman systems using the same samples. Two main factors are responsible for this difference: Raman microscopes use single-mode laser excitation for a tighter focus and confocal detection for out-of-focus rejection. This restricts the signal detection to a small sample region with the highest laser power density, which leads to saturation and bleaching of fluorescence, while the Raman response remains linear in excitation power.

For compactness, the Wasatch Photonics 785 XS system integrates a single-mode laser diode for Raman excitation. With its tight focus, it covers the first aspect of the Raman microscope advantage. As the top-left panel of Figure 3 illustrates, the focus generated by the 19-mm focal length front optic is not tight enough to improve the Raman-to-fluorescence ratio to the desired levels. A shorter-focal-length, 50x microscope objective improves this ratio further (Figure 3, top-right panel).

To reduce the detected volume in sync with the excited volume, we reduce the vertical height, known as the region of interest (ROI), which is read from the CMOS detector to just 4 pixels (15 μm). As the non-astigmatic lens system of the XS spectrometer bench images a point from the slit 1:1 onto a point on the detector, this dramatic reduction in effective detector height constitutes a confocal detection scheme, equivalent to using a pinhole in the optical path.

As shown in the bottom-left panel of Figure 3, this confocal detection leads to a significant improvement in the overall Raman-to-fluorescence signal ratio. However, the reduction in detector height also decreases the overall signal-to-noise ratio of the spectrum. To counteract this loss, the confocal spectra are post-processed with the DALAI neural network to generate the spectrum shown in the bottom-right panel of Figure 3.

We needed to assess whether this captured and post-processed spectrum consists of typical Raman bands seen in cell cultures. To evaluate, we overlay the observed peaks with the Raman shifts listed for various biochemical compounds typically observed in Raman spectra of tissues and cell cultures.⁹ The comparison shown in Figure 4 illustrates a good alignment of expected and observed peaks. This shows that the XS Raman system with tight focus, combined with advanced post-processing, can reveal details in the resulting spectra for a cell culture that are typically associated with research Raman microscopes.

Conclusion

We present two approaches that allow us to record high-quality Raman spectra for biological samples. For a broth of glucose with yeast, as an illustration for a biotechnology application, we demonstrate effective reduction of fluorescence background and noise. Combining this post-processing with a tight single-mode focus and confocal detection in a miniature Raman system yields higher clarity spectra for a cell culture, with a quality commonly associated with research Raman microscopes.

References

1. Wei, D.; Chen, S.; Liu, Q. Review of fluorescence suppression techniques in Raman spectroscopy. Appl. Spectrosc. Rev. 2015, 50 (5), 387–406. DOI: 10.1080/05704928.2014.999936
2. Liland, K. H.; Almøy, T.; Mevik, B.-H. Optimal choice of baseline correction for multivariate calibration of spectra. Appl. Spectrosc. 2010, 64 (9), 1007–1016. DOI: 10.1366/000370210792434350
3. Galloway, C. M.; Le Ru, E. C.; Etchegoin, P. G. An iterative algorithm for background removal in spectroscopy by wavelet transforms. Appl. Spectrosc. 2009, 63 (12), 1370–1376. DOI: 10.1366/000370209790108905
4. Zhang, Z. M.; Chen, S.; Liang, Y. Z.; Liu, Z. X.; et al. An intelligent background-correction algorithm for highly fluorescent samples in Raman spectroscopy. J. Raman Spectrosc. 2010, 41 (6), 659–669. DOI: 10.1002/jrs.2500
5. Schulze, H. G.; Foist, R. B.; Okuda, K.; Ivanov, A.; Turner, R. F. B. A model-free, fully automated baseline-removal method for Raman spectra. Appl. Spectrosc. 2011, 65 (1), 75–84. DOI: 10.1366/10-06010
6. Cadusch, P. J.; Hlaing, M. M.; Wade, S.A.; McArthur, S. L.; Stoddart, P. R. Improved methods for fluorescence background subtraction from Raman spectra. J. Raman Spectrosc. 2013, 44 (2), 1587–1595. DOI: 10.1002/jrs.4371
7. Liu, H.; Zhang, Z.; Liu, S.; et al. Joint baseline-correction and denoising for Raman spectra. Appl. Spectrosc. 2015, 69 (9), 1013–1022. DOI: 10.1366/14-07760
8. Crocombe, R. A.; Kammrath, B. W.; Leary, P. E.; et al. LEGO blocks as “standard” samples for evaluation of fluorescence avoidance and mitigation in Raman spectroscopy. Appl. Spectrosc. 2024, 78 (3), 340–348. DOI: 10.1177/00037028231221585
9. Chen, Y.; Dai, J.; Zhou, X.; et al. Raman spectroscopy analysis of the biochemical characteristics of molecules associated with the malignant transformation of gastric mucosa. PLoS One 2014, 9 (4), e93906. DOI: 10.1371/journal.pone.0093906