Detecting Cancer Biomarkers in Canines: An Interview with Landulfo Silveira Jr.

Detecting biomarkers in cancer is a hot topic in clinical and biological analysis. Spectroscopy, particularly techniques like infrared (IR) spectroscopy, mass spectrometry (MS), and Raman spectroscopy, is being increasingly applied to detect cancer biomarkers in canine sera (1) and urine (2), as well as to detect cancer in dog’s brain tissues intraoperatively using portable instruments (3). These methods provide a non-invasive approach to analyze tissue and fluid samples and identify subtle molecular changes associated with cancer (1–3).

Landulfo Silveira Jr. is exploring this question. Silveira Jr. received his master’s degree in biomedical engineering from Universidade do Vale do Paraíba in 1996 after a 1-year period appointed as a Research Assistant fellow in the Spectroscopy Laboratory at MIT, in Cambridge, MA, under Professor Michael Feld’s supervision, and his Ph.D. at Pathological Anatomy and Clinical Pathology from the Faculty of Medicine from Universidade de São Paulo in 2001. His research has mostly concentrated in using Raman, fluorescence, and reflectance spectroscopy, focusing on application areas such as the processing of biological signals and biomedical instrumentation aiming at clinical and laboratorial diagnosis.

Landulfo Silveira Jr. of Universidade Anhembi Morumbi-UAM and Center for Innovation, Technology and Education-CITÉ (São Paulo, Brazil). Photo Credit: © Landulfo Silveira Jr.

Recently, Silveira Jr. and his team used Raman spectroscopy to detect cancer biomarkers in the blood serum of domestic dogs (3). Spectroscopy sat down with Silveira Jr. to discuss his team’s research and what these findings means for the future of detecting cancer in canines.

Can you explain the principle behind Raman spectroscopy’s application in detecting biomarkers of cancer in canine sera?

Raman spectroscopy is an optical technique that relies on the inelastic scattering of light, known as the Raman effect, to analyze the molecular composition of a sample. When light interacts with the molecules in a biological sample such as blood serum, proteins, and lipids could be identified and, in most cases, estimated (quantified). Other biological compounds could also be detected in serum such as free amino acids, nutrients absorbed by the intestine from the food, liver-synthesized metabolites, residues from cell and tissue metabolism, and other constituents that circulate for healthy body functioning. Despite cancer as a disease normally impacting specific organs and tissues, there could be changes in cells and tissue dynamics occurring after the beginning of a neoplastic change. This leads to the presence of specific metabolites, which can be considered “biomarkers” of cancer. In the context of our recent study, Raman spectroscopy was used to detect differences in the biochemical composition of serum from healthy dogs and those with cancer, identifying potential biomarkers related to the neoplastic alterations already diagnosed in the sample dogs.

How does the “redshift” phenomenon in Raman spectra relate to conformational changes in proteins and amino acids in the context of cancer detection?

The “redshift” phenomenon refers to the displacement of Raman peaks to lower energy (higher wavelength) regions in the spectra; that is the opposite of the “blueshift” (displacement of Raman peaks to higher energy and lower wavelength regions in the spectra). These shifts can occur in some cases (for example, heating or cooling a crystal that causes a decrease or increase in its crystallinity). In the context of the serum constituents related to the cancer detection, the redshift observed in the peaks at 621, 642, 1003, and 1032 cm^-1, which correspond to peaks of free (circulating) amino acids or amino acids containing aromatic rings (phenylalanine and tyrosine) in proteins, may suggest that the structure of the protein molecules has been altered, likely because of changes in the chemical environment, such as protonation, conformational changes (protein folding), or bonding near aromatic rings. In fact, more research is needed to confirm these alterations in protein structure and the amount of redshift. It is interesting that we found blueshift in the same peaks in a recent study with serum from children and adolescents with diagnosed cancer compared to subjects without cancer (4). At the beginning of the data analysis of the cited study, we thought that the Raman instrument was miscalibrated, so we checked the calibration and repeated the measurements (4). When we faced the study of canine sera, we confirmed that the shift is a real phenomenon that needs further investigation, because it seems to be a key marker of cancer detection by Raman spectroscopy in serum.

What is the significance of the peaks at 451 cm⁻¹ (tryptophan) and 1441 cm⁻¹ (lipids) being higher in the Healthy group?

The higher intensity of the peaks at 451 cm^-1 (assigned to tryptophan) and 1441 cm^-1 (assigned to lipids) in the Healthy group suggested a greater concentration of these molecules in the sera of healthy dogs compared to those with cancer. Tryptophan is an essential amino acid involved in protein synthesis, and its higher concentration in healthy dogs could indicate normal protein absorption and metabolism. Similarly, the higher lipid content in healthy dogs may reflect the normal nutritional and metabolic state, as cancerous cells often exhibit altered protein and lipid metabolisms. The reduction of these peaks in the Cancer group indicates that cancer may disrupt the normal biochemical processes involving these molecules. For lipids, cachexia and increased catabolism may be present, which promotes an increased use of lipids (accumulated and from diet) as an energy source. For proteins, the systemic inflammatory response caused by the cancer may redirect the production of acute phase proteins such as C-reactive protein, thus reducing the production of other proteins. Cachexia is a classical issue related to cancer patients.

Could you elaborate on the role of principal component analysis (PCA) in feature extraction and how it aids in differentiating between healthy and cancerous sera samples?

I have been working with PCA for so many years and one thing that I am impressed with the technique is the possibility of “finding a needle in a haystack,” where the “needle” can be interpreted as a small peak of a compound of interest or a small difference in the sample composition that is not seen when the mean spectra of different groups are compared. The “haystack” refers to the collection of compounds common to all samples, including spectral noise. As PCA is a multivariate statistical technique used to reduce the dimensionality of the data while retaining most of the variation present in the data set, it can perceive the spectral differences in the composition of one group sample compared to another if this difference is present, even if this difference in from a spectral feature with low intensity. This occurs because of the nature of the PCA—the variance is well captured when a large data set is used. Therefore, one needs to have a considerable amount of spectral data to accomplish the task. In this study, PCA was applied to the Raman spectra to extract features that capture the most significant biochemical differences in the serum samples of healthy dogs compared to dogs with cancer. By identifying the principal component variables (“loadings” and “scores”) that explain most of the variance in the data set, PCA helps to differentiate between the two groups by highlighting spectral features seen in the PCA score that vary significantly (seen by the intensity of the loadings) in the group with cancer compared to the Healthy group. These features can then be used to identify potential biomarkers by associating the spectral features found in the score with the biocompounds expected to be present in the sample. Therefore, diagnostic models based on the loadings values for each group can be developed. Please note that I am using the MATLAB’s nomenclature for the PCA variables, “loadings” and “scores”, that may differ from other sources.

How does partial least squares (PLS) regression work in the discrimination between Healthy and Cancer groups, and why was it chosen for this study?

PLS regression is a statistical method used to model the relationship between predictor variables (Raman spectra) and response variables (group labels regarded to the health status—healthy or cancer). In this study, PLS was used for discriminant analysis (PLS-DA) to classify each one of the sera samples into the group Healthy or Cancer. PLS-DA works by finding latent variables that maximize the covariance between the spectra and the group labels, thereby improving the classification accuracy (often better than the PCA-DA). The method was chosen because it can handle large, multicollinear data sets (that is the case of Raman spectra), and it is particularly effective in cases where the number of predictors exceed the number of observations (another characteristic of the Raman spectra). In the PLS algorithm, the Raman spectra are the X (predictors) and the group labels are the Y (responses). Because the PLS-DA uses the group labels to build the predictive model, it is said to be a supervised classification technique, opposite to PCA that is said to be an unsupervised technique. We used “leave-one-out” cross-validation in the PLS-DA model, meaning that the model is built with n – 1 samples, and the left-out sample is tested prospectively. Then, the model is rebuilt where each sample is left-out and tested one at a time.

The study mentions a classification accuracy of 78% using PLS-DA. What factors could influence this accuracy, and how might it be improved in future studies?

The classification accuracy achieved in this study could be influenced by several factors, including:

• Sample variability: Differences in age, breed, and gender among the dogs could introduce variability in the serum composition, affecting the accuracy. Different food intake and nutritional status of each animal could also influence the accuracy results by introducing variability in the serum composition.
• Data preprocessing: The effectiveness of baseline correction, normalization, and cosmic ray removal could impact the quality of the spectra and the subsequent data analysis.
• Number of latent variables (LVs): The choice of the number of LVs used in PLS-DA can affect the model’s performance. Overfitting or underfitting (which means more or less variables than needed) could reduce accuracy.
• Instrument’s spectral resolution and sensitivity: Spectrometers with higher resolution and higher sensitivity could capture more detailed biochemical information, potentially improving accuracy. Increasing the time exposure for collecting Raman signal with better signal-to-noise ratio could also be a key factor.
• Data processing and modeling: The use of more advanced processing techniques based on artificial intelligence and machine learning could also improve the accuracy.

Future improvements in the data collection aiming at increasing the model’s accuracy include increasing the sample size, controlling for demographic variables and standardizing food intake, optimizing the preprocessing steps, and exploring more advanced processing techniques.

What potential implications do these findings have for early cancer detection in domestic dogs, and how might similar techniques be applied to human cancer diagnostics?

Comparative oncology is a research field that studies cancer across different species, particularly between animals and humans, to gain insights into cancer biology and its development, as well as its treatment and prevention. This comparative approach leverages the fact that many companion animals share similar environments and disease patterns with humans, making them valuable models for studying cancer. By examining how cancer develops and progresses in animals, such as dogs and cats, which naturally develop cancers like humans as their age increases, researchers can identify commonalities and differences that help advance cancer studies for both species in diagnostics and therapy. This study fits in this concept as the findings of this study are close to others that showed spectral changes in serum of patients that were diagnosed with cancer (4–7). The study demonstrated the potential of Raman spectroscopy as a non-invasive, rapid, and label-free diagnostic tool for cancer detection in dogs. This approach could be particularly useful for routine screenings and monitoring treatment responses. Given the similarities in the biochemical processes between dogs and humans, the technique could be potentially applied to human cancer diagnostics as in fact there are many studies including the cited above. This could lead to the development of early detection tools for various cancers in different species, improving prognosis and treatment outcomes.

What further research needs to be done in this area?

Further research in this area could focus on:

• Larger and more diverse sample populations: To better understand the variability in serum composition of both healthy subjects and cancer patients and improve the robustness of the diagnostic models;
• Longitudinal studies: To assess the ability of Raman spectroscopy to monitor the disease progression and the treatment response over time, as the patients can be evaluated over time by a single blood sample withdrawn;
• Screening test: The Raman-based classification model applied to serum samples could be evaluated as a screening tool for cancer diagnosis by collecting blood samples in routine tests (check-ups) and vaccine campaigns;
• Integration with other diagnostic tools: Combining Raman spectroscopy with other available diagnostic methods, such as anamnesis and clinical examination, x-ray and ultrasound imaging, or more expensive molecular assays, the diagnostic accuracy could be improved;
• Exploration of other biological fluids: Investigating the application of Raman spectroscopy to other biological fluids, such as blood and urine, could expand its diagnostic utility; and
• Machine learning approaches: Exploring more sophisticated machine learning algorithms and algorithms based on artificial intelligence for better classification and prediction.

These steps could enhance the diagnostic capabilities of Raman spectroscopy and facilitate its broader adoption in veterinary medicine.

References

1. Lopes, D. F.; Silverio, A.; Schmidt, A. K. A.; et al. Characterization of Biomarkers in Blood Serum for Cancer Diagnosis in Dogs Using Raman Spectroscopy. J. Biophotonics 2024, 17 (3), e202300338. DOI: 10.1002/jbio.202300338
2. Robertson, J. L.; Dervisis, N.; Rossmeisl, J.; et al. Cancer Detection in Dogs Using Rapid Raman Molecular Urinalysis. Front Vet. Sci. 2024, 11, 1328058. DOI: 10.3389/fvets.2024.1328058
3. Doran, C. E.; Frank, C. B.; McGrath, S.; Packer, R. A. Use of Handheld Raman Spectroscopy for Intraoperative Differentiation of Normal Brain Tissue From Intracranial Neoplasms in Dogs. Front Vet. Sci. 2022, 8, 819200. DOI: 10.3389/fvets.2021.819200
4. Lima, A. M. F.; Daniel, C. R.; Pacheco, M. T. T.; et al. Discrimination of Leukemias and Non-leukemic Cancers in Blood Serum Samples of Children and Adolescents Using a Raman Spectral Model. Lasers Med. Sci. 2022, 38 (1), 22. DOI: 10.1007/s10103-022-03681-2
5. Luo, H.; Zu, R.; Li, L.; et al. Serum laser Raman spectroscopy as a potential diagnostic tool to discriminate the benignancy or malignancy of pulmonary nodules. iScience 2023, 26 (5), 106693. DOI: 10.1016/j.isci.2023.106693
6. Githaiga, J. I.; Angeyo, H. K.; Kaduki, K. A.; et al. Quantitative Raman Spectroscopy of Breast Cancer Malignancy Utilizing Higher-order Principal Components: A Preliminary Study. Sci. African 2021, e01035. DOI: 10.1016/j.sciaf.2021.e01035
7. Gonzalez-Solis, J. L.; Martinez-Espinosa, J. C.; Torres-Gonzalez, L. A.; et al. Cervical Cancer Detection Based on Serum Sample Raman Spectroscopy. Lasers Med. Sci. 2014, 29 (3), 979–985. DOI: 10.1007/s10103-013-1447-6