“When the Indian physicist Chandrasekhara Venkata Raman discovered the spectroscopic method named after him, Raman spectroscopy, in 1928,” said Jürgen Popp, Professor at Friedrich-Schiller-University in Jena, Germany (where he is Director of the Institute of Physical Chemistry) and scientific director of the Leibniz Institute of Photonic Technology in Jena, “nobody, probably not even he himself, would have predicted the great success of this spectroscopic method almost 100 years after its discovery. Without exaggeration, it can be said that Raman spectroscopy has become one of the most important analytical methods, with applications in all fields of natural sciences, medicine, but also in various other disciplines. Raman spectroscopy has even left the earth and flies to Mars.”
Popp is the recipient of the 2023 Charles Mann Award, presented to, as stated on the Federation of Analytical Chemistry and Spectroscopy Societies (FACSS) webpage, an individual who has demonstrated advancement(s) in the field of applied Raman spectroscopy, and/or demonstrated dedication to the advancement of the Raman spectroscopy program. Popp shared his thoughts about the technique with Spectroscopy.
Your research particularly focuses on the development of innovative linear and non-linear Raman approaches in a multimodal approach together with other optical or spectroscopic methods to address important biomedical questions in areas such as infectious diseases and cancer diagnostics, as well as environmental applications and food safety. What benefits do the Raman approaches offer that other techniques do not?
In general, the Raman effect describes the inelastic scattering of photons by a quantized molecular system. In most cases, the vibrational states of molecules are used as the scattering system, which is why Raman spectroscopy is often referred to as vibrational spectroscopy. Since molecular vibrations are specific to each molecule, vibrational spectra can be interpreted as a kind of characteristic "molecular fingerprint" of a studied inorganic, organic, or biological molecule or more complex systems such as biological cells and tissues.
Especially in the fields of life sciences and medicine, Raman-based technologies have demonstrated their great potential and increasingly complement established techniques such as fluorescence spectroscopy or microscopy. Raman spectroscopy enables label-free detection of the molecular composition as well as the morphology of complex samples such as biological cells or tissues with little or no sample preparation. This is because a Raman spectrum consists of many dozens of independent parameters ranging from concentration to the three-dimensional arrangement of macromolecules and biopolymers in biological samples.
The advantages of Raman spectroscopy are its unprecedented high molecular specificity and its versatility, but it suffers from its low sensitivity, which limits the detection of molecules at very low concentrations. This drawback can be overcome by using special Raman signal amplification techniques such as resonance Raman spectroscopy, surface-enhanced Raman scattering (SERS), or nonlinear coherent Raman scattering phenomena such as CARS = coherent anti-Stokes Raman scattering or stimulated Raman scattering (SRS).
The application of Raman approaches in biomedical research and clinical diagnostics has grown rapidly over the past decade, entering a new era due to advances in instrumentation and, more importantly, increased interdisciplinary dialogue between spectroscopists and end users such as clinicians. Raman spectroscopy enables non-invasive morphochemical characterization of a wide range of different biological samples, from prokaryotic and eukaryotic cells, fungi, biofilms, to tissue sections and whole organs. In this way, Raman studies have provided insights and knowledge on, disease mechanisms for early diagnosis of diseases, for microbial diagnostics, for onsite detection of pathogens, for visualization of metabolic, defense, or chemical communication processes in cells and plant tissues, for on-site environmental and soil monitoring, in forensics, and in pharmaceutical process analysis.
In a recent paper, you discuss developments of deep learning algorithms for Raman spectroscopy as a response to the challenges associated with Raman spectra analysis (1). Please summarize these challenges and explain how these algorithms can assist in overcoming them.
A Raman spectrum of a biological cell monitors the sum of the individual molecular fingerprints of the molecular cell constituents which are water, proteins, RNA, phospholipids, DNA, polysaccharides, and various small molecules to name the most important ones. The sum of these individual molecular Raman fingerprints results in a typical cellular Raman fingerprint. In case of a microbe this cellular Raman spectroscopic fingerprint is characteristic of this microbe and shows specific changes due to, for example, metabolic reactions or treatment conditions.
Thus, different biochemical compositions of cells and tissues lead to spectral differences, which can be used for the differentiation of bacteria or for the differentiation of benign and cancerous tissue. However, these spectral differences are often very subtle, as bacterial cells show quite similar spectra, which are indistinguishable to the naked eye, since the biochemical composition of different bacterial species or genera does not necessarily differ in the amount of the molecular constituents, but often only by small variations within the proteome.
To identify these small spectral differences, to allow for a Raman spectroscopic differentiation of different bacterial species or to differentiate between healthy and cancerous tissue sophisticated photonic data science approaches are required. As mentioned above the utilization of Raman methods in biomedical research and clinical diagnostics has grown rapidly over the past decade, entering a new era due to advances in instrumentation and an interdisciplinary dialogue between spectroscopists and clinicians. However, this is only half the truth equally important is the fruitful interplay between artificial intelligence (AI) and Raman spectroscopy. There are great synergies between Raman spectroscopy and AI. First and foremost, is the automated interpretation of biomedical Raman data sets with powerful AI methods instead of the “naked eye.” So far, mostly classical machine learning approaches have been successfully applied for a qualitative and quantitative analysis of biomedical Raman datasets. But now as the number of Raman datasets and biomedical Raman studies are increasing, the application of deep learning approaches using neural networks is rapidly entering the field of Raman spectroscopy.
However, the application of classical machine or deep learning approaches to Raman spectroscopy represents various challenges. The biggest challenge is appropriately preparing the Raman spectra for model training and preprocessing the experimentally measured spectra so they can be used for building a “general” model via classical or machine learning approaches. This preprocessing involves various steps like spike correction, wavenumber and intensity calibration, baseline correction, and smoothing, to name a few. (See DOI: 10.1021/acs.analchem.5b04665). In this context we have developed a universally applicable Raman data analysis software called RAMANMETRIX (see: https://docs.ramanmetrix.eu/). This software allows for a one-click data analysis of Raman spectroscopic data in a robust and reliable way.
Coming back to the challenges arising when applying deep learning approaches in analyzing Raman spectra, still the most important challenge or issue is the fact that deep learning algorithms are highly data-demanding. Large independent Raman data sets are required since small sample sizes of Raman datasets might lead to low algorithm performance. This challenge can be overcome by generating large open-source Raman datasets via for example, international ring trial experiments (see https://dx.doi.org/10.1021/acs.analchem.0c02696). On the other hand, the more popular Raman spectroscopy becomes, the more Raman data are generated, also increasing the number of Raman datasets. Here, the generalization ability of the trained models must be guaranteed because a model that performs quite well on one dataset might produce disappointing results on another dataset due to overfitting. In this context, the preprocessing routines to prepare the measured spectra for model classification must be optimized in a way that generalization is ensured.
However, I am convinced that the unique possibilities AI, and in particular deep learning, approaches offer is the missing part to finally pave the way for Raman spectroscopy to be used as a routine clinical method. The success of Raman spectroscopy in biomedical diagnosis is inherently connected with the development of customized Raman data evaluation algorithms for translating Raman measurement data (spectral data sets and image data, for example) into qualitatively and quantitatively usable information for end users.
Can you explain the significance of developing tailored artificial intelligence-based spectral analysis routines in conjunction with clinically usable spectroscopic devices for real-time spectral analysis and selective identification and removal of tumors during surgery?
Reliable diagnosis of cancer or tumors after initial suspicion is nowadays a complex process involving a range of imaging modalities such as ultrasound, computed tomography, magnetic resonance imaging, endoscopy, and so on. All these techniques, which are used as a standard part of a preparation for a tumor resection can "only" visualize the pathological correlate of the tumor disease - the mass. They show the tumor location and its extent, but specific information on the tumor type cannot be retrieved. In an intraoperative imaging setting, precise diagnostics are not at hand. Magnifying glasses, endoscopy and microscopy are employed to visualize the tumor area with higher magnification, in addition to direct visual inspection by the surgeon.
The gold standard for confirming a definitive tumor diagnosis, and for pathological staging and grading, is histopathological diagnosis of the removed tumor tissue following surgery. Modern clinical pathology is a highly developed field of diagnostic medicine. Tissue biopsy diagnosis is very time-consuming and complex. In most cases, the pathologist works with formalin-fixed and paraffine-embedded (FFPE) tissue samples, which is why it is often not possible to obtain histopathological findings while the patient is still in the operating room. In principle, it is possible to evaluate unfixed snap frozen tissue sections while the patient is still undergoing surgery. However, since the quality of intraoperatively prepared sections is not comparable to that of FFPE tissue, the diagnostic possibilities in the operation room are limited per se, and the results of frozen section diagnostics may deviate from those of an examination of FFPE sections, which is why subsequent diagnostic confirmation by means of FFPE sections is still necessary.
In summary, "conventional" pathological diagnostics requires new methods and approaches to obtain additional diagnostically relevant information quickly and reliably, directly in the operation room, for targeted surgical interventions. Here, new intraoperative imaging technologies are particularly needed that are capable of precisely localizing the tumor and its boundary, to remove it as completely as possible. The targeted detection of malignant tissue during curative surgery is the most important prerequisite for complete tumor removal. The treatment of patients with a tumor disease is primarily performed surgically whereby tissue-saving surgery is often required, although a purely clinical assessment of the required safety distance is often difficult. However, tumor-free margins represent the decisive prognostic factor for the recurrence of a tumor. Thus, reliable tumor margin detection during surgery is key to an effective tumor treatment.
The combination of linear and nonlinear Raman approaches with tailored spectral image analysis routines allows for a label-free detection of the molecular composition as well as the morphology of complex biological samples, such as cells or tissues, with little or no sample preparation. Because pathological abnormalities are associated with changes in the biochemical composition and the structure of biomolecules, the Raman spectrum of a tissue provides a sensitive and specific fingerprint of its nature and condition. This makes Raman spectroscopy in combination with tailored AI-based spectral analysis routines a perfect tool for label-free histopathological examination of tissue. To take full advantage of Raman spectroscopic approaches a major step forward would be the implementation of Raman spectroscopic-guided femtosecond ablation in a “seek and treat manner.” In this way, it would be possible to have real-time monitoring of ablated features and enable ‘seek and treat’ applications.
How does this work differ from what has been previously done by yourself or others?
Various groups in the world are working on utilizing Raman spectroscopy for medical diagnosis. I leave it up to others to comment on how our work differs from what is done by others in terms of biomedical Raman spectroscopy. I only want to comment on how our work differs from what we have done previously. Our previous work mainly consisted in proof-of-concept studies with complex laboratory equipment. These studies showed the potential of Raman spectroscopy for microbial analysis or spectral histopathology. However, by these laboratory experiments it is difficult to convince medical doctors from the benefits of biophotonic approaches. I still remember 10-15 years ago when we showed our non-linear Raman lab to a medical doctor he was quite impressed by the complexity of the optical setup (dozens of mirrors, various delay stages, bulky lasers, and complex laser scanning microscopes) but said at the same time, “How do you plan to transfer this into routine clinical applications?,” meaning “How should this be applied in an operating room?” That is exactly what we worked on in the last years translating Raman approaches into clinically applicable systems which can be operated by clinicians and used outside specialized optics labs. Within the last year, we have realized various clinically usable spectroscopic devices (both microscopy and endoscopy based) for spectral histopathology and microbial analysis. In doing so we have introduced novel multimodal labelfree spectroscopic instrumentation like innovative Raman fiber probes, clinically usable non-linear multimodal microscopes or endospectroscopic probes and so forth, for precise surgical guidance and intraoperative histopathological examination of tissue (staging and grading) under in-vivo or near in-vivo conditions to initiate an individualized therapy plan tailored to the patient as quickly as possible. Furthermore, we recently introduced a multimodal nonlinear microendoscope, which also allows for the ablation of biological tissue with femtosecond lasers. Thus, we have achieved translating our spectroscopic approaches into clinically applicable systems providing clinicians and pathologists with new diagnostic and therapeutic tools to address currently unmet medical needs.
In summary, our research covers the entire range from fundamental research towards the development of new Raman spectroscopic instruments for clinical translational research. Bridging this gap may be regarded as a token of quality of our work.
In this context, I must mention that this work would not have been possible without having a network of fruitful collaborations. Within the last 10-15 years, we have built up a very strong biophotonics network in Jena with both academic and industrial cooperation partners. Here, I would particularly like to mention our clinical collaborators who always supported us from the very beginning. Developing a medical photonic approach without having clinicians on board from the very beginning is not purposeful.
Were there any challenges or limitations encountered in your work? Are there still challenges that remain to be overcome?
Besides the technological challenges one always faces in biophotonics which, however, can be mostly overcome via pursuing an interdisciplinary research strategy, such as having various scientific disciplines on board. The biggest challenge we are currently facing are regulatory issues. We have now reached a stage where we are ready for initiating clinical trials with our Raman instrumentation. Before the clinical trials can begin, however, regulatory approval must be obtained. In terms of logistical and financial resources, this stage usually exceeds by far what academic groups can master. Since new technology is usually considered very risky, it will not easily be taken up by industry or commercial investors.
There is a general issue regarding, translating biophotonic research results which poses major challenges, especially regarding the EU Medical Device Regulation (MDR) for translational research. Currently, the regulation makes it significantly more difficult or impossible to test biophotonic approaches on patients in the form of preclinical or clinical studies. Besides Raman, various other photonic technologies have proven their potential for certain diagnostic and therapeutic questions in proof-of-principle studies, but the actual performance has not yet been demonstrated under routine clinical conditions in the form of comparative studies on a large cohort of patients. Here, funding for such validation studies is urgently needed to generate a marketable product. It only makes economic sense for industry to support a biophotonic proof-of-concept approach if there is a regulation-compliant study on a large cohort of patients that clearly demonstrates the added value.
What such translational research could look like in concrete terms is demonstrated by the Leibniz Centre for Photonics in Infection Research (LPI) (see https://lpi-jena.de/en/), which was recently included in the national roadmap of research infrastructures by the German government. LPI will become a translation infrastructure that accompanies new solution approaches from the idea to validation and significantly shortens the development time required for this. As an open user platform, the center is available to national and international users alike. At LPI, compact devices for the rapid and unambiguous diagnosis of infections and new therapeutic approaches are researched and developed. The diagnostic and therapeutic approaches realized in the form of prototypes are directly submitted to multicenter clinical validation by means of flying study nurses during the value creation process. This concept is unprecedented and an ideal approach to efficiently tap the potential of photonic technologies for routine clinical processes much faster than before. The LPI will be located on the premises of the Jena University Hospital near the clinical facilities. A "first-in-patient" unit for the LPI will be located directly in the clinic. The LPI complements state-of-the-art technologies with new photonic methods that are not yet commercially available in this form. In the future, users from science and industry will have access to a broad spectrum of unique light-based and molecular biological methods in combination with all the necessary technologies to accelerate the translation of new methods for the diagnosis and treatment of infectious diseases. In 2021, the establishment of the technological infrastructure of the LPI was started: The German ministry of education and research (BMBF) is funding five projects designated as basic technologies with around 50 million euros. Researchers from the four supporting institutions are working in the inter-institutional projects to increase the technical maturity of the basic technologies (TRL - Technology Readiness Level) to prepare them for use in the LPI. LPI can thus serve as a blueprint for other medical problems, such as cancer or neurodegenerative diseases, to overcome the “valley of death” of clinical translation in these areas as well.
LPI is a result of the unique interdisciplinary biophotonic research network we have set up in the last years here in Jena. Overall, translation is a major challenge that we are currently tackling within the framework of the LPI. I am sure that the LPI will succeed in overcoming the so-called “valley of death” from idea to product.
Can this technique be applied to any other spectroscopy techniques? If so, which might find the technique to be especially useful?
Raman spectroscopy can be easily combined with various other spectroscopic approaches. Various research efforts exist in this direction to extend the potential of Raman spectroscopy by combing it with other spectroscopic approaches in a multimodal approach. Multicontrast spectroscopy has become a powerful tool of biomedical research. One can think of two end-member strategies for combination. On one end, modalities with similar acquisition speed and resolution can be combined— for example, nonlinear Raman imaging methods like CARS or SRS can be combined with multi-photon excited autofluorescence or higher harmonic microscopy (like SHG = second harmonic generation or THG = third harmonic generation), all of which are efficiently excited by the same laser source and can be detected in parallel. On the other end, modalities of different speed and tissue penetration can be synergistically combined, so that a fast but chemically less specific method provides an overview on the tissue volume, while a slower, molecule-specific second method is used to classify tissues detected by the faster modality in suspicious areas. One such approach would be to combine Raman with optical coherence tomography (OCT) or fluorescence lifetime imaging microscopy (FLIM).
Can you please summarize the feedback that you have received from others regarding this work?
The feedback is in general very positive, which can be best seen by the publications we have published in high impact journals which are very well cited, by the great number of invited or keynote lectures I received in the last years or the prices and honors for this work, like the Charles Mann award or an honorary doctoral degree from the state university New York in Albany which I received very recently.
In this context, I would like to mention once again that this is not a one-man show, but the result of a unique collaboration with outstanding colleagues. In my opinion, research is always teamwork. Here I would like to express my sincere thanks to the numerous cooperation partners as well as to my working group.
What are the next steps in this research?
As mentioned above the next steps are initiating clinical trials with the developed Raman instrumentation. I am not satisfied with having successfully demonstrated an idea in the lab and then publishing it. For many researchers that is the end, but not for me. I am always driven by the basic idea of further developing a successfully published idea and ultimately translating it. This is also the slogan of the Leibniz Institute for Photonic Technologies, which I am heading as scientific director "from ideas to instruments". Translation is a major challenge, which my colleagues and I are currently tackling.
What does your being named the recipient of the Mann Award mean to you professionally? Personally?
As mentioned above, I am driven by the motivation to get a research idea beyond publication, and getting it translated in a clinical device. It is also exactly what drove the award's namesake, Charles Mann. Professor Mann's research areas range from basic research, including data analysis (chemometrics and databases) to applied research in polymers, inorganics, and other areas. Therefore, it means a lot to me to receive this distinguished award and to continue, so to speak, the legacy of Professor Mann's philosophy and promote Raman spectroscopy as a powerful analytical technique from basic research to routine applications. This award perfectly reflects what I and my group stand for with their research. Personally, the award means to me a confirmation that the hard work with all its ups and downs is worth it. This award gives me strength to continue and shows me that I, or rather we, are on the right track. FACSS, or now SciX, is without question the conference I most enjoy attending and have been doing so regularly for more than 20 years. To receive this award at SciX among all the many dear colleagues and friends I meet there regularly means an extraordinary amount to me.
(1) Luo, R.; Popp, J.; Bocklitz, T. Deep Learning for Raman spectroscopy: A Review. Analytica 2022, 3 (3), 287-301; https://doi.org/10.3390/analytica3030020