Raman fingerprints as promising markers of cellular senescence and aging.

Due to our aging population, understanding of the underlying molecular mechanisms constantly gains more and more importance. Senescent cells, defined by being irreversibly growth arrested and associated with a specific gene expression and secretory pattern, accumulate with age and thus contribute to several age-related diseases. However, their specific detection, especially in vivo, is still a major challenge. Raman microspectroscopy is able to record biochemical fingerprints of cells and tissues, allowing a distinction between different cellular states, or between healthy and cancer tissue. Similarly, Raman microspectroscopy was already successfully used to distinguish senescent from non-senescent cells, as well as to investigate other molecular changes that occur at cell and tissue level during aging. This review is intended to give an overview about various applications of Raman microspectroscopy to study aging, especially in the context of detecting senescent cells.

Introduction

Demographic changes in the industrialized world lead to an increased occurrence of medical conditions for which biological age is the major risk factor (GBD 2015 DALYs and HALE Collaborators 2016; GBD 2015 Mortality and Causes of Death Collaborators 2016). In addition to arising complications and onset of multimorbidity due to decreased resilience, frailty severely limits the quality of life at advanced age. Thus, a better understanding of biological aging processes and identification of markers will promote the design of interventions, which target biological aging and thereby increase fitness, decrease frailty, and improve resilience at advanced age (Bellantuono 2018; Cardoso et al. 2018; Figueira et al. 2016).

One of these processes, cellular senescence, is defined not only as an irreversible growth arrest induced by either serial passaging, which causes the shortening of telomeres to a critical length (replicative senescence) (Bodnar et al. 1998; Hayflick and Moorhead 1961), or by exposure to stress (stress induced premature senescence = SIPS) (Toussaint et al. 2002), but also as a consequence of chemo- and radiation therapy (Demaria et al. 2017; Schosserer et al. 2017) or oncogene activation (Collado and Serrano 2006). Senescent cells accumulate in the body during normal aging and occur predominantly at sites of age-associated pathologies, which include atherosclerosis (Erusalimsky and Kurz 2005; Gorenne et al. 2006; Minamino 2002; Vasile et al. 2001), osteoporosis (Kassem and Marie 2011), neuroinflammation (Bitto et al. 2010), and liver cirrhosis (Wiemann et al. 2002). While considered being a beneficial tumor suppressor mechanism in the young (Campisi et al. 2011; Campisi 2005), cellular senescence is by now well accepted to contribute to in vivo aging (Baar et al. 2017; Baker et al. 2016, 2011; Xu et al. 2018) and even tumor progression in the elderly (Campisi et al. 2011; Campisi 2005). These deleterious effects are caused at least in part by the senescence-associated secretory phenotype (SASP) (Coppe et al. 2010), which was already shown to promote chronic inflammation and thereby fuel several aging-associated pathologies including atherosclerosis, kidney fibrosis, and cancer progression (Demaria et al. 2017; Schosserer et al. 2017). Thus, one of the major goals of current aging research is the development of compounds that specifically eliminate senescent cells (“senolytics”) or inhibit the SASP and thereby alleviate deleterious effects caused by senescent cells (Baar et al. 2017; Xu et al. 2018; Zhu et al. 2017, 2015).

However, although a prerequisite for screening and evaluation of senolytic compounds, the detection of senescent cells, especially in vivo, is still one of the challenges in the field. Currently, flattened cell morphology, activation of p16INK4a (Baker et al. 2016; Tchkonia et al. 2013) and p53 (Tchkonia et al. 2013), activity of SA-β-Galactosidase (Debacq-Chainiaux et al. 2009), staining with Sudan Black B (Georgakopoulou et al. 2013), presence of ɣH2AX foci at the telomeres (Fumagalli et al. 2014) and senescence-associated heterochromatin foci (Narita et al. 2003), High Mobility Group Box 1 (HMGB1) secretion (Davalos et al. 2013), and growth arrest as measured by BrdU-incorporation (Lämmermann et al. 2018) are considered to be senescence markers. The drawback is that none of them is specific for senescence and some of them can only be detected in vitro. Therefore, combinations of these markers have to be used. Raman microspectroscopy could thus offer a non-invasive and label-free method that allows to monitor the progression of senescence in real time in vitro and in vivo.

Raman microspectroscopy distinguishes cellular states in a label-free and non-invasive manner

Raman spectroscopy is based on the interaction between light that is focused on a sample and the chemical bonds within the material to be analyzed. Compared to elastic or Rayleigh scattering, inelastic or Raman scattering is a rare and comparatively weak phenomenon. Depending on the direction of the energy shift (Raman shift in cm−1), scattered electrons are either at lower (Stokes Raman) or higher (anti-Stokes Raman) energy levels (Raman 1928).

Modern Raman microspectrometers consist of a confocal microscope equipped with one or more lasers, an efficient longpass filter to remove highly abundant Rayleigh-scattered light, a spectrometer with different gratings, and a sensitive CCD line detector (Fig. 1). UV and blue lasers go along with high energy, which might damage biological samples, and induce significant levels of autofluorescence. Thus, green and red lasers (e.g., 532 nm or 785 nm) are most commonly used for the analysis of cells and tissues. Most current Raman microscpectrometers offer automated mapping applications, whereby the laser scans over the specimen and a spectrum is recorded at every single pixel to generate a multi-dimensional hyperspectral image. While acquiring spectra is relatively simple, data processing poses a major challenge and usually consists of background removal and normalization steps, followed by multivariate statistic approaches including principal component analysis (PCA), linear discriminant analysis (LDA), classical least square (CLS) fitting, multivariate curve resolution (MCR), among others (Butler et al. 2016; Notingher et al. 2005).

Fig. 1

Raman microspectroscopy of cells. Photons emitted from a laser light source are differentially scattered by chemical bonds of cellular constituents. Rayleigh and anti-Stokes scattered light is filtered, and remaining Raman Stokes scattered photons are recorded by the spectrometer. A typical Raman spectrum of mammalian cells is shown. Spectral regions explaining the most prominent differences between cellular states are depicted

Raman spectra from biological materials, typically recorded in the region of 400–2000 cm−1 (Fig. 1), provide chemical fingerprints detecting even subtle changes in the biochemical composition of cells (Beattie et al. 2013; Brauchle and Schenke-Layland 2013; Charwat et al. 2015; Rösch et al. 2006; Swain and Stevens 2007), tissues (Ashtikar et al. 2013; Bocklitz et al. 2013; Movasaghi et al. 2007), and whole organisms (Lau et al. 2012). The advantage of Raman microspectroscopy compared to traditional staining approaches lies in the fact that this technique can be used on living specimen without prior fixation and does not require any label that might interefere with normal physiology. Raman signatures of in vitro cultured cells were already successfully recorded and used for the characterization and identification of various specific cell types, as for example for endothelial (Szafraniec et al. 2018) and human lung (Surmacki et al. 2018) cell lines. Our lab was also able to distinguish different Chinese Hamster Ovary (CHO) host and production cell lines by Raman microspectroscopy (Prats Mateu et al. 2017). In chondrocytes (Pudlas et al. 2013), hematopoetic stem cells (Ilin et al. 2015), and hematopoetic progenitor cells (Choi et al. 2018), it has been shown that Raman microspectroscopy is capable of monitoring the dynamic process of cell differentiation. Also, different cellular states, such as apoptosis and necrosis, were successfully distinguished (Brauchle et al. 2014b), and cell progression through mitosis was followed by Raman microspectroscopy (Matthäus et al. 2006).

Biochemical deviations occuring in cancer have been extensively studied using Raman spectroscopy not only at cellular level in vitro (Brauchle et al. 2014a; Duraipandian et al. 2018; Lee et al. 2018; Managò et al. 2018; Terentis et al. 2013), but also in tissues ex vivo (Bocklitz et al. 2013; Santos et al. 2016). These promising results pave the way for clinical use of Raman spectroscopy for analysis of extracted specimen and identification of markers for tumor resections (Santos et al. 2017; Shipp et al. 2018). A fiber-optic Raman probe was already used during brain surgery and allowed differentiation between cancer and healthy tissue (Jermyn et al. 2015). Another putative clinical application of Raman spectroscopy is the detection of fragility fractures by using Spatial Offset Raman Spectroscopy (SORS) (Buckley et al. 2015).

Raman microspectroscopy enables distinction of senescent and non-senescencent cells

Only few studies were conducted to investigate cellular senescence using Raman microspectroscopy so far. Bai and coworkers acquired Raman signatures of mesenchymal stem cells obtained from human umbilical cord tissue during serial passaging (Bai et al. 2015). The authors found that the ratio of peaks at 1157 cm−1 vs. 1174 cm−1, both corresponding to vibrations of proteins, could serve as a marker for late population doubling levels (PDLs). Other notable, but not significant, differences between Raman spectra of late and early PDL cells were found within the amide II (1480–1575 cm−1) region.

Eberhardt and coworkers analyzed four different human dermal fibroblast cell strains using Raman spectroscopy as well as Fourier transform infrared spectroscopy (FTIR) (Eberhardt et al. 2017a). Comparing Raman signatures of early PDLs, middle PDLs, and senescent cells, peak intensities at 1580 cm−1 and 1658 cm−1 assigned to nucleic acids and proteins, respectively, were found to be decreased, while lipid associated peaks at 1732 cm−1, 2850 cm−1, and 2930 cm−1 were increased in senescent cells. Partial least squares-linear discriminant analysis (PLS-LDA) was able to distinguish these three groups. Analysis of the difference spectra obtained through PLS-LDA again revealed changes in the amide I region (1600–1800 cm−1), at high wavenumbers (> 2800 cm−1), as well as in the amide III region (1220–1300 cm−1), and below 1200 cm−1. Raman-based classification models set up for each cell strain separately revealed an overall sensitivity of 93% and specificity of 90%, although outcomes from the four cell strains differed. Senescence was confirmed by morphological changes, cell proliferation in different PDLs, as well as SA-β-galactosidase activity.

In another study, Eberhardt and coworkers expanded their Raman and FTIR-based detection of senescent dermal fibroblasts towards a 3D model of human skin (Eberhardt et al. 2017b). Fibroblast-derived matrices (FDM) were built by seeding fibroblasts in PDL 4 and PDL 20 for the young and senescent model, respectively. In 3D, Raman peaks between 600 and 900 cm−1 and a peak at 1260 cm−1 associated with the amide III region were decreased in senescent cells. The spectral region between 930 and 1230 cm−1 showed increased intensities in the spectra of senescent cells. Comparison of fibroblasts from passages 4, 7, and 20 in 2D culture showed alterations below 1250 cm−1 and in the amide I and II region. PLS-LDA for cells cultivated in 2D and 3D revealed differences in the amide I and II region, as well as at 788 cm−1, a peak that can be assigned to ring breathing modes in nucleic acids. A classification model trained with proliferating and senescent cells grown in 3D was then indeed able to predict these cellular states in 2D culture. However, vice versa, classification of 3D data was not successful when a 2D training set was used, underlining the fact that other differences became more obvious in the 3D environment. Senescence was confirmed by SA-β-galactosidase staining.

Oncogene-induced senescence was studied in MCF-7/NeuT cells (Mariani et al. 2010). Senescence was induced by doxycycline treatment, leading to oncogenic ErbB2 overexpression and consequently p21 induction (Trost et al. 2005). Raman spectra of nuclei from senescent cells showed a single peak at 1652 cm−1, whereas two peaks at 1652 cm−1 and 1666 cm−1 were found in control cells. These two peaks were assigned to cis and trans unsaturated fatty acid isomers, respectively. However, as the amide II band is also located in this region, interpretation of the signal beeing protein-derived seems also possible. Mariani and coworkers concluded that mainly cis isomers can be found in senescent cells, leading to instabilities in the nuclear membrane. Furthermore, peaks at 1313 cm−1 and 1339 cm−1 assigned to glycoproteins were found in control, but not in senescent cell spectra. Accordingly, mRNA levels of nuclear pore complex glycoprotein Nucleoporin 210 (NUP210) were significantly decreased in senescent cells. The assignment of the glycoprotein peak was based on a publication measuring Raman spectra from an isolated antifreeze glycoprotein (Tomimatsu et al. 1976). In case of cell-based Raman signatures, proteins and nucleic acids might also be worth considering for being the source of chemical interactions located in that wavenumber region.

As summarized in Table 1, in all the studies comparing Raman signatures between young and senescent cells (Bai et al. 2015; Eberhardt et al. 2017a, b; Mariani et al. 2010), peaks assigned to the amide II region were subject to substantial changes. The amide II band between 1480 and 1575 cm−1 refers to C–N stretching and N–H bending occuring in peptides (Movasaghi et al. 2007). The amide I band between 1600 and 1800 cm−1, related to C=O stretching and the amide III band from 1220 to 1300 cm−1 depicting C–N stretching and N–H bonding (Movasaghi et al. 2007) also constantly recured with the exception of the studies done by Bai and coworkers. Differences in the biochemical fingerprint between young and senescent cells could thus be explained by varying occurences of proteins. However, chemical interactions associated with glycoproteins, lipids, and nucleic acids also contribute to the variations located in the range of the three dominant amide bands.

Table 1

Most prominent peaks and spectral regions contributing to differences in spectra from senescent versus non-senescent cells. The corresponding studies, as well as peak assignments differing from Movasaghi et al. (2007), are indicated

Spectral region	Peak assignment (Movasaghi et al. 2007)	Bai et al. (2015)	Eberhardt et al. (2017a)	Eberhardt et al. (2017b)	Mariani et al. (2010)
600–900 cm−1	Ring breathing modes in nucleic acids (among others)			x
788 cm−1	Ring breathing modes (nucleic acids)			x
930–1230 cm−1	Proteins, lipids, glycogen, glucose, nucleic acids			x
1156/7 cm−1	C–C, C–N stretching (proteins)	x
1174 cm−1	C–H bending (tyrosine, phenylalanine)	x
1220–1300 cm−1	Amide III region: C–N stretching, N–H bonding		x	1260 cm−1
1313 cm−1	CH3CH2 twisting (lipids)				x (glycoproteins)
1339 cm−1	CH3CH2 twisting (lipids), C–C stretching (phenyl)				x (glycoproteins)
1480–1580 cm−1	Amide II region: C–N stretching, N–H bending, ring breathing modes (nucleic acids)	x	1580 cm−1 (nucleic acids)	x
1652 cm−1	C=C stretching (lipids)				x (cis unsaturated fatty acids)
1666 cm−1	C=C stretching (proteins)				x (trans unsaturated fatty acids)
1600–1800 cm−1	Amide I: C=O stretching, C=C stretching (proteins, lipids)		1658 cm−1 (proteins), 1732 cm−1 (lipids)	x
2850 cm−1	CH2 stretching (lipids, fatty acids)		x
2930 cm−1	CH2 stretching (lipids)		x

Raman spectroscopy is able to visualize molecular changes occurring in skin aging

Raman-based in vivo investigations have been performed to analyze age-related changes in human skin and its components. Especially the stratum corneum (SC), the outermost part of the epidermis, has been subject to these studies (Boireau-Adamezyk et al. 2014; Choe et al. 2018; Egawa and Tagami 2008). Differences between young and aged female subjects regarding water content in the SC of forearm skin were found using Raman signatures (Egawa and Tagami 2008). Also, changes in the barrier function of SC were observed, especially a decreased lipid/protein ratio, as well as an increased transepidermal water loss with age and an increased SC thickness, though the condition of the barrier function also strongly depended on the site of measurement (Boireau-Adamezyk et al. 2014). Contradictorily, another Raman-based study, including subjects from a smaller age range, showed that the lipid/protein ratio stayed constant with increasing age, while the expansion in SC thickness was confirmed (Choe et al. 2018). The dermis, the skin layer underneath the epidermis, was examined regarding the water content by using a prediction model, followed by Raman-based analysis, pointing out higher water content in the dermis of healthy aged and diabetic women compared to healthy young women (Téllez et al. 2015).

Special interest was given to photoaged skin which was investigated ex vivo (Gniadecka et al. 1998; González et al. 2012). Raman spectra recorded from chronologically aged skin as well as photoaged skin obtained by punch biopsies from a total of 20 individuals, showed a shift towards lower wavenumbers in the amide I band compared to young skin. In photoaged skin, the amide III region and C–H stretching bands higher than 2800 cm−1 were also shifted towards lower wavenumbers, possibly indicating an increase in protein folding in photoaged skin. However, in chronologically aged skin, only the peak at 1658 cm−1 in the amide I region was different from young individuals (Gniadecka et al. 1998). Raman spectroscopy was used to study intrinsic aging and photoaging in vivo in 15 subjects between 28 and 82 years of age, divided into three different groups (de Vasconcelos Nasser Caetano et al. 2017). The authors pointed out the proline-hydroxyproline region (intensities of peaks at 855 cm−1 and 938 cm−1) as suitable for the evaluation of intrinsic skin aging. Similarly, Villaret and coworkers showed that the 938/922 cm−1 peak ratio was decreased in spectra from aged photo-protected skin compared to aged exposed, young photo-protected, and young exposed skin obtained via punch biopsies from 14 female individuals (Villaret et al. 2018). Results from another study (Nguyen et al. 2013) show that the proline-hydroxyproline region, more specifically the 938/922 cm−1 peak ratio, turned out not to be able to distinguish between resected skin samples from the dermis of four females classified in two different age groups (40 years, 70 years). Nguyen and coworkers also found that the 1658/1668 cm−1 peak ratio, assigned to reflect interactions of water with collagen, was able to differentiate between the two age groups. However, considering the biological variability, the relatively small number of analyzed samples in these studies might not be sufficent to draw generalizable conclusions. Furthermore, the penetration depth of Raman probes for in vivo use is relatively low, allowing analysis of just the upper skin layers.

Interestingly, C–H stretching bands that were found to be shifted in photoaged skin (Gniadecka et al. 1998) were among the regions that also contributed to differences in the spectra of senescent cells in comparison to non-senescent cells (Eberhardt et al. 2017a). Furthermore, the amide I band was responsible for spectral differences at both cellular (Eberhardt et al. 2017a, b) and tissue level (Gniadecka et al. 1998; Nguyen et al. 2013).

The application of Raman spectroscopy to study aging in various tissues

Apart from studies in the skin, research in the field of ophthalmology already made use of Raman microspectroscopy for studying processes that occur during aging, when analyzing dried human Bruch’s membranes for the quantification of advanced glycation end products (AGEs) and advanced lipoxidation end products (ALEs) that accumulate with age (Beattie et al. 2013; Glenn et al. 2007). Resonance Raman spectroscopy, a specialized Raman technique, was used to investigate age-related effects on macular pigment optical density (MPOD) (Obana et al. 2014) as well as differences in carotenoid levels in healthy subjects compared to patients with age-related macular degeneration (Bernstein et al. 2002).

Other studies focused on Raman-based analysis of bone tissue (Ager et al. 2005; Akkus et al. 2003; Gamsjaeger et al. 2010; Milovanovic et al. 2018; Toledano et al. 2018), providing insights into compositional changes that occur during aging. Ager and coworkers used deep-ultraviolet Raman spectroscopy and found significant age-related differences in the shape and intensity of the amide I band from excised cortical bones in humans (Ager et al. 2005). As reported recently, AGEs might also contribute to the aging process of bones and show a specific Raman signal (Toledano et al. 2018). Raman spectroscopy has also been applied to investigate age-related structural changes of human teeth (Ager et al. 2006; Tramini et al. 2001). Similarly, Tramini and coworkers found that the chronological age of an individual could be predicted by analysis of the dentin’s Raman spectra (Tramini et al. 2001). Apart from the importance of understanding aging-related mechanisms, this approach might also be of interest for forensic investigations.

The same applies to another study showing successful classification into three age groups (< 1 year, 11–13 years, 43–68 years), based on Raman spectra of human peripheral blood (Doty and Lednev 2018). Apart from blood, other biofluids might provide suitable substrates for Raman-based investigation of aging phenomena as well. Erythrocyte aging has recently been studied with the help of Raman spectroscopy, revealing changes in lipids and membrane proteins (Dinarelli et al. 2018).

Multivariate statistics were able to classify spectra from human oral buccal mucosa into young and physiologically aged individuals, without inferering with the classification of Raman spectra based on tobacco-related changes (Sahu et al. 2012). Alterations in lipid composition due to aging were also examined in murine perivascular adipose tissue using Raman microspectroscopy and a Raman fiber optic probe (Czamara et al. 2018). Aging-related oxidative damage in mouse oocytes leading to developmental abnormalities was studied via Raman microspectroscopy aiming towards the use of this method in assisted reproductive treatment in humans (Bogliolo et al. 2013).

When comparing these data to the Raman-based investigation of skin aging and cellular senescence, it becomes evident that peaks responsible for the differences in the analyzed spectra are frequently located in the three prominent amide regions, depicting mostly proteins, lipids, and nucleic acids.

Summary and perspectives

As shown here, the field of aging research has just begun to make use of the label-free, non-invasive technology of Raman microspectroscopy. For studying Raman fingerprints of senescent cells, caution must be given to the precise definition and characterization of the senescent state, which was neglected by some of the previous studies complicating their interpretation. Furthermore, as senescence is by now considered to occur progressively and to show heterogeneity between indiviual cells and tissues (Hernandez-Segura et al. 2017), it would be interesting to compare early to late senescent cells during development of the characteristic SASP. Coupling Raman microspectroscopy to microfluidic systems, as reviewed by Li and coworkers (Li et al. 2012), will pave the way for investigation of heterogeneity within a large cell population. It also remains to be seen if Raman bands explaining the differences between senescent and non-senescent cells vary between different cell types and tissues, and if these fingerprints might match to in vivo data.

The challenges of Raman microspectroscopy lie in the fact that the peak assignment to chemical interactions and further to biochemical structures is challenging and has to be conducted with great care. Moreover, the current instrumentation and data analysis require higher speed and simplification, since only thereby Raman microspectroscopy will become widely applicable to biologists and clinicians not specialized in biophotonics. These insights will not only help to efficiently identify senescent cells in a label-free and non-destructive manner with large potential for in vivo and ex vivo applications including compound screenings, but also to gain insights into intracellular biochemical changes that occur during aging.