Raman, Infrared and Brillouin Spectroscopies of Biofluids for Medical Diagnostics and for Detection of Biomarkers

Abstract

This review surveys Infrared, Raman/SERS and Brillouin spectroscopies for medical diagnostics and detection of biomarkers in biofluids, that include urine, blood, saliva and other biofluids. These optical sensing techniques are non-contact, noninvasive and relatively rapid, accurate, label-free and affordable. However, those techniques still have to overcome some challenges to be widely adopted in routine clinical diagnostics. This review summarizes and provides insights on recent advancements in research within the field of vibrational spectroscopy for medical diagnostics and its use in detection of many health conditions such as kidney injury, cancers, cardiovascular and infectious diseases. The six comprehensive tables in the review and four tables in supplementary information summarize a few dozen experimental papers in terms of such analytical parameters as limit of detection, range, diagnostic sensitivity and specificity, and other figures of merits. Critical comparison between SERS and FTIR methods of analysis reveals that on average the reported sensitivity for biomarkers in biofluids for SERS vs FTIR is about 10³ to 10⁵ times higher, since LOD SERS are lower than LOD FTIR by about this factor. High sensitivity gives SERS an edge in detection of many biomarkers present in biofluids at low concentration (nM and sub nM), which can be particularly advantageous for example in early diagnostics of cancer or viral infections.

Highlights

• Raman, Infrared spectroscopies use low volume of biofluidic samples, little sample preparation, fast time of analysis and relatively inexpensive instrumentation.

• Applications of SERS may be a bit more complicated than applications of FTIR (e.g., limited shelf life for nanoparticles and substrates, etc.), but this can be generously compensated by much higher (by several order of magnitude) sensitivity in comparison to FTIR.

• High sensitivity makes SERS a noninvasive analytical method of choice for detection, quantification and diagnostics of many health conditions, metabolites, and drugs, particularly in diagnostics of cancer, including diagnostics of its early stages.

• FTIR, particularly ATR-FTIR can be a method of choice for efficient sensing of many biomarkers, present in urine, blood and other biofluids at sufficiently high concentrations (mM and even a few µM)

• Brillouin scattering spectroscopy detecting visco-elastic properties of probed liquid medium, may also find application in clinical analysis of some biofluids, such as cerebrospinal fluid and urine.

Keywords:

• SERS
• infrared spectroscopy
• Brillouin spectroscopy
• biomarker
• biofluid
• urine analysis
• blood
• serum
• cancer biomarkers
• LOD
• sensitivity
• specificity

Introduction

Blood components like plasma and serum, and urine tests provide plethora of information about patients’ health. Health conditions detected by these tests range from cardiovascular and kidney diseases to various types of cancers.^[1–3] Moreover, those biofluids are easily accessible and minimally invasive for patients. Other biofluids, like tears, bile and cerebrospinal fluid are also of interest to medical diagnostics purposes.^[4–6]

Infrared and Raman spectroscopies along with their modifications such as SERS, FTIR represent very widely used methods of vibrational spectroscopies. Infrared spectroscopy, Raman/SERS are promising tools to be adopted in routine clinical tests in the future because it is fast, cheap, label-free. Big advantage of these techniques that they could provide information about biofluid constituents, by not only focusing on individual analytes and their corresponding peaks in the spectrum but rather on the spectrum as a whole. In comparison, the fluorescence spectroscopy techniques like fluorescence resonance energy transfer (FRET) or surface-enhanced fluorescence (SEF), have also been used in detection of analytes and clinical studies.^[7,8] However, FTIR and SERS unlike fluorescence are label free, direct techniques, that do not require analyte modification, both of them unlike fluorescence have a great capacity for multiplexing due to relatively narrow Raman peaks (typical full width at half maximum (FWHM), about 5–15 cm⁻¹ in typical Raman range 400 to 1800 cm⁻¹) and relatively narrow IR peaks.^[9,10] Also, they can be combined with other methods to perform multilayered simultaneous analysis.^[11,12] In addition, these spectroscopies can be used for the treatment of various diseases by using methods like photoinduced cleavage of bonds or photothermal cell elimination.^[13–15] Studies performed with the means of vibrational spectroscopies where analytical parameters are determined such as the limit of detection, linear range etc., are focused on analysis of individual peaks in the spectrum corresponding to a specific molecule, a biomarker of a disease. Many molecules of various types such as peptides, DNAs, proteins, glycopeptides, low molar mass metabolites etc. are known to be indicative of specific health conditions and are referred to as biomarkers.^[16] On the other hand, majority of studies mentioned in this review related to medical diagnostics of health conditions utilize statistical analysis to find differences between vibrational spectra of biofluids, for example serum or urine, obtained from a control (healthy) group and an unhealthy group.^[17,18] Since biofluids contain hundreds of constituents and their quantity vary widely between individuals, even for healthy individuals, their spectra vary. Rather than focusing on individual peaks, they utilize multivariate analysis to take into consideration hundreds of variables (peaks).

A number of reviews are available on the similar topic of the usage of vibrational spectroscopies for the analysis of biofluids: general overview of vibrational spectroscopies,^[2] protein biomarkers in urine,^[3] cancer diagnosis by vibrational spectroscopies,^[19] cancer diagnosis by SERS,^[20] overview of SERS,^[21] application to point of care medicine,^[22] Raman spectroscopy for plasma and serum,^[23] overview of various methods including vibrational spectroscopies,^[24] FTIR spectroscopy for medical diagnostics,^[25] ATR-FTIR for biofluids.^[26] While there is some overlap between mentioned reviews and the current review, this review puts more emphasis on structuring of experimental papers into easy-to-follow format of tables along with their critical analysis and comprehensive comparison between applications of two major methods of vibrational spectroscopy: IR and Raman (SERS). This analysis is done by combination of two prospective: one describing literature about SERS/Raman and FTIR analytical methods with LOD as a major figure of Merit (FoM) and another one describes reports of clinical trials, where infected (unhealthy) patients are typically distinguished from control group of healthy patients with SERS/Raman or FTIR measurements and where sensitivity specificity and accuracy are often reported. Also, this review discusses Brillouin spectroscopy applied to non-contact viscoelastic diagnostics of bio-relevant fluids.

This review summarizes and provides insights on recent advancements in research within the field of vibrational spectroscopy for medical diagnostics and its use in detection of many health conditions such as kidney injury, cancers, cardiovascular and other infectious diseases. Firstly, this review explains theory behind vibrational spectroscopies, typical experimental procedures, common experimental methods, and mathematical analysis of spectra in the next paragraph. Then, in corresponding paragraphs blood, urine and other biofluids are discussed, and respective experimental studies summarized in tables. Important part of this review are tables where experimental papers are summarized with their analytical parameters, such as limit of detection, range, diagnostic sensitivity and specificity, and other figures of merits.

Methods of Infrared and Raman spectroscopies

Theory behind Infrared and Raman spectroscopies

Vibrational spectroscopies relate to Infrared, Raman and Brillouin spectroscopies. The Brillouin spectroscopy and its application to biofluid analysis discussed in another chapter in this review. Infrared and Raman are discussed in the current chapter. Analytical signal in both techniques is due to vibrations of chemical bonds. Therefore, both techniques provide molecular fingerprint of materials. The Infrared spectrum arises from absorption of Infrared light at resonant frequencies when the absorbed radiation matches the vibrational frequency, while the Raman spectrum arises from inelastic scattering of the laser light interactions with molecular vibrations, phonons or other excitations in the system, resulting in corresponding emission of photons. Figure 1(A) contains Jablonski energy diagram that shows the energy transitions involved during infrared absorption, Rayleigh, Raman Stokes, anti-Stokes and Resonance Raman scattering.^[27]

Figure 1. (A) Jablonski energy diagram showing the transitions involved during infrared absorption, Rayleigh, Raman Stokes, anti-Stokes and Resonance Raman scattering. The vibrational states (Vn) of a molecule in the ground electronic state (S0) can be probed either by directly measuring the absolute frequency (IR absorption) or the relative frequency or Raman shift (Stokes and anti-Stokes) of the allowed transitions. Resonance Raman also involves the vibrational states (V’n) of the excited electronic state (S1). Hν0 = incident laser energy, hνvib = vibrational energy, Δν = Raman shift and νvib = vibrational frequencies. (B) Raman and IR spectra of liquid ethanol. Reprinted from open access publication under CC BY license.^[27,28]

Raman and Infrared spectroscopy are complementary techniques and usually both are required to completely measure the vibrational modes of a molecule.^[29] Although some vibrations may be active in both Raman and IR, these two forms of spectroscopy arise from different processes and different selection rules. The same molecular vibrations can appear in both Raman and Infrared spectroscopy. The example showcasing the difference of Raman and IR spectra of the same compound can be seen in Figure 1(B).^[28] However, the fundamental difference between Infrared and Raman Spectroscopy is in their working principles: Raman reacts to change in polarizability, while Infrared is sensitive to dipole moment. Infrared spectroscopy can detect compounds with heteronuclear functional group with high dipole moments (-OH, -COOH, -NH, -CO etc.). While, Raman spectroscopy can detect with homonuclear, low-dipole (-C = C-,-C = N-,-S = C-, -N = N- etc.) features of the compound. Therefore, Raman spectroscopy and Infrared spectroscopy often considered to be complementary techniques rather than analogous.

Spontaneous Raman signal is inherently weak. To overcome this limitation, surface-enhanced Raman spectroscopy (SERS) was developed. SERS enhances Raman scattering by molecules adsorbed on rough metal surfaces or by nanostructures. Simple flat metallic surface can already serve as an “amplifier” of Raman signals for molecules deposited on it, albeit achieving a much lower level of amplification than that reached normally in metallic nano-structures.^[30] The enhancement factor can be as high as 10¹⁰–10¹⁴ that enables to detect a single molecule.^[31,32]

The enhancement is achieved by electromagnetic effect or chemical mechanism. The electromagnetic effect arises from optical excitation of surface plasmon resonances in metal nanoparticles or nanostructures , which leads to a significant increase in the electromagnetic field strength at the particle surface.^[33] The enhancement factors due to plasmon excitation on the metallic surface, typically gold or silver, are commonly reported in range of 10⁴ to 10^8.[34] In the chemical enhancement mechanism, molecules adsorbed at certain surface sites (such as atomic clusters, terraces, and steps) are believed to couple electronically with the surface, leading to an enhancement effect like resonance Raman scattering. However, the magnitude of chemical enhancement is in the 10–100 range.^[9] The SERS enhancement is maximized in so called hot spots, the areas of high electromagnetic field enhancement, which are located near the sharp tips of nanoparticles or nanostructures and/or in the nanometer scale gaps between plasmonic (usually made of gold or silver) nanoparticles/nanostructures.^[35] The SERS enhancement can be stronger for nanoparticle dimers and trimers, if compared to SERS enhancement for Raman reporter molecules on single gold or silver nanoparticles, when those nanoparticles are drop casted on the surface of various substrates such as gold, silver, aluminum films or silicon.^[36,37]

Typical experimental procedures

In this chapter of the review, typical experimental procedures are described. These procedures are summarized after analysis of dozens of papers that are given in the tables further down in the review. However, only some of them will be mentioned here where appropriate, to show examples of how certain experimental procedures are utilized. For Infrared spectroscopy experimental procedures are given for ATR-FTIR only since it is the dominant method. Sample collection, storage and spectra analysis are the same for ATR-FTIR and Raman/SERS. Procedures for ATR-FTIR, Raman/SERS is mostly the same for serum, urine or other biofluids. Schematic diagram for SERS procedure is given in Figure 2.^[17]

Figure 2. (A) Schematic diagram of the procedure used to prepare the urine-Ag nanoparticle mixture and SERS measurements. (B) TE micrograph of the Ag colloid surface. (C) Enhancement effect of the Ag colloid on Raman spectroscopy in urine. The red spectrum is the background Raman signal of the Ag colloid; the blue spectrum is the spectrum of the urine sample from a normal volunteer obtained by mixing urine and the Ag colloid; the black spectrum is the regular Raman spectrum of the same urine sample without the Ag colloid. Copyright of Elsevier. Reprinted with permission from Elsevier from article of Hu et al.^[17]

Procedures for sample collection and storage are the same for ATR-FTIR and Raman/SERS:

1. Taking serum/urine samples. All serum/urine samples are taken from a medical facility by researchers. Firstly, a written consent from the patient is taken. For urine collection, depending on health conditions of interest, first morning urine sample or 24-hour urine are taken by a medical facility. As for blood samples, serum is usually extracted from blood for further analysis. It is prepared by blood centrifugation for 10 minutes at 1000–1200 g.

2. Sample storage. Samples are frozen immediately and stored until further analysis at −80 °C, as in,^[17,38] at −35 °C.^[39] For SERS and ATR-FTIR as low as 2–5 µL of sample is required, for Raman around 2 mL. Samples are thawed at room temperature before analysis.

ATR-FTIR

Next steps for ATR-FTIR experiments after sample collection and storage are:

1. Spectra acquisition. The liquid sample is directly placed on the ATR crystal that is usually made of zinc selenide. The spectrum is obtained in the region of 600–4000 cm⁻¹. The resolution for FTIR spectroscopy ranges from 2 to 8 cm⁻¹. Several dozen scans are made.

2. Spectra analysis. At this stage, researchers could investigate individual peaks in spectra, otherwise, if multiple peaks, sometimes thousands, are indicative then statistical methods are used. Usually, the latter happens when patients with health conditions must be distinguished from healthy group, as in.^[11] PCA (principal component analysis) algorithm is used for such research. PCA is used to reduce numbers of variables so that they still contain the most of original information. Then PCA is followed by another algorithm—LDA (linear discriminant analysis). By LDA, researchers find a linear combination of features that characterizes or separates two or more classes of objects, in our case—healthy and unhealthy groups.

Raman/SERS

Next steps for Raman/SERS experiments after sample collection and storage are:

Preparation of nanoparticles for SERS

For SERS, nanoparticles (NPs) are synthesized or purchased. Usually, Au NPs and Ag NPs are utilized, and could be prepared by these procedures^[40] and,^[38] respectively. If NPs are not synthesized by common methods, then they are characterized by instrumental methods such as TEM to determine size of NPs and UV-VIS to determine concentration of NPs.^[41] Alternatively, suspensions of commercially available AuNPs are used for preparation of SERS substrates or/and in SERS assays of biomarkers.^[42,43] However, sometimes Raman measurements in solutions do not require NPs.^[39]

Preparation for Raman/SERS spectra acquisition

For SERS, the NPs solution is mixed with urine, usually, in 1:1 ratio.^[44] Then a droplet of the obtained mixture is put on gold or aluminum slides and dried in air. For liquid Raman, the sample is put into glass vials as bulk liquids. However, if urine contain precipitates, it could be vortexed to resuspend and dissolve these prior to spectra acquisition.^[39]

Raman/SERS spectra acquisition

The most frequently used parameters for taking Raman/SERS spectra for biosensing and sensing of biomarkers are: 785 nm excitation laser, magnification of ×10 or ×20, range 400–1800 cm⁻¹, for example as in.^[17,45,46]

Fluorescence background correction and spectra analysis

Polynomial fitting is used to remove fluorescence background.^[44] At this stage, researchers could investigate individual peaks in spectra, otherwise, if multiple peaks, sometimes thousands, are indicative then statistical methods are used. Usually, the latter happens when patients with health conditions have to be distinguished from healthy group, as in.^[47] PCA (principal component analysis) algorithm is used for such research. PCA is used to reduce numbers of variables so that they still contain the most of original information. Then PCA is followed by another algorithm—LDA (linear discriminant analysis). By LDA, researchers find a linear combination of features that characterizes or separates two or more classes of objects, in our case—healthy and unhealthy groups, for example as in.^[48–50]

Experimental parameters

Different types of analytes can be detected by Raman/SERS and Infrared spectroscopy, ranging from low-molar mass substances such as urea to proteins and even to bacteria and cells.^[51] They can provide a fingerprint information on biochemical molecules. Moreover, body fluids themselves can be distinguished by Raman/SERS. For instance, Vyas et al. publication, while updating the work reported by Muro et al., utilized SERS for the forensic identification of body fluid stains commonly found at crime scenes that include blood, saliva, semen, sweat, vaginal fluid and urine.^[52,53] Big advantage for Raman/SERS and Infrared spectroscopy is that they are label free, and usually require no pre-processing or modification of the samples. Even though, currently Raman/SERS and Infrared spectroscopy are hardly frequently used in clinical analysis, they have a strong potential for commercial clinical applications, due to their relative competitiveness with established methods on such parameters as cost, time to results, accuracy, sensitivity/LOD (particularly for SERS) etc. Further chapters of this review summarize SERS/Raman/IR experimental studies with the corresponding analytical and experimental parameters for blood, urine and other biofluids. With wide range of analytes, Raman/SERS can be used in routine clinical analysis, drug and forensic analysis, bacteria analysis. Moreover, the SERS analysis has the potential to become a common point of care test for blood plasma and blood, which means taking the patient’s medical testing at the time and place of patient care.^[54] Raman/SERS can be successfully performed in vivo, which means analysis performed on living humans, as opposed to a tissue extract or bodily fluids.^[55]

In the most cases for Raman and Infrared spectroscopy, the procedure is quite straightforward—a liquid sample is directly irradiated by laser or lamp and a Raman or IR spectrum is acquired. Sometimes first step in SERS is to mix a sample and nanoparticles. Then the obtained mixture is left to dry on metal surface, such as gold or aluminum.^[42,56] Lastly, the dried sample is irradiated by laser and a SERS spectrum is acquired. Substantial effort of researchers in SERS is focused on obtaining the right types of nanoparticles that show the best analytical parameters. Nanoparticles can vary in materials, shapes and sizes. Most frequently, gold and silver nanoparticles are used, however, other materials can offer competitive Raman enhancement as well.^[9] For example Markina et al. showed that copper nanoparticles (Cu NPs) as compared to Au and Ag NPs offered competitively high Raman enhancement efficiency for detection of cephalosporin antibiotics in spiked human urine.^[57] Multicomponent nanoparticles are used as well, for example, Gao et al. developed a glucose sensor for SERS based on Ag nanodendrites/Cu mesh substrate.^[58] SERS detection can be categorized into direct detection (label-free) and indirect detection based on how signal is generated from interaction of analytes and nanoparticles.^[59] In the indirect detection, nanoparticles are coated with ligands that selectively binds with the necessary analytes, on the other hand, in the label-free detection, nanoparticles do not specifically bind to molecules. The key issue of indirect detection is to develop multifunctional SERS nanoparticles with high stability, sensitivity, and specificity. SERS Nanoparticles can have various shapes—spheres, hollow shells and cages, rods, nanostars, nanoflowers and etc.^[60] Shapes and sizes of nanoparticles are characterized by transmission electron microscopy, for example, as in the work by Lu et al. for detection of glucose in urine.^[41]

Important spectra acquisition parameters for Raman/SERS are laser excitation wavelength and laser power, acquisition (integration) time, spectral resolution, scans per sample, magnification lens, and analyte concentration. For contemporary Raman/SERS equipment common wavelength numbers are 532, 633, 785, 1064 nm. Alajtal et al. compared the effect of laser wavelength on the Raman spectra of phenanthrene, chrysene, and tetracene.^[61] They showed that fluorescence background drops in the order—514, 633, 785, 1064 nm, which means that higher excitation wavelength equates to less fluorescence background. However, there are two disadvantages associated with moving to higher excitation wavelength; the Raman scattered intensity is inversely proportional to the fourth power of the laser wavelength, so the intensity of the Raman signal is weaker than with visible excitation; hence, practically, the acquisition time must be increased. In addition, the efficiency of the typical silicon-based CCD detector decreases significantly in the NIR (near-infrared) region. Parameters such as acquisition (integration) time and laser power should be adjusted to avoid photodissociation that results from overly high laser power and/or long exposure time.^[62] Total Raman/SERS spectra acquisition time is mostly dependent on acquisition (integration) time, spectral resolution, scans per sample. Laser polarization is another factor for consideration. Compared with the non-polarized laser excitation and linear-polarized laser excitation, the use of circularly polarized laser excitation greatly improved the diagnostic sensitivity and specificity.^[46] Lastly, concentration of analyte, should be adjusted carefully, because the signal enhancement will be inefficient when adding highly concentrated analyte solutions.^[63]

According to Xiong et al. despite its high specificity and sensitivity, the SERS technique has not been established as a routine analytic method most likely due to the low reproducibility of the SERS signal.^[64] Aggregation of the nanostructures in solution and deposition on substrate can greatly contribute to the resonance condition resulting in low reproducibility if left uncontrolled. Grys et al. experimentally showed that it is possible to nearly eliminate variance (the relative standard deviation below <1% for different batches and days as a measurement of reproducibility) if parameters such as interparticle spacing and aggregation time are controlled.^[65] A minireview by Bell et al. contains information about good analytical practice of SERS that address the problem with reproducibility.^[66]

Attenuated Total Reflectance-Fourier Transform Infrared Spectroscopy (ATR-FTIR) is a dominant method of performing Infrared spectroscopy. ATR-FTIR has been shown to be advantageous for its sample thickness independent measurement, ability to probe highly IR absorbing materials, especially aqueous solutions due to strong adsorption at 2800–3700 cm⁻¹, without significant sample preparations and the demonstrably improved spatial resolution as compared to other FTIR imaging approaches.^[67] ATR-FTIR spectroscopy involves directing the infrared light at an interface between an infrared transparent material with a high refractive index called the internal reflection element (IRE, e.g., a prism made of ZnSe, diamond, silicon or germanium, depending on the required wavelength range) and a sample on the surface of the IRE.^[68] The angle of incidence of the IR beam is greater than the critical angle and as such, total internal reflection occurs. Typically, signal strength is relatively low in FTIR. Like Raman, the signal can be enhanced by coating or making contact of the crystal with a metal film.^[69] The Molecules adsorbed on this metal layer show enhancement in infrared absorption by factor 10–1000 times and sometimes even higher enhancement up to 4 × 10⁴ on NIR plasmonic nanoparticles, for instance nanocrescents.^[70] This effect is called Surface Enhanced Infrared Absorption (SEIRAS) and results from the electromagnetic interactions of the incident photon field with the metal and the molecules,^[69] illustrated in Figure 3.

Figure 3. (A) Sketch of the experimental setup for Surface Enhanced Infrared Absorption Spectroscopy (SEIRAS) with the prism for attenuated total reflection (right) and the spectro-electrochemical cell with LED for illumination (left). (B) Optical setup for Infrared Reflection Absorption Spectroscopy (IRRAS). (C) Conventional attenuated total reflection unit with multiple internal reflections. Copyright of Elsevier. Reprinted with permission of Elsevier from Ataka et al.^[69]

Costs of the analysis by Raman and Infrared spectroscopies are determined to some extent by the cost of equipment, and for the SERS, it also depends on the cost of nanoparticles and substrate. Analysis by Raman and ATR-FTIR do not require many consumables since it is a label free method, and usually samples require minimum amount of preparation. Modern Raman and ATR-FTIR equipment are available in online stores, and our search indicate that equipment usually priced from 5000 to 30000 USD. Nanoparticles for SERS are commercially produced, or they could be synthesized from available chemicals such as solution of HAuCl₄, for example as in.^[41] Moreover, substrates are also commercially available, or they could be as simple as aluminum foil.^[71] Paper-based SERS platform can further reduce costs. Typically, these “paper fluidics” are cellulose based, allowing for the flow or imprinting of nanoparticles embedded within its matrix.^[72] Overall, relatively low cost of consumables per each sample analysis (particularly for FTIR and Raman) and ambient experimental conditions (no need for vacuum or high pressure) would position the Raman/SERS/ATR-FTIR methods of analysis among relatively inexpensive analytical methods.

Spectra analysis

Fluorescence background correction

Many biological samples, including urine, exhibit strong fluorescence that represent challenge to Raman/SERS methods. Raman spectra requires fluorescence background correction to demonstrate accurate results. Fluorescence background correction could be done by experimental and computational approaches. Experimental techniques could be generally grouped into time-domain, frequency domain, wavelength-domain according to Wei et al.^[73] All three of them requires certain changes in instrumentation, particularly of how substances are excited by the laser. An example, of fluorescence background correction is given in Figure 4. One drawback of these methods is the relatively complex instrumentation, the long acquisition times, and alterations in the sample that could make the analysis of biological samples difficult.^[75] As a recent example of fluorescence background removal specifically for urine, Dutta et al. by photobleaching urine samples improved signal-to-noise ratio by 67% and signal-to-background ratio by 47%.^[76] On the other hand, fluorescence background could be corrected by computational methods, that are often based on simple or modified polynomial fitting. Polynomial fitting can be enhanced by peak removal, methods to account for signal noise effects, iterative models, smoothing algorithms.^[77–79] Lastly, the obtained spectrum after fluorescence correction can be further enhanced by smoothing algorithm to improve signal to noise ratio. Radzol et al. utilized Savitzky-Golay smoothing filter on SERS spectra to improve signal to noise ratio.^{[80, p. 1]} They were able to preserve 98% of maximum intensity while 33.49% of unwanted features were removed.

Figure 4. (A) Block diagram of the time-gated Raman spectrometer. (B)Curve 1 (blue): uncompensated Raman spectrum measured with a time gate window of 600 ps; curve 2 (red): calculated fluorescence background based on the measured fluorescence counts within a window of 600 ps located 3 ns after the laser shot; curve 3 (green): fluorescence-compensated Raman spectrum, i.e., the differences between curves 1 and 2. Reprinted from open access publication under CC license.^[74]

Multivariate analysis

Biological samples are complex in nature consisting of various inorganic and organic molecules, such as peptides, proteins, lipids, carbohydrates etc. Because of that, analysis of their Raman spectra often requires to consider many variables, in other words, many peaks, not a single peak. Thus, analysis of a single peak often is not enough, as a result, to utilize the complete information of the complex spectra and to handle the large data set, multivariate analysis is needed. Multivariate data (MVA) analysis refers to data analytical methods that deal with more than one variable at a time. As an example, consider a research group that wants to detect protein in urine by SERS to determine proteinuria in patients. Patients and control group will contain uncorrelated to protein, and different amount of creatinine, urea. Creatinine and urea, along with protein will influence resulting SERS spectra. This shows that other substances (variables) need to be considered for the correct interpretation of spectra. Gautam et al. summarized multidimensional data processing approaches for Raman and infrared spectroscopy.^[81] Usually MVA analysis, if applied to urine or serum samples, is used to discriminate between patients with health conditions and healthy group. In Tables 2, 4 and 5 such experiments are summarized.

For MVA analysis fluorescence background correction is the first step for analysis of Raman spectra. It follows, by spectra normalization methods and lastly by MVA computations.^[125] The same MVA methods are applied to ATR-FTIR. Raman/SERS/ATR-FTIR spectra could have different intensities either within the same sample or within a group of similar samples due to the nature of samples or experimental conditions. Normalization process deals with differences in intensity levels—a simple and useful approach is the normalization to area. In this method, the intensity at each frequency in the spectrum is divided by the square root of the sum of the squares of all intensities, so that the total area of the spectrum equals to 1.

Finally, MVA analysis usually consist of principal component analysis and linear discriminant analysis. Principal component analysis (PCA) is a technique for reducing the dimensionality of such datasets, increasing interpretability but at the same time minimizing information loss.^[126] It does so by creating new uncorrelated variables that successively maximize variance. As an exploratory study, principal component analysis (PCA) is used to visualize differences and similarities between spectra, for example as in Figure 5. Then PCA is followed by another algorithm—LDA (linear discriminant analysis). By LDA, researchers find a linear combination of features that characterizes or separates two or more classes of objects.^[127] For example, LDA can be applied to the intensities of Raman spectra to construct a statistical model that separates healthy and unhealthy groups. Then diagnostic sensitivity and specificity of the method is calculated. In addition to traditional approaches, machine learning methods applied to Raman spectra can provide feasible alternatives for the recognition and quantification.^[128]

Figure 5. Discriminant analysis of principal components (DAPC) of peritoneal dialysis (PD) patient urine, spent dialysate, and urine from healthy individuals. Principal components are new variables that are constructed as linear combinations of the initial variables. DAPC results for models made with 50 principal components. (A) 362 urine specimens obtained from patients receiving PD therapy for end stage kidney disease (ESKD) and 395 spent dialysate specimens. (B) 362 urine specimens obtained from patients receiving PD therapy for ESKD and 235 urine specimens from healthy individuals. Reprinted from open access publication under CC BY license.^[82]

Medical diagnostics tests

Overall, experimental papers related to human urine analysis by Raman/SERS/Infrared spectroscopy can be divided into two groups. The first group develops new methods for detection of individual substances and provides information about analytical parameters, such as LOD, range, recovery etc. Usually, these papers firstly test their methods in the buffer matrix, and then apply to real samples, urine and blood serum in our case. The second group of papers is mainly focused on medical diagnostics. These papers aim to distinguish patients with health conditions from healthy individuals based on characteristics of spectra of urine, serum and other biofluids as whole. In this case, diagnostic sensitivity and diagnostic specificity between subject group and control (healthy) are important measurements to determine efficacy of the given method. Diagnostic sensitivity is the percentage of individuals who have a given health condition who are identified by the method as positive for the condition. Specificity is the percentage of individuals who do not have a given condition who are identified by the method as negative for the condition. Diagnostic sensitivity and specificity of a new clinical method should be compared with the most widely clinically adopted method for a given health condition.^[129] For example, in a general population-based study, the accuracy of urine dipstick for diagnosis of proteinuria was relatively low.^[130] When albumin/creatinine ratio ≥30 mg/g was set as the reference standard for proteinuria, the sensitivity and specificity were 37.1% and 97.3% (cutoff: trace) and 23.3% and 98.9% (cutoff: 1+). According to explanations by Akobeng, a relatively low sensitivity of 37.1% means that the method will be falsely positive for individuals who do not have that condition.^[131] A highly sensitive test is, therefore, most helpful to the clinician when the test result is negative. However, a test with a high specificity is useful for “ruling in” a disease if a person tests positive.^[132] Akobeng states that “the sensitivity and specificity of a test have limited clinical usefulness as they cannot be used to estimate the probability of disease in an individual patient”. Instead, predictive positive and negative predictive values should be used, which in itself depends not only on the method but also the prevalence of the condition for which the test is used, and cutoff values, as in the work.^[87]

Diagnostic sensitivity and specificity do not only depend on the experimental method used but also on the detection of a particular biomarker, since several biomarkers may indicate the similar health condition. For example, Verring et al. compared diagnostic sensitivity/specificity for breast cancer biomarkers.^[133] The following results were found: CA 549 (sen.: 40%/spec.: 90%), MSA (sen.: 22%/spec.: 96%), and CA 15-3 (sen.: 33%/spec.: 86%). Budman et al. in their review compared various biomarkers for the detection of bladder cancer where sensitivity of the bladder cancer test based on nuclear matrix proteins as biomarkers ranged from 34.6% to 100%, and specificity from 60.0% to 100%.^[134] There are similar works devoted to the detection of biomarkers for a particular disease, for example, Jakobsen et al. summarized detection of prostate cancer biomarkers;^[135] Necula et al. for gastric cancer;^[136] Li et al. for the breast cancer.^[137]

Urine analysis

Detection of individual components

Urine matrix is complex, it consists of various inorganic and organic compounds, from low-molar mass molecules to polymers; urine could contain cells, such as blood cells, or bacteria, which changes the composition of urine in time rapidly.^[138] It is widely accepted that analysis of urea taken throughout a day is the best representation of any illnesses in the human body if any, the so-called—the urine 24-hour volume test, that measures the amount of urine the human body produces in a day. Most abundant organic compounds in urine are urea, creatinine, uric acid.^[139] The urine of a healthy individual contains up to 150 mg of protein in total measured throughout a day.^[140] Among proteins, Tamm–Horsfall protein (also known as uromodulin) is the most abundant protein in urine (50%), followed by albumin (20%) and immunoglobulin (5%).^[141] Albumin excretion of 30 to 300 mg a day, which is called microalbuminuria, is an early and sensitive marker of diabetic nephropathy,^[142] cardiovascular and renal disease.^[143] Though one of the main proteins in urine is albumin, there are thousands of other types of proteins.^[144] Many low abundant proteins in urine are biomarkers of various diseases including cancers that are summarized in our other review.^[3] Other organic compounds in urine, such as peptides, RNAs, DNAs, low molar mass metabolites are also biomarkers for various types of cancers.^[145] In conclusion, besides renal diseases, urine can also be a useful source of information related to cancers and non-renal diseases.

Clinical urine tests (urinalysis) involves physical examination (odor, color, concentration), microscopic analysis and chemical analysis.^[146] Chemical analysis of urine usually measures blood, leukocyte esterase, nitrite, protein, glucose, ketones, bilirubin, urobilinogen, ascorbic acid. A commercial reagent strip or dipstick are used for those analyses. They consist of reagent-impregnated test pads that are fixed to an inert plastic strip. After the strip has been appropriately wetted in a urine sample, chemical reactions cause the reaction pads to change color. Modern automated urine analyzers are based on similar principles and are widely used in clinical routine tests.^[147] Many models contain urine strip readers, a type of reflectance photometer that can process several hundred strips per hour. Some analyzers can perform up to 240 tests an hour. In research setting, various types of immunoassays are common.^[3] High-performance liquid chromatography coupled with mass-spectroscopy is powerful tool for protein and non-protein biomarker detection in urine.^[145]

Table 1 summarizes some of the experimental papers on detection of various analytes that includes, but not limited to, low and high abundant substances, drugs, bacteria, enzymes in the human urine matrix by SERS spectroscopy from 2017 to 2020 to give a reader overview of what can be studied in the human urine by SERS. There are much more articles on human urine analysis by SERS that can be found online. Our literature search in scopus.com by the keywords “human urine,” “infrared” within date range of publications from 2017 to 2020 lead to some articles loosely connected to our objective, for example by Sharaha et al.,^[165] Takamura et al.^[166] We have found predominantly SERS in our literature research on the topic of Raman/SERS/IR spectroscopy for analysis of the human urine hence Table 1 consists only of them. The example of findings from such SERS human urine analysis can be observed from Figure 6, where we can see the schematic of experimental section, TEM image of probe, and results of SERS measurement.^[153]

Figure 6. (A) Illustration of the SERS-based methods for the detection of albumin. (B) TEM image of SERS tag. (C) SERS spectra (from 950 to 1200 cm⁻¹) of albumin with different concentrations. (D) Linear relationship between SERS intensity and albumin concentration. Copyright of Elsevier. Adapted with permission of Elsevier from Huang et al.^[153]

Table 1. Summary table of experimental papers on detection of various analytes in the human urine matrix by SERS spectroscopy from 2017 to 2020.

Method	Analyte	Analytical parameters	Reference
SERS with AuNPs after separation by extraction. Portable Raman equipment.	Creatinine	LOD 1.28 × 10^−5 M Experiment time 2 min	Zhu et al.^[^148^]
SERS nanocomposites of cucurbit[7]uril (CB7) and Au Nps in synthetic urine	Creatinine	LOD 1.10 × 10^−7 M Reproducibility <5%	Chio et al.^[^149^]
SERS, graphene oxide (GO)-decorated silver-coated gold nanobones	Glucose	LOD 2.61 × 10^−9 M Linear range from 10 × 10^−9 M to 1.05 × 10^−7 M	Zhu et al.^[^150^]
SERS, Raman (the buffer matrix)	Glucose	LOD 8.00 × 10^−8 M Range from 1.0 × 10^−6 M to 6.1 × 10^−3 M	Yuan et al.^[^151^]
SERS	Glucose	LOD 1.0 × 10^−4 M Linear range from 1.0 × 10^−4 M to 6.1 × 10^−3 M, R^2 = 0.922	Lu et al.^[^41^]
SERS coupled with a miniaturized array gas membrane separation	Albumine (inference from the average sulfur content in it)	LOD 3.0 × 10^−8 M Linear range from 1.5 × 10^−7 M to 7.5 × 10^−7 M, R^2 = 0.9915 Recovery 83.0%–105%	Chen et al.^[^152^]
Polydopamine bi-functionalized SERS immunoassay	Albumine	LOD 3 × 10^−9 M Linear range from 1.5 × 10^−7 M to 4.5 × 10^−6 M	Huang et al.^[^153^]
SERS with AuNPs on blu-ray DVD	Albumine Creatinine Urea	LOD 1.5 × 10^−9 M LOD 1.768 × 10^−6 M LOD 1.0 × 10^−5 M Raman signal stable over 45 days EF 6.96 × 10^7 (Malachite green standard)	Chamuah et al.^[^56^]
SERS AgNPs/ZnO/Fe3O4	Uric acid	LOD 3.6 × 10^−7 M Linear range from 5 × 10^−7 M to 4 × 10^−6 M EF 1.6 × 10^6 (in the buffer matrix)	Alula et al.^[^154^]
SERS	Uric Acid	Range from 1 × 10^−4 M to 7 × 10^−4 M RSD 8.5%	Westley et al.^[^155^]
Solid-phase extraction (SPE) with Aluminum oxide loaded with silver nanoparticles followed with SERS	Methotrexate (anticancer drug)	LOD 9.2 × 10^−6 M Range from 4.4 × 10^−5 M to 6.6 × 10^−4 M RSD 11%–19% Recovery 97%–104% Experiment time 15 min	Markina et al.^[^156^]
SERS nitrosonaphthol reaction-assisted	5-hydroxyindole-3-acetic acid (marker for carcinoid tumors)	LOD 6.3 × 10^−9 M RSD <5% Recovery 89.6%–106.3% Experiment time 12 min	Zhang et al.^[^157^]
SERS based lateral flow immunosensor in humans and animals	Aflatoxin M1	LOD 5.44 × 10^−12 M Recovery 93.8%–111.3% Experiment time 20 min	Wang et al.^[^158^]
SERS (artificial urine)	Ketamine	LOD 7.5 ppm Linear range 7.5–150 ppm (R^2 = 0.977) Recovery 95.7%–104.9%	Han et al.^[^159^]
SERS with silver nanocube coupling with a nanoporous silver film	Dopamine (neurotransamitter)	LOD 4 × 10^−14 M Linear range from 5 × 10^−13 M to 5 × 10^−4 M EF 5.5 × 10^10	Lu et al.^[^160^]
SERS, gold nanorods	Escherichia coli (human uropathogen)	10 CFU/mL Experiment time 3 hours	Andrei et al.^[^161^]
SERS and colorimetric, lateral flow immunoassay strips	Escherichia coli (human uropathogen)	45 CFU/mL Linear range 10^3 and 10^6 CFU/mL	Gumustas et al.^[^162^]
SERS, gold nanorods	Hyaluronidase (enzyme related to tumor)	LOD 1.7 U/mL Range 5–70 U/mL	Si et al.^[^163^]
SERS, silver nanocubes	Hemoglobin (hematuria)	LOD 5.35 × 10^−13 M Linear range from 5 × 10^−9 M to 5 × 10^−8 M Recovery 96.6%–101.2%	Swaminathan et al.^[^164^]

SERS was successful to determine various analytes: low and high abundant substances, drugs, bacteria, enzymes. Geometric mean for LOD 3 × 10⁻⁸ M, LOD range 4 × 10⁻¹⁴ M for dopamine to 1 × 10⁻⁵ M for urea. Geometric mean for EF 1.83 × 10⁸, with range from 1.6 × 10⁶ to 5.5 × 10¹⁰ for EF.

Medical diagnostics tests of urine

Table 2 summarizes experimental papers of medical diagnostic studies on human urine (from 2017 to 2021). We tried the table to be as thorough as possible for this kind of studies. In order, to keep this paragraph focused, a reader could find more information about typical experimental procedures, spectra analysis, and diagnostic sensitivity and specificity above in this review. There are significantly less paper on this topic that published before 2017. However, it should be noted, that Raman spectroscopy was used not only to determine diseases but also to determine metabolomic states in healthy individuals. For example, Bernal-Reyes et al. compared urine of volleyball and soccer players before and after exercise by Raman spectroscopy.^[167] Their analysis showed peaks associated with the presence of urea, uric acid, creatinine, and ketone compounds with different intensity before and after exercise. A similar study but for swimmers was performed by Moreira et al.^[168] Ding et al. collected spectra of athletes with temperature-dependent near-infrared spectroscopy (NIRS) technology to assess the effect of nutritional supplementation.^[169]

Table 2. Summary table of medical diagnostic studies (from 2017 to 2021) on human urine performed by SERS/Raman spectroscopy arranged in chronological order.

Health condition	Method and spectra processing algorithm	Patient count: Total/Sick/Healthy	Diagnostic sensitivity/specificity/accuracy	Reference
Bladder cancer	SERS, PCA-LDA	248/161/87	100%/98.85%/- high-grade tumors 97.53%/90.80%/- low-grade tumors	Hu et al.^[^17^]
Prostate cancer	SERS, GA-PCA-LDA	75/12/63	86.0%/87.1%/86.6%	Ma et al.^[^38^]
Colorectal cancer	SERS, PCA-DFA	116/63/53	95.8%/94.3%/- normal 80.9%/95.0%/- stages I-II 84.3%/94.2%/- stages III-IV	Lin et al.^[^40^]
Gastric cancer, Breast cancer	SERS, PCA-LDA	141/93/48	90.0%/93.8%/- gastric cancer 96.0%/93.8%/- breast cancer 62.0%/81.4%/- gastric versus breast cancer	Lin et al.^[^44^]
Bladder cancer, Prostate cancer	SERS, PCA-LDA	158/101/57	(given in graphs in the article)	Cui et al.^[^47^]
Bladder cancer	Raman, PCA	56/49/7	82.4%/79.5%/80.4%	Huttanus et al.^[^39^]
End-stage kidney disease	Raman, PCA-DAPC	144/96/48	97%/94%/96%	Senger et al.^[^82^]
Chronic kidney disease	SERS, PCA-LDA	187/100/87	78.0%/86.0%/-	Zong et al.^[^83^]
Gastric cancer Breast cancer	SERS, PCA-LDA	41/50 + 43/48	84.0%/95.8%/- gastric cancer 76.7%/87.5%/- breast cancer 82.0%/90.7%/- gastric cancer vs breast cancer	Lin et al.^[^84^]
Breast cancer	SERS, PCA-LDA	75/53/22	81%/95%/88%	Moisoiu et al.^[^85^]
Prostate cancer	SERS	120/90/30	87%/90%/-	Koo et al.^[^86^]
Coronary heart disease	SERS, PCA-LDA	127/87/20	90%/78.9%/-	Yang et al.^[^48^]
Microalbuminuria	SERS, PCA-LDA	27	Accuracies between high and low concentration with cutoff values: 9%–3 μg mL^−1 93%–6 μg mL^−1 89%–10 μg mL^−1	Stefancu et al.^[^87^]
Endometrial cancer Ovarian cancer	ATR-FTIR	30/20/10	95%/100%/95% endometrial cancer 100%/96.3%/100% ovarian cancer	Paraskeaivaidi et al.^[^88^]
Acute kidney injury	NIRS	38/15/23	79%/82%/-	Chock et al.^[^89^]
Urinary tract infection	SERS, PCA	108/108/0	Accuracy ∼95%	Tien et al.^[^90^]
Nasopharyngeal cancer Esophageal cancer	SERS, PLS-DA		95.2%/97.2%/- nasopharyngeal cancer 90.9%/98.2%/- esophageal cancer 98.1%/95.7%/- normal group	Feng et al.^[^91^]
DNA/RNA biomarker detection (no specific health conditions)	SERS	43	93.0%/95.3%/94.2%	Wang et al.^[^92^]
Oral cancer	Raman, PCA-LDA	197/117/80	95.6%/94.9%/- oral premalignant 96.4%/92.1%/- oral malignant	Brinhda et al.^[^93^]
Acute kidney injury (ATN) Delayed graft function (DGF)	SERS, PCA-LDA	30	Sensitivity of >90% in differentiating the ATN class from the DGF and control classes.	Chi et al.^[^50^]
Chronic kidney disease	Raman, Mahalanobis distance	72	Classification: 89% for normal subjects 86% for diabetic patients with hypertension 79% for patients with chronic renal failure doing dialysis	De Souza Vieira et al.^[^94^]

Diagnostic sensitivity is % of individuals identified by the method as positive for the condition among individuals who really have a given health condition. Specificity is the percentage of individuals who do not have a given condition and who are identified by the method as negative for the condition.

Below are some observations regarding Table 2 about the content of the listed research papers: Number of Patients: average—106.6 patients, median—116, range 27 to 248.

• Raman acquisition conditions: Though it is not mentioned in the table directly, the Raman studies performed on the human urine utilize SERS with 785 nm laser, magnification ranging from ×10 to ×20. Majority of papers performs PCA-LCA analysis on Raman spectra to distinguish between patients with health condition and healthy.

• The prevalent method was SERS, as 15 out of 21 utilized it. Liquid Raman was used in 4 works, and Infrared spectroscopy in 2 works. Once again, we tried the table to be as thorough as possible for collection for this kind of studies.

• Diagnostic sensitivity and specificity between the subject group and the sick group obtained by Raman/SERS/Infrared show comparable and promising value for all 3 methods. For SERS average sensitivity, specificity, accuracy (SSA) are 89%, 93%, 91%, respectively; and median values of the same parameters are 90%, 94%, 91%, respectively. For Raman, average SSA values are 93%, 90%, 88%, respectively; and median values are 96%, 93%, 88% respectively. For IR average SSA values 91%, 93%, 98%, respectively; and median values are 95%, 96%, 98%, respectively. Both diagnostic sensitivity and specificity have median values over 90% simultaneously in the same method for all 3 types of vibrational spectroscopies. For comparison, some other methods used in clinical practice might have high sensitivity and low specificity, and vice versa. For example, the dipsticks for proteinuria show low sensitivity and high specificity of 37.1%/97.3%, respectively,^[130] contrary immunoassays for prostate cancer show high sensitivity and low specificity of 95%/18%, respectively.^[170] Overall the average of 3 mean sensitivities and 3 mean specificities for all reported vibrational methods (SERS, Raman, IR) is 94%, which shows a strong potential.

• Diagnostic sensitivity and specificity between two subject groups were measured in 2 works by SERS.^[44,84] Average sensitivity and specificity are 72%, 86%, respectively.

Blood analysis

Detection of individual components

Blood has four main components: plasma, red blood cells, white blood cells, and platelets. The liquid component of blood is called plasma. Plasma, which constitutes 55% of blood fluid, is mostly water (92% by volume), and contains proteins, glucose, mineral ions, hormones, carbon dioxide (plasma being the main medium for excretory product transportation), and blood cells themselves. Albumin is the main protein in plasma, and it functions to regulate the colloidal osmotic pressure of blood. The blood cells are mainly red blood cells (also called RBCs or erythrocytes), white blood cells (also called WBCs or leukocytes) and platelets (also called thrombocytes). In a laboratory blood plasma can be collected by putting a blood sample in a centrifuge. Because blood cells are heavier than the liquid matrix, they are packed in the bottom of the tube by the centrifugal force. The light-yellow colored liquid on the top is the plasma. Another important component of blood for medical analysis is serum. Serum is the fluid and solute component of blood which does not play a role in clotting. To obtain serum, a blood sample is allowed to clot (coagulation). The sample is then centrifuged to remove the clot and blood cells, and the resulting liquid supernatant is serum. Together, blood serum and plasma are some of the largest sources of biomarkers, whether for diagnostics or therapeutics.

Our literature search related to blood analysis by Infrared spectroscopy led to a single paper. Jessen et al. simultaneously determined glucose, triglycerides, urea, cholesterol, albumin and total protein in human plasma by FTIR.^[171] LOD values were estimated to be 5.5 × 10⁻⁴ M for glucose; 3.4 × 10⁻⁴ M for triglycerides; 5 × 10⁻⁴ M for urea; 2.5 × 10⁻²mmol/L for cholesterol; 5 g/L (or 7.5 × 10⁻⁵ M) for albumin, and 5 g/L for total protein. Our calculations lead to geometric mean of 7.1 × 10⁻⁴ M. A single work related to determination of analytes in human blood or its components by IR can be explained by low signal intensities, and high wavenumber region being effectively unusable due to water absorption.^[172] On the other hand, there are abundance of work related to SERS determination. The example of SERS study of blood can be seen in Figure 7 where we can observe the schematic for the preparation of Raman probe, their respective TEM images, and results of Raman measurement.^[173] Table 3 summarizes articles related to SERS detection of low abundant analytes in human blood and its components.

Figure 7. Design scheme of (A) ARANPs synthesis, and (B) SERS-based strategy for quantitative detection of exosomal miRNA extracted from human blood. (C-D) TEM images of Au@R6G@AgNPs and ARANPs, respectively. (E) Performances of SERS detection of miRNA-21 in buffer solution. Inset: dependence of SERS intensity enhancement as functions of different concentrations of miRNA-21. (F) SERS signal enhancement upon miRNA-141, TM, SM, miRNA-21 addition. Copyright of Elsevier. Adapted with permission of Elsevier from Ma et al.^[173]

Table 3. SERS detection of low abundant analytes in human blood and its components.

Matrix	Analyte	Analytical parameters	Raman parameters	Reference
Blood serum	Hepatitis C viral rna	LOD 2.55 log IU/mL LOQ 7.71 logIU/m	785 nm laser 50 mW laser power ×40 objective	Nasir et al.^[^45^]
Blood plasma	Pseudoephedrine (drug)	LOD 10^−12 M RSD 8.4% EF ∼10^8	Mesopores with diameter 26.2 nm 785 nm laser, laser power of 70 mW	Guselnikova et al.^[^174^]
Whole blood	Bilirubin (BR)	LOD 2.5 × 10^−8 M Recovery values over 80% RSD lower than 15% EF 1.26 × 10^8	Ag NRs/M-BN 785 nm laser, laser power was 5.0 mW, 20× magnification resolution less than 4.5 cm^−1	Geng et al.^[^175^]
Blood serum	C-reactive protein (CRP)	LOD 10^−14 M (1.14 × 10^−12 g/mL) Linear range 10^−14–10^−8 M Recovery above 3% RSD lower than 10%	reporter-labeled Au nano-bridged nanogaps particles, Au NNPs 633 nm laser, 50×	Hu et al.^[^17^]
Blood plasma	TNF-α (tumor necrosis factor)	LOQ 10^−13 M R^2 0.994 RSD 3.8% for different locations 5.4%; for three different substrates; 9.47% between days. Reproducibility 0.17 pg/mL	Gold-coated silicon nanopillar substrate 785 nm laser, laser power 5 mW	Gholami et al.^[^176^]
Blood plasma	Carcinoembryonic antigen (CEA)	LOD 0.033 pg/mL LOQ 0.0001 ng/mL RSD 0.49%, RSD 0.68%	4-mercaptobenzoic acid labeled MoS2 NFs@Au NPs (tag) ;638 nm laser	Medetalibeyoglu et al.^[^177^]
Blood serum and plasma	Exosomal miRNA	LOD 5 × 10^−15 M EF 2 × 10^7	AuNPs 13 nm diameter, Au@R6G@ AgNPs 25 nm, 632.8 nm laser, ×50	Ma et al.^[^173^]
Blood serum	Free bilirubin	LOD 10^−12 M Range 10^−12–10^−7 M EF 10^9	gold nanoparticles (AuNPs) decorated MoS2 nanoflowers 488 nm 200 μW, 633 nm 100 μW; ×50	Singha et al.^[^178^]
Blood serum	luteinizing hormone (LH)	LOD 0.036 IU/L LOQ 0.1108 IU/L Recovery 90% RSD lower than 4%	785 nm, ×10 objective, 30 μm laser spot area, 150 mW laser power	Gjergjizi et al.^[^179^]
Whole blood	Ricin B Chain	LOD 1 μg/mL EF ∼10^6	AgFON substrate 633 nm	Campos et al.^[^180^]
Blood plasma	Hepatitis B virus antigen (HBsAg)	LOD 0.01 IU/mL R^2 0.997 RSD less than 10%	Fuchsin-labeled immuno-Au nano- flowers; 785 nm, ×10, beam D = 5 μm, laser power 5 mW	Kamińska et al.^[^181^]

Geometric mean for LOD 8.1 × 10⁻¹⁰ M, range from 0.03 × 10⁻¹⁴ M for CRP to 2.5 × 10⁻⁸ M for bilirubin. Geometric mean for EF 4.8 × 10⁷, with EF range from 10⁶ to 10⁹.

Medical diagnostics tests of serum and plasma

Tables 4 and 5 summarize medical diagnostic studies on human blood and its components serum and plasma performed by Infrared spectroscopy and Raman/SERS arranged in chronological order. Below each table discussions are given, and average values of sensitivity/specificity/accuracy are calculated.

Table 4. Summary table of medical diagnostic studies on human serum and plasma performed by Infrared spectroscopy arranged in chronological order.

Health condition	Method and spectra processing algorithm	Patient count: Total/Sick/Healthy	Diagnostic sensitivity/specificity/accuracy (%)	Reference
SARS-CoV-2 infection	ATR-FTIR, GA-LDA	38/20/18	95/89/-	Barauna et al.^[^95^]
Type 2 diabetes	ATR-FTIR, Savitzky-Golay, XG Boost and PCA	113/51/62	95.24/96/95.65	Guang et al.^[^96^]
Breast cancer	ATR-FTIR, PCA and PCR	146/66/80	92.3/87.1/-	Sitnikova et al.^[^97^]
Abnormal thyroid function	ATR-FTIR, MLP and LSTM and CNN	382/199/183	93.1/97.2/95.1	Yue et al.^[^98^]
Parkinson’s disease	Transmission FT-IR	83/61/22	93.44/77.27/91.94	Wang et al.^[^99^]
Opioid use disorder	ATR-FTIR (diamond), PCA-LDA-SVM	34/20/14	92.85/100/95.7	Guleken et al.^[^100^]
HIV infection	ATR-FTIR, GA-LDA	120/40/80	83/92/-	Silva et al.^[^101^]
Brain cancer	ATR-FTIR, SVM	724/104/620	93.2/92.8/-	Butler et al.^[^102^]
Dengue infection	ATR-FTIR, PCA-LDA-PLSR	77/57/20	89/95/-	Naseer et al.^[^103^]
Hepatitis B and C	ATR-FTIR, PLS-DA	120/80/40	87.5/81.6/-	Roy et al.^[^104^]
Dengue, chikungunya, zika viral infections	ATR-FTIR, SPA-LDA and PCA-LDA and GA-LDA	50/40/10	100/100/100 (starting from 10^3 PFU/Ml)	Santos et al.^[^105^]
Alzheimer’s disease	ATR-FTIR, Savitzky-Golay, PCA-LDA	82/56/26 (WBC)	75–82/86–90/83	Mordechai et al.^[^106^]
		73/50/23 (plasma)	64–70/79–87/89
Breast cancer	ATR-FTIR, PCA-LDA	86/43/43	83/84/74	Elmi et al.^[^107^]
Gastric cancer	Transmission FT-IR (barium fluoride), CDA	129/74 + 41/14	95.0/70/84.2	Liu et al.^[^108^]
Bipolar and schizophrenia	ATR-FTIR (diamond), PLS-DA and PCA	90/60/30	84.6–100/100/-	Ogruc Ildiz et al.^[^109^]
Brain cancer	ATR-FTIR (diamond), RF and 2 D correlation	433/differentiated	91.5/92.8/-	Smith et al.^[^110^]
Brain tumor	ATR-FTIR, LIBSVM	433/311/122	91.5/83/-	Hands et al.^[^111^]
Colon cancer	ECD and ROA and Raman and ATR-FTIR, LDA	55/28/27	93/81/100	Tatarkovic et al.^[^112^]
Basal cell carcinoma	ATR-FTIR, filtering-QDA and GA-QDA	75/40/35	85.1/75/100	Khanmohammadi et al.^[^113^]
Hepatocellular carcinoma	HT-FTIR, SVM	79/39/40	82.05/82.5/82–86	Zhang et al.^[^114^]
Bladder cancer	Transmission FTIR, RF-LDA	135/89/46	93.1/46.18/-	Ollesch et al.^[^115^]
Ovarian cancer	ATR-FTIR, LDC and QDC and eClass, k-NN, SVM	90/60/30	96.7/81.7/-	Gaijar et al.^[^116^]

Table 5. Summary table of medical diagnostic studies on human serum and plasma performed by Raman/SERS spectroscopy arranged in chronological order.

Health condition	Method and spectra processing algorithm	Patient count: total/sick/healthy	Diagnostic sensitivity/specificity/accuracy	Reference
Breast cancer	SERS, PCA-LDA	29/17/12	90%/98.4%/-	Nargis et al.^[^117^]
Malaria	SERS	25/15(5- P. falciparum infection; 10 - P. vivax infection)/10	-/100%/-	Wang et al.^[^118^]
Chronic lymphocytic leukemia (CLL); Diffuse large B-cell lymphoma (DLBCL)	SERS, OPLS-DA	102/72 (33 -DLBCL; 39- CLL)/30	92.86%/100%/- (CLL) 80%/92.31%/- (DLBCL)	Bai et al.^[^119^]
Breast cancer	SERS, PCA-FDA	26/18/8	87%/89%/-	Nargis et al.^[^18^]
Ovarian cancer	Raman SERS	72/27- ovarian cancer; 28—benign gynecological conditions; 17—early ovarian cancer	Ovarian cancer Raman 94%/96%/- SERS 87%/89%/- Early ovarian cancer Raman 93%/97%/- SERS 80%/94%/-	Paraskevaidi et al.^[^120^]
Chagas disease	SERS, PCA-LDA	110/65/45	-/-/96.3%	Perez et al.^[^121^]
Liver disease	SERS, OPLS-DA	736/333—hepatology; 99 - esophageal cancer/304	-/-/81.76%	Shao et al.^[^122^]
Prostate cancer (PCa) Benign prostatic hyperplasia (BPH)	SERS, PCA-LDA	40/20—PCa; 20 - BPH	60%/76.5%/68%	Shao et al.^[^123^]
Nasopharyngeal cancer	SERS, PCA-LDA	160/100(25 -T1 stage; 75: T2-T4 stages)/60	T1: healthy 84%/83.3%/85.5% T2-T4: healthy 92%/95%/93.3% T1: T2T4 64%/62.4%/63%	Lin et al.^[^49^]
Gastric cancer	SERS, PCA	64/31/33	100%/100%/-	Lin et al.^[^124^]
Gastric cancer	SERS, PCA-LDA	65/32/33	100%/97%/-	Feng et al.^[^46^]

Here are some observations/conclusions regarding Table 4 about the content of the listed research papers: Number of patients: average—143.5, median—84.5.

• Diagnostic sensitivity and specificity between the subject group and the sick group obtained by Infrared spectroscopy show promising value of those methods. From 22 papers, average sensitivity, specificity, and accuracy are 91.5%, 86.7% and 95.1%, respectively. Median values for the same parameters are 93.0%, 89.0%, 95.1%, respectively.

• Matrix. Out of 22, 12 papers used serum. Ollesch et al., and Gajjar et al., used both plasma and serum.^[115,116] Only 3 papers used whole blood due to protein complexity of the blood. More specific approaches were taken by Liu et al., and Mordechai et al., which used red blood cells and white blood cells, respectively.^[106,108]

• Methods. 19 out of 22, used FT-IR in Attenuated Total Reflectance mode. Due to cost efficiency, most of them used ZnSe and some of them used diamond as a crystal. However, in an article by Butler et al, there is novelty on that point: they evaluated silicon as IRE, which is relatively cost effective and provides no less accuracy.^[102] Only 3 papers were based on Transmission FT-IR and they used barium fluoride as a crystal plate. From the table, it can be noticed that those papers have relatively lower specificity parameters than others (46%–77%).

Below are some observations and conclusions regarding Table 5 about the content of the listed research papers: Number of patients: average—72 patients, median—72, range 26 to 736.

Raman/SERS acquisition conditions: Though it is not mentioned in the table directly, the Raman studies performed on the human blood utilize SERS with 785 nm laser (except for^[182,183]), magnification ranging from ×10 to ×100. Majority of papers performs PCA analysis on Raman spectra to distinguish between patients with health condition and healthy. The prevalent method was SERS, as 10 out of 13 utilized it. Raman was used in 3 publications works, and Infrared spectroscopy in 2 publications. Diagnostic sensitivity and specificity between the subject group and the sick group obtained by Raman/SERS show promising value of those methods. Average sensitivity, specificity, and accuracy are 86.0%, 91.3% and 81.3%, respectively. Median values are 88.5%, 95.0%, 83.6%, respectively. Diagnostic sensitivity and specificity between two subject groups were measured in only one work.^[49]

Other biofluids

Saliva

Saliva is a biofluid produced by the salivary glands in the oral cavity consisting of a complex mixture of different organic and inorganic components.^[184] In more detail, it contains active inorganic cations (Na⁺, K⁺, Ca²⁺) and anions (Cl⁻, HClO₃⁻), organic amino acids, hormones, proteins.^[185] It is becoming a common biofluid for detection and analysis of a variety of oral, respiratory, systemic, and genetic diseases, including oral cancer, lung cancer, and 2019-nCov, due to its complex structure, noninvasive nature, and simple sampling technique.^[186,187] Also, saliva can be used for the detection of trace drug amounts, as it is direct contact with major intake pathway.^[188,189] In context of our review, there is a sizable number of analytical and clinical studies conducted with saliva by using SERS and significantly less for FT-IR.

SERS is a method that provides high sensitivity and high selectivity. Thus, the detection of illicit drugs is quite popular among analytical studies involving saliva and SERS. For example, the research conducted by Inscore et al. studied the detection of 5 illegal drugs by using glass capillaries with gold and silver doped sol-gels.^[190] Before the measurement they performed solid-phase extraction of drugs from saliva to increase the sensitivity of the method. They were able to consistently detect five drugs at concentrations of less than 1 ppm. In the same year, this group improved the methodology and performed the detection of 152 illicit drugs by using fused gold colloids in a porous glass matrix sensor.^[191] They achieved 5000 times improvement in sensitivity compared with the previous study with confident detection of 50 ng/ml cocaine (0.17 µM), 1 µg/ml PCP (4.1 µM) and diazepam (3.5 µM), and 10 µg/ml acetaminophen (66 µM). A similar study was conducted by Andreou et al. on the detection of methamphetamine with a glass microfluidic device containing silver nanoparticle clusters.^[192] The device detected methamphetamine concentrations well below physiological thresholds, with the lowest being 10 nM. A different approach for the detection of drugs was proposed by Yang et al. in 2015.^[193] The substrate system consisted of Au nanoparticles dotted magnetic nanocomposites (AMN) responsible for signal enhancement and controlled aggregation, modified with inositol hexakisphosphate (IP6) on the surface of Al foil. They confidently detected 50 nM of nicotine marker (cotinine) and 29 ppb of cocaine marker (benzoylecgonine—10 µM) in saliva. Other examples of analyte detection in saliva by SERS include detection of pyocyanin with Teflon capillary chip (LOD <0.5 µM),^[194] silver (I) (0.17 nM) and mercury (II) (2.3 pM) ions with Au nanostar@Raman-reporter@silica sandwich structure,^[195] pH1N1/H275Y mutant virus (down to 1 PFU) with urchin Au nanoparticles,^[196] and Thioflavin‐T (10 pg/ml—31 pM) with multibranched polycarbonate (PC) nanopillar arrays.^[197]

Clinical studies involving the SERS technique and saliva are mostly concentrated on the diagnosis and differentiation of cancer diseases. As an example, Feng et al. conducted a study involving 97 patients, in which 33 have benign and 31 have malignant breast tumor.^[198] SERS detection was performed with saliva samples after membrane protein purification by using Al plate substrate with silver nanoparticle colloids. The spectra then underwent PCA-DA analysis, which yielded following results for selectivity, specificity and accuracy: healthy 75,75%, 93,75%, 87,63%; benign 72,73%, 81,25%, 78,35%; malignant 74,19%, 86,36%, 82,47%. A different approach to breast cancer detection was taken by Hernández-Arteaga et al. by using sialic acid in saliva as an indicator.^[199] In this case, the patient group consisted of 106 healthy women and 100 with cancer. Statistical analysis of SERS spectra showed the sensitivity of 94%, specificity 98%, and accuracy of 92% for patients with breast cancer. But with the assumption that the sialic acid concentration for positive results is >7 mg/dL. These two studies are unique as the rest of the clinical studies are primarily about oral and respiratory diseases. For example, the saliva of lung cancer patients was studied by two groups with different approaches for statistical analysis and substrates. Li et al. used common PCA-LDA analysis with silver nanoparticles gold nanoparticles as a substrate to differentiate between 20 healthy and 21 cancer patients.^[200] Result of spectra analysis showed that this method has 78% sensitivity, 83% specificity, and 80% accuracy toward lung cancer samples. Qian et al. significantly enhanced these results by utilizing support vector machine (SVM), leave-one-out, random forest algorithms, and gold nanomodified chip.^[201] The study of 66 healthy and 61 cancer patients resulted in a sensitivity of 95.08% and specificity of 100% for the worst performing algorithm. The rest of the examples used PCA-LDA analysis for the detection of oral squamous cell carcinoma and asthma. The sensitivity of 89%, the specificity of 57%, and accuracy of 73% were achieved for the differentiation between 18 oral cancer patients and 18 healthy patients.^[202] Better results were obtained for the detection of 26 asthma patients with 85%, 82%, and 84% for selectivity, specificity, and accuracy, respectively.^[203]

The main difference between the SERS and FT-IR methods in saliva analysis is the higher interference of water in IR spectroscopy, as it is the main component of saliva. Thus, the number of analytical and clinical studies is considerably lower compared to SERS. Most of them are related to the detection of illicit drugs, which is also the case for SERS analytical studies. For example, the research paper published by Armenta et al. describes the detection of ecstasy by ATR-FTIR after liquid-liquid extraction.^[204] Time to result of this method was 5 minutes with LOD of 320 µg/L (1.66 µM) and RSD of 11%. Real-time results for the detection of cocaine were shown by Wagli et al. by utilizing a novel chip combining microfluidic droplet-based liquid − liquid extraction and quantum cascade laser source IR spectroscopy.^[205] More common methodology for the detection of cocaine was published by Hans et al.^[206] They used ATR-FT-IR with a one-step extraction method to achieve 20 µg/ml (66 µM) LOD in saliva and 10 µg/ml (33 µM) LOD in TCE after extraction. The only clinical study that we were able to find utilizing IR spectroscopy with saliva was conducted by Scott et al. on the differentiation of diabetes patients by detection of lipid, thiocyanate, protein, and glucose levels.^[207] The study was conducted on 22 healthy and 39 ill patients and used LDA statistical analysis. Results of the spectra analysis across 6 spectral regions of choice showed 100% accuracy and specificity for training set and 88.2% accuracy and specificity for sample set.

Cerebrospinal fluid

Cerebrospinal fluid is a complex mixture of inorganic ions (K⁺, Na⁺, Ca⁺², Mg⁺², Cl⁻), enzymes, glucose, neurotransmitters, and a low concentration of proteins. This biofluid is mostly generated in the choroid plexus and acts as a protective and regulatory medium for the brain and central nervous system (CNS).^[208] Thus making it an ideal candidate for the detection and analysis of diseases affecting the brain and nervous system.^[209,210] The sampling method for this biofluid is highly invasive, which makes it more difficult to perform analytical and clinical studies.^[211] Consequently, the number of studies involving SERS and IR methods is limited.

The increasing number of analytical studies utilizing SERS with CSF in recent years can be mainly attributed to the high sensitivity and selectivity of the method. One such study was published by Liu et al. for the detection of glucose level in CSF of patients with CNS infection by metabolic assay and bio-nanoreactor sensor.^[4] With just 20 µL of sample volume they were able to detect glucose levels, which agreed with the clinical glucose assay kit. Also, they were able to monitor the increase in glucose concentration of the patient undergoing anti-infectious therapy. The results of these measurements can be seen in Figure 8. The detection of CNS bacterial infection was also studied by Kamińska et al., particularly meningococcal meningitis Neisseria meningitides. In their 2016 study, they used Si/ZnO substrate with a sputtered gold layer for the detection of neopterin levels, which is a new biomarker for bacterial infections.^[212] They achieved LOD of 1.6 nM which is below normal neopterin levels (5 nM) and showed similar results as the commercial ELISA kit. Also, they conducted PCA analysis for several clinical samples and achieved a specificity of 95% and sensitivity of 98%. In their next study, they proposed a simpler and faster substrate production method and applied its results in the detection of neopterin.^[213] Analysis of CSF samples confirmed high sensitivity and selectivity of these substrates with LOD for a healthy group of 4.3 nM and an infected group of 54 nM. In comparison with their previous methodology, they achieved a similar level of sensitivity and selectivity with a more cost-efficient substrate. Similar studies on the detection of CNS and brain diseases in CSF with SERS were conducted in the following years. A good example of such a study was done by Kim et al. in 2018.^[214] They studied CSF after subarachnoid hemorrhage (SAH) by using label-free cellulose substrate with gold nanoparticles. With crystal violet as an analyte for small molecules, they achieved a LOD of 0.74 pM and a relative standard deviation (RSD) of 8.5%. Moreover, by using a combination of 5 biomarkers they were able to distinguish SAH-induced complications with 87%–93% reliability and 11%–29% RSD. Another study utilizing CSF with SERS was conducted on Alzheimer’s disease detection with polyA aptamer AuNPs.^[215] This study is based on the specific binding of aptamer and core disease biomarkers, Tau protein, and Aβ(1-42) oligomers. Because of this binding, they achieved LOD of 4.2 × 10⁻⁴ pM for Tau protein and LOD—3.7 × 10⁻²nM Aβ(1-42) oligomer with recoveries higher than 96% for CSF samples. Other examples of SERS application in CSF analysis concentrate on the understanding of brain process through detection of dopamine levels, which is an important neurotransmitter for the normal function of hormonal, nervous, and vascular systems. A study by Ranc et al. used nanocomposite of silver and magnetite nanoparticles modified by dopamine-specific chemical.^[216] They achieved two-digit femtomolar limits of detection similar to electrochemical methods, but with better selectivity. However, the RSD values were greater than for conventional HPLC-MS method, although this method is more cost-efficient, faster and easier. Different approach for detection of dopamine was taken by Zhang et al. utilizing dual-recognition-induced hot spot generation.^[217] In more detail, they used ester to immobilize dopamine molecules on gold film which was further modified with silver nanoparticles. This substrate structure selectively detected dopamine with LOD of 0.3 pM and recoveries for real life samples between 93.7% and 109.3%. This methodology further cuts cost for fabrication and reduces use of complicated processes retaining good sensitivity and selectivity. Literature search for clinical studies involving CSF yielded only several publications for FT-IR and no publications for SERS

Figure 8. (A) Scheme of monitoring glucose in patient CSF with plasmonic nanoreactor. (B) Glucose concentrations in patient CSF determined with nanoreactor and clinical test kit. (C) glucose concentrations of brain infection patient and noninfection control measured with plasmonic nanoreactor. (D) Dynamic monitoring of CSF glucose of a brain infection patient during anti‐infectious therapy. Reprinted from open access publication under CC BY license.^[4]

The good examples of successful FTIR applications are demonstared by Nabers et al.^[218] They firstly introduced a method for the detection of secondary structure change in Abeta peptide, which is an early sign of Alzheimer’s disease. They proposed the use of a novel immuno-infrared sensor consisting of Aβ peptide selective antibody attached to the internal reflection element of the ATR setup. This structure provided sensitive detection of 1 ng/ml Aβ peptide (S/N 41-52) without any isolation or purification stages. Later, they performed a clinical study of this setup among 141 patients: 80 healthy, 50 with dementia Alzheimer type (DAT), and 11 with mild cognitive impairment due to Alzheimer’s disease (MCI-AD).^[219] Results of data evaluation and statistic tests showed an accuracy of 90%, sensitivity of 94%, and specificity of 88% for differentiation between DAT and healthy patients and accuracy of 86%, the sensitivity of 73%, and specificity of 88% for differentiation between MCI-AD and healthy patients. Based on these results, they proposed further follow-up studies on a bigger set of patients and the possibility for detection of other proteins. Another example of an FT-IR application was published by Yonar et al.^[220] They studied relapsing-remitting multiple sclerosis (RRMS) and clinically isolated syndrome (CIS) multiple sclerosis (MS) by using multivariate analysis methods and IR spectra of CSF. The study was conducted on 89 patients: 24 healthy, 35 RRMS, and 30 CIS. Results of HCA analysis showed that discrimination between healthy and diseased groups (sensitivity 95%, specificity 92%) is significantly better than between RRMS and CIS (sensitivity 88%, specificity 40%).

Tears, bile juice, etc

Tears is a body fluid produced by the lachrymal gland and contains approximately 100 proteins, electrolytes, lipids, and metabolites.^[72,221] Because of its composition and ease of sampling tear is a biofluid with potential applications in disease detection. For example, detection of urea and uric acid in tears for detection of kidney problems and gouty arthritis. Leordean et al. conducted SERS detection of urea by using glass slides modified with gold colloids.^[5] They found a distinctive urea band at 1004 cm⁻¹ in a teardrop, which was reproduced several times. So, the use of tears is also viable for the detection of urea. On the other hand, Park et al. performed detection of uric acid with Schirmer strip modified with Au nanoislands.^[222] They achieved quantitative physiological level (25–150 µM) detection of uric acid in tears, which were found to correlate with gouty arthritis diagnose in 16 volunteers. A similar analysis of healthy and ill patients was conducted by Kim et al. but on the breast cancer patients.^[223] They showed the potential of SERS with gold-decorated, hexagonal-close-packed polystyrene (Au/HCP-PS) nanosphere monolayer for the detection of Raman reporter and further use in breast cancer diagnostics via PC-LDA analysis. The scheme of the whole study can be observed in Figure 9. The results of performance analysis showed EF of 5.4 × 10¹⁰ and a linear correlation curve for the concentrations of 2-naphthalenethiol (2-NAT) between 100 fM and 1 pM. The PCA-LDA analysis resulted in a sensitivity of 92%, specificity of 100%, and accuracy of 96%. Similar research involving PCA analysis was performed by Choi et al. on the detection of adenoviral conjunctivitis.^[224] They utilized drop-coating deposition surface-enhanced Raman scattering (DCD-SERS) method for the analysis tear fluid. Their PCA analysis showed sensitivity of 93.3% and specificity of 94.5% for three spectral biomarkers at 1242, 1342, and 1448 cm⁻¹. However, both studies were conducted on the patient set lower than 20. On the other hand, the study conducted by Travo et al. analyzed 63 patients of which 11 have nonspecific inflammatory ocular disease and 21 have keratoconus.^[225] They diagnosed these patients by using FT-IR and subsequent PCA analysis of spectral differences in lipid and carbohydrate vibrations. Results showed that there is a significant statistical difference between healthy and ill patients, and between patients with keratoconus and nonspecific inflammatory ocular disease.

Figure 9. A scheme of a study of the breast cancer identification from human tears by SERS. Copyright ACS. Reprinted with permission of ACS from Kim et al.^[223]

Bile juice is a biofluid produced by the liver and contained in the gallbladder. Its composition is quite complex, containing different bile salts, proteins, fatty acids, etc.^[226] Due to its storage place and composition bile juice is a major biofluid for detection and analysis of gallbladder diseases both by SERS and IR. Examples utilizing SERS were conducted by Vu et al. The first example showed the applicability of voltage-induced SERS sensor for differentiation of gallbladder (GB) polyp and GB cancer.^[6] The concentrations of bilirubin and urobilinogen were detected by using Au nanodendrites constructed on a screen-printed electrode (AuND@SPE) at voltages from −300 to 300 mV. With this setup, they achieved EF 1.4 × 10⁸ and found that the concentration of urobilinogen is higher for patients with GB cancer at −100 mV. Next, they proposed the use of different substrates for the detection of bilirubin in GB stone and GB polyp patients.^[227] The study was performed on 6 patients with GB polyp and 21 patients with GB stone. The results for a paper strip bridging Au nanodendrite-encaged nickel foam (PS-AuND@NF) and subsequent two-trace two-dimensional (2T2D) correlation analysis showed an accuracy of 88.9%, sensitivity of 100%, and selectivity of 85.7%. Other studies on bile juice were performed by the FT-IR method. Hwang et al. also performed FT-IR analysis on the differentiation of GB polyp and GB stone patients.^[228] They analyzed bile juice from 9 GB polyp and 27 GB stone patients and found that three distinct spectral differences between patient types. Five principal components from PCA analysis were fed to support vector machine (SVM), which gave 78.6% discrimination accuracy. Remaining IR study was conducted by Untereiner et al. on 19 patients with malignant and 38 with benign biliary structures.^[229] PCA analysis with SVM model and randomized leave-one-out cross-validation (LOOCV) procedure was applied to the IR spectra of different bile juice purity levels. Results of this analysis showed that additional purification and extraction stages for bile juice does not necessarily increase sensitivity, specificity and accuracy.

Other examples of biofluids used with SERS include feces and sweat. Sweat is body fluid produced by sweat glands and feces are a by-product of digestion and both of them are excretions of bio-waste.^[230,231] Thus, they can be used for the detection of various diseases and different body states. For example, Xing et al. used to sweat for the analysis of drug effect on the cancer patients by SERS.^[232] They utilized silver colloids on a glass slide as a substrate for the monitoring of capecitabine concentration. The linear correlation was found for the concentrations from 2 to 20 µg/ml, with an R² of 0.9933. Thus, this methodology has a future in the analysis of drug concentrations in sweat samples. An example for the analysis of feces was published recently and focused on the detection of rotavirus by SERS immunochromatographic assay (ICA).^[233] The results showed an 80 pg/ml detection limit for naked-eye detection and an 8 pg/ml detection limit for SERS signal detection. Also, results showed a 100% correlation with conventional RT-qPCR. Therefore, the scope of applications for vibrational spectroscopy is not limited to just blood and urine and encompasses most of the body fluids.

Comparison of SERS and Infrared spectroscopy

As mentioned in earlier chapters there are several fundamental differences between Raman/SERS and IR spectroscopies, which produce perceptible disparity between these analytical methods in terms of number and type of applications. Nowadays, SERS is the dominant method of vibrational spectroscopies for analytical studies that we define as studies aimed for quantitatively detect individual analytes, and for clinical studies that we define as studies aimed for distinguishing the control (healthy) group and the group with a specific health condition.

Analytical studies studying human urine the literature search showed only papers related to SERS detection (Table 1). As the detection of trace analytes in such a complex system requires a highly sensitive method like SERS. Which, according to found studies, shows the average limit of detection of 3 × 10⁻⁸ M for various analytes in human urine. In more detail, the highest LOD is 1 × 10⁻⁴ M for detection of glucose and the lowest LOD is 4 × 10⁻¹⁴ M for detection of dopamine.^[41,160] Which are significantly lower than physiological levels for glucose and dopamine in urine, 2.8 × 10⁻³ M and 4.9 × 10⁻⁷ M.^[234,235] These values are the direct consequence of high signal enhancement produced by the SERS effect, up to 5.5 × 10¹⁰. The same pervasiveness of the SERS method can be seen in clinical studies investigating human urine, with 15 publications for SERS and 3 publications for IR methods, with more details available in Table 2. However, there is no significant difference between these methods in terms of clinical parameters except for accuracy as can be seen in Table 6. This is probably the result of a low number of IR studies compared to SERS.

Table 6. Quantitative comparison of SERS vs IR spectroscopy of biofluids.

Medium/biofluid	Method	Geometric mean LOD, M	Average/median sensitivity	Average/median specificity	Average/median accuracy	N reports for LOD, sensitivity, specificity, accuracy
Urine	IR	NA	91%/95%	93%/96%	98%/98%	0, 3, 3, 3
Urine	SERS	3 × 10^−8	89%/90%	93%/94%	91%/91%	19, 25, 25, 11
Blood, plasma, serum	IR	7.1 × 10^−4	91.5%/93.0%	86.7%/89.0%	95.1%/95.1%	4, 24, 24, 13
Blood, plasma, serum	SERS	0.81 × 10^−9	86.0%/88.5%	91.3%/95.0%	81.3%/83.6%	25, 10,11, 4
Other biofluids	IR	2.5 × 10^−6	87.3%/90.0%	90.3%/88.2%	87.5%/88.2%	4, 6, 7, 7
Other biofluids	SERS	3.5 × 10^−10	88.4%/91.5%	88.9%/90.1%	88.2%/88.3%	17, 12, 12, 12

Average and median diagnostic sensitivity, selectivity, and accuracy (between the healthy control group and the sick patient group) calculated from the tables summarizing medical diagnostics/clinical studies (Tables 2–5 and Tables S2–S4 in the supplementary information). The geometric means of LODs are calculated from the tables reporting about analytical methods (Tables 1 and 3; and Tables S1, S3, and S5 in the supplementary information).

The same prevalence of the SERS method can be observed for analytical studies involving human blood plasma and serum, with 12 analytical studies for SERS compared to 1 analytical study for IR which can be observed in more detail in Table 3. The same level of difference can also be seen from average LODs of these methods, 7.1 × 10⁻⁴ M for IR and 8.1 × 10⁻¹⁰ M for SERS producing almost 8.8 × 10⁵ difference in sensitivity. In more detail, the single study found for IR detected glucose, triglycerides, urea, cholesterol, albumin, and total protein in human plasma.^[171] Which are analytes that are present in high millimolar concentrations in blood. In comparison, studies using the SERS method are detecting analytes that are present in blood in micromolar to picomolar concentrations, like tumor necrosis factor-alpha (TNF-α) (0.44 × 10⁻¹² M) or pseudoephedrine drug (3 × 10⁻⁷ M).^[236,237] On the other hand, the IR technique is more predominant for clinical studies with 22 publications compared to 10 publications for SERS. More detailed information about these studies can be seen in Tables 4 and 5. Also, the average performance of the IR method was significantly better in terms of accuracy, as the SERS method (81.3%) was three times more likely to produce wrong diagnostic results in comparison to FTIR (95.1%), which can be seen in Table 6. However, in a direct comparison of both methods used for the detection of breast cancer, the SERS method outperforms the IR method for about 20% in terms of accuracy (74% vs 94.2%), resulting in approximately 5 times lower chance for misdiagnosis.^[107,117]

So, the use of SERS as a primary tool for the detection of drugs, biomarkers, metabolites, and especially for cancer diagnostics is evidently becoming more popular. As the extremely high sensitivity of this method produces a significant advantage for the detection of low concentration analytes like drugs,^[190,192] biomarkers,^[197] metabolites,^[193] toxins^[195] compared to IR. Moreover, it retains intrinsic advantages of Raman like nondestructive analysis, minimum sample preparation, multiplexing, measurements in different biofluids, etc.^[238]

Brillouin spectroscopy of biofluids

Brillouin light scattering (BLS) spectroscopy, like Raman spectroscopy, is based on measurement of inelastically scattered incident laser light by thermally generated or coherently excited elemental phonons in Stokes and anti-Stokes events.^[239–241] Contrary to Raman spectroscopy, which probes primarily optical phonons (∼ 10–3000 cm⁻¹) in the medium, the BLS spectroscopy probes light scattered from acoustic phonons or magnons (< 10 cm⁻¹) in condensed inorganic, organic or biological medium.^[242] Figure 10 shows schematic diagram of Brillouin and Raman inelastically light scattering spectra along with elastically scattered Rayleigh light.^[239]

Figure 10. Brillouin and Raman scattering events. Reprinted from open access publication.^[239]

For biofluid applications this technique is attractive due to its ability to probe in non-contact and nondestructive manner viscoelastic properties of transparent and semi-transparent, elastically isotropic biological fluids via detection of back-scattered laser light from propagating bulk longitudinal acoustic phonons. Fluids, including biofluids, typically do not sustain shear stresses. Therefore, Brillouin scattering from transverse (shear) phonons in probed liquids is expected to be negligible. With measured Brillouin shifted peak position (phonon frequency) and the knowledge of refractive index of the interrogated biofluid medium, one can access the longitudinal sound velocities in fluids. An additional knowledge of the medium density can also provide information of longitudinal bulk “storage” modulus. With measured spectral width of the Brillouin shifted peak one can assess the longitudinal bulk “loss” storage (i.e., viscosity) of the fluid. With inherently small signal intensities associated with spontaneous BLS from fluids, there have been efforts to enhance Brillouin signals using surface plasmon enhancement^[243] and stimulated BLS spectroscopy.^[244] Recently BLS spectroscopy has started to gain momentum in biomedical field.^[245–247] The spectrometers used to detect BLS light are based primarily on multi-pass tandem Fabry-Perot interferometers (TFPI) and virtual image phase arrays (VIPA).^[246,248]

BLS spectroscopy has shown a promise for screening the increased total protein content in cerebrospinal fluid (CSF) during bacterial meningitis quantifying protein concentration without alteration of the fluid obtained from the body,^[249] as the protein concentration of a solution typically scales with the fluid elasticity.^[250] Brillouin technique has also provided the path for urinary protein probing as biomarkers of kidney diseases,^[251,252] studying thermal denaturation in aqueous solutions of lysozyme^[253] and bovine serum albumin (BSA) proteins.^[254] This technique has shed light on bio-relevant aqueous fluids of glycine, triglycine, glycerol, sucrose, lysozyme, BSA and gelatin in relation to their viscosity and solute concentration,^[255] on cell culture^[256] and in-vivo Brillouin microscopy of the human eye.^[257] Biofluids are largely transparent or semi-transparent, where the penetration depth of laser light is sufficiently larger than the incident laser light wavelength, so that BLS is observed due to bulk longitudinal phonons (acoustic waves) in the interrogated fluid medium. At the same time this technique can be applied to small-size samples and in combination with microscopy the imaging of biomechanical properties of biofluids can be accomplished, i.e., noncontact three-dimensional mapping of intracellular hydromechanical properties.^[258] Stimulated Brillouin scattering microscopy and imaging has been demonstrated for bio-relevant fluids^[244] in quartz microchannels,^[245] which opens the avenue for future viscoelastic micro-bio-fluidic diagnostic devices.

Conclusion

In this review, vibrational spectroscopy techniques were discussed in the light of their applications for biofluids. Those techniques are mainly presented by Raman spectroscopy and Infrared spectroscopy. Nowadays, SERS and ATR-FTIR became predominant methods for Raman and Infrared spectroscopy, respectively. There are some disadvantages in necessity for SERS substrate preparation, while achieving reproducibility of the signal enhancement on those substrates might be a challenge. The limited shelf life for nanoparticles and substrates also adds up to a bit higher cost for SERS as compared to FTIR . However, those cons of SERS as opposed to FTIR are usually more than generously compensated by much higher sensitivity to low concentrations of analyte. This high sensitivity makes SERS a method of choice for detection and diagnostics of many conditions and drugs, particularly in diagnostics of cancer, including diagnostics of its early stages, which might amount to saving a few thousands lives per year worldwide. According to Sung et al., 10 million cancer deaths occurred in 2020, and it is expected that cancer burden is expected to rise by 47% in 2040.^[259]

Still there some biomarkers (glucose etc.), present at biofluids at significant concentrations of several mM and sometimes even a few µM that can be detected efficiently by FTIR at probably lower cost and time than any alternatives. Both of those vibrational spectroscopy methods are direct, which mean that they require no labeling of analyte with tag molecules, such as for instance labeling/modification with fluorophores in fluorescence. FTIR and SERS also typically use low volume of samples, no or very little sample preparation, fast time of analysis and relatively compact and inexpensive instrumentation (at least relative to most MS and HPLC instruments). Moreover, due to relatively narrow spectral peaks, both Raman and IR spectroscopic methods have a significant capacity for direct multiplexed detection of analytes, unlike some other rmolecular spectroscopy methods such as fluorescence or UV-vis spectrophotometry. Therefore, there is place for both SERS and FTIR applications in clinical diagnostics and in the next decade we may witness a wave of commercialization in diagnostic applications of those techniques. Brillouin scattering spectroscopy may also find application in clinical analysis of some human biofluids, for example, CSF and urine.

Declaration of competing interest

The authors declare that they have NO known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Additional information

Funding

The authors of this review acknowledge funding from the Nazarbayev University Collaborative Research Program (CRP) for 2020–2022 (Funder Project Reference: 091019CRP2105).