Why bone quality?

The main clinical tool in the effort to estimate/predict bone strength and potential fracture risk is the evaluation of clinical risk factors together with bone mineral density (BMD) measured by dual X-ray absorptiometry (DXA). Although BMD is an important contributing factor to bone strength, it is not the sole factor that determines bone fragility [1]. Clinical data show a substantial overlap in BMD values between populations that do and do not develop fractures [2–4]. Based on BMD, it has been suggested that for a certain bone mass, an individual’s fracture risk increases with age [5] while the 1-year relative risk associated with a given change in BMD has been shown to be independent of age [6]. Many studies have also revealed that mechanical variables directly related to fracture risk are either independent [7] or partly accounted for, by bone mass itself [8, 9]. The combination of BMD outcomes with biomarkers corresponding to bone turnover rates has improved the clinical situation, but even the combination of these two clinically utilized outcomes does not fully account for the observed bone strength. As a result, bone quality was introduced as an important determinant of bone strength [10]. Bone quality’s pivotal role becomes evident when one considers several clinical paradigms such as the finding that a prior fracture is a strong predictor of future ones independently of BMD suggesting intrinsic skeletal defects [11], or the clinical realization that BMD gains are disproportionate to the fracture risk reduction evident in osteoporotic patients receiving antiresorptives treatment [12]. Finally, conditions such as premenopausal idiopathic osteoporosis (IOP) where widely employed clinical tools such as DXA and biochemical markers fail to either account for the demonstrated bone strength further emphasize the importance of bone quality [13–15]. Bone quality is a broad term encompassing the structural and material properties of bone [16]. Although structural properties are equally important with the material properties of bone in the determination of its mechanical characteristics in health and disease, they are beyond the scope of the present review, and the reader is encouraged to consider review articles that are focusing on the role of bone architecture as well as methods to assess it [17–21].

Need for microscopic imaging techniques when assessing matrix and mineral composition/quality and bone strength

When assessing matrix and mineral quality, one has to account for tissue heterogeneity at the microscopic level. In healthy humans, bone surfaces may be undergoing formation or resorption, or they may be relatively inactive, and these bone turnover processes occur throughout life in both cortical and trabecular bone [22]. Bone remodeling is essential to the ability of bone to repair damaged tissue and successfully adapt to variable loading conditions. In humans, it occurs on both cortical (periosteal and endosteal surfaces, Haversian canals) and trabecular compartments, with trabecular rates higher than cortical ones throughout life [22]. The rate of cortical bone turnover, which is due to rapid growth, modeling and remodeling reaches values of about 50% per year in the mid-shaft of the femur during the first 2 years of life and eventually declines to a rate of 2–5% per year in the elderly, where maintaining the skeleton is of utmost importance. Rates of remodeling in trabecular bone are higher throughout life [22–24]. As a matter of fact, mineral and matrix composition/quality in healthy humans has been shown to vary as a function of both patient age as well as tissue age within the same patient [25]. Since organic matrix and mineral chemistry are dependent on turnover rates but also factors other than bone turnover [26, 27], it becomes evident then that microscopic imaging techniques are appropriate since they are capable of accounting for both types of age: patient and tissue age [16, 28]. This attribute is critical as published data have shown the correlation between tissue age, bone tissue composition/quality, and mechanical properties [25, 29, 30].

Healthy bone is characterized by adequate stiffness and strength combined with sufficient toughness [31]. Bone’s mineral component is responsible for its stiffness, while the organic matrix diminishes its inherent brittleness. Although bone stiffness is an average (albeit not linear) of all microanatomical values, strength and toughness depend on minuscule material defects [31]. For fracture resistance, the most important attribute of bone is its aptitude to absorb impact-generated energy without reaching complete failure, i.e., its toughness [31], which is inversely related to its stiffness [31]. Bone material toughness depends on two interrelated factors: the quality of its building blocks (mineralized collagen fibers) at multiple organizational scales and the interface quality between these blocks (one example would be the cement line) [16, 31]. The study of both toughness determining factors, and especially the quality of interfaces, is therefore optimized when microscopic analyses are utilized.

As a result of all these considerations, analysis of matrix and mineral properties as a function of specific microanatomical location (based on selection criteria such as tissue age, or mineralization extent determined by either microradiography or quantitative backscattered electron imaging [32]) is desirable as it affords the discrimination between bone turnover-dependent and independent effects. Techniques such as μCT, small angle X-ray scattering (SAXS), microradiography, quantitative backscattered electron imaging (qBEI), scanning acoustic microscopy, atomic force microscopy, Raman microspectroscopy, and Fourier transform infrared microspectroscopy (FTIRM) and imaging (FTIRI) are particularly suitable for studies aiming at exploring bone quality [16, 28, 32–36]. In the present review, we will focus on FTIRM, FTIRI, and Raman microspectroscopic imaging techniques.

Fourier transform infrared and Raman spectroscopy

The mutual principle underlying both techniques is the transition between vibrational energy states of molecules; infrared spectra arise directly from absorption of energy in the infrared range, while Raman spectra from the scattering of visible or ultraviolet photons that have either gained or lost part of their energy upon interaction with the vibrating molecular bonds. Each molecule has its own unique vibrational characteristics. Additionally, the neighboring molecular environment influences the vibrational features, thus offering additional information on the “molecular neighborhood.” The position, intensity, integrated area, and width of a vibrational band can be used for monitoring a particular functional group or regions of a particular chemical species [28]. Since information on both the molecule and its neighborhood is readily available, most vibrational spectroscopic bands are composite ones, consisting of underlying peaks specific to a specific molecule-molecular neighborhood state. Thus, resolution of these broad bands to their underlying peaks offers the most detailed molecular picture possible. This is achieved through the utilization of second derivative spectroscopy (to decipher the exact number and position of these underlying peaks in a mathematically objective manner) coupled with curve-fitting routines (to determine their relative contribution) [37–40].

Vibrations are for the most part only either infrared or Raman active; thus, the two techniques offer complimentary information regarding the analyzed sample [41]. In the 1970s and 1980s, spectra-structure correlations were developed for proteins, nucleic acids, and lipid assemblies as well as molecules that are vital for the biological functions of the macromolecular moieties [28]. It was simultaneously realized that the proper spatial organization of the structural components in both cells and tissues is crucial for their optimum function, deviation from which results in pathological states [28, 42–44]. The earliest attempt to address this insight occurred in 1949 [45], when a reflecting microscope was coupled with an IR spectrometer to acquire utilizable spectra of nerve fibers and small crystals. On the other hand, the challenging signal-to-noise ratio of common Raman scattering delayed the introduction of Raman microscopic and imaging applications until 1974 [46]. Ever since, wide-ranging technical advances have resulted in a broader application of vibrational microscopy and imaging [28]. Recent new technological gains have greatly enhanced the afforded spatial resolution and rapid data acquisition. A contemporary infrared setup based on a scattering-type scanning near-field optical microscope equipped with a coherent-continuum infrared light source allows the acquisition of infrared spectra with an AFM-comparable spatial resolution (20 nm) [47]. Similar results are achievable through a second setup utilizing a pulsed, tunable infrared laser which excites molecular absorption in a sample in attenuated total reflection (ATR) mode, on a ZnSe prism [48]. Tip enhanced Raman scattering (TERS) combines surface enhanced Raman spectroscopy (SERS) with atomic force microscopy (AFM) analysis allowing for analyses with spatial resolution in the nanometer scale [49]. The development of stimulated Raman scattering microscopy permits the acquisition of images at specific Raman-shifted frequencies at video rates [50]. Finally, spatially offset Raman spectroscopy (SORS) is being actively developed so as to offer a method for noninvasive in vivo Raman measurements with custom-designed fiber optic probes placed on the skin of a subject [51].

Unique advantage of vibrational spectroscopic techniques

Bone is a composite material consisting of mineral, organic matrix, and water, with a highly complex and amazing hierarchical structure [16]. All three components contribute, both individually and in concert, to the determination of bone mechanical properties such as stiffness (essential for preventing bone bending under the load of body; often measured by the elastic modulus (or Young’s modulus) [31]), strength (maximum stress the bone material can sustain prior to failure), and toughness (absorption of impact energy without reaching complete failure; roughly measured by the energy to failure) [31]. Since bone is a heterogeneous material, its overall stiffness is a nonlinear average of all the local stiffness values [31]. On the other hand, bone strength and toughness may not be calculated as an average of bone constituents’ local values, as they depend on the nucleation and progression of cracks; thus, they are dependent on miniscule material defects (pores, microdamage, interfaces, fiber organization) [31]. Interestingly, toughness and stiffness are inversely related [31]. Although plenty of analytical techniques offer a wealth of information on individual bone components, to the best of our knowledge, thermogravimetric and vibrational spectroscopic analyses are the only two methods that provide simultaneous information on all three bone components. The former requires tissue homogenate and is destructive (no further analyses are feasible on the same tissue), while the latter offers both qualitative and quantitative information in a nondestructive, spatially resolved manner, thus allowing their combinatorial use with other techniques, enabling the gaining of a wealth of information on the same bone surface.

Tissue requirements

In principle, both FTIR [52] and Raman [42, 53] techniques can analyze fresh bone tissue provided there is an adequately flat surface. On the other hand, reflection FTIR is hampered by challenges such as subtle changes in the raw spectral bands, the significance of which remains to be determined, whereas blood in the fresh bone tissue may result in uncorrectable fluorescence interference with the bone characteristic Raman bands, rendering the spectra unusable. The overwhelming majority of tissue available for spectroscopic analysis (especially human tissue) has undergone tissue dehydration through a series of acetones or alcohols, followed by embedding in a plastic medium, most often poly methyl methacrylate (PMMA). The choice of fixative solution as well as embedding material is of utmost importance for subsequent spectroscopic analysis as the various media have different, profound effects on bone tissue [54, 55]. The most commonly used (due to the fact that the least amount of artifacts is introduced) protocol involves fixation of tissue in alcohol and embedding in PMMA resin. Thin tissue sections (2–5 μm) are required for FTIR analysis, whereas Raman can analyze both bone blocks with a polished surface as well as sections that are more than 5 μm thick. Details of spectral acquisition and processing have been published elsewhere [56, 57].

Parameters assessed by vibrational spectroscopic analysis and their importance

Below, we show typical examples of FTIR and Raman spectra of bone tissue (Fig. 1).

Fig. 1
figure 1

Typical FTIR (top) and Raman (bottom) spectra, with various bands of utility for the analysis of bone tissue appropriately labeled

Full size image

(i) Mineral to matrix ratio

This parameter has been validated against ash weight measurements, as well as mineral content by quantitative backscatter electron imaging [58–60].

The mineral to matrix ratio is the most widely reported parameter of vibrational spectroscopic analysis and is easily determined in experiments utilizing either infrared or Raman methods [28]. Its physical principle is that in vibrational spectroscopy, the integrated area of a band is directly proportional to the concentration of the specific molecular moiety giving rise to the specific band. Most commonly, the ratio of the integrated areas of the v1,v3PO4 (stretching and bending vibrations) to amide I (mainly from the peptide bond C=O stretching vibration with minor contributions from the out-of-phase CN stretching vibration, the CCN deformation, and the NH in-plane bend) or II (out-of-phase combination of the NH in-plane bend and the CC and NC stretching vibration with smaller contributions from the CO in-plane bend and the CC and NC stretching vibrations) bands in FTIR, and either v1 or v2PO4 to either amide I or amide III band in Raman, provides what is commonly reported as the mineral/matrix ratio. This is a form of bone density that, unlike other measures such as BMD by DXA, or tissue mineral content (TMC) or BMD or degree of mineralization of bone (DMB) by μCT, directly measures and accounts for the amount of the organic matrix in the sample volume analyzed. It strongly correlates with ash weight measurements by thermogravimetric analysis [58] and with mineralization data by qBEI in secondary mineralized bone packets [59]. The fact that this ratio encompasses the amount of organic matrix is of great importance and utility, as the mechanical properties of bone are determined by the quantity and quality of both its mineral and the matrix (primarily type I collagen fibrils) constituents [61]. The importance of the organic matrix amount in the determination of bone strength has been shown in published reports which demonstrate that bone displays a region of elastic (recoverable) deformation, followed by a region of plastic (permanent) deformation. Utilizing tensile testing of gradually decalcified bovine bone tissue, Burstein and Reilly suggested that the mineral offers mainly compressive, while the collagen tensile strength [61], though it is more likely that the interaction between the two is responsible for the unique mechanical characteristics of bone. The spectroscopically determined mineral/matrix ratio is directly proportional to bending stiffness and failure moment and a superior predictor of bone-bending stiffness compared to BMD alone [62]. At actively forming trabecular surfaces of healthy females, this ratio increases as a function of tissue age and decreases as a function of patient age at constant tissue age [25]. It is unaltered in juvenile osteoporosis [63], decreased in osteoporosis [64] compared to premenopausal healthy controls, and modified by various osteoporosis therapies [65–69]. Unlike outcomes based on either DXA, μCT, or qBEI, in premenopausal patients diagnosed with IOP, this ratio at actively forming trabecular surfaces was altered between IOP patients that sustained fragility fractures vs. those who did not [14].

An interesting option is inherent with Raman analysis, as this ratio can be calculated based on either the ratio of the integrated areas of the v1 to amide I or v2PO4 to amide III band; the former is tissue organization independent, while the latter is dependent on both the amount of mineral and organic matrix present in the sample analyzed, as well as collagen fiber and mineral crystallite orientation. All four peaks are present in the Raman spectra of bone tissue; thus, they offer the option to combine quantitative data with tissue organization considerations, without the need of extra analyses [70, 71]. Such information is also available through FTIR analysis, but since a polarizer needs to be introduced in the beam path, the required time for analysis is tripled as several spectra of the same pixels need to be acquired to be able to incorporate tissue orientation considerations [72].

(ii) Mineral maturity/crystallinity

This parameter has been validated against X-ray diffraction outcomes [37, 73–75].

Bone mineral consists of poorly crystalline, highly substituted, apatite crystals. Pure apatites consist of calcium (Ca2+), phosphate (PO4 3−), and hydroxyl (OH−) ions. Biological apatites are characterized by various crystal lattice substitutions such as magnesium (Mg2+), potassium (K+), and/or sodium (Na+) for Ca2+, acidic phosphate (HPO4 2−) for PO4, and carbonate (CO3 2−), which can actually substitute for either OH− (type A), or PO4 3− (type B), or be loosely bound on the crystal surface (labile [76]). Additionally, crystal vacancies may be present in the crystallites. The type and extent of substitutions influences the crystal solubility as well as the size and shape. In the literature, the chemical makeup of these crystals (and how close it is to that of pure, highly crystalline hydroxyapatite) is usually referred to as crystal maturity, while the size and shape as crystallinity. Interestingly, it is changes in the chemical makeup that drive changes in size and shape; thus, the term mineral maturity/crystallinity is often used.

Vibrational spectroscopic techniques such as FTIR and Raman [28] are particularly useful in establishing the chemical makeup of the poorly crystalline apatitic crystals in bone, i.e., the presence of impurities, and based on comparison to X-ray line broadening analysis, on their shape and size [75, 77]. Bone mineral maturity/crystallinity is significantly dependent on both subject and tissue age within the same subject. Healthy bone consists of crystallites whose size and shape fall within a certain range. Values outside this range have been encountered in clinical cases with prevalent fragility fractures such as osteoporosis and fluorosis [16, 64, 78, 79]. In FTIR, the information is attainable through the resolution and quantification of the underlying peaks in the v1,v3PO4 band, using second derivative spectroscopic and curve-fitting analysis [37, 39]. In Raman spectroscopic analysis, this bone quality index is assessed from either the full width at half height of the v1PO4 band or the wavelength at maxima of the same band [75].

(iii) Carbonate content

This parameter has been validated against chemically determined carbonate content in synthetic analogs of bone mineral, as well as bone tissue [76, 80, 81].

As stated in the previous subsection, bone mineral crystallites are poorly crystalline, highly substituted apatites. One of the most abundant substitutions is carbonate, which may be in the form of type A, type B, or labile [76], as outlined previously [82]; the type and extent of the substitution influences the crystals’ solubility and size and shape. In healthy bone, the average carbonate content is about 6 wt%, although it is known to vary according to subject and tissue age, as well as anatomical compartment (trabecular vs. cortical) [81]. The extent of carbonate substitution has been reported to be altered in bone susceptible to fragility fractures in both animal models and humans [83–92].

Determination of relative carbonate content in bone tissue is readily available in both FTIR and Raman analyses. In FTIR, two bands are evident that are due to carbonate: one ∼850–890 cm−1 and one ∼1400–1500 cm−1. Both bands are composite ones, overlapping with either acidic phosphate peaks (the former) or noncollagenous protein peaks (the latter); thus, they require further resolution to their respective underlying peaks using second derivative spectroscopy followed by curve-fitting routines so as to gain information on the extent as well as the type of carbonate substitution [37, 81]. In Raman spectroscopy, the most frequently used spectral band to report type B carbonate content is ∼1070 cm−1, while for type A, a weak band ∼1103 cm−1 is used [42]. Both of these bands are composite ones as well, overlapping with phosphate, lipid, and PMMA peaks; thus, as is the case with FTIR, further processing is required so as to resolve the carbonate-related peak.

Using homogenized human iliac crest biopsy tissue, it has been shown that carbonate content slightly increases with increasing subject age, even in the absence of accompanying changes in crystallinity after the age of 30 [93]. Unfortunately, since homogenized tissue was used, no correlation to tissue age within the same subject was feasible. Later studies employing vibrational spectroscopic techniques and attempting to establish the correlation of carbonate content with tissue age have been conflicting, with some reports alluding to a direct relationship between the two, while others showing the opposite [37, 81, 94]. This apparent conflict is most likely due to the fact that in most studies, no strict tissue age control was applied (for example using fluorescent double labels to determine the exact tissue age), or the broad raw bands were used in the calculations unjustifiably ignoring the spectral contribution of other moieties to these unresolved bands. Moreover, although most frequently the carbonate content based on vibrational spectroscopic analyses is reported as the carbonate to phosphate ratio, a few reports have introduced the ratio of carbonate to organic matrix [44]. As the normalization is to the organic matrix content, this ratio is obviously not reflecting the extent of carbonate substitution in the mineral crystallites, but is rather another, crude way of expressing mineral to matrix ratio. Nevertheless, since altered carbonate content has been reported in bone tissue susceptible to fragility fractures, future studies are needed so as to explain and consolidate the glaring discrepancies that exist in the published literature concerning this bone mineral quality metric.

(iv) Relative tissue water content/nanoporosity

The direct method in fresh bone tissue has been validated against gravimetric analysis results [95].

Recently published reports have reintroduced to the spotlight the importance of bone tissue water content in the determination of its mechanical behavior [96, 97]. Vibrational spectroscopic techniques and in particular Raman analysis can provide information on tissue hydration. Unal et al. recently identified four Raman peaks (∼3220, 3325, 3453, and 3584 cm−1) that correlated with water loss determined by thermogravimetric analysis in a series of experiments involving the gradual dehydration of fresh, demineralized bovine femur bone [95]. The advantage of this approach is that fresh tissue is utilized, and water is directly measured through quantification of its hydroxyl groups. On the other hand, potential limitations include potential spectral interference from apatite when undemineralized bone is analyzed and the fact that such analysis requires a custom-made, noncommercially available instrument. Moreover, although fresh bone tissue may be secured during experiments involving animal models, it should be kept in mind that most of the human bone tissue available for spectroscopic analysis has already been dehydrated and embedded in plastic (most often PMMA), as part of standard clinical practice.

There is also an indirect way of estimating tissue hydration during Raman analysis of fixed, embedded, mineralized bone tissue. Since PMMA will infiltrate bone tissue in spaces that were occupied by tissue water prior to fixation, it follows that its relative content may be used as a tissue water surrogate, or nanoporosity within the mineralized tissue [25]. It may be expressed as PMMA/amide III based on the integrated area of the spectral slice ∼494–509 cm−1 (PMMA)/amide III. In healthy females, this parameter is greatly dependent on tissue age, inversely correlates with healthy subject age, has the lowest values in subjects aged 20–45 years [25], and eventually increases in patients diagnosed with postmenopausal osteoporosis. This spectroscopic parameter has been recently shown to be altered in children diagnosed with mild forms of osteogenesis imperfecta [98], in agreement with thermogravimetric analysis results in a mouse animal model [60].

(v) Collagen cross-links

This parameter has been validated against biochemically characterized collagen cross-linked peptides, as well as biochemically analyzed model tissues [26, 38, 99, 100]. It has also been correlated with mechanical properties at the nano- and whole bone organ levels [26, 101].

The organic matrix of bone consists of collagen and a series of noncollagenous proteins and lipids. Some 85–90% of the total bone protein consists of collagen fibers [102]. Type I collagen, the principal component of the organic matrix of bone, as well as other connective tissues, is a large fibrous protein with a highly repetitive amino acid sequence [103] {Gly (glycine) – X – Y}n (often X is proline and Y is hydroxyproline), allowing three polypeptide chains (called α chains; type I collagen is composed of two α1 and one α2 chains) to fold into a unique triple helical structure, consisting of three domains: the –NH2 terminal nontriple helical, the triple helical, and the –COOH terminal nontriple helical domains. The single uninterrupted triple helical domain represents more than 95% of the molecule [103]. The most distinctive feature of mineralizing type I collagen in mineralized tissues is its cross-linking chemistry and molecular packing structure [103]. At present, seven major collagen cross-links have been established as naturally occurring intermolecular cross-links. They are (1) dehydrodihydroxylysinonorleucine (deH-DHLNL), (2) dehydrohydroxylysinonorleucine (deH-HLNL), (3) dehydrohistidinohydroxymerodesmosine (deH-HHMD), (4) pyridinoline (Pyr), (5) deoxypyridinoline (d-Pyr; lysyl analog of Pyr), (6) pyrrole, and (7) histidinohydroxylysinonorleucine (HHL). The first three are NaBH4-reducible (their reduced forms are referred to as DHLNL, HLNL, and HHMD, respectively) and the rest are nonreducible compounds [103]. Cross-linking is initiated with the conversion of the specific peptidyl Hyl, mainly at the C domain of the collagen molecules soon after the molecule is synthesized. These aldehydes are located at the boundaries of the “hole” zones and react with the juxtaposed Hyl on neighboring molecules to form bifunctional cross-links (deH-DHLNL) prior to mineralization. These cross-links are capable of tautomerizing between an iminium and a keto form. When two such cross-links are able to contact each other, they can condense to form a complex cross-link, pyridinoline. As a minor form, the lysyl analog of pyridinoline (deoxypyridinoline or lysylpyridinoline) is present in some tissues. It comprises two Hyls and one Lys and appears to be present mainly in mineralized tissues. The ratio of pyridinoline to deoxypyridinoline cross-links in human bone is 3 or 4 to 1, and their content appears to reach the maximum concentration by ∼10–15 years of age and remain in the same range throughout adult life [103]. Because cross-link condensation reactions, except for the initial oxidation step, are spontaneous, turnover rate is an important factor (but not the sole one) in regulating cross-link maturation [103].

The intermolecular cross-linking affords the fibrillar matrices with mechanical properties such as tensile strength and viscoelasticity. There have been suggestions that not only is the matrix content in osteoporotic bone decreased relative to age- and sex-matched controls [104] but that biochemical alterations in the collagen molecules also exist [105, 106]. The decreased mechanical strength and the change in the thermal stability can be directly related to changes in posttranslational modifications, i.e., lysine hydroxylation and the intermolecular cross-link profile. The increased hydroxylation and change in cross-linking are consistent with increased turnover of the collagen, possibly in an attempt to initiate a repair mechanism which, in fact, leads to an acceleration in the increase in fragility of bone. Increased collagen hydroxylation in osteoporotic as opposed to osteoarthritic femoral heads has also been reported [107]. Hip fracture patients are frequently vitamin B6 deficient suggesting that B6 deficiency may be a risk factor for osteoporotic fractures [108] (Vitamin B6 is an essential co-factor for lysyl oxidase, the enzyme needed to make reactive aldehydes on collagen [109, 110]). The importance of collagen intermolecular cross-links to the mechanical performance of bone is very apparent in lathyrism, where it has been shown in animal models that alterations in enzymatic collagen cross-links, even when restricted to microanatomical locations, are sufficient to influence bone strength [99, 101].

Vibrational spectroscopy has been used previously for analyzing three-dimensional structures of collagen and other proteins. There are many vibrational bands characteristic of peptide groups and side chains from which information on protein structures may be obtained. Among these, the amide I band is especially sensitive to secondary structures [111]. In such studies, information on protein structures is extracted from resolution of broad envelopes consisting of component bands arising from the amide I modes of various secondary structures. This may be achieved through employment of resolution enhancement techniques such as such as second derivative spectroscopy, deconvolution, and differential spectroscopy. We would strongly recommend the use of second derivative spectroscopy as it is mathematically objective (no initial assumptions are required to be made by the operator). This technique allows enhanced identification of small and overlapping absorption peaks that are not resolved in the original spectrum, albeit at a loss in signal-to-noise ratio [112–114]. Using such techniques, spectral parameters have been developed that correlate with biochemically determined enzymatic collagen cross-links (specifically pyridinoline, deoxypyridinoline, and divalent) for FTIR [26, 38, 99], and by extrapolation for Raman spectroscopy [42, 101], and which enable the description of them in a spatially resolved manner, a unique feature of spectroscopic analysis of mineralized tissues. Although spectroscopic theory allows in principle this extrapolation from FTIR to Raman, the considerably lower signal-to-noise ratio of Raman compared to FTIR in the spectral region of amide I should be kept in mind when minor underlying peaks shown in the FTIR amide I spectra to correlate with deoxypyridinoline and divalent enzymatic collagen cross-links are utilized in Raman spectra without prior proper validation. On the other hand, we have recently validated the use of the pyridinoline-correlating subpeak in Raman amide I spectral regions [100].

Based on such analyses, it has been shown that in healthy females, the relative Pyd content of trabeculae is significantly dependent on tissue age while exhibiting a Gaussian distribution as a function of subject age at constant tissue age [25]. Moreover, it was the most discriminant bone quality index discerning between changes due to aging and postmenopausal osteoporosis [25]. The spectroscopically monitored collagen cross-links (and their spatial distribution) may be an underestimated, yet major contributor to bone strength and/or toughness. In cases where bone resistance to fracture cannot be explained by either BMD or architecture, or predicted by finite element analysis, the ratio of pyridinoline to divalent collagen cross-links correlates with fracture incidence [13–15, 68, 115]. Its potential clinical value may be seen in the case of IOP, a rather rare condition in premenopausal women, in which fragility fractures and/or low BMD occur in otherwise healthy women with normal gonadal function [14]. The IOP-diagnosed population includes women with abnormally low BMD and no fragility fractures, as well as women with normal BMD and fragility fractures. Although different techniques (histomorphometry, qCT, DXA, qBEI, biochemical to determine advance glycation endproducts, finite element analysis) are able to discriminate healthy controls and IOP women, to date, only vibrational spectroscopic techniques, and in particular the pyridinoline/divalent collagen cross-link ratio, are capable of further discrimination within the IOP group based on fracture incidence [13–15].

(vi) Relative proteoglycan content

This parameter has been validated against a series of standard proteoglycans, as well as model tissues [116–121].

Proteoglycans are noncollagenous constituents of the extracellular matrix, present in both cartilage and bone among many other tissues. They are characterized by the presence of one or more glycosaminoglycan polymers attached to a protein core via an O-linkage through a serine residue [122]. In bone, they fulfill multiple roles involving the organic matrix assembly, the modulation of both organic matrix mineralization and remodeling rates [123–126], as well as preventing mineralization of the perilacunar matrix around the osteocyte lacunae, and the canaliculi in compact lamellar bone, safeguarding the unhindered interstitial fluid movement [127]. They are subject to subtle posttranslational modifications, some of which are subject and tissue age dependent [128], in size, sulfation, and/or charge density, which are crucial for their specific role [123]. Finally, an important chemical property of proteoglycans is their capacity to swell by binding large amounts of water (through their GAG chains) and fill in spaces [129].

While information on proteoglycans is available with both FTIR [130] and Raman [116] spectroscopic analyses, only Raman (spectral band ∼1375 cm−1; CH3 symmetric deformation of glycosaminoglycan groups) can readily provide this in mineralized tissues [116] as the marker band for proteoglycans in FTIR ∼1060 cm−1 (indicative of sugar rings) is overlapped by the v3PO4 band. To date, vibrational spectroscopic analysis is not discriminant enough between different proteoglycan species, but the Raman spectral signature of proteoglycans arises from the glycosaminoglycan (GAG) chains [116], and in bone, chondroitin 4-sulfate accounts for ∼90% of the total GAG content, while being predominantly encountered in biglycan and decorin [131].

In healthy humans, at actively forming trabecular surfaces, the relative glycosaminoglycan content of bone is strongly dependent on tissue age and shows a weaker correlation with subject age [25]. It exhibits a Gaussian distribution when considered as a function of subject age, and its values are significantly lower in diseases with a prevalence of fragility fractures such as postmenopausal osteoporosis and osteogenesis imperfecta [25, 98].

(vii) Relative lipid content

This parameter has been validated against a series of chemically characterized lipids [132, 133].

Lipids have been identified in the published literature as nucleators of collagen fiber mineralization, revealed to be present in a layer just immediately behind the first deposited bone mineral [134, 135]. They are also part of the matrix vesicle membranes, implicated in endochondral and bone mineralization [136–148]. Although information on their relative content is available during both FTIR [149] and Raman analysis [135] of bone tissue, Raman is better suited as in the relevant spectra a band ∼1300 cm−1 (arising from methylene twisting vibrations) is present, which although relatively weak, is not overlapped by the spectral signature of any other bone constituent [135]. The ratio of the integrated area of this band normalized to the integrated area of the amide III band may be used to describe the relative lipid content in mineralized tissues [28]. Establishing the relative lipid content in bone tissue may be very important for another, yet often overlooked reason. Oxidized lipids serve as a substratum for advanced glycation endproducts (AGEs) formation and accumulation [150]. Since increased accumulation of AGEs has been related with elevated fracture risk [151], it becomes evident that including considerations pertaining to lipid content and their spatial distribution is desirable when attempting to understand factors contributing to the determination of bone strength. In healthy females, in the area of trabecular bone, it has been shown to be dependent on tissue age while exhibiting a Gaussian behavior as a function of subject age, at constant tissue age [25].

(viii) Other parameters

Using vibrational spectroscopic analysis, it is feasible to obtain information on several other bone quality indices.

Some single-detector infrared instruments (mostly using synchrotron radiation) provide access to the v4PO4 spectral region (500–650 cm−1). This spectral region offers the advantage of a much more straightforward (is easier to discern the underlying constituent peaks of the raw band) determination of mineral maturity/crystallinity, calculated either as a peak splitting function calculated as a ratio of the areas above and below the double peak of the phosphate anti-symmetric bending frequency (v4) at approximately 550 to 650 cm−1 [73] or through curve-fitting routines based on second derivative spectral resolution [74], and acid phosphate content (∼530–560 cm−1) [74, 152].

The peak width at half height of the pixel distribution of any of the previously described bone quality indices, often referred to as “heterogeneity” [44], has been reported in FTIR imaging and shown to decrease with bisphosphonate treatment [153] and in bones from patients with fractures as compared to nonfractured controls [154]. This may be an important piece of information in the puzzle of factors leading to bone fragility.

Acid phosphate substitution into the apatite crystal lattice can be measured with FTIR imaging [39] and has been shown to be inversely proportional to crystallinity, thus is expected to be an important determinant of mineralized tissues’ mechanical properties and their response to treatment [64, 155–157]. Using uni- and multivariate analyses of FTIR data from model compounds and spectral images from mineralized tissues, Spevak et al. were able to define two peak height ratios in the v1,v3 phosphate spectral band (namely the 1112/1096 cm−1 and the 1127/1096 cm−1) that provide information on the acid phosphate substitution within these samples. Both of these ratios correlated significantly with the previously published mineral maturity/crystallinity spectral index [37, 158], based on the ration of the 1030 cm−1 and the 1020 cm−1 subpeaks of the v1,v3 phosphate band. Moreover, since acid phosphate substitution is known to be associated with new mineral deposition [155, 159], Spevak et al. suggested that it may be feasible to use images of acid phosphate substitution in place of tetracycline labeling to locate the most recent site of bone deposition [39], although this suggestion has not been validated to date and is unclear how it may be utilized in cases where the acid phosphate content is altered due to either disease and/or treatment.

Nonenzymatic collagen cross-links (AGEs) are evaluable by Raman spectroscopic analysis. AGEs are a diverse group of compounds, formed through the nonenzymatic glycation or glycoxidation of proteins, lipids, and nucleic acids [160] with the best-studied being carboxymethyl-lysine (CML) and pentosidine (PEN) [161–164]. Using Raman spectroscopic analysis of tissues, the relative content of CML and PEN may be monitored based on the integrated area ratio of bands at 1150 (CML) cm−1 or 1495 (PEN) cm−1/1450 cm−1 (methylene side chains (CH2)) [165–170]. In mineralized tissues, AGEs formation and accumulation has been shown to result in more brittle bone that is less capable to deform prior to fracturing, and thought to be a pivotal culprit for the increased fracture incidence encountered in diabetes [151, 170]. In animal models, the spectroscopically determined AGEs have been shown to correlate with inflammation and affiliated bone loss, as well as propagation toughness (a measure of the ability of the bone matrix to resist crack propagation and avoid catastrophic failure) [169, 170]. This measurement is better performed using fresh, nondehydrated or embedded bone tissue, as the fixation and embedding processes usually result in loss of lipids, which may result in subsequent reduction of the AGEs spectroscopic signal below instrumental detection limit [170].

Examples

Due to its excellent spatial resolution (∼1 μm), Raman analysis is not particularly well suited for large imaging analyses, but rather smaller, more anatomically focused one. An example is provided in Fig. 2, in the cortical (endosteal surface) bone compartment of an iliac crest biopsy from a 65-year-old postmenopausal osteoporosis patient who had been treated with teriparatide for 2 years. Raman microspectroscopic instruments can be readily coupled to a fluorescent microscope, thus enabling the selection of areas for analysis based on the presence of double fluorescent labels. Since the protocol for the administration of the double labels is precise, the analyses may be performed as a function of tissue age. (a) Picture through the coupled fluorescent microscope of a Raman instrument, with the tetracycline double labels evident. The green box indicates the area of Raman imaging. Interestingly, the integrated area of the whole, uncorrected Raman spectra accurately shows the presence of fluorescence labels due to the sensitivity of the Raman detectors to fluorescence (b; top image). The tissue water content (b; middle image) and mineral/matrix (b; bottom image) can be readily calculated and, as seen, are inversely related. When an anatomically equivalent bone surface (cortical, endosteal) surface is analyzed (c) that is not undergoing active bone formation (no fluorescent double labels present), Raman imaging analysis shows no presence of fluorescent labels (c; top image), low tissue water content (c; middle image), and high mineral/matrix values (c; bottom image) (Figs. 3 and 4).

Fig. 2
figure 2

a–c An example of information that may be gained through Raman imaging of bone tissue

Full size image
Fig. 3
figure 3

FTIR imaging is well suited for imaging both small and large areas with a spatial resolution of ∼6.3 μm. The advent of faster detectors as well as the evolution of the associated software allows the analysis of a whole iliac crest biopsy surface within 6–8 h depending on spectral resolution used. An example is given in this figure, in which the whole cortex of a thin (∼4 μm) section from the same iliac crest biopsy used in Fig. 2 is analyzed. a Photomicrograph of the cortex. b The spatial distribution of the mineral content (blue = no mineral, red = maximum mineral content). c The spatial distribution of the organic matrix content (blue = 0, red = maximum mineral content). d The spatial distribution of PMMA (blue = 0, red = maximum)

Full size image
Fig. 4
figure 4

An example of further information available through FTIR imaging analysis, in the area of trabecular bone in a thin section from the iliac crest biopsy used in Figs. 2 and 3. a A photomicrograph of the trabecula analyzed. b The calculated spatial distribution of the organic matrix content, showing the lamellar structure. c The calculated spatial distribution of the mineral maturity/crystallinity. d The calculated spatial distribution of the pyridinoline/divalent enzymatic collagen cross-link ratio

Full size image

Potential pitfalls

Much like every other analytical technique, potential pitfalls are associated with vibrational spectroscopic analysis, that the user and/or reader ought to be aware of when selecting specific methods and interpreting the results.

Lasers used in Raman spectroscopy are inherently polarized; thus, certain spectral bands are dependent exclusively on the concentration of the chemical species giving rise to them, while others are dependent on both concentration and tissue organization. The best example of the latter is the most commonly reported mineral/matrix ratio calculated from the v1PO4 and amide I bands, as opposed to the ratio calculated from the v2PO4 and amide III bands [70].

Spectral bands are usually composite ones. Their constituent underlying peaks are unique to specific chemical/biochemical environments. As a result, valuable information may be gained by resolving the raw bands into their constituent peaks. This can be achieved through the utilization of spectral resolution techniques coupled with curve-fitting routines. When engaging in such activity, one should always rely on spectroscopically sound resolution methods such as deconvolution (though minimal initial parameter input by the operator is required) and second derivative spectroscopy (mathematically objective, thus strongly recommended) [38]. Moreover, the calculation and further processing of average spectra (from individual ones acquired over a large tissue area) should be avoided, as this results in smoother average spectra of superior S/N, at the expense of biological variability (which is then underestimated). We also strongly advise against choosing a set number of peaks at specific wavelengths to curve-fitting series of spectra acquired in different tissues and/or different anatomical areas within the same mineralized tissue surface/block/section, especially if organic matrix is being analyzed. Organic matrix composition may be different even between two neighboring 1-μm3 voxels [171]; as vibrational spectroscopic analysis is sensitive to protein secondary structure (thus, its results are greatly dependent on tissue composition heterogeneity), fitting all spectra with a set number of peaks at predetermined wavelengths will result in mathematical solutions that may not necessarily be biologically relevant [38].

As mentioned earlier, the peak width at half height of the pixel distribution of the previously described bone quality indices (during vibrational spectroscopic imaging analysis) is often used as a surrogate of bone tissue heterogeneity for the specific parameter considered [44, 153, 154]. Although tissue heterogeneity is accepted as a major determinant of bone mechanical properties, the reader should always keep in mind that to date, spectroscopically derived heterogeneity is based on analysis of a few, randomly selected areas of the available bone tissue, and no thorough validation exists thus far as to the validity of such an approach (whether the values of these randomly selected areas are adequately representing the heterogeneity of the whole bone tissue available for analysis). Moreover, they treat the distributions as Gaussian ones, even in cases where it is evident that the distribution of values is not normal.

Finally, both FTIR and Raman microspectroscopic imaging methods report on tissue properties that may be dependent on bone turnover (thus changes in mean tissue age are driving potential changes in monitored tissue properties), and/or factors independent of bone turnover [172, 173]. As such, we would advise against random selection of anatomical areas for analysis. Either the whole bone surface should be considered for analysis (a time-consuming endeavor), or microanatomical areas should be chosen for analysis based on robust criteria, such as tissue age or mineralization extent among others.

Conclusions/outlook

Vibrational spectroscopic techniques offer a wealth of unique information on bone material properties and structure. Yet, at the time of this review, they are not amenable for routine use in the clinic, as they require a bone tissue biopsy to be excised from the patient, something that is not a routine diagnostic procedure in the clinical practice. On the other hand, they have proven to be invaluable analytical tools in instances where pathophysiologic mechanisms are investigated, especially when the routinely used clinical outcomes cannot explain or predict fracture incidence, or when the effect of various therapies on bone material properties is investigated (especially involving clinical trials). They are also particularly suitable for studies involving animal models. Attempts are currently underway to further develop these techniques so as to circumvent the need for a biopsy [174, 175], though it should be kept in mind that only the periosteal bone surface would be available for analysis.

Both FTIR and Raman outcomes are thought to be important contributors in the determination of bone’s mechanical characteristics. Thus, it is of great interest to correlate the outcomes of the spectroscopic analysis with those of mechanical tests at the nano-, micro-, as well as the whole organ levels. This is already actively pursued between nanoindentation and vibrational spectroscopic analysis [176–214], although not all of the publications report on analysis of the same area by both techniques. On the other hand, the published reports to date show an excellent correlation between spectroscopically determined bone quality indices, nanomechanical, and whole bone mechanical properties, especially when tissue age and mineralization extent were accounted for when selecting areas of analysis [26, 101]. Although few reports already exist on the subject [83, 84, 170, 188, 215–218], great potential is also envisioned for the correlation of spectroscopically determined bone quality indices with outcomes of microindentation, as the latter is actively being developed to become an everyday clinical tool in the effort to identify patients at increased fracture risk based on mechanical rather than mineral density considerations.

Recently, reports have been published entailing the coupling of an infrared spectrometer to an AFM microscope, allowing the study of bone material properties with a lateral spatial resolution of 20–100 nm [219, 220]. This is a considerable improvement over the 6.3- or 0.6-μm spatial resolution capability afforded by FTIR imaging and Raman analysis, respectively, thus promising novel bone compositional information at the nanometer level. This should become readily achievable once a couple of concerns are addressed regarding the actual spot of analysis (to date, a range rather than a precise value is provided), the partial spectral shift evident in the published spectra (the extent of which seems to be instrument dependent), and whether there are heat-related alterations induced on the analyzed areas.

Great promise also holds for the use of these techniques in combination with other imaging modalities such as time-lapse micro-CT [34] (thus correlating material and structural properties), scanning acoustic microscopy (correlating with mechanical outcomes) [35], and backscattered electron imaging (to obtain the distribution of mineral) [32], offering both structural and material outcomes on the very same bone surface.