Causative or Associative: A Critical Review of the Role of Advanced Glycation End-products in Bone Fragility

Abstract

The accumulation of advanced glycation end-products (AGEs) in the organic matrix of bone with aging and chronic disease such as diabetes is thought to increase fracture risk independently of bone mass. However, to date, there has not been a clinical trial to determine whether inhibiting the accumulation of AGEs is effective in preventing low-energy, fragility fractures. Moreover, unlike with cardiovascular or kidney disease, there are also no pre-clinical studies demonstrating that AGE inhibitors or breakers can prevent the age- or diabetes-related decrease in the ability of bone to resist fracture. In this review, we critically examine the case for a long-standing hypothesis that AGE accumulation in bone tissue degrades the toughening mechanisms by which bone resists fracture. Prior research into the role of AGEs in bone has primarily measured pentosidine, an AGE crosslink, or bulk fluorescence of hydrolysates of bone. While significant correlations exist between these measurements and mechanical properties of bone, multiple AGEs are both non-fluorescent and non-crosslinking. Since clinical studies are equivocal on whether circulating pentosidine is an indicator of elevated fracture risk, there needs to be a more complete understanding of the different types of AGEs including non-crosslinking adducts and multiple non-enzymatic crosslinks in bone extracellular matrix and their specific contributions to hindering fracture resistance (biophysical and biological). By doing so, effective strategies to target AGE accumulation in bone with minimal side effects could be investigated in pre-clinical and clinical studies that aim to prevent fragility fractures in conditions that bone mass is not the underlying culprit.

1. Introduction

The organic matrix primarily influences the inelastic deformation of bone, while the hard, inorganic phase composed of calcium (Ca²⁺)-phosphate (PO₄³⁻) mineral influences elastic deformation [1]. Plasticity refers to the ability of a material to dissipate energy through the generation of permanent damage during inelastic deformation, and post-yield toughness is one mechanical property characterizing this ability. Characterized by an apparent elastic modulus, elasticity pertains to the reversible response of a material to loading. Marking the transition between elastic and inelastic (plastic) deformation, yield strength characterizes the ability of a material to resist permanent deformation. Determination of these and other mechanical properties from a stress vs. strain curve can be found in Section 2.5. In cases of minimal organic matrix content (by mass or by volume fraction), bone is brittle (i.e., it will fracture with little to no deformation after yielding). With the typical amounts of organic matrix (approximately 45% by volume) in human cortical bone that is properly organized into inter-connected collagen fibrils and infused with mineral as well as arranged into multi-scale interfaces to absorb energy and divert cracks, healthy bone can sustain occasional loads that exceed its yield strength. Being a living tissue, the damage generated during an overloading event is repaired by remodeling. Also in homeostasis, it is repaired faster than it can accumulate to cause frank fracture of a bone. When the distribution of microdamage overwhelms the ability of this targeted remodeling to repair bone tissue, as occurs with sudden increases in the intensity of loading (e.g., shift from not running to daily running), stress fractures can occur in otherwise healthy tissue.

From a clinical perspective, osteoporosis ensues from a loss of bone mass or bone mineral density, thereby increasing fracture risk. Higher fracture risk however is not always attributable to lower bone mass. From another perspective then, the end-result of osteoporosis (i.e., a low-energy, fragility fracture) occurs because of an inability of unhealthy bone to repair microdamage via remodeling [2], limit the accumulation of microdamage throughout the tissue, and hinder, if not arrest, the subsequent initiation and propagation of cracks [2]. Why the latter occurs is complex due to the hierarchical arrangement of collagen type I (col I) and other collagens into mineralized collagen fibrils (MCFs), which i) bind to non-collagenous proteins (NCPs), bound water, and extrafibrillar mineral and ii) are arranged in shifting orientations between lamellae [3]. Undoubtedly though, age- and disease-related changes in the organic matrix – whether cell-mediated, biochemical, or biophysical in nature – contribute to the occurrence of fragility fractures.

At present, widely prescribed medications to treat osteoporosis were developed to either inhibit bone resorption, thereby impeding a loss in bone mass (e.g., bisphosphonates, selective estrogen receptor modulators, anti-RANKL antibody) or promote bone formation (recombinant human parathyroid hormone (1-34), parathyroid hormone-related protein (1-34), and anti-sclerostin antibody), thereby countering bone mass loss. Intermittent injections of rhPTH(1-34) or rhPTHrP(1-34) also stimulate remodeling or bone turnover, whereas the affinity of nitrogen-containing bisphosphonates for mineral suppresses turnover. Arguably then, these therapies indirectly affect the organic matrix, but whether there is sufficient turnover stimulation to restore the inherent quality of the organic matrix or sufficient turnover suppression to degrade the organic matrix is clinically unknown. Unknown too is whether the quality of newly formed bone tissue is degraded in age and disease. As such, targeting specific chemical moieties in the extracellular matrix (ECM) of bone requires a better understanding of the biology underlying aging, disease, and drug effects on the organic matrix. Regardless, all current osteoporosis medications primarily decrease fracture risk by acting on osteoblasts including osteocytes (anabolic) or osteoclasts (anti-catabolic), not on correcting deleterious changes that occur within the organic phase with aging and disease progression (Table 1).

Table 1.

A summary of age- and disease-related changes in the organic matrix that are thought to contribute to bone fragility. Relationships derived from studies that tested correlations in human bone, augmented with some evidence from animal studies.

Change In Organic Matrix	Change in Apparent-level Mechanical Properties of Cortical Bone	Also Correlates with	Notes
Increase in AGE crosslinking (PEN) and in fluorescent AGEs	Loss of strength, ductility, toughness, and fracture toughness [7,8,12,34] No correlation: [162]	Loss of LOX crosslinking and increase in AGE adducts [32]	Studies reporting negative correlations between PEN and mechanical properties of bone also report significant correlations with other characteristics.
Increase in AGE adduct content (specifically CML) [12, 34]	Loss of fracture toughness [87,89]	Increase in PEN and loss of LOX crosslinking [32]	This is the latest area of interest for correlation studies between AGEs and bone mechanical properties.
Loss of LOX crosslinking (especially immature crosslinks) or decrease in ratio of immature to mature enzymatic crosslinks [162]	Weak relationship with loss of cortical bone fracture toughness in adult humans [162] Supported by a) the comparisons between young and adult human bone [163,164] and b) by loss of strength and toughness in lathyrism models [128,129]	Increase in PEN and CML [32]	This area was explored early-on and new results redirect attention to immature LOX crosslinks again. Immature LOX crosslinks are more difficult to quantify. Mature enzymatic crosslink content tend to be relatively constant in adults.
Denatured collagen content	Loss of strength and work-to-fracture [130,165]	Increase in PEN [130,165]	It is very difficult to measure denatured collagen without creating more denaturation during extraction
Degradation of network integrity of demineralized bone	Loss of work-to-fracture and fracture toughness [13,86]		Willett et al. [13] were unable to confirm the correlation between thermal stability and J-integral reported by Zioupos et al. [86]
Degradation of collagen in terms of the strength of the demineralized bone	Loss of work-to-fracture [7]	Increase in PEN [7]	Consistent with the loss of nativity and integrity of the bone collagen, the strength of the bone collagen is also degraded.

Over the lifespan of humans, post-yield toughness of cortical bone decreases to a greater extent than the strength (yield or ultimate) of cortical bone at the apparent level (i.e., independent of macrostructure but not microstructure), an observation initially made by Burstein et al. in the 1970’s [4] and subsequently confirmed in the 90’s [5,6] and 2000’s [7,8]. Increase in tissue mineralization combined with an increase in intra-cortical porosity with aging could be a possible explanation for this observation as there is an inverse relationship between the mineral volume fraction and bone toughness across multiple species encompassing a large range in mineral content relative to dry mass of bone [9]. Also, mineral density and porosity are positively and negatively correlated with strength, respectively. However, the mean degree of mineralization does not appear to correlate with age between 20 years and 99 years of age [5,10], presumably because bone remodeling continues throughout the lifespan of humans. In the largest sample of donors to date (99 females and 94 males), analysis of sections of the femur mid-shaft by contact micro-radiography revealed a weak correlation between degree of tissue mineralization and donor age (female only) in which the correlation was negative with mineralization peaking during the 5^th decade when menopause occurs [11]. Thus, hypermineralization is likely not the primary cause of the age-related decline in bone toughness.

The age-related decrease in bone strength at the apparent level is a manifestation of the decrease in apparent bone density, a product of porosity and mineralization. On the other hand, the age-related decrease in toughness is independent of porosity [5,12] and apparent volumetric bone mineral density (vBMD) [13]. Given the importance of col I and other collagens to the toughness of bone, a reasonable supposition then is that the age-related decline in bone toughness, namely post-yield toughness, is due to a loss in the integrity of the organic matrix.

Of the possibilities explaining the age-related decrease in the inherent quality of the organic matrix of bone, the accumulation of advanced glycation end-products (AGEs) has garnered considerable interest with multiple review articles published over the years [14–19]. For one, they accumulate in cortical bone [7,8,20,21] and trabecular bone with age [22], and the presence of AGEs in proteins has been implicated in the pathogenesis of multiple complications associated with diabetes [23,24]. Despite decades of research into the role of AGEs in mechanical properties of bone, inhibition of AGEs with compounds such as pyridoxamine, a B₆ vitamer, or alagebrium chloride (ALT-711), an analog of phenacylthiazolium (PTC), is not actively being investigated in clinical trials to prevent fractures (Table 2). As such, we critically examined the evidence that AGE accumulation contributes to bone fragility with the goal of working toward a comprehensive understanding of their role in the mechanical behavior of bone.

Table 2.

AGE inhibitor drug candidates that to Phase 3 clinical trials

Drug candidate	Clinical trial and outcome	Current status	Refs.
Aminoguanidine	Phase 3 trial in patients with Type 2 diabetic nephropathy; terminated due to toxic side effects	No longer in development	[166]
Pyridoxamine (Pyridorin)	Phase 3 trial in patients with diabetic nephropathy; showed efficacy in patients with early disease stages	Other trials are ongoing1	[167]
Thiamine	Pilot clinical trials in patients with Type 2 diabetes and micro-albuminuria; demonstrated efficacy	Phase 3 trial is not planned	[56,168]
ALT-711	Phase 3 trial in sedentary adults with arterial stiffness; demonstrated no efficacy.	No longer in development2	[169]
TRC4186	Phases 1 and 2 trials in diabetic patients with heart failure; demonstrated efficacy	Phase 3 trial is planned3	[170]

¹A pilot clinical trial “AGEs and bone material strength in Type 2 diabetes” is underway (ClinicalTrials.gov: NCT03778580);

²Thirteen trials involving Alagebrium (ALT-711) (ClinicalTrials.gov: NCT00739687, NCT00516646, NCT00557518, NCT00662116, NCT00089713, NCT00043836, NCT01417663, NCT01913301, NCT01014572, NCT00045981, NCT00045994, NCT00302250, NCT00277875)

³a Phase 3 clinical trial in patients with heart failure, diastolic dysfunction and Type 2 diabetes has been announced (ClinicalTrials.gov: NCT04507347).

2. The interest in AGEs as a potential contributor to bone fragility

Dictated by the mechanical loading environment of the anatomical site and requirements of mineral homeostasis, bone tissue can exist for years to decades before being replaced by the coupled actions of osteoclasts and osteoblasts. Since the formation of AGEs is a result of time-dependent, non-enzymatic reactions, AGE accumulation was an early candidate for the cause of the decline in bone toughness with advanced aging. Then, when the disproportionate increase in fracture risk among patients with type 2 diabetes (T2D) for a given areal bone mineral density (aBMD) was established in epidemiology studies [25,26], AGE accumulation was postulated to contribute to the elevated fracture risk in diabetes [16]. Here, we describe why AGEs, in particular the crosslinking type, are commonly thought to be an important contributor to bone fragility (i.e., osteoporosis).

2.1. AGE formation

AGEs form via enzyme-independent pathways. Since AGE formation involves multiple pathways with many intermediates [27,28], the study of their functional role in pathophysiology is challenging. Broadly, their formation involves reactions between amino groups in protein or lipids, on one hand, and glucose and/or products of glucose degradation and metabolism, on the other [27]. As a result, modifications are made to lysine and arginine residues and to amino groups in certain lipids such as phosphatidylethanolamine [29]. In bone, AGEs accumulate at multiple Lys and Arg sites of col I, other collagens, and various non-collagenous proteins (NCPs) found in the ECM.

Multiple AGEs have been identified over the years and are either crosslinks (i.e., covalent bonds between neighboring proteins or lipids) or adducts (i.e., a modification to the sidechain of a protein or lipid) [30]. While crosslinks bind or interconnect proteins and lipids together, adducts change the structure of the amino acid, which can alter the microenvironment of the modified protein sites and subsequently alter protein function [31]. For example, the formation of N^ε-(carboxymethyl)lysine (CML) adds a carboxyl group to the residual ε-amine of a lysine resulting in a switch from positive ionic charge to negative charge under physiologic pH. Furthermore, there is growing evidence that the formation of AGEs interferes with normal enzymatic post-translational modifications (PTMs). In 2020, Arakawa et al. [32] demonstrated negative correlations between AGE content and the levels of the immature, lysyl-oxidase-derived collagen crosslinks in human trabecular bone supporting the idea by Hudson et al. [33], who analyzed mouse tendons, that glycation hinders enzymatic crosslinking.

AGE crosslinks have received much of the attention among the bone community studying bone quality over the last two decades because of the putative idea that such crosslinks would over-stabilize the collagen and other proteins and result in reduced “plasticity” (stable irreversible deformation) of the collagen fibrils, thereby degrading the inherent toughness of the tissue [34,35]. However, recent work by Arakawa et al. [32], using mass spectroscopy methods, has demonstrated that, of the measurable AGEs, adducts are much more abundant than crosslinks by at least one or two orders of magnitude. Currently, only the non-enzymatic crosslink pentosidine (PEN) is easily quantified as it survives acid hydrolysis and naturally fluoresces [36]. Another AGE crosslink glucosepane (GSPN) is not fluorescent nor easily isolated, but may be more abundant than PEN in bone since it is much more abundant than PEN in skin (Fig. 1) and increases in col I with age (GSPN = 301 x e^{(0.016 x age in years)}) and diabetes [37]. Unlike PEN, GSPN has not been quantified in bone even though increasing levels of GSPN have been found to be more strongly correlated with microvascular diseases than PEN [38].

Figure 1: Relative amount of non-enzymatic modifications to skin col I from the Diabetes Control and Intervention Trial.

AGEs were quantified in skin biopsies of subjects with type 1 diabetes with either conventional or intensive control of glycemia. The amount of glucosepane in collagen I is 70 times greater than the amount of pentosidine in collagen I (Reproduced from Monnier et al. Glycoconjugate J. 2016).

2.2. Different types of non-enzymatic post-translational modifications

PTM is a final step in biosynthesis and conditioning of protein macromolecules. Following translation, the individual amino acid residues within the protein are structurally modified by specific enzymes to optimize protein functionality and/or stability. Involving specific hydroxylases and lysyl oxidases, two examples of enzymatic PTMs are hydroxylation of lysine and proline residues and the formation of lysine and hydroxylysine crosslinks in col I, respectively [48,49]. In addition to enzymatic modifications, non-enzymatic (NE) environmental conditioning of protein structure – particularly via oxidation, nitration, deamidation, halogenation and glycation of amino acid side chains – causes numerous NE-PTMs (Table 3).

Table 3.

Major physiologically relevant non-enzymatic post-translation modifications ^a

Modification	Modifying agent	Targeted residues/groups	Comment	Ref
Fructosyllysine	glucose	Lys, N-terminal and PE amino groups	AGE	[145]
CML	glucose, glyoxal, glycolaldehyde, PUFA	Lys, N-terminal amino groups	oxidative AGE, ALE	[171]
CEL	methylglyoxal, PUFA	Lys, N-terminal amino groups	AGE, ALE	[172]
G-H1	glyoxal	Arg	AGE	[73]
MG-H1	methylglyoxal	Arg	AGE	[59]
Argpyrimidine	methylglyoxal	Arg	AGE	[173]
Oxidation	hydroxyl and superoxide radicals	Cys, Met, Trp	NE-PTM	[174–176]
Halogenation	HOCI, HOBr	Tyr, Trp, Lys	NE-PTM	[177,178]
Nitration	peroxynitrite	Tyr	NE-PTM	[176]
Deamidation	n/a	Asn and Gln	NE-PTM	[53]
MOLD/GOLD b	methylglyoxal, glyoxal	Lys	AGE	[179,180]
Pentosidine b	glucose	Lys and Arg	oxidative AGE	[181]
Glucosepane b	glucose	Lys and Arg	AGE	[182]

^aThis table includes only major physiologically relevant NE-PTMs and is not a comprehensive list of known NE-PTMs. CML, carboxymethyllysine; CEL, carboxyethyllysine; G-H1, 5-hydroimidazolone; MG-H1, 5-hydromethyl imidazolone; MOLD/GOLD, methylglyoxal/glyoxal lysine dimer; PUFA, polyunsaturated fatty acids; HOCI, hypochlorous acid; HOBr, hypobromous acid; PE, phosphatidylethanolamine; AGE, advanced glycation end-products; ALE, advanced lipoxidation end-products.

^bAGE crosslinks

Even though NE-PTMs are formed spontaneously, they preferentially target sites within a given protein structure as dictated by the microenvironment (e.g., pH, oxidative stress, glucose concentration) [50–53]. The presence of low levels of NE-PTMs is indicative of a normal and healthy condition, but further accumulation of NE-PTMs may affect protein functionality and lead to pathology. Long-lived ECM proteins such as col I and other collagens are especially vulnerable to functional damage as they can accumulate more NE-PTMs than proteins with a high turnover rate and are not protected by robust cellular antioxidant defenses.

AGEs constitute a subclass of NE-PTMs that derive from glycation (indirect) and glycoxidation (direct) pathways where glucose and/or its degradation products react with lysine, arginine, and N-terminal residues in proteins as well as with amino groups in lipids (Table 3). Accumulation of AGEs has been implicated in pathogenesis of diabetic complications, including nephropathy, retinopathy, and neurodegenerative disorders [54–56], mobility [57], and in aging [37]. Because formation of many AGEs involves oxidative reactions (Table 3), their accumulation often accompanies inflammatory diseases that are not directly associated with hyperglycemia (e.g., cancers) [58].

2.3. Detecting and quantifying AGEs

Analysis of AGEs under physiological conditions can be challenging because of the relatively low levels of these modifications in vivo. Nevertheless, several analytical approaches have been successfully applied to AGE quantitation in human and animal tissues. Immunodetection techniques (immunohistochemistry, Western blotting, and ELISA) allow for a highly sensitive analysis of AGE localization within tissues and semi-quantitative measurements of AGE levels [59,60]. The variety of commercially available anti-AGE antibodies however is limited. These antibodies are either nonspecific (i.e., recognize multiple AGE structures) or specific to CML, MG-H1 or PEN. More recently, a polyclonal anti-glucosepane antibody has been developed and approbated in aging mouse retinae [61]. The immunodetection techniques do not allow for discovery research as the antibodies are developed against known AGEs.

High-performance liquid chromatography (HPLC) with absorbance and/or fluorescence detection allows for analyses of individual AGEs following tissue hydrolysis. The chromatographic step prior to analysis enables AGE measurements in complex mixtures and improves sensitivity. The technique is usually limited to fluorescent AGEs such as PEN and argpyrimidine [62]. In the bone field, the HPLC assessment of pyridinoline crosslinks (PYD and deoxy-PYD or DPD) and PEN is a variation of the method by Bank et al. [63]. Pre-column derivatization can expand the range of detectable structures [62], but with the advances of mass spectrometry applications, derivatization is now rarely used in AGE analysis.

A more general approach to quantifying total fluorescent AGEs (fAGEs) involves the use of a spectrofluorometer or similar instruction to measure the fluorescence (370-nm Ex, 440-nm Em) of tissue hydrolysates [64]. This technique is often used in conjunction with the more specific HPLC methods because it lacks specificity [65]. It of course cannot detect and quantify non-fluorescent AGEs.

Fourier transform infrared and Raman spectroscopy (FTIR and RS) can detect molecular vibrations of specific chemical groups and are thus sensitive to modifications of proteins due to glycation [66]. In particular, RS has been applied to glycated human bone [67] and quantification of glycation-mediated changes in diabetic mouse bone [68–71]. The technique has the potential to provide non-invasive assessments of AGEs in patients. However, direct detection of individual AGEs in the complex in vivo environment can be challenging due to relatively low sensitivity and specificity [72].

Mass spectrometry (MS) technology has been applied to the analysis of AGEs in tissues using several different approaches: 1) liquid-chromatography tandem mass-spectrometry (LC-MSMS) after exhaustive tissue digestion or hydrolysis, 2) LC-MSMS following digestion with specific proteases, and 3) imaging mass spectrometry (IMS). In the first approach, the total tissue levels of multiple AGEs can be analyzed simultaneously and with very high sensitivity using isotopically labeled standards [73]. In the second approach, the levels of the AGE-modified residues at specific positions within the protein primary structure can be determined [53]. IMS allows for simultaneous determination of tissue localization of multiple AGEs without the use of antibodies [74]. Gas chromatography coupled with mass spectrometry (GC-MS) has also been used in tissue AGE analyses [75]. The MS approaches are amenable to discovery research as the MS data can be interrogated to search for mass-charge shifts corresponding to unknown AGEs.

2.4. Correlations between AGE content and mechanical properties of human bone

In support of the possibility that AGE accumulation contributes to bone fragility, measurements of AGE levels, mainly PEN and fAGEs, in bone matrix negatively correlate with measures of bone plasticity, toughness, and fracture toughness (Table 1). Upon extracting uniform, ‘engineering’ specimens from cadaveric bone, bone fragility can be assessed in terms of the relevant mechanical properties measured using a variety of mechanical tests.

When probing the effects of PTMs in the organic phase, such as AGEs, on the mechanical properties of cortical bone, tests that involve tensile stresses such as bending and some fracture tests are more sensitive because they engage mechanisms provided and controlled by the organic matrix and water [76]. In common mechanical tests of uniform specimens, material properties of bone are derived from engineering stress vs. engineering strain curves (Fig. 2A). The elastic or proportional limit normally marks the onset of permanent deformation that starts once the applied stress is sufficient to cause microdamage and can be detected as the onset of non-linearity by a loss in secant modulus. This is more commonly quantified using the concept of the yield strength (YS) of the material, following the conventional engineering use of the 0.2% offset rule (Fig. 2A). However, bone does not exhibit a clear transition from elastic to inelastic behavior because it is not a purely elastic material. Rather, before the onset of permanent damage, the mechanical behavior is viscoelastic meaning the modulus and indeed the yield strength depend on loading rate, the initial stress vs. strain response is not perfectly linear, and hysteresis, albeit small, occurs during cyclic loading. Post-yield strain and secant modulus characterize the ability of the tissue to deform beyond the yield point and before fracturing when damage irreversibly but stably forms (Fig. 2A). The area under the curve captures the ability of the bone tissue to absorb mechanical work and is most often referred to as toughness (T) or the work-to-fracture (W_f). The maximum stress achieved during the test is the ultimate stress also known as the ultimate strength (σ_u). The strengths of cortical bone are lower in tension than compression (Fig. 2B).

Figure 2. Mechanical testing of cortical bone in tension and compression.

A typical mechanical test (monotonic load-to-failure in tension) generates force vs. displacement data that is converted to an engineering stress vs. strain curve from which mechanical properties are determined. Toughness is the area under the entire curve up to the point of fracture (ultimate stress), while post-yield toughness is the area under the curve after a yield point that is conventionally identified by the 0.2% offset method (intersection of the curve and a linear line with a slope equal to apparent modulus and extending from 0.2% strain) (A). In load-to-failure tests of human cortical bone at 5 mm/min, bone typically has a lower ultimate stress in tension than in compression. To mark the yield point on each curve, the long dashed lines are parallel to the apparent modulus of each test and start from 0.2% strain (B).

Fracture toughness is the resistance of a material to the growth of a pre-existing crack or flaw driven by mechanical stresses [77]. Fracture toughness in materials like cortical bone is best measured in terms of R-curves from which meaningful single point fracture toughness values can be extracted [78–80]. In bending tests of single-edge notched beams (Fig. 3A), K_IC represents the intensity of the stress field around a crack tip required to initiate crack growth [77] and is often calculated using the load at the onset of non-linearity in force vs. displacement curves (Fig. 3B) or the load at the onset of detectable crack growth [81]. J_Ic represents the incremental energetic cost of crack growth at the point of crack growth initiation (Fig. 3C). J- integral (J-int) represents the incremental energetic cost of crack growth at the point of instability or catastrophic fracture after a non-negligible about of stable crack growth [12]. Unlike K_IC which originates from linear elastic fracture mechanics, J_Ic and J-int are derived from non-linear fracture mechanics [77] and have been shown to better represent the fracture behavior of cortical bone because J captures the non-negligible effects of post-yield behavior [82]. As noted above, organic phase modifications, such as PTMs and AGEs, affect the post-yield behavior more than the pre-yield behavior of cortical bone. Therefore, J-based parameters are more sensitive to these changes than are K-based parameters.

Figure 3. Fracture toughness testing of cortical bone.

The single-edge notched beam (SENB) is typically loaded in three-point bending until a crack emanating from a micro-notch propagates through the specimen (A). In the magnified photo, the sharpened notch creates a stress concentration to start crack growth. As the loading rate of human cortical SENBs increases from low and high, the final fracture displacement decreases (B). In the schematic of an R-curve, the determination of J_Ic is based on a construction line (e.g., a slope = yield strength, σ_y, of cortical bone) and characterizes fracture toughness at crack growth initiation while J-int is fracture toughness at fracture instability (C).

Analyzing fresh cortical bone from 30 unique cadaveric femurs, Wang et al. reported that ultimate bending stress (σ_u,B), total work-to-fracture (W_f), and the critical stress intensity factor (K_Ic) to propagate a crack through a uniform beam of bone were all lower in the elderly group (>70 years; 4 males & 6 females) than in the young adult group (19-49 years; 8 males & 2 females) while the concentration of PEN (mmol/mol of col I) was higher in the elderly than the young group [7]. From tensile tests of demineralized bone, the investigators also observed that the failure strength of the organic matrix was also lower with age. Among the 30 donors, PEN negatively correlated with σ_u,B (r=−0.55), W_f (r=−0.62), and K_Ic (r=−0.48); whereas the tensile strength of collagen only correlated with W_f (r=+0.54) [7]. In subsequent tensile testing of cortical bone specimens from cadaveric tibia (8 females 53-90 years and 9 males 49-87 years) by the same lab, the overall tensile toughness and post-yield toughness (i.e., energy dissipated during failure and after permanent deformation occurs) were negatively correlated with PEN (r=−0.602 & r=−0.671, respectively) [8].

As for trabecular bone, Karim and Vashishth analyzed cores from the proximal tibia of 23 donors between 18 and 97 years (11 males and 12 females) [22]. Determining mechanical properties from unconfined compression tests to an apparent strain of 1.1%, PEN positively correlated with yield strain (Spearman r = +0.46) and ultimate strain (Spearman r = +0.44) as did fAGEs content. Correlations between toughness and either PEN or total fAGEs were negative but not significant. In these prior studies, mature enzymatic collagen crosslinks (PYD and DPD) did not correlate with the mechanical properties of bone.

With respect to femoral neck strength, Abraham et al. [83] analyzed cadaveric tibiae and femurs from 28 elderly donors by multiple techniques. Using backwards, stepwise regression, they reported that the combination of areal BMD of the femoral neck (macrostructural level), indentation distance increase from cyclic reference point indentation of the tibia (material level), cortical porosity of the tibia (microstructural level), and fluorescent AGEs of tibia cortical bone (ultrastructural level) were independent contributors to the prediction of the maximum force that the proximal femur could withstand in single stance configuration. Characterizing bone at a similar length scale as the force to break the proximal femur, aBMD had the strongest association with femoral neck strength (r=+0.755). Despite characterizing bone at the nanometer length scale, bulk fAGEs (tibia) was also significantly associated femoral neck strength (r=−0.336).

We conducted a comprehensive study of femoral cortical bone from a general population of 54 human donors [13]. We applied hydrothermal isometric tension (HIT) testing to evaluate the connectivity of the collagen network. HIT testing, while mechanistically relevant, involves decalcification of the tissue and super-physiological temperatures to melt the collagen and generate the measured tension. The connectivity parameter, Max Slope, positively correlates with crosslinking [65,84] and other factors that increase connections within the collagen and negatively correlates with fragmentation of the collagen [81,84,85]. In this study [13], there was a relatively strong, positive correlation between the bone collagen network connectivity measure Max Slope and the fracture toughness of the cortical bone indicating that the degree of connectivity within the collagenous matrix is related to the tissue’s ability to resist crack growth – a finding similar to those of a study by Zioupos et al. approximately twenty years earlier [86]. Notably, in our study, PEN content was not a significant factor in the statistical models when attempting to explain variation in the fracture toughness properties and did not correlate with the network connectivity of the demineralized bone either. This suggests that loss of fracture resistance of cortical bone is at least partially affected by a loss in the connectivity or integrity of the collagenous matrix which is independent of amount of AGE crosslink PEN present in the tissue.

Recently, the hypothesis of a negative relationship between AGE adducts and cortical bone mechanical properties has been tested. Matrix-bound CML correlated with crack propagation toughness of cortical bone as determined for 2 female donors (23 and 97 yo) and 4 male donors (25, 50, 61, 79 yo) [87] in which the mechanical properties were determined in a previous study [88]. More recently, the authors of this review recently found a negative, weak correlation between CML content and fracture toughness in a set of femoral cortical bone specimens from 53 donors [89]. CML increased with donor age (r = 0.560, p < 0.0001), and weakly correlated with transverse crack growth initiation fracture toughness (K_Ic; r = −0.284, p = 0.0293) and transverse J- at crack instability (J-int; r = −0.281, p = 0.0313). In multivariate general linear models (GLMs) with the fracture toughness as the dependent variable, CML content and our bone collagen network connectivity parameter, Max Slope, suggested a small negative contribution for CML and stronger positive contribution for Max Slope to the prediction of the fracture toughness measures.

While PEN content is often found to negatively correlate with the mechanical properties of bone, bone collagen network connectivity appears to be more predictive of cortical bone resistance to fracture, possibly because of generally very low PEN levels and mechanistic contributions from other crosslinks. Also, the non-crosslink CML adduct may contribute to these correlative effects by accumulating in tandem with other AGEs but disrupting normal enzymatic crosslinking (See Section 4).

One key challenge in conducting these types of studies using human cadaveric tissues is the large inter-donor variability. This therefore requires large sets of specimens to provide sufficient statistical power to detect strong correlations between parameters. This may explain differences in findings between studies; some with 10 to 20 donors versus others with 50 or more.

2.5. Age-related decrease in toughness and increase in PEN of rodent bone

Analysis of bones in pre-clinical models of aging support the negative association between mechanical properties and AGEs, namely PEN. Between 6-mo. (or 12-mo.) and 20-mo. of age, the cortical bone of the radius diaphysis lost toughness in male Fischer F344 rats, while PEN and the enzymatic PYD in cortical bone increased between 6-mo. and 20-mo. [90]. Between 6- mo. and 20-mo. of age, the cortical bone of the femur diaphysis lost toughness in male and female BALB/c mice, while PEN and PYD in the same bone increased between these two age groups [91]. In both pre-clinical studies, other changes to bone such as an increase in tissue mineral density and a decrease in bound water were observed. Also, in an early study of demineralized femora from male Wistar rats, the thermal stability, as determined by shrinkage temperature and melting temperature, decreased with age (2-, 5-, 15-, and 25-mo.) [92]. Lastly, we found that the relative abundance of deamidation, a spontaneous, non-enzymatic modification of Asn or Gln, at select sites of col I triple helix negatively correlated with toughness in BALB/c mice and that the resulting age-related increase in negative charge altered hydrogen bonding between col I and water [53]. This is consistent with the accumulation of CML observed in studies mentioned above. Nonetheless, the cause of the apparent decline in the molecular stability and connectivity of collagen I in human bone with age remains unknown.

2.6. Inconsistencies in whether PEN or AGE content increases as toughness of bone decreases in pre-clinical rodent models of type 1 diabetes and type 2 diabetes

The injection of streptozotocin (STZ) to kill islet cells, reduce the production of insulin, and increase circulating blood glucose in growing, male F344 (130 ± 25 mg/dl vs. 436 ± 72 mg/dl) and Sprague-Dawley rats (126 ± 38 mg/dl vs. 410 ± 69 mg/dl) caused an increase in PEN as determined for cortical bone of the femur but did not affect the energy-to-fracture (U_f) of the femur diaphysis subjected to three-point bending [93]. In this study, U_f was the area under the moment (Force x Span / 4) vs. span-adjusted displacement (12 x Displacement / Span²) curve meaning bone structure contributed to its determination, but post-yield displacement of the femur, an indicator of brittleness, was also not lower in the diabetic rats compared to their controls. In fact, the primary effect of type 1 diabetes (T1D) in this pre-clinical model on mechanical behavior of cortical bone was a reduction in the ultimate moment endured by the femur mid-shaft during the load-to-failure bending test. This measurement of bone strength is dependent on the structure and material strength of the mid-diaphysis. The loss in bone strength occurred in both strains of rats and was likely due to the STZ-induced loss of circulating insulin reducing periosteal expansion, thereby resulting in a reduced cross-sectional moment of inertia of the diaphysis [93].

In a rat model of spontaneous T2D (male WBN/Kob rat), fasting blood glucose increased from 100 ± 10 mg/dl to 450 ± 50 mg/dl between 10-mo. and 16-mo. of age as did PEN in the cortical bone of the femur diaphysis [94]. These levels were significantly higher in the diabetic WBN/Kob rats than in the non-diabetic, age-/sex-matched Wistar rats. Coincident with onset of hyperglycemia, the work-to-fracture (Wf) and the ultimate load of the femur diaphysis during three-point bending were lower in diabetic than in non-diabetic animals. However, the decline in the W_f and ultimate load as the duration of diabetes increased (10- to 18-mo.) in the WBN/Kob rats was modest unlike the rise in PEN over the same age range. Moreover, pooling all the WBN/Kob rats from the 10 age groups (1-mo. and 2-mo. to 18-mo. in 2-mo. intervals), PEN was positively correlated with W_f (Pearson r = +0.312) and ultimate load (Pearson r = +0.460). In another rat model of spontaneous T2D in which a high fight diet (HFD) was used to time the onset of hyperglycemia (ZDSD rat), toughness and fracture toughness (K_Ic), but not ultimate bending stress, progressively decreased between 16-wk., when the male rats were switched to the HFD for 7 weeks, and 29-wk. of age, while PEN and the enzymatic PYD in cortical bone increased [95]. These mechanical properties and PEN did not change between 16-wk and 29- wk in the non-diabetic, male CD(SD) rats (PYD similarly increased). Compared to CD(SD) rats with glucose levels of 86 ± 8 mg/dl, the ZDSD rats with glucose levels of 320 ± 51 mg/dl lost weight as they aged suggesting other organs were also being deleteriously affected by the unchecked diabetes.

There are multiple mouse models of T2D, each with advantages and disadvantages in their ability to phenocopy human complications of the disease [96], but few mouse studies have reported differences in PEN or fAGEs. Devlin et al. [97] found a trend toward higher fAGE in the cortical bone of diabetic TallyHO femurs compared to the non-diabetic SWR/J femurs (p=0.074). Blood glucose was variable among the TallyHO mice with some having fasting levels below 250 mg/dl, but on average, the levels were 2.5 times higher than those of SWR/J mice. Three-point bending tests revealed that the femur diaphysis of the TallyHO mice was clearly more brittle than the femur diaphysis of the SWR/J mice with 59% and 48% lower post-yield displacement in the diabetic animals at 8-wk. and 17-wk. of age. In a follow-up study of the brittle bone phenotype in the TallyHO model of juvenile-onset T2D with similar variance in glucose levels (non-fasting: 126 to 205 mg/dl for SWR/J and 206 to 578 mg/dl for TallyHO), Creecy et al. [69] found that the low bone toughness (femur diaphysis) at 16-wk. did not progressively worsen at 34-wk. of age (18 more weeks of hyperglycemia). Moreover, fAGE significantly increased with age but the difference between SWR/J and TallyHO only trended toward being higher in the diabetic animals (p=0.095). PEN, on the other hand, did not depend on the age or strain of the animal. As for the KK-Ay mouse model of spontaneous T2D, PEN in de-embedded proximal femurs was not significantly different between the diabetic A^y/a mice and the non-diabetic a/a littermates at 20 weeks of age, which is approximately 12 weeks of hyperglycemia at euthanasia (A^y/a: 426.5 ± 49.2 mg/dL vs. a/a: 234.5 ± 5.1 mg/dL) [98]. Interestingly, an infrared spectroscopy marker of mean collagen maturity or mature-to-immature crosslinking ratio was 12% higher in the proximal femur from the diabetic mice compared to the non-diabetic mice. Unfortunately, the study did not include mechanical testing of long bones. Lastly, in a diet-induced obesity (DIO) model of T2D, male C57BL/6 mice fed a HFD between 10-weeks and 32-weeks of age developed significantly higher blood glucose levels after 23- weeks of age reaching levels between 260 mg/dl and 340 mg/dl compared to the mice fed a low fat diet with levels between 190 mg/dl and 240 mg/dl [71]. In the femurs, there was an increase and decrease in fAGEs and crack initiation toughness (K_Ic) that negatively correlated (Pearson r = −0.80, p=0.0017) in addition to a negative correlation with interfibrillar spacing of mineral crystals (r = − 0.62, p = 0.032).

In summary, rodent models of diabetes partially support the idea that bone toughness decreases and AGE content in the matrix increases as the duration of hyperglycemia advances over time. Of note, the relationships between mechanical properties and AGE levels of bone depend on i) the degree to which the animals are hyperglycemic relative to control animals, ii) the age of the animal at which glucose levels increase and the duration of hyperglycemia before euthanasia, and iii) whether animals in each group are littermates, whether the animal prone to diabetes carries a mutation, or whether the control animals are similar genetically but a different strain nonetheless.

2.7. Clinical associations between urinary or serum AGEs and fracture risk

There is evidence that patients with osteoporosis have elevated levels of AGEs in their skeleton or in their circulation. In two publications by Saito and colleagues [99,100], discarded proximal femurs were acquired from hemiarthroplasty cases involving a fragility fracture of the femoral neck (females between 72 and 86 years of age) and compared to cadaveric proximal femurs from age-matched donors (females between 69 and 89 years of age). Cortical bone or trabecular bone, depending on the study, was extracted from the femoral neck region, fractionated into low- (1.7-2.0 g/ml) and high-density (>2.0 g/ml) mineral, and then hydrolyzed for collagen crosslink analysis. PEN in the low-density fraction of cortical bone was higher in the fracture case compared to the cadaver control [99], whereas PEN in both the low- and high-density fractions of trabecular bone was significantly higher in the fracture group than in the cadaver, non-fracture group [100].

In a follow-up study from the same group of investigators but involving 432 post-menopausal Japanese women (45-85 years) who were selected from the prospective Nagano Cohort Study after confirming they were not being treated for osteoporosis, baseline urinary pentosidine levels – determined by the same HPLC assay as bone PEN but normalized to mg of Cr instead of mol of collagen and log transformed – was a significant predictor of incident vertebral fractures over 5-year follow up period (5.2±3.3 years) [101]. Moreover, the highest quartile of PEN was an independent risk factor for a vertebral (VB) fracture (fx) after adjusting the Cox’s proportional hazard model for baseline age, lumbar aBMD, and number of previous vertebral fractures (Table 4). In a similar study involving 396 post-menopausal Caucasian women living in France, baseline urinary PEN was significantly higher by ~8% in women who experienced a fracture (vertebral and non-vertebral) than women who did not experience a fracture over the 10-year follow-up (10.1±2.6 years) [102]. However, the association between urinary PEN and increased fracture risk was not significant after adjusting for age, previous fracture, and hip T-score (Table 4). When the Cox proportional hazards model with multiple covariables was limited to 5 years of follow-up to better match the study in Japan, urinary PE was still not an independent predictor of fracture risk (for possible reasons for the discrepancy between the studies, see [102].)

Table 4.

Associations between fracture risk and AGE measurements as indicated by hazard ratio (HR), relative risk (RR), or odds ratio (OR). Red font indicates a non-significant association.

Population	Sample size	Follow up (years)	Predictor	Unadjusted risk ratio (95% Cl)	Adjusted risk ratio (95% Cl)	Fx site	P-value	Ref
Post-menopausal Japanese women	432	5.2±3.3	A log increase in urinary PE	HR: 5.16 (2.95-8.78)	Not reported	VB	<0.01	[101]
			highest quartile of urinary PE over other PE quartiles	Not reported	1.33 (1.01-1.76)a	VB	0.04
Post-menopausal French women	396	10.1±2.6	A log increase in urinary PE	HR: 2.65 (1.23-5.76)	HR: 1.23 (0.54-2.82) b	Any	0.01; 0.62	[102]
			highest quartile of urinary PE over other PE quartiles	HR: 1.55 (0.98-2.48)	HR: 1.16 (0.72-1.90) b	Any	0.06; 0.56
Post-menopausal Japanese women	765	5.1 c	1 SD increase in urinary PE	Not reported	HR: 1.18 (1.05-1.33)d	VB	0.005	[103]
					HR: 1.11 (0.88-1.41)d	Long bone	0.381
Elderly Caucasian and African American men and women (≥65 yo)	3373	9.22 (5.12, 11.42) c	1 SD increase in serum CML	HR: 1.27 (1.16-1.40	HR: 1.25 (1.13-1.38) e	Hip	<0.001;<0.001	[105]
					HR: 1.18 (1.06-1.31)f		0.003
Post-menopausal Japanese women	517	No (−) prevalent fx vs. (+) prevalent fx	1 SD increase in urinary log-PE	Not reported	OR: 1.93 (1.09-3.41)g	Any	0.024	[104]
Elderly females and males (70-79 yo) in the US with and without diabetes	427 (non- DM)	7.5±2.7	1 SD increase in urinary log-PE	RR: 0.97 (0.72-1.30)h	RR: 1.08 (0.79-1.49) i	Any	0.817; 0.630	[106]
				501 (T2D)	RR: 1.50 (1.22-1.85)h		RR: 1.42 (1.10-1.83) i		<0.00; 0.007
Japanese men and women (50-85 yo) with T2D	77 (M)	No (−) VB fx vs. (+) VB fx	1 SD increase in serum PE	Not reported	OR: 0.79 (0.41-1.52)j	VB	0.475	[107]
	76 (F)			Not reported	OR: 2.50 (1.09-5.73)j		0.030
Age-matched females and males in the US with and without diabetes	2332 (non-DM)	Clinical Fx: 10.9±5.2	1 SD increase in log CML	1.07 (0.98–1.16)k	1.03 (0.94–1.13)l	Any	0.16; 0.50	[109]
	712 (T2D)	Clinical Fx: 9.6±5.1	1 SD increase in log CML	1.45 (1.22–1.73)	1.49 (1.24– 1.79)		< 0.001<0.001

^acovariates in model were baseline age, lumbar aBMD, and number of prevalent vertebral fractures

^bcovariates in model were baseline age, T-score of hip aBMD, and number of prevalent fractures

^cmedian follow up with or without interquartile range being reported (otherwise, mean±SD)

^dcovariates in model were baseline age, body weight, diabetes mellitus, lumbar aBMD, prior fracture, and presence of back pain

^ecovariates in model were age, gender, race/ethnicity, and clinic site

^fcovariates in model were age, gender, race/ethnicity, and clinic site plus prevalent coronary heart disease, smoking, body mass index (BMI), alcohol use, level of physical activity, and baseline eGFR

^gcovariates in model were age, gender, body weight and height, aBMD, smoking, alcohol drinking, urine NTx, serum hCRP, diabetes mellitus, and hypertension

^hcovariates in model were age, gender, and race/ethnicity

ⁱcovariates in model were age, gender, race/ethnicity, current smoker, baseline aBMD of hip, baseline weight, weight loss, cystatin-C, HbA1C, and use of vitamin D supplements, calcium supplements, and relevant medications.

^jcovariates in model were age, body weight and height, HbA1C, eGFR, duration of diabetes, duration of postmenopausal state (F), presence of retinopathy or neuropathy, alcohol consumption, smoking, prevalent nonvertebral fractures, lumbar aBMD, and use of insulin/pioglitazone.

^kcovariates in “unadjusted” model were age, race, sex, and clinic site.

^lcovariates in “adjusted” model were age, race, sex, clinic site, current smoking status, total hip BMD, weight, weight loss of 5+ pounds in year before baseline, cystatin-C, A1c, and medication use.

Collecting urine from post-menopausal women in the Nagano cohort (N=765), urinary PEN was associated with an elevated VB fx risk but not an elevated long bone fx risk (Table 4). Based on receiver operating characteristic (ROC) analysis, there was a moderate improvement in risk classification when urinary PEN was added to 10-year probabilities of a fragility fracture: C statistic = 0.690 (95% CI: 0.639-0.741) using FRAX alone to classify vertebral body fracture vs. C statistic = 0.732 (95% CI: 0.686-0.778) combining FRAX and PEN [103]. In the most recent analysis involving the cohort of post-menopausal women in Japan, urinary PEN, as determined by ELISA instead of HPLC, was significantly associated with the prevalence of fracture before and after adjusting for known risk factors (Table 4) [104].

Circulating PEN is an imprecise marker of AGE content in bone because the urine and serum levels depend on resorption activity. To the best of our knowledge, there isn’t a published study testing the ability of combining markers of AGEs, bone resorption (CTX), and/or bone formation (P1NP) to predict fracture risk.

Pentosidine is not the only AGE to be associated with fracture risk. Applying an ELISA approach to the measurement of CML, a non-crosslinking AGE, in serum from participants (N=3373) in the US-based, prospective Cardiovascular Health Study, CML was significantly associated with increased risk of a hip fracture with and without adjustment for known risk factors (Table 4) [105]. Like urinary PEN (Fig. 4), the proportion of subjects who remained fracture free decreased as the quartile of CML at baseline increased (i.e., going from the lowest CML levels in the <25^th percentile to the highest CML levels in the >75^th percentile).

Figure 4. Kaplan-Meier curves of fracture occurrence for different quartiles of AGEs.

In three independent studies, those individuals with a circulating AGE level in the highest quartile (>75^th percentile) were much more likely to experience a fracture over a 10-year period than individuals in the lowest quartile (<25^th percentile): (A) Proportion of subjects without a hip fracture vs. time to follow-up for serum CML (Reproduced from Barzilay et al. J Bone Miner Res. 2014;), (B) Fraction of subjects without vertebral or peripheral fracture vs. time to follow-up for urinary PE (Reproduced from Gineyts et al. Osteoporos Int. 2010), and (C) Cumulative incidence of a vertebral fracture vs. time to follow-up for urinary PE (Reproduced from Shiraki et al. J Bone Miner Metab. 2008).

An association between pentosidine and elevated fracture risk has been observed in the elderly with T2D. Using the same HPLC analysis as Gineyts et al. [102], Schwartz et al. [106] compared urinary PEN between 427 subjects without diabetes (73.4±2.9 years of age) and 501 subjects with T2D (73.6±2.9 years of age). There was no difference in urinary PEN between the 2 groups with or without adjusting for age, gender, kidney function (eGFR), and previous weight loss. However, in logistical regression models, the log of PEN was significantly associated with a higher fracture risk but only for those with diabetes (Table 4). The significant relative risk (RR) in the T2D group and the non-significant RR in the non-diabetes group remained nearly the same after adjusting for additional factors such as glucose-lower medications and osteoporosis medications.

When serum PEN was measured in 77 Japanese men (50-80 years of age) and 76 Japanese post-menopausal women (50-85 years of age) with type 2 diabetes (HbA1c = 9.1±2.5% and 9.1±1.8%, respectively), there was not a significant difference in PEN between those men without vertebral fractures and those with a VB fx (Table 4). However, serum PEN was significantly higher in women with a VB fx than in women without a fracture [107]. The 28 men with vertebral fracture(s) were older than the men without a fracture, but the duration of diabetes was not significant between the two groups. The women with a VB fx, on the other hand, were not only older but had diabetes for a longer time (18.0±11.4 years) than women without a fracture. Upon adjusting the logistic regression for risk factors, serum PEN was significantly associated with the presence of vertebral fractures in women but not in the men (Table 4). In a similar study with a small sample size (7 males and 33 females), serum PEN was 72% higher in the VB fx group than in the non-fx group (p=0.04) [108]. In a recent study of fracture risk in T2D with the largest cohort to date [109], 1 SD increase in log CML (serum) was associated with ~1.5 times higher clinical fracture incidence for adults with T2D but not adults without diabetes (Table 4). Baseline CML was significantly higher in T2D than in non-diabetes. Once confined to prevalent VB fx, there was no significant association between log CML and fracture risk [109]. For those interested in how AGEs may cause bone fragility in chronic kidney disease, see the review article by Damrath et al. [110].

For the most part, clinical measurements of PEN in the previously cited studies either did not correlate or weakly correlated with aBMD (hip or spine), resorption marker (CTX or NTX), and duration of diabetes. PEN correlated with the age of the subjects, but there was considerable scatter in the relationship, which was not always linear. The fracture risk marginally increases with an increase in circulating PEN (Table 4), but this increase does not appear at the present time to be clinically meaningful since PEN is currently not part of the diagnosis of osteoporosis. A high PEN or CML value though is concerning as the likelihood of remaining fracture free is significantly reduced when the level is higher than 75^th percentile (Fig. 4). Prospective studies likely need to demonstrate that an AGE marker adds value to the prediction of fracture risk by FRAX before serum or urinary measurements of PEN or CML become the standard of care in diagnosing osteoporosis regardless of whether a patient has diabetes.

2.8. Diabetes-related changes in bone that coincide with AGE accumulation

There are multiple diabetes-related changes to bone that possibly portend a loss of fracture resistance. For example, five different groups reported that the ability of the tibia mid-diaphysis to resist impact micro-indentation using an instrument called the OsteoProbe was lower in adults with type 2 diabetes than those without T2D [111–115]. While the determinants of Bone Material Strength Index (BMSi), the primary measurement of the OsteoProbe, are debatable [116,117], Samakkarnthai et al. [118] observed negative correlations between BMSi and skin AGEs (via autofluorescence) in 171 subjects with T2D (r=−0.30, p<0.0001) and 108 age-matched subjects without diabetes (r=−0.23, p=0.020). In a previous study of postmenopausal women with (N=16) and without T2D (N=19), BMSi was only correlated with skin autofluorescence in the T2D group (r=−0.65, p=0.006 compared to control: r=−0.24, p=0.92) [112]. As another example, two studies acquired discarded proximal femurs from patients who elected to undergo total hip arthroplasty, due to osteoarthritis of the hip, so that differences in cortical and trabecular bone between non-T2D (HbA1c < 6.5%; n≥19) and T2D (HbA1c > 6.5%; nࣙ12) could be ascertained [119,120]. In both studies, fluorescent AGEs (fAGEs) in the trabecular bone were not significantly different between the 2 groups. However, fAGEs in cortical bone [119] and PEN in trabecular bone [120] was significantly higher in T2D than in non-T2D subjects. In one study, trabecular fAGE was positively and negatively correlated with yield stress and post-yield displacement of trabecular bone (r=+0.56 and r=−0.66, respectively) [119], while in the other study, trabecular fAGE was not a significant predictor of ultimate stress and post-yield strain of trabecular bone [120]. The latter study involved stepwise linear regression in which trabecular BV/TV, mineral-to-matrix ratio, and age were positive contributors to the prediction of ultimate stress (adj-R²=0.694) and age, matrix maturity ratio, and PEN were negative contributors to the prediction of post-yield strain (adj-R²=0.116). The challenge in quantifying the relative contribution of AGEs to bone fragility is the multiple, hierarchical characteristics that all simultaneously contribute to the toughness and strength of bone.

3. Effect of in vitro glycation/ribation models on the mechanical properties of bone

One approach to the challenge of quantifying the relative contribution of AGEs to bone fragility has been to conduct in vitro experiments with controlled variables and conditions. The most common approach involves the incubation of bone specimens in physiological solutions (phosphate buffered saline, PBS, Hanks’ balanced salt solution, and HBSS, simulated body fluid, SBF) with and without reducing sugars, such as glucose and ribose. See Table 5 for a comprehensive listing of studies. These reducing sugars diffuse into the matrix of the bone and form a multitude of reaction products (i.e., adducts, crosslinks) through the process of non-enzymatic glycation (NEG; aka ribation or glycation). Commonly used ribose as well as the less reactive glucose leverage the Maillard reaction pathway [121]. Incubating bone in ribose at neutral pH readily forms a robust accumulation of PEN and fAGE relative to controls [64,65]. After the incubation process, glycated/ribated specimens are mechanically tested along with their incubated controls, and sometimes with non-incubated ‘native’ controls [65,122] to elucidate the effect of NEG on the mechanical behavior of bone. Widely used over the last two decades (Table 5), this approach continues today.

Table 5.

The effect of in vitro glycation on the mechanical properties of bone. Studies reporting no effect of glycation on a mechanical property shaded in blue grey.

Species (age)	Incubation protocol (buffer, pH, anti-microbial)	Effect on AGEs	Effect on mechanical properties a	Ref
Cortical	bone
Human tibia (34-to 85-yr.)	0.6 M ribose (HBSS with 30 mM HEPES, pH 7.2-7.4, PI) for 7 days at 37 °C	104% higher fAGEs	No significant difference in longitudinal crack initiation toughness	[88]
			35% lower longitudinal crack growth toughness (compact tension)
			42% reduction in failure strain, 72% lower creep rate, and 40% lower residual strain for stress levels > 100 MPa in multicyclic creep tests
Human tibia (57-to 97-yr.)	100 mM glucose (HBSS with 10 mM HEPES, pH 7.2-7.6, PI & 100 μg/mL streptomycin) for 3 and 7 days at 50 °C	175% higher fAGEs after 7 days of incubation	~40% higher indentation distance increase (IDI) after 3 days	[151]
			~18% higher IDI after 7 days
			No other cyclic reference point indentation (cRPI) properties reported
Human tibia (57-to 86-yr.)	0.6 M ribose (HBSS with 30 mM HEPES, pH 7.2-7.6, PI); 10 days at 37 °C	136% higher fAGEs	56% lower transverse crack initiation toughness (single edge notched beam)	[183]
			55% lower microindentation modulus
			No differences in cRPI properties including IDI
Bovine tibia (20-mo.)	0.67 M ribose (HBSS pH 7.3-7.6, 100 mL/L of toluene & chloroform & 0.5 mg/mL of gentamicin) for 15 days at 37 °C	Whitish compared to brownish color	No significant differences in elastic modulus and yield stress, but higher secant stiffness	[184]
			Lower post-yield strain and damage fraction
			No significant difference in J-integral but lower plastic component of J-int with quasi-static loading rate
			No significant differences in the elastic and plastic component of the J-int under fall-like loading rate
Bovine tibia (1.5- to 2-yr.) irradiated at 33 kGy	1.8 M ribose (PBS, pH 7.4, NS) for 24 h at 60 °C	No effect due to irradiation 100-fold increase in PEN	Relative to irradiated control, no significant difference in modulus	[185]
			10% and 20% higher yield and ultimate stress
			4% and 39% higher yield and failure strain
			73% higher work-to-fracture (three-point bending tests of machined specimens)
Bovine femur /tibia (18-mo.)	0.67 M ribose (HBSS b NS c, 10 mL/L of toluene & chloroform & 0.5 mg/mL of gentamicin) for 3, 8, 11, 17, 29, 38 days at 37 °C	Whitish compared to dark brownish color fAGE = 1416 × [1– e−time/12]	No significant difference in compressive & tensile modulus, regardless of time	[64]
			Tensile properties after 29-days
			4.2% higher yield strain
			9.3% higher yield stress
			12.3% lower DF d
			no difference in post-yield strain
Bovine femur (3-mo.)	0.2 M ribose (PBS, pH neutral, 1% penicillin-streptomycin & 1% fungizone & PIf) for 15 days at 37 °C *	PEN not detected in control and ~250 mmol/mol in ribose	No significant differences in yield stress, yield strain, ultimate stress, ultimate strain, post-yield strain, energy to yield, and post-yield energy to fracture (three-point bending of machined specimens), and micro-hardness.	[186]
Bovine femur (18-mo.)	0.67 M ribose (HBSS, NS, 10 mL/L of toluene & chloroform & 0.5 mg/mL of gentamicin) for 21 days at 37 °C	NS	No significant difference in thermal stability of mineralized bone. (Significantly higher melting temperature in ribated bone after demineralization in 45% formic acid)	[187]
Bovine metatarsal (18-to 24-mo.)	0.6 M ribose (SBF g with HEPES h, pH 7.4, 10 mL/L of toluene & chloroform & 0.5 mg/mL gentamycin) for 14 days at 37 °C	88% higher fAGEs PEN not detected in control and ~16 mmol/mol in ribose	No significant differences in modulus, yield stress, ultimate stress, and yield strain.	[65]
			26% and 27% lower post-yield strain and failure strain
			21% and 29% lower toughness and post-yield toughness
			11.5% and 11% DF and post-yield strain/failure strain (3pt bending of machined specimens)
Rat femur/ tibia (4-mo.)	0.2 M glucose (PBS e, pH 7.4, 0.1% sodium azide) for 60 days at 29 °C *	Whitish compared to brownish color	No significant differences in ultimate force, stiffness, displacement, energy-to-yield, and energy-to-fracture (three-point bending of diaphysis).	[188]
Trabecular	bone
Human proximal femurs (42-to 97-yr.)	0.6 M ribose (HBSS with 30 mM HEPES, pH 7.3-7.6, PI) for 7 days at 37 °C	Whitish compared to brownish color; 90% higher fAGEs	No significant differences in apparent modulus, yield strain, yield stress, ultimate strain, and ultimate stress 92% lower post-yield strain energy and 38% lower DF (compression tests of cores)	[189]
Human tibial plateau (64-yr.)	0.6 M ribose (HBSS with 30 mM HEPES, pH 7.3-7.6, PI) for 7 days at 37 °C	NS	No significant differences in apparent modulus at either 0.6% strain or 1.1% strain.	[190]
			No difference in toughness at 0.6% strain
			30% lower toughness at 1.1% strain (compression tests of trabecular cores).
Porcine mandibular condyles	0.2 M ribose (PBS; pH 7.4; 0.1% sodium azide & PI) for 15 days at 37 °C *	200% higher PEN	No significant difference in tissue-level modulus (nanoindentation on dry trabeculae).	[191]

^aHigher or lower change with glycation reported as 100 × (sugar – control) / control

^bHanks’ balanced salt solution (HBSS) is typically Hanks buffer and 1.3 mM CaCI₂

^cNot specified in the paper (NS)

^dDamage fraction (DF) is 1 – secant modulus at failure / initial modulus

^ePhosphate buffered saline (PBS)

^fProtease inhibitor (PI) cocktail is typically either a tablet (e.g., Roche Diagnostics) or 25 mM ε-amino-η-caproic acid, 5 mM benzamide, and 10 mM N-ethylmalemide

^gSimulated body fluid (SBF)

^hHEPES is a zwitterionic buffer

^*changed incubation media every day, every 3 days, every 4 days (volume not always specified), 5 times per week

3.1. Findings from in vitro AGE accumulation studies involving bone

In the first study to use in vitro ribation, Vashishth et al. [64] incubated bovine cortical bone specimens in 0.67 M ribose solution and found that the treatment increased fAGE, stiffened the bone col I (measured after decalcification) and led to a reduction in the ability of the tissue to plastically deform before fracture (detected as a ‘damage fraction’ parameter based on the secant modulus, Fig. 2A). Over the two decades since this publication, variations of this approach have been applied to both human and animal bone specimens leading to mixed and inconsistent results (Table 5). Confounded by differences in buffered solutions, concentration of sugar, incubation time and temperature, species of bone, antimicrobial agents, and in mechanical testing, studies either report no effect of AGE accumulation on the mechanical properties of bone or report a significant decrease in post-yield (plastic) properties of bone (Table 5).

Despite discrepancies in the in vitro glycation/ribation studies (Table 5), the consensus includes suppressed plasticity (Fig. 5A), suppressed ability to sustain damage stably (damage fraction based on the secant modulus at fracture), and suppressed fracture toughness (Fig. 5B). Interestingly, these effects seem to depend on the development of extensive browning of the ribated specimens.

Figure 5. The mechanical effects of the non-enzymatic glycation/ribation model on bone.

The effects of the non-enzymatic ribation model on mechanical behavior of bovine cortical bone tissue tested in three-point bending (A). Note the loss of post-yield strain in the Ribated group. (Reproduced from Willett et al. Bone. 2013). Schematic of the effects of the non-enzymatic ribation model on the R-curve of cortical bone (B), based on measurements reported in Jia et al. JMBBM 2021.

3.2. Mechanism by which in vitro glycation/ribation affects bone tissue

In the Willett et al. study (Table 5), ribation of bovine cortical bone significantly decreased post-yield strain (Table 5) while it increased PEN content and thermomechanical measures of collagen network connectivity, which are dependent on crosslinking and measured by hydrothermal isometric tension testing (HIT). Interestingly, within the ribated group, PEN and half-life of stress decay during HIT testing were positively correlated with measures of strain accommodation (post-yield strain) and energy absorption (work-to-fracture) before failure [65]. Clearly, the non-enzymatic ribation model of AGE accumulation reduced the cortical bone tissue’s toughness through a reduced capacity for post-yield strain accommodation. However, the positive correlations suggest that increased crosslinking may not provide a complete explanation for this ‘embrittlement’.

The AGE-related reduction in post-yield strain of cortical bone is often attributed to increased pathological inter-helical (helix-to-helix) collagen crosslinking that suppresses bone tissue plasticity [16,123]. Similar conclusions have previously been reached for soft tissues in which tendon viscoelasticity [124] and fibril-level sliding and plasticity mechanisms [125] are inhibited. Interestingly, work on individual collagen fibrils from tendons treated with methylglyoxal, a potent reactive metabolite that produces AGEs, demonstrated that treated fibrils were stiffer, stronger and less viscoelastic, but not necessarily more brittle [126]. Unfortunately, it is not clear how this translates to bone tissue, which is composed of mineralized collagen fibrils with presumably different mechanics due to their different nanocomposite structure and hierarchical organization.

Unlike normal collagen crosslinking catalyzed by the lysyl oxidase (LOX) enzyme, which forms crosslinks from a telopeptide to an helical site [16], it was thought that AGE crosslinks form between helical sites and prevent intra-fibrillar sliding. Computational models have been developed to explain this mechanism in which enzymatic crosslinks are at the terminal end of each collagen triple helix and non-enzymatic crosslinks are at random sites between collagen interfaces conferring and reducing inelastic deformation, respectively [123]. However, Chiue et al. [127] found that incubating rat tail tendons in ribose (<30 mM) mainly forms crosslinks at the same sites as the LOX crosslinks (i.e., telopeptide region to helical region), supporting the idea that AGEs can hinder enzymatic collagen crosslinking [32,33] and strengthening concerns regarding the role of NEG/AGE crosslinks in the age- and diabetes-related decline in fracture resistance. Greater certainty regarding the amount and location of AGE crosslinks and adducts in the organic matrix of bone is needed to inform such computational models of bone’s mechanical behavior. This calls for standardization of protocols to modify AGEs in bone and to detect different AGEs with respect to their location.

3.2.1. Insufficient control precludes accumulation of known AGEs

In addition to the questions regarding the location and consequence of the crosslinks formed during incubation of bone in ribose, this model of AGE accumulation is not sufficiently controlled to isolate the effects of specific crosslinks or adducts or other products alone. The spontaneous NEG reactions form a multitude of products that are difficult to measure. It is unclear which, if any, adducts are formed with ribation and what their effects are on the mechanical behavior of the tissue. In vivo, the formation of adducts, such as CML, may result in meaningful alterations in the profile of enzymatic and non-enzymatic collagen crosslinks [38] and in the charge distribution along the surface of a collagen molecule [59], thereby disrupting critical ionic and hydrogen bonding interactions among col I, mineral, water, and NCPs [53]. Certainly, the in vitro model does not recapitulate a loss of total crosslinking. In fact, total crosslinking increases. It is unknown if the surface charge of the col I and other collagens is modified by in vitro ribation.

Furthermore, Garnero et al. [122] noted that col I in ‘younger’ bovine bone tissue will mature when incubated in vitro in similar conditions to the controls in the ribation model. They used fetal bovine cortical bone and detected increased mature enzymatic collagen crosslink levels along with increased PEN levels. Willett et al. [65] also noted that, in bovine specimens from older steers than those used by Garnero et al. [122], the mature pyridinoline crosslinks (PYD and DPD) increased in both incubated controls and ribated specimens relative to non-incubated controls. In both studies, the bone tissue gained toughness due to incubation alone. Notably, this is consistent with the results of lathyrism studies that demonstrate the importance of the mature enzymatic collagen crosslinks towards bone tissue strength and toughness [128,129].

3.2.2. In vitro models are not consistent with the pathology

The in vitro model of AGE accumulation increases crosslinking, stabilization, stiffening, and strengthening of col I and other collagens in bone. Using differential scanning calorimetry and hydrothermal isometric tension (HIT) testing, incubating bovine cortical bone in 0.6 M ribose for 7 days at 37 °C (Table 5) increased thermal stability and connectivity of the organic matrix respectively [65]. This increase in stability and connectivity is not consistent with studies of human and primate cortical bone over a wide range of ages. Early studies of cadaveric bone found that the collagen network degrades. For example, Wang et al. observed that an age-related loss of tensile strength coincided with increased susceptibility to proteolysis by α- chymotrypsin, which is indicative of collagen network degradation and denaturation [7,130]. In the analysis of baboon femurs (5-26 years of age) and human femurs (19-89 years of age), the fracture toughness of the bone tissue negatively correlated with the percent of denatured collagen [130,131]. In their analysis of cadaveric femurs (35-92 years of age), Zioupos et al. found a negative correlation between network connectivity of demineralized bone (measured using a technique similar to HIT) and age and a positive correlation between the network connectivity and work-to-fracture as determined by three-point bending tests of uniform samples [86]. We found similar results in our recent study [13] (mentioned in Section 2.4) in which there was a relatively strong, positive correlation between bone collagen network connectivity measures and the fracture toughness of human cortical bone.

Based on this correlation, increased collagen network connectivity following AGE accumulation by non-enzymatic ribation could be expected to improve strength, plasticity and fracture toughness, but it does not. The non-specific accumulation of AGEs in in vitro glycation studies may promote and degrade the toughness of bone such that the observed difference, or lack thereof, in the mechanical behavior of bone depends on incubation conditions. Therefore, the findings from in vitro ribose-based, non-enzymatic glycation models should be treated with caution. They have led to a putative but possibly incorrect understanding of the role of AGEs in the pathology of bone fragility. There are important limitations and concerns that must be considered and addressed. There is a gap between the in vitro models and the known in vivo pathology. This represents an opportunity for the development of new in vitro models that are more pathophysiology relevant and mimetic.

4. The relative sparsity of AGE crosslinks and abundance of non-crosslink AGE adducts

Largely due to the in vitro ribation model and the elevated levels of PEN in aging and diseases, such as diabetes, the field has been focused on the concept of extraneous crosslinking-related suppression of bone tissue plasticity. Unfortunately, there is little empirical evidence that the effect observed in the model is consistent with the in vivo mechanism, as discussed above. Furthermore, and until recently, there has been a challenging issue with the quantity of the AGEs measured, specifically PEN [132].

The amount of LOX crosslinks in mature human bone is on the order of 1 to 1.3 moles per mol of collagen [32,133]. Of this, 1 mol per mol of collagen are the immature, divalent, LOX-derived crosslinks, and 0.1 to 0.3 mol per mol are the mature, trivalent enzymatic crosslinks. Strikingly, PEN content in bone is two to three orders of magnitude lower, in the 0.001 to 0.010 mol per mol range, than these enzymatic crosslinks. This low amount and the evidence that PEN may primarily form at the same sites as LOX crosslinks, rather than interconnecting helical regions of two neighboring col I molecules [127], challenges the idea that PEN plays a clinically meaningful role in bone fragility and suggests that it may simply be indicative of other AGEs.

Recently, a comprehensive mass spectrometry study by Arakawa et al. [32] demonstrated again that PEN content in bone is very small compared to the LOX crosslinks. Furthermore, PEN was orders of magnitude less abundant than four AGE adducts (Fig. 6). These AGE adducts were found to strongly correlate with PEN in a positive manner and occur at levels comparable to the LOX crosslinks. Therefore, while PEN, and perhaps GSPN, likely play a minor mechanistic role in bone fragility, PEN may be indicative of other AGEs that are not necessarily crosslinks but can affect bone mechanics. This alternative mechanism which does not involve crosslinks has yet to be tested.

Figure 6. AGEs and enzymatic crosslinks in human cancellous bone from lateral tibia plateau.

The filled bar and error bar reflect mean value and standard deviation. Abbreviations: Hyp, hydroxyproline; CML, Nε-(carboxymethyl)lysine; CEL, Nε-(carboxyethyl)lysine; MG-H1, Nδ-(5-hydro-5-methyl-4-imidazolon-2-yl)-ornithine 1; CMA, Nω-(carboxymethyl)arginine; DHLNL, dihydroxylysinonorleucine; HLNL, hydroxylysinonorleucine; LNL, lysinonorleucine; PYD, pyridinoline; DPD, deoxypyridinoline. (Reproduced from Arakara et al., Scientific Reports, 2020;)

Interestingly, Arakawa et al. demonstrated negative correlations between AGE adduct content and LOX crosslink content [32]. If AGE adducts compete for sites of enzymatic crosslinking on col I and other collagens (lysine, arginine), they could cause a net loss of enzymatic crosslinking (or maturation) and an overall decline in the stability, connectivity and force-carrying capacity of bone’s organic matrix, similar to those observed in human bone (Table 1) and to the effects measured in lathyrism models in which LOX crosslinking is inhibited [128,129]. One challenge to this hypothesis is that these AGE adducts would need to form early in the synthesis of new osteiod (i.e., bone collagenous matrix before mineralization) in order to compete with LOX for these reactive sites. At this time, the reaction kinetics of the formation of AGE adducts are poorly understood. Furthermore, it might require many years of deposition of compromised osteoid via osteonal remodeling before the mechanical properties of the bone tissue would be compromised enough to lead to bone fragility and perhaps fracture. Bone turnover in diabetes is thought to be slower than normal [134–136]. This may partially explain the gradual time course of bone fragility in diabetic patients and other disease groups.

In summary, measurable AGE crosslinks are not sufficiently abundant to affect bone fragility while non-crosslink AGE adducts may be. (Note: the levels of the most abundant physiological crosslink glucosepane have not yet been measured in bone.) The mechanisms of bone collagen stability and connectivity degradation do not appear to be directly related to PEN or CML but may be linked to a) loss of total LOX crosslinking due to competition with AGE adduct formation and b) other forms of collagen network degradation.

5. Mechanisms of AGE inhibition and challenges in clinical application of AGE inhibitors

Extensive AGE research over the past several decades resulted in the following proposed mechanisms of AGE inhibition:

1. Blocking adduction of glucose to lysine residue either via scavenging glucose or through protection of protein amino groups (e.g., polyamines [137] and aspirin [138], respectively).

2. Blocking oxidation of Amadori intermediate to AGEs (e.g., pyridoxamine [139]).

3. Scavenging of reactive dicarbonyl compounds derived from glucose autoxidation or oxidation of glucose protein adducts (e.g., aminoguanidine [140], pyridoxamine [141]).

4. Scavenging of reactive oxygen species (e.g., pyridoxamine [142]).

5. Breaking the existing AGE crosslinks (e.g., N-phenacylthiazolium bromide [143]).

However, since practically all the mechanistic studies have been conducted in vitro with or without cultured cells, it remains to be seen whether these mechanisms are also present in vivo. For example, it has been suggested that most AGE inhibitors in vivo act primarily via chelation of redox-active transition metal ions, mainly iron and copper [144]. Also targeting an intermediate of glycation reactions is a potential strategy to inhibit AGE accumulation in vivo. When glucose reacts with lysine, the resulting adduct undergoes intramolecular rearrangement forming the Amadori product, a key intermediate in AGE pathways [145]. An enzyme like amadoriase can de-glycate lysine residues and thus could be an in vivo treatment to inhibit AGE accumulation [146].

Dozens of AGE inhibitors have been proposed as drug candidates based on in vitro and animal studies, including synthetic compounds, phytochemicals, and plant extracts (for recent reviews, see [24,147]). In preclinical studies, inhibition of AGEs by pyridoxamine has been shown to inhibit AGE levels in skin and kidney while limiting progression of renal disease in diabetic rats [148]. Similarly, benfotiamine treatment inhibited retinal AGE levels and ameliorated diabetic retinopathy in the rat model [149]. Yet, despite significant research and clinical efforts in the past 35 years, only a few AGE inhibitors progressed to Phase 3 clinical trials (Table 2) and none of these potential therapeutic treatments received FDA approval. A challenge in translating the inhibition of AGE formation is understanding the biological effects of these inhibitors in vivo.

To date, there have been in vitro studies of AGE inhibitors to prevent bone fragility. With respect to human trabecular bone, inhibition of AGEs was investigated using phenacyl thiazolium bromide (PTB), a crosslink breaker [150]. Adding 2 different concentrations of PTB during ribation (0.6 M ribose at 37 °C for 3 days or 7 days) reduced fAGEs in trabecular bone from young donors (19-49 years) but less so in bone from old donors (50-80 years). In a similar study involving glycation (100 mM glucose at 50 °C for 7 days) of cortical human bone from female donors (57-97 years), the addition of either aminoguanidine or pyridoxamine prevented the accumulation of fAGEs [151]. Finally, upon collecting femurs from male C57BL/6J mice fed a high fat diet from 10-weeks to 32-weeks of age, femur mid-shafts were incubated for 7 days at 37 °C in phenacyl thiazolium chloride (PTC). Compared to non-PTC-treated femurs, this AGE breaker reduced fAGEs by 41.2% and increased maximum fracture toughness (K_Ic,max) by 35% [71]. The in vitro treatment with PTC however did not affect crack initiation toughness (K_Ic,init). While these results are encouraging, there are no published in vivo studies demonstrating effects of AGE inhibitors on AGE levels or pathophysiology of bone. Research is needed to understand the relative effects of AGE inhibitors on the material behavior of bone and on bone metabolism (e.g., resorption).

6. Limitations and Future Directions

One key limitation is that most human cadaveric tissue and in vivo animal model studies report negative correlations between AGEs and toughness rather than demonstrate causation or mechanistic insight. A second pertains to the lack of validated protocols to detect and quantify the different types of AGEs. Microplate readers and liquid chromatography can only quantify the bulk levels of fluorescent AGEs in a piece of bone tissue and so miss numerous non-enzymatic modifications of the organic matrix. These traditional techniques also do not indicate where in proteins the AGEs occur. These limitations may be overcome by the standardization of protocols between labs and the adoption of mass spectroscopy and proteomics techniques. Unfortunately, this requires greater labor and expense.

As another important limitation, in vitro glycation and ribation models meant to reveal mechanism are inconsistent in results between studies and with the established hallmarks of the pathology observed in human bone specimens. This includes the type of reactant and the AGEs formed, and their relative quantity, location and specific effects on the organic matrix. Again, these models and their protocols could be standardized and modified to include other factors involved in the pathology such as alternative reactive sugars and metabolites and established inflammatory factors such as oxidative stress. One possible avenue of research to elucidate mechanisms is the incubation of bone in glucose or ribose with different types of AGE inhibitors such that specific AGEs (e.g., CML over glucosepane) form [152]. This strategy can be couple with mechanical testing of human bone or rodent bone to determine which AGEs affect which mechanical properties.

The idea of competition between AGE adducts and LOX-derived crosslinks is a compelling hypothesis that to be tested requires interdisciplinary studies spanning multi-scale mechanics and biochemistry. For example, if overall collagen crosslinking is decreased due to adduct formation instead of enzymatic crosslink formation, how does this affect strength and toughening mechanisms such as collagen unravelling [153]? Rodent models may be useful addressing such questions. The maturation of enzymatic collagen crosslinking occurs with age in rodents and can be disrupted with a toxin β-aminopropionitrile [129], while AGE formation can be manipulated with high sugar or high fat diets and/or AGE inhibitors. The challenge again with in vivo studies is separating biological effects of any exogeneous compound and biophysical effects of PTMs.

Since non-collagenous proteins (NCPs) contribute to toughening mechanisms of bone at the molecular scale [154–156], there is the possibility NE-PTMs of NCPs (i.e., AGE accumulation on other proteins in bone) affect fibril-level deformation and sliding in a manner that is detrimental to the ability of bone to resist fracture. NCPs, such as osteopontin and osteocalcin, are small proteins that may be modified by AGE formation [157]. Though not as abundant as the collagens, NCPs are present in bone tissue and are thought to facilitate mineralization and adhesion between collagen and mineral [155]. The effects of AGEs on NCP structure and function, and subsequent effects on bone mechanics, have not been thoroughly investigated to date.

Since clinical trials of AGE inhibitors would require a large sample size to capture a difference in fracture incidence between a placebo group and a treatment group, a surrogate would be beneficial if it tracked treatment-related changes in bone. The standard surrogate, DXA-derived aBMD, is not expected to show efficacy unless the biological effects of AGE inhibitors prove to include a reduction in bone resorption and/or a promotion of bone formation. Thus, research is also needed to validate new biomarkers or surrogates of bone matrix quality. Two possible candidates are the OsteoProbe that assesses the ability of a person’s tibia to resist impact microindentation [158] and ultrashort-time-to-echo magnetic resonance imaging (UTE-MRI) that assesses bound and pore water concentrations of a person’s tibia as positive and negative correlates, respectively, of mechanical properties of human cortical bone [159]. In a study of cadaveric tibiae from female donors, BMSi correlated with fAGEs (r=−0.613, p<0.001) [160]. When cadaveric cortical bone samples were incubated in 0.1 M or 0.5 M ribose for 4 weeks, a large accumulation of AGEs decreased bound water, as determined by ¹H nuclear magnetic resonance relaxometry, the basis of UTE-MRI [161]. Studies are needed to establish whether BMSi or bound water concentration are related to fracture incidence in the context of elevated AGEs.

7. Conclusions

Given the importance of collagen type I to the fracture resistance of bone, it is widely accepted that the age- and disease-related declines in bone toughness, namely post-yield toughness, are due to degradation of the organic matrix. Because glycation-mediated, non-enzymatic post-translation modifications (AGEs) accumulate in the organic matrix of bone with aging and chronic diseases such as diabetes, AGE accumulation is thought to be one reason why an individual’s fracture risk is higher than expected for a given X-ray-based measurement of bone mass at the hip or spine (T-score). While there is plenty of in vitro evidence that AGEs lower bone toughness, though not all glycation studies are supportive, in vivo evidence of AGE accumulation contributing to bone fragility is lacking. For example, the role of AGE adducts in the mechanical behavior of bone is unknown despite being far more abundant than AGE crosslinks. This gap in knowledge is problematic because the accumulation of AGE adducts has the potential to hinder enzymatic crosslinking and to negatively alter stabilizing interactions among collagens, water, non-collagenous proteins, and mineral.

Before collagen degradation can be translated to the clinic, several key gaps in knowledge must be addressed: i) causal relationships between AGEs and degradation in bone toughness and ii) use of methods to locate and quantify various AGEs in collagens and non-collagenous proteins. Otherwise, targeting AGE accumulation in bone to prevent or treat bone fragility is a difficult proposition. In addition, adopting serum or urinary measurements of AGEs as a clinical tool to diagnose osteoporosis or bone fragility likely requires a prospective study showing such AGE markers add value to the prediction of fracture risk by FRAX. Lastly, the opportunity is there for pre-clinical studies to assess the efficacy of AGE inhibitors or breakers in their ability to promote the ability of bone to resist fracture.

Highlights.

• Accumulation of AGEs may contribute to bone fragility with aging and disease.

• Whether crosslinking AGEs or AGE adducts are detrimental is unknown.

• There is a need to quantify various AGEs and their location on proteins in bone.