Mass spectrometric quantitation of AGEs and enzymatic crosslinks in human cancellous bone

Advanced glycation end-products (AGEs) deteriorate bone strength. Among over 40 species identified in vivo, AGEs other than pentosidine were roughly estimated as total fluorescent AGEs (tfAGEs) due to technical difficulties. Using LC-QqTOF-MS, we established a system that enabled the quantitation of five AGEs (CML, CEL, MG-H1, CMA and pentosidine) as well as two mature and three immature enzymatic crosslinks. Human bone samples were collected from 149 patients who underwent total knee arthroplasty. Their clinical parameters were collected to investigate parameters that may be predictive of AGE accumulation. All the analytes were quantitated and showed significant linearity with high sensitivity and precision. The results showed that MG-H1 was the most abundant AGE, whereas pentosidine was 1/200–1/20-fold less abundant than the other four AGEs. The AGEs were significantly and strongly correlated with pentosidine, while showing moderate correlation with tfAGEs. Interestingly, multiple linear regression analysis revealed that gender contributed most to the accumulation of all the AGEs, followed by age, tartrate-resistant acid phosphatase-5b and HbA1c. Furthermore, the AGEs were negatively correlated with immature crosslinks. Mass spectrometric quantitation of AGEs and enzymatic crosslinks is crucial to a better understanding of ageing- and disease-related deterioration of bone strength.

www.nature.com/scientificreports/ marker of blood glycemic status over several weeks to months 40 , tartrate-resistant acid phosphatase-5b (TRACP-5b) as the measure of bone resorption, and estimated glomerular filtration rate (eGFR).

Results
Clinical characteristics of the patients are summarized in Table 1. There were no gender differences in age, BMI and laboratory data. The prevalence of hypertension and dyslipidemia was higher in females.  www.nature.com/scientificreports/ Typical extracted ion chromatographic peaks of the AGEs and enzymatic crosslinks in human bone hydrolysate are shown in Fig. 3. Linearity, sensitivity, precision and recovery validations are shown in Table 2. All individual AGEs demonstrated linear response when dissolved in 0.1% (v/v) formic acid with coefficients of correlation (r 2 ) above 0.998 ranging from 0 to 1000 nmol/L. All individual enzymatic crosslinks also demonstrated linear response both in 0.1% (v/v) formic acid and in bone hydrolysate with r 2 above 0.998 ranging from 0 to 250 nmol/L. Limit of detection (LOD) and limit of quantitation (LOQ) of AGEs were below 14 and 42 nmol/L, respectively. LOD and LOQ of enzymatic crosslinks were below 12 and 35 nmol/L in 0.1% formic acid, and below 16 and 48 nmol/L in bone hydrolysate. Validation of precision showed that intraday coefficients of variations (CVs) were below 5% for all analytes, whereas interday CVs were slightly greater, especially for enzymatic crosslinks in bone hydrolysate (6-12%). Recoveries of AGEs were 82-100% except for MG-H1 (60%). Recovery rates of enzymatic crosslinks were approximately 50-65%.
MG-H1 was the most abundant AGE among the AGEs quantitated (Fig. 4). Pentosidine was 1/200-1/20-fold less abundant than the other AGEs. Analysis of gender and comorbidities revealed that male gender was associated with statistically higher AGE contents (Supplemental Table S1). The AGEs were higher in hypertensive patients but did not reach statistical significance. There were strong correlations between the AGEs; Spearman's coefficient (r s ) reached 0.9, except for CMA (Table 3). There were moderate correlations between the AGEs and tfAGEs with r s of around 0.3.
Linear regression analyses were performed to evaluate the relative contributions of clinical parameters to AGE amounts. The result showed that age and HbA1c were positively correlated, while tartrate-resistant acid phosphatase-5b (TRACP-5b) 41,42 , which is an indicator of bone resorption, was negatively correlated with the amount of each AGE (Supplemental Table S2). Subsequent multiple stepwise linear regression analysis, using the amount of each AGE as the dependent variable, with age, body mass index (BMI), estimated glomerular filtration rate (eGFR), TRACP-5b, and gender as independent variables, showed that all parameters except eGFR were independently associated with the amount of each AGE (Table 4). Gender contributed most to the accumulation of all AGEs, followed by age and TRACP-5b. HbA1c was not an independent determinant for pentosidine. Independent contributions of BMI were only significant for CML and MG-H1.
The contents of the enzymatic crosslinks are summarized in Fig. 4. PYD was the most abundant enzymatic crosslink, followed by DHLNL and HLNL. Except for HLNL, there were no differences based on gender (Supplemental Table S3). The maturation index, defined as the ratio of total mature (PYD + DPD) to total immature crosslinks (DHLNL + HLNL + LNL), was positively correlated with age (r s = 0.271; p < 0.001) and negatively with TRACP-5b (r s = − 0.237; p < 0.01).

Discussion
In this study, the amounts of CML, CEL, MG-H1, CMA and pentosidine were quantified using LC-QqTOF-MS. Until recently, three methods were applied for the evaluation of AGEs. The first is HPLC equipped with a fluorescence detector (HPLC-FLD) 12 . This method has been reported for the quantitation of pentosidine; however, whether the quantitation of pentosidine is suitable for the elucidation of other AGEs had been unclear. The second method estimated tfAGEs 17 using a spectrofluorometer. This method cannot detect non-fluorescent AGEs such as CML, and overestimation can occur in the presence of fluorophores other than AGEs. The third method characterizes AGEs using ELISA 16 . Although ELISA is well-characterized and conventional, a cross-reaction can occur and possibly leads to inaccurate analyses especially for AGEs. For example, 6D12, a commercially available anti-CML monoclonal antibody, can cross-react with CEL because CML and CEL are structurally similar 43 . LC-MS can potentially overcome the limitations that HPLC-FLD, spectrofluorometer and ELISA had for two reasons. First, LC-MS detects analytes in their ionized forms, thus does not need analytes to be fluorescent. Second, LC-MS can quantitate analytes with similar structures individually by the combination of retention time and mass-charge ratio (m/z). Although the five AGEs, whose authentic standards and isotope-labelled internal standards (ISTDs) were commercially accessible, were analyzed in this study, LC-MS will serve to the quantitation of other AGEs in the future. Among them, glucosepane is of great interest because it is thought to be a major crosslinking AGE present in various collagenous tissues 22,44 . Table 3. Correlation analyses between the AGEs and tfAGEs. CML N ε -(carboxymethyl)lysine, CEL N ε -(carboxyethyl)lysine, MG-H1 N δ -(5-hydro-5-methyl-4-imidazolon-2-yl)-ornithine 1, CMA N ω -(carboxymethyl)arginine, tfAGEs total fluorescent AGEs; r s Spearman's coefficient. *p < 0.05; **p < 0.01; ***p < 0.001.    Table 5. Correlation analyses of the lys-derived AGEs (i.e. CML, CEL and pentosidine) and the enzymatic crosslinks. CML N ε -(carboxymethyl)lysine, CEL N ε -(carboxyethyl)lysine, DHLNL dihydroxylysinonorleucine, HLNL hydroxylysinonorleucine, LNL lysinonorleucine, PYD pyridinoline, DPD deoxypyridinoline, r s Spearman's coefficient. *p < 0.05; **p < 0.01; ***p < 0.001. www.nature.com/scientificreports/ The developed method was validated for linearity, sensitivity, precision and recovery. All the AGEs and the enzymatic crosslinks demonstrated linear response even beyond physiological ranges. LOD and LOQ of AGEs were comparable with previous studies which utilized triple quadrupole MS 45 and quadrupole-Orbitrap MS 46 . LOD and LOQ of enzymatic crosslinks were also satisfactory compared with the previous study using single quadrupole MS 13 . Precision validation showed that intraday CVs were below 5%, but continuous analysis lasting over three days resulted in elevated CVs. This may be due to increased solvent concentration. Therefore in this study, all samples were analyzed within 2 days. Decreased recoveries especially for enzymatic crosslinks were possibly caused by ion suppression 47 . Stable isotope dilution method and standard addition method used in this study are desirable for accurate quantitation using LC-QqTOF-MS 47,48 .
AGEs were analyzed in human bone samples in this study. It was important to take a bone sample undergoing normal remodeling because tissue remodeling is a determinant of AGE accumulation 49,50 . Cadaveric bone is ideal but is not readily available in Japan. Therefore, we took samples from the dissected tibia in total knee arthroplasty. The turnover of cancellous bone in osteoarthritic area might be disturbed, as is the case for subchondral bone 51 . We adopted the sampling method of Oren et al. 52 and took only the cancellous bone from the central one-third and at a depth of 10 mm of the lateral tibial plateau where the superficial cartilage was macroscopically intact. Besides the sampling limitation, it was noteworthy that significant correlations were observed here between the AGE contents and age as shown in the previous studies 12, 16 .
As described in the introduction, and the alternative pathways that involve reactive carbonyls (i.e. methylglyoxal, glyoxal, glycolaldehyde and glyceraldehyde) are postulated to be the dominant mechanism compared to the classical pathway (Fig. 1) 4 . To support this, the results of this study showed that MG-H1, formed from methylglyoxal and arginine, was the most abundant AGE. On the other hand, pentosidine was much less abundant. MG-H1 content in bone has never been examined previously; however, reports on serum protein 18 , muscle 18 and Achilles tendons 53 have shown that MG-H1 accumulation is 200-9,000 times greater than pentosidine accumulation. Methylglyoxal is 20,000 times more reactive than glucose 54 , and this high reactivity could explain the abundance of MG-H1 in various tissues. CML was the second most abundant AGE, accumulating at approximately 100 times the rate of pentosidine. Thomas et al. quantitated CML in human cortical bone using ELISA and compared it with the amount of pentosidine 16 . Although sample type (cancellous versus cortical bone) and analytical methods were different, their results were in accordance with ours in terms of the higher abundance of CML over pentosidine. For a crosslinking AGE to form, two amino acids need to be located at an adequate distance, as shown by an in silico analysis of glucosepane, one of the well-known crosslinking AGEs 22,55 . This stipulation should also apply for pentosidine and thus would have a lower chance of formation when compared with non-crosslinking AGEs such as MG-H1 and CML.
It is technically difficult to quantitate all the individual AGE compounds accumulating in bone collagen, though it is presumed that multiple AGE structures, rather than one specific structure, are involved in the deterioration of bone strength. Thus a surrogate marker such as pentosidine and tfAGEs have been widely used to estimate total AGEs accumulation in bone collagen. However, whether pentosidine or tfAGEs accurately represents other AGEs had not been thoroughly investigated. Thomas et al. found strong correlation (r 2 = 0.87, r < 0.05) between pentosidine and CML in human cortical bone with limited sample volume (n = 5) 16 . Thus in this study, using 149 human cancellous bones, correlations between five AGEs including pentosidine and CML, and tfAGEs were analyzed. As a result, there were strong correlations between the AGEs quantitated in this study, and pentosidine was a better surrogate of other AGEs compared with tfAGEs, at least in human cancellous bone from the tibia. This result was unexpected because AGEs are formed by different pathways 4 as described earlier, and thus prompted us to perform regression analyses to investigate potential determinants of AGEs accumulation. Surprisingly, in single and multiple regression analyses, gender was the strongest determinant of the AGEs, followed by age, TRACP-5b, HbA1c and BMI. Aging 12,38 , bone turnover 49,56 and glycative stress 38,57 , which are closely linked, are shown to be determinants for pentosidine accumulation in bone. To the best of our knowledge, there is only one study, reporting gender difference, in which the pentosidine content of human cancellous bone from the femoral head was higher in males 58 . Barp et al. demonstrated that the level of oxidative stress was higher in male rats than in female rats 59 . Data have also been accumulated to show that women are less susceptible to oxidative stress 60 . It was also reported that levels of carbonyl compounds in skin were higher in male turkeys than in females, which raises the possibility that males are more prone to carbonyl stress 61 . The differences of oxidative and carbonyl stress between genders may explain the result that gender was a strong determinant of AGEs accumulation. BMI was a weak but independent determinant of CML and MG-H1. This might be explained by the obesity-related acceleration of oxidative stress seminal to the accumulation of AGEs 62 . Compared with the other AGEs, the adjusted R 2 of CMA was less than 0.2, which implies the presence of other major determinants for CMA that were not examined in this study. It was interesting to find that, unlike other AGEs, HbA1c did not become an independent determinant of pentosidine. HbA1c is a glycated hemoglobin and used as a serological marker of glycemic status over several weeks to months 40 . Although pentosidine formation via the alternative pathway has been implicated in vitro 35 , pentosidine is still presumed to form predominantly through the classical pathway which is a longer process compared to other AGEs pathways 4 . It can be speculated that pentosidine accumulates in bone collagen in association with a longer period of hyperglycemia than required for the accumulation of other AGEs quantitated in this study. Thus HbA1c, which represents hyperglycemic status over only a few months, did not become an independent determinant of pentosidine accumulation. Overall, the strong correlations observed between individual AGE contents indicate that gender, age and bone turnover were strong independent determinants of individual AGEs, irrespective of their pathways. This may be true for the other AGEs not analyzed in this study. From the results of this study, pentosidine is shown to be a more suitable surrogate marker for AGEs accumulation in the human cancellous bone.
The amounts of PYD and DPD detected in this study were consistent with previous reports documenting human cancellous bone in which HPLC-FLD and LC-MS methods were used in conjunction with a conversion Scientific Reports | (2020) 10:18774 | https://doi.org/10.1038/s41598-020-75923-8 www.nature.com/scientificreports/ factor which assumes that collagen weighs 7.5 times the measured weight of Hyp which has a molecular weight of ~ 300,000 daltons 12,13 . The DHLNL amount of human cancellous bone observed by Gineyts et al. 13 were slightly higher than ours (0.345 ± 0.063 vs 0.231 ± 0.057 mol/mol of collagen). This discrepancy might be attributable to differences in sample types (cancellous tibial bone versus cancellous bone of lumbar vertebrae) and sample extraction methods. The maturation index was positively correlated with age and negatively correlated with TRACP-5b. These results are consistent with the findings that immature crosslinks transform into mature crosslinks time-dependently 12,13 . Interestingly, the results of this study showed that the contents of immature crosslinks were negatively correlated with the contents of Lys-derived AGEs. We previously observed that the contents of femoral pentosidine were elevated in spontaneously diabetic rats whereas immature crosslinks decreased 38 . It was postulated that a hyperglycemia-induced vitamin B6 deficiency caused the down-regulation of lysyl oxidase (LOX) and an eventual reduction in immature crosslinks. Recently, Hudson et al. reported using tail collagen from spontaneously diabetic mice that the sites where immature crosslinks began forming were almost identical to the glycation sites 36 . Thus, the formation of immature crosslinks can be blocked by glycation. In this regard, our findings support the notion of Hudson et al. In terms of abundance, the AGEs other than pentosidine were present to the same order of magnitude as enzymatic crosslinks. Crosslinking AGEs, such as pentosidine and glucosepane, crosslink randomly and can deteriorate the biochemical and mechanical properties of collagen fibrils. Non-crosslinking Lys-derived AGEs, such as CML, on the other hand, can weaken collagen fibrils by blocking the formation of immature crosslinks. With regard to non-crosslinking Arg-derived AGEs such as MG-H1, there has been no study that tried to investigate their effects on bone strength. Given their abundance, future research focusing on their effects on bone strength would be significant.
Several study limitations warrant addressing. First is regarding acid-labile AGEs. In order to determine total AGE contents in proteins using LC-MS, all peptide bonds need to be hydrolyzed. For this, acid hydrolysis in presence of 6 N HCl at 110 °C for 18-24 h is widely used 12,16,52 . However, acid-labile AGEs, such as glucosepane, break down during this process. Enzymatic hydrolysis is an alternative in this situation, although it is time consuming, expensive, and impractical for routine analysis 18 . Moreover, Antonova, Frolov et al. 63 pointed out the possibility that compromised solubility of proteins could lead to insufficient enzymatic digestion and result in inaccurate quantitation of AGEs. We opted to add ISTDs before acid hydrolysis so that the losses of partially acid-labile AGEs (i.e. MG-H1 and CMA) during hydrolysis could be corrected. Secondly, the strong correlations between the AGEs in this study may be limited to cancellous bone. Thomas et al. found strong (r 2 = 0.87, p < 0.05, n = 5) correlations between pentosidine and CML in human cortical bone 16 . Meanwhile, Karim et al. found weak correlations (r = 0.226, p < 0.05, n = 91) between pentosidine and tfAGEs in human cortical bone 17 . Factors other than bone turnover are more influential to the accumulation of AGEs because cortical bone turnover is relatively slower 64 . The quantitation of various AGEs in cortical bone is currently in progress. Thirdly, the spatial distribution of AGEs cannot be evaluated using our method of analysis. The efficient use of immunohistochemical techniques may improve the distributional evaluations of AGEs in bone tissue.
In conclusion, mass spectrometric quantitation of AGEs and enzymatic crosslinks in human cancellous bone revealed that pentosidine was 1/200-1/20-fold less abundant than the other AGEs. Unexpectedly, there were strong correlations between the individual AGEs, whereas moderate correlations were observed between the individual AGEs and tfAGEs. Pentosidine, though in itself accumulates in a relatively small quantity, may be a more suitable surrogate marker of other AGEs compared with tfAGEs. In single and multiple regression analyses, gender was the strongest determinant of the AGEs, followed by age, TRACP-5b, HbA1c and BMI. The gender difference in oxidative and carbonyl stress may explain this. In addition, CML and CEL, the non-crosslinking Lys-derived AGEs, were negatively correlated with the immature crosslinks. This result raises the possibility that non-crosslinking AGEs attribute to the deterioration of bone strength by inhibiting the formation of enzymatic crosslinks. Mass spectrometric approach has enabled a detailed analysis of individual AGEs and crosslinks, and is crucial to a better understanding of ageing-and disease-related deterioration of bone strength. Sample collection. Patients recruited to this study included 149 patients with medial osteoarthritis of the knee who underwent total knee arthroplasty at Jikei University Hospital between 2014 and 2017. Patients with diabetes mellitus, rheumatic and chronic kidney disease on hemodialysis were excluded. Patients taking statins and vitamin D supplementation were also excluded. During surgery, excised tibia samples were immediately frozen and stored at − 80 °C until biochemical analysis. Blood samples were collected from patients one month prior to surgery and creatinine concentrations and HbA1c were determined subsequently. The estimated glomerular filtration rate (eGFR) was calculated using the following formula: 194 × serum creatinine − www.nature.com/scientificreports/ (× 0.739 if female). Serum TRACP-5b was quantitated using an ELISA kit from DS Pharma Biomedical (Osaka, Japan) 65 as an indicator of bone resorption 66 . The baseline height and weight of patients was measured for the calculation of BMI. This study was conducted according to the Declaration of Helsinki and received approval from the Ethics Review Committee of Jikei University Hospital. Each patient provided written informed consent before participating in this study.

Methods
Sample preparation. Cancellous bone samples were harvested from the central one-third of the lateral tibia where the superficial cartilages were macroscopically intact and at a depth of 10 mm from the cartilage surface, and pulverized using a Multi-beads Shocker (Yasui Kikai, Osaka, Japan). Subsequently, bone powders were washed with phosphate-buffered saline (0.15 mol/L NaCl in sodium phosphate buffer, pH 7.4), delipidated with chloroform/methanol (2:1, v/v) mixture during 24 h at 4 ℃ and demineralized with 0.5 mol/L EDTA in 50 mmol/L Tris buffer (pH 7.4) for 96 h at 4 ℃.
Delipidated and demineralized bone powders were reduced by sodium borohydride in sodium borate buffer (100 mmol/L of boric acid and 1 mmol/L diethylenetriamine pentaacetic acid; pH 9.1) for 4 h at ambient temperature. Isotope-labelled internal standards (ISTDs) were spiked to reduced bone powders in the following quantities: 5 nmol of (v/v) formic acid. After an isocratic step at 80% eluent B for 2 min, analytes were separated in the gradient to 10% eluent B for 14 min. After a second isocratic segment (3 min at 10% eluent B) for wash, a third isocratic segment (3 min at 80% eluent B) was run for re-equilibration. For analysis of pentosidine, 5 mg of hydrolysate was loaded on a SunShell RP-AQUA column (2.6 µm, 150 × 2.1 mm) by ChromaNik Technologies (Osaka, Japan), installed on a same LC system and with the same eluents. The separation was performed at a flow rate of 200 µL/ min, at 40 °C in an isocratic mode. Pentosidine was separated in an isocratic step at 0% eluent B for 4 min. After a second isocratic segment (6 min at 60% eluent B) for wash, a third isocratic segment (3 min at 0% eluent B) was run for re-equilibration.
Detection was performed using a micrOTOF-QII quadrupole time-of-flight mass spectrometer (QqTOF-MS) by Bruker Daltonics (Bremen, Germany), equipped with an electrospray ionization source. The instrument was operated in positive ion mode, using a m/z range of 50-1000. Capillary voltage of ion source was set to 4500 V, nebulizer gas flow was 1.6 bar, and dry gas flow was 8 L/min. Dry temperature was set at 200 °C.
Post-run calibration for each sample was performed by injecting 20 µL of mass calibrator at the beginning of each run via the 6-port diverter valve. At each calibration, the measured masses of sodium formate cluster ions were compared with theoretical ones (m/z 90.976645-974.813156) to achieve 5 ppm mass accuracy. Target analytes and their ISTDs are summarized in Supplemental Table S4. The extracted ion chromatograms (theoretical m/z ± 0.005) were constructed for target analytes and ISTDs, and peak area ratios (target/ISTD) were calculated. The amounts of target analytes in samples were determined by comparing peak area ratios with the 6-point calibration curves of the external standards spiked with ISTDs, respectively. For enzymatic crosslinks, the standard addition method was used because the ISTDs of these crosslinks were commercially unavailable 48 . A separate hydrolysate (0.05 mg) and a hydrolysate (0.05 mg) spiked with 10 pmol of LNL, 20 pmol of DHLNL, HLNL and DPD, and 30 pmol of PYD standards were analyzed. The contents of AGEs and enzymatic crosslinks were standardized to Hyp amounts to surrogate collagen amounts 12 and expressed as µmol/mol of Hyp.

Method validation.
The LC-QqTOF-MS method was validated for linearity, sensitivity, precision recovery.
Linearity of response for individual AGEs was validated in 0.1% (v/v) formic acid, while linearity for enzymatic crosslinks was validated both in 0.1% (v/v) formic acid and bone hydrolysate because enzymatic crosslinks were quantified using standard addition method. LOD and LOQ as measures of sensitivity were calculated both in 0.1% (v/v) formic acid and bone hydrolysate based on the residual standard deviation of a regression line. Results were expressed as nmol/L. To calculate intraday and interday precisions, authentic standards of all the analytes comparable to physiological concentrations spiked in 0.1% (v/v) formic acid, and a single bone hydrolysate without standard spike were measured 10 times during the same day and further five times each in two consecutive