Association of Urinary Metabolites with Radiographic Progression of Knee Osteoarthritis in Overweight and Obese Adults: An Exploratory Study

Summary

Introduction

Metabolic factors may contribute to osteoarthritis (OA).This study employed metabolomics analyses to determine if differences in metabolite profiles could distinguish people with knee OA who exhibited radiographic progression.

Methods

Urine samples obtained at baseline and 18 months from overweight and obese adults in the Intensive Diet and Exercise for Arthritis (IDEA) trial were selected from two subgroups (n=22 each) for metabolomics analysis: a group that exhibited radiographic progression (≥ 0.7mm decrease in joint space width, JSW) and an age, gender, and BMI matched group who did not progress (≤0.35mm decrease in JSW). Multivariate analysis methods, including orthogonal partial least square discriminate analysis, were used to identify metabolite profiles that separated progressors and non-progressors. Plasma levels of IL-6 and C-reactive protein were evaluated as inflammatory markers.

Results

Multivariate analysis of the binned metabolomics data distinguished progressors from non-progressors. Library matching revealed that glycolate, hippurate, and trigonelline were among the important metabolites for distinguishing progressors from non-progressors at baseline whereas alanine, N,N-dimethyglycine, glycolate, hippurate, histidine, and trigonelline, were among the metabolites that were important for the discrimination at 18 months. In non-progressors, IL-6 decreased from baseline to 18 months while IL-6 was unchanged in progressors; the change over time in IL-6 was significantly different between groups.

Conclusion

These findings support a role for metabolic factors in the progression of knee OA and suggest that measurement of metabolites could be useful to predict progression. Further investigation in a larger sample that would include targeted investigation of specific metabolites is warranted.

Introduction

Risk factors for osteoarthritis (OA) include age, obesity, prior joint injury, joint shape, and genetics¹. Concerning body mass index (BMI) as a risk factor, a recent meta-analysis found a pooled odds ratio for knee OA of 2.02 for overweight adults (BMI 27–29.9 kg/m²) and 3.91 for obese adults (BMI ≥30 kg/m²)². Weight loss studies also support a link between body weight and knee OA. Weight loss in overweight and obese women without OA is associated with a reduced risk of developing symptomatic knee OA³. In overweight and obese adults with established knee OA, diet-induced weight loss reduces pain and improves physical function^{4, 5} as well as reduces levels of systemic inflammatory mediators^{5, 6}. The mechanisms by which excessive body weight contributes to the development or progression of OA are not well understood but are of increasing interest, in part due to the growing numbers of overweight and obese older adults worldwide.

Although the obesity-knee OA link is well established, it is not clear how much of this is due to the effects of greater body weight on joint loads, promotion of a pro-inflammatory state, or altered metabolism. In addition to knee OA, increased BMI is also associated with OA of the hips⁷ and the hands^{8, 9}. Data from the Netherlands Epidemiology of Obesity (NEO) study show that BMI and fat mass were associated with both knee¹⁰ and hand OA⁹. The association of BMI and fat mass with hand OA in particular suggests that inflammatory and metabolic factors, rather than simply excessive joint loading, may be contributing to the higher risk of development of OA with obesity.

An association of fat mass with OA suggests that factors produced by adipose tissue, such as cytokines and adipokines, may promote OA. There are several cytokines and adipokines, such as IL-1β, IL-6, TNFα, and leptin, produced by adipose tissue that have been hypothesized to contribute to OA^{11, 12}, although it is not known if elevated systemic levels of these factors directly impact joint tissues. Systemic metabolic changes that result from excessive adiposity may also contribute to OA. Studies suggesting a metabolic contribution to OA have shown links between OA and other obesity-related metabolic conditions including hypertension and diabetes^13–15.

The studies demonstrating that increased BMI is a risk factor for OA have been studies of incident or prevalent disease. The role that obesity and associated metabolic factors might play in the progression of established OA is less clear¹⁵. There is some evidence that lower extremity alignment influences the risk of radiographic progression, defined by joint space narrowing, in obese adults with OA such that an increased risk is seen in people with neutral or valgus but not varus alignment^{16, 17}. If progression of OA in overweight and obese adults is mostly driven by biomechanics, then metabolic factors might be expected to be much less important in progression of OA than in the early development of the disease, though this is not yet known.

This study examines the association between metabolic and inflammatory factors and radiographic progression of knee OA, defined by joint space loss, in overweight and obese adults with established knee OA. We sought to determine if a profile of metabolites could be found that distinguished progressors from non-progressors. A few cross-sectional studies have reported a different metabolomic profile between individuals with and without knee OA^18–20 but an association of metabolites with disease progression in people with established OA has not been reported.

Methods

Study Participants

The participants in this study were a subset of those enrolled in the Intensive Diet and Exercise for Arthritis (IDEA) trial. Details of the inclusion and exclusion criteria as well as the interventions have been previously published²¹ as have the main clinical outcomes⁵. Briefly, IDEA was a prospective, single-blind, randomized controlled trial that enrolled 454 overweight and obese (27.0 ≤ BMI ≤ 40.5 kg/m²), older (age ≥ 55 yrs) adults with knee pain and radiographic evidence of tibiofemoral osteoarthritis (Kellgren Lawrence grade = 2 or 3). Participants were randomized to one of three 18-month interventions: intensive dietary weight loss-only; intensive dietary weight loss-plus-exercise; or exercise-only control. Knee X-rays and blood and urine samples used for the present study had been obtained at baseline and 18 (FU18) months.

Radiographic measure of OA progression and selection of subjects

A decrease in joint space width was used as a measure of OA progression. A detailed description of the radiographic procedures used to measure joint space width in the IDEA study has been recently published²². In brief, bilateral, posteroanterior, weight-bearing, knee x-rays were obtained with the participants’ knees flexed at 15 degrees using a positioning device (Synaflexer), and the beam centered on the joint space. The minimum joint space width (mJSW) was measured in digitized radiographic images using an automated software application. Based upon superior responsiveness from prior analyses, we selected the JSWx=0.225 as the optimal location for the fixed coordinate JSW measure. Positioning accuracy was measured by examining the distance from the floor of the tibial plateau to the most proximal tibial rim. We defined malpositioning as a change in this distance of >2mm between the baseline and 18 month radiograph. For the present study malpositioned subjects were excluded. A radiographic “progressor” was defined as those exhibiting a decrease in JSW of ≥ 0.7mm in the medial compartment from baseline to 18 months which was based on the OARSI-OMERACT definition of relevant radiological progression²³. A “non-progressor” was defined as a decrease in JSW of ≤0.35mm. Progressors and non-progressors were chosen based on meeting the radiographic criteria without regards to intervention group assignment.

From the subset of IDEA participants who had radiographs available at both baseline and 18 months and met these criteria, we had 45 progressors and 150 non-progressors and from these selected 22 progressors who could be matched to 22 non-progressors for age, gender, and baseline BMI. The baseline KL scores did not differ significantly between the progressors and non-progressors (p=0.36). Because alignment may influence the risk of progression, we measured baseline alignment using the femur-tibia angle measure (method B) reported by McDaniel et al²⁴. There was a difference in the mean baseline alignment angle but it was not significant due to the large variability (progressors 0.96±5.06, mean±SD, non-progressors −1.46±3.78; p=0.08). The intervention group assignment for the non-progressors included 11 participants in diet plus exercise, 10 in diet only, and 2 in exercise only, while the progressor group included 9 in diet plus exercise, 7 in diet only and 6 in exercise only. We have recently published the radiographic outcomes for the entire IDEA cohort that showed there was no difference among the three intervention groups in radiographic progression²².

Blood and Urine samples

Blood and urine (second am void) were collected from all subjects in the early morning after a 10-hour fast. The 18-month samples were collected at least 24 hours after the last acute bout of exercise training and sampling was postponed (1–2 weeks after recovery from symptoms) in the event of an acute respiratory, urinary tract, or other infection. Samples were aliquoted and stored in cryovials at −80°C until analyzed.

Analysis of inflammatory biomarkers

Interleukin-6 and high sensitivity C-reactive protein (CRP) were chosen as biomarkers of inflammation for the IDEA trial. Plasma IL-6 levels were measured using an ELISA assay and the results for the entire cohort have been published⁵. CRP was measured using an automated immunoanalyzer (IMMULITE; Diagnostics Products Corporation, Los Angeles, CA) and the results for the entire cohort have also been published²⁵.

Metabolomics analysis

Sample preparation, data acquisition, statistics, and pathway analysis were performed as previously described^26–29. Briefly, each urine sample was prepared by mixing an aliquot (400 μL) of each of the study urine samples with 230 μL of D₂O and 70 μL Chenomx Internal Standard solution (Chenomx, Edmonton, Alberta, Canada). In addition, phenotypic pooled urine samples were made by combining 200 μL aliquots from each of the study samples belonging to the same phenotype (progressors-baseline, progressors-follow up, non-progressors-baseline, and non-progressors-follow up). A combined phenotypic pooled sample was also prepared by using 400 μL aliquot from each of the phenotypic pooled sample. Three 400 μL aliquots of each of the pooled urine samples (phenotypic and combined) were prepared identical to the individual urine samples. Metabolomics data were acquired for each of the individual study samples and the pooled samples. ¹H NMR spectra of urine samples were acquired on a Bruker Avance III 950 MHz NMR spectrometer using a 1D NOESY presaturation pulse sequence (noesypr1d, Beckonert et al, 2007). NMR spectra were pre-processed using ACD NMR software (Advanced Chemistry Development, Toronto, ON, Canada). Spectral region of 0.20 to 10.00 ppm was binned²⁷ into 228 integrated segments (NMR bins) of ppm regions (0.04 ppm bucket width) after excluding water (4.65–5.00 ppm), urea (5.50–6.00 ppm), and imidazole (7.20–7.50 ppm) using Intelligent Bucketing Integration with a 0.04 ppm bucket width and a 50% looseness factor. Each of the NMR bin integrals was normalized to total integral of each of the spectrum.

Statistical Analyses

Normalized binned NMR data were Pareto scaled by dividing the integral of each bin by the reciprocal of the square root of the standard deviation for the bin and centered prior to multivariate analysis. These pattern recognition methods are commonly used to analyze high dimensional multicollinear data such as metabolomics data^{30, 31}. The advantage of these methods is that they reduce the dimensionality by creating new variables which are not correlated, and also allow for the data to be visualized^{28, 29}. SIMCA 13 (Umetrics, Umeå, Sweden) was used to conduct the principal component analysis (PCA), orthogonal partial least squares (OPLS), and orthogonal partial least squares discriminant analysis (OPLS-DA). PCA is an unsupervised analysis (the outcome is not used in the analysis) which reduces dimensionality by projecting the data onto a new coordinate system that allows for visualizing any clustering in the data and identifying outliers³¹. The PCA scored plots of the data in the new coordinate system were inspected to ensure that the phenotypic pooled samples were tightly clustered in the center of phenotypic groups and that the pools of all individual study samples clustered in the center of all samples, a quality control method that is widely used in metabolomic studies³².

OPLS and OPLS-DA are supervised analyses for continuous and categorical outcomes respectively. OPLS-DA was used to determine the metabolites that were important for differentiating the progressors and non-progressors. Loadings plots and variable influence on projection (VIP) plots were inspected, and bins that had a VIP ≥ 1.0 with a jack-knife confidence interval that did not include 0 were determined to be important for differentiating the study groups. The VIP statistic summarizes the importance of the bin in differentiating the phenotypic groups³¹. OPLS was used to examine the relationship between the continuous outcomes (changes in IL-6 and joint space). All models used a 7-fold cross-validation to assess the predictive variation of the model (Q²). Chenomx NMR Suite 7.7 Professional software (Edmonton, Alberta, Canada), which has a concentration library of approximately 350 compounds, was used to match the signals in the identified bins to metabolites. Metabolic pathway analysis was performed using the MetaCore® module in GeneGo software (GeneGo, St. Joseph, MI).

Statistical analyses were conducted using SAS 9.4 (SAS Institute Inc, Cary, NC). Statistical tests for continuous variables were conducted using a two-sided t-test with the Satterthwaite correction for unequal variances. The chi-square test was used to test for differences in gender, and the Fisher’s Exact test was used to test for differences in race due to small cell counts. In this exploratory metabolomics study, p-values < 0.05 were considered to be statistically significant and were not adjusted for multiple testing as described^33,34.

Results

Participant demographics are provided in Table 1. There were no differences between progressors and non-progressors by age, gender or race. Clinical outcome measures at baseline, 18 months, and change from baseline are presented in Table 2. BMI was not significantly different between progressors and non-progressors at baseline, 18 months, or 18 months compared to baseline. Progressors had significantly higher (p=0.004) minimum joint space at baseline (mean=4.1mm, SD=2.0) compared to non-progressors (mean=2.4mm, SD=1.5). At 18 months, there was no difference between the groups (p=0.53). By design, progressors had significantly greater (p < .0001) decline over 18 months in joint space width (mean= −1.3mm, SD=0.7) than non-progressors (mean=0.1mm, SD=0.4).

Table 1.

Baseline Demographics

Characteristic	Progressors (n=22)	Non-Progressors (n=22)	p-value
Age, mean (SD), years	66.5 (5.3)	66.4 (5.3)	0.94*
Gender
Female	15 (68.2%)	15 (68.2%)	1.0**
Male	7 (31.8%)	7 (31.8%)
Race/Ethnicity
Caucasian	18 (81.8%)	19 (86.4%)	1.0†
African American	4 (18.2%)	3 (13.6%)

SD, standard deviation;

^*t-test with Satterthwaite correction,

^**Chi-Square Test,

^†Fisher’s Exact Test

Table 2.

Clinical Data

Clinical Measurement	N*		Progressors (n=22)	Non-Progressors (n=22)	p-value**
BMI, mean (SD), kg/m2
	44	Baseline	31.9 (4.6)	33.2 (3.8)	0.33
	41	Month 18	30.5 (5.2)	30.0 (3.9)	0.75
	41	Change	−1.6 (2.6)	−2.9 (2.7)	0.14
Minimum joint space width, mean (SD), mm
	44	Baseline	4.1 (2.0)	2.4 (1.5)	0.004
	44	Month 18	2.8 (1.8)	2.5 (1.5)	0.53
	44	Change	−1.3 (0.7)	0.1 (0.4)	< .0001
CRP, mean (SD), mg/L
	44	Baseline	5.4 (5.2)	12.3 (14.8)	0.049
	40	Month 18	4.2 (4.7)	5.0 (6.1)	0.66
	40	Change	−0.6 (3.4)	−7.1 (13.4)	0.052
IL-6, mean (SD), pg/mL
	44	Baseline	2.9 (1.9)	3.4 (2.2)	0.38
	41	Month 18	3.0 (2.3)	2.4 (2.3)	0.38
	41	Change	0.1 (1.8)	−0.9 (1.1)	0.04

^*Three subjects were excluded from the month 18 metabolomics analysis due to high levels of glucose in the urine samples. All subjects were included in the month 18 and change calculations for minimum joint space width (JSW) since JSW was modeled using the baseline metabolomics data. SD, standard deviation; CRP, C-reactive protein

^**t-test with Satterthwaite correction

Baseline CRP tended to be higher in non-progressors than progressors (p=0.049), but CRP at 18 months was not statistically different between groups (p=0.66). Non-progressors also tended to have a larger decrease in CRP than progressors (p=0.052). IL-6 was not significantly different between groups at baseline or 18 months; however, the change in IL-6 was significantly different (p=0.04) with the mean IL-6 for non-progressors decreasing (mean= −0.9pg/ml, SD =1.1) and progressors increasing (mean=0.1pg/ml, SD=1.8).

The binned metabolomics data of urine collected at baseline was analyzed using OPLS-DA, which is a commonly used method for high dimensional multicollinear data, such as metabolomics data, and allows for visualization of the data^{30, 31}. OPLS-DA distinguished radiographic progressors from non-progressors (Figure 1). NMR bins that were deemed to be important (VIP ≥ 1.0 with a jack-knife confidence interval that did not include 0) for distinguishing progressors from non-progressors were library matched for metabolites. NMR signals that could not be library matched were classified as unknown. Collectively, glycolate, hippurate, trigonelline, and two unknowns were important for differentiating progressors from non-progressors at baseline (Table 3).

Fig. 1.

Scores Plot of OPLS-DA showing separation of progressors (blue circles, right) from non-progressors (green circles, left) at baseline [R2X = 0.2; R2Y= 0.5; Q2=−0.2].

Table 3.

Bins and library matched metabolites that distinguished progressors from non-progressors at baseline.

NMR Bin	Library Matched Metabolite(s)	VIP	P-value	Fold Change*
[1.49 .. 1.52]	Unknown A	1.7	0.097	−1.4
[3.95 .. 4.00]	Hippurate, Glycolate	3.0	0.182	1.2
[4.12 .. 4.15]	Unknown B	1.5	0.041	−1.1
[4.39 .. 4.44]	Trigonelline	1.7	0.036	−1.2
[8.55 .. 8.60]	Hippurate	1.6	0.134	1.8

^*Positive fold change means the peak intensity of progressors is higher than non-progressors.

VIP: variable influence on projection

Three of the 18 month follow up urine samples were found to be strong outliers in the PCA for the follow up samples (but not in the baseline samples) and high amounts of glucose were evident in the ¹H NMR spectra of these samples and were excluded from further analysis of follow up data. Since these samples were not outliers at baseline they were included in the analysis for predicting OA progression. OPLS-DA distinguished progressors from non-progressors at the 18 month follow-up visit and library matching revealed differences in glycolate, hippurate, histidine, N,N-dimethylglycine, trigonelline, and two unknowns (Table 4). OPLS regression analysis showed that the baseline metabolomics data and the change in joint space width from baseline to 18 months were highly correlated in both progressors (R²=0.94, Q²=0.3, Figure 3) and non-progressors (R²=0.96, Q²=0.14, Figure 4). In addition, OPLS regression analysis showed that there was a greater correlation between the change in IL-6 and the baseline metabolomics data for non-progressors (R²=0.81, Q²=−0.28) than for progressors (R²=0.66, Q²=−0.69).

Table 4.

Bins and library matched metabolites that distinguished progressors from non-progressors at the 18 month follow up visit.

NMR Bin	Library Matched Metabolite(s)	VIP	P-value	Fold Change*
[1.45 .. 1.49]	Alanine	1.2	0.624	−1.3
[1.73 .. 1.79]	Unknown C	1.4	0.193	−1.1
[2.88 .. 2.93]	N,N-Dimethylglycine	1.5	0.202	−1.1
[3.95 .. 4.00]	Hippurate, Glycolate	4.8	0.043	1.3
[4.39 .. 4.44]	Trigonelline	1.6	0.026	−1.3
[7.14 .. 7.16]	Histidine	1.1	0.016	1.3
[7.28 .. 7.32]	Unknown D	7.9	0.023	−4.6
[8.55 .. 8.60]	Hippurate	1.4	0.159	2.4

^*Positive fold change means the peak intensity of progressor is higher than non-progressors.

VIP: variable influence on projection

Fig. 3.

Plot of observed decrease in joint space width (JSW) from baseline to 18 months vs predicted decrease in JSW using OPLS regression of baseline metabolomics data for OA progressors. [R²=0.94, Q²=0.3]

Fig. 4.

Plot of observed decrease in joint space width (JSW) from baseline to 18 months vs predicted decrease in JSW using OPLS regression of baseline metabolomics data for OA non-progressors [R²=0.96, Q²=0.1].

Pathway enrichment analysis using the library-matched metabolites that distinguished OA progressors from non-progressors indicated that metabolic pathways related to amino acid metabolism, lipid metabolism, glycosphingolipid metabolism, and the GalNAcbeta1-3Gal pathway were perturbed. The amino acid metabolism pathways included alanine, glycine, cysteine metabolism and transport; alanine, serine, cysteine, methionine, histidine, proline, glycine, glutamate, glutamine metabolism and transport; tryptophan metabolism and transport, the glycine pathway; glycine pathways and transport; and the (L)-alanine pathways and transport.

Discussion

There is mounting evidence that metabolic factors are important contributors to a host of conditions associated with obesity including OA^{13, 14, 35, 36}. The association of obesity and metabolic factors with disease has been explained, at least in part, to associated systemic inflammation or “metainflammation” stemming from inflammatory mediators produced by an increased fat mass^{37, 38}. Findings from the IDEA study have shown that an exercise and weight loss intervention in overweight and obese adults with knee OA improved pain and function in association with a reduction in fat mass as well as systemic inflammatory markers including IL-6 and CRP^{5, 25}. In the present study, using urine samples from a sub-set of IDEA participants, we found that differences in urine metabolites correlated with radiographic progression of knee OA and this correlation was much stronger than with the two measures of inflammation. Interestingly, NMR metabolomics at both baseline and at the 18-month follow-up distinguished progressors from non-progressors who were matched for baseline BMI and who did not exhibit significant differences in BMI at 18 months, suggesting that differences in metabolism between the two groups contributed to the progression of their OA rather than simply a change in BMI.

It is important to note that the urine samples used for the metabolomics analysis were from IDEA participants selected based on meeting the radiographic criteria of disease progression, independent of intervention group assignment, with matching by age, gender, and baseline BMI to participants who met the criteria defined for the “non-progressor” group. Progressors (n=22) and non-progressors (n=22) were fairly evenly distributed among intervention groups, and analysis by intervention group could not be performed due to sample size constraints. For the full cohort, we recently reported²² that the radiographic measures of disease progression did not differ by intervention group assignment suggesting that the differences in the metabolite profiles between progressors and non-progressors were likely due to factors separate from the intervention group assignment. The finding that the non-progressors had a decrease in plasma IL-6 levels over the course of the study while the progressors had an increase suggests an association of progression with systemic inflammation and is consistent with a prior study that reported a positive association between IL-6 levels and development of knee OA³⁹. A study with a larger sample size would be needed to determine if the change in IL-6 correlates with a change in metabolomics.

All prior metabolomic studies in humans have been cross-sectional in nature comparing individuals with or without OA using either urine^{18, 40}, serum¹⁹, synovial fluid⁴¹ or plasma²⁰. An NMR-based urine metabolomics study that used samples from the Johnston County Osteoarthritis project found differences in participants with radiographic hip or knee OA from controls and noted a correlation of a metabolite profile with radiographic OA severity¹⁸. The NMR signals were not fully characterized (no mass spectroscopy was performed), but the NMR patterns suggested higher levels of hydroxybutyrate, pyruvate, creatine/creatinine and glycerol and lower levels of histidine and methylhistidine in those with OA relative to those without OA. We found that progression of joint space narrowing in adults with OA is associated with a different pattern of metabolites that included baseline levels of glycolate, hippurate, trigonelline, and at 18 months glycolate, hippurate, histidine, N,N-dimethylglycine, and trigonelline.

A more targeted approach using mass spectrometry to measure ratios of specific metabolites in serum noted that the ratios of branched-chain amino acids to histidine were significantly associated with knee OA when compared to healthy controls¹⁹. In a similar approach using plasma samples, patients with knee OA were found to have lower levels of arginine²⁰. Finally, a metabolomics study that examined serum samples from healthy controls and patients with osteoarthritis, rheumatoid arthritis (RA), ankylosing spondylitis and gout, found that the patients with arthritis could be differentiated from the controls and that each form of arthritis had a distinct metabolite profile⁴².

Other studies have examined the profile of metabolites in synovial tissue or synovial fluid from subjects with knee OA. Adams et al⁴³ analyzed conditioned media from cultures of synovial explants obtained from early and end-stage OA joints and noted differences in metabolites associated with collagen degradation, amino acid and branched-chain amino acid catabolism, energy metabolism and lipid and carbohydrate metabolism. Metabolic profiling using synovial fluid collected from 55 knee OA patients and 13 cadaveric controls, revealed 11 metabolites that separated the two groups⁴¹. None of the 11 metabolites identified in synovial fluid matched those found in the urine samples from our progression study or those found in plasma samples from a cross-sectional study of knee OA²⁰.

Although these previous studies were cross-sectional and could not associate metabolic differences with OA progression, like our study they do support a role for metabolic factors in the development and progression of OA. Differences in the methodologies used for analysis, as well as the different patient populations studied, source of samples, and how the results were reported, preclude a direct comparison of specific metabolite profiles associated with OA in these studies and our own. In general, there was evidence that OA may be associated with differences in amino acid and collagen metabolism, the TCA cycle, and fatty acid metabolism. Our finding that hippurate and trigonelline, potential gut-flora derived metabolites⁴⁴, differed between progressors and non-progressors suggests that differences in the gut microbiome could also be contributing to metabolic differences associated with OA progression.

There are several limitations of the present study. We were not able to include a replication cohort because of an insufficient number of radiographic progressors in the IDEA cohort to use for replication of the original metabolomics analysis and we did not have access to urine samples from an independent cohort of overweight and obese older adults with knee OA that would meet the radiographic progression criteria we used here. Also due to the limited number of progressors, we included participants independent of the intervention group assignment and did not have sufficient numbers in each group to analyze for an intervention effect on the metabolites.

In summary, our finding that profiles of metabolites could distinguish overweight and obese adults with knee OA who exhibited radiographic progression from those who did not experience progression supports a role for metabolic factors in the progression of OA. These results also suggest that measurement of metabolites that include amino acids and lipids may be useful for evaluation of the efficacy of interventions for OA, as was recently reported for metabolic profiling predicting the response to anti-TNFα therapy in patients with RA⁴⁵. Our results merit further investigation in a larger sample set including targeted analysis of identified marker metabolites.

Fig. 2.

Scores Plot of OPLS-DA showing separation of progressors (blue circles, right) from non-progressors (green circles, left) at follow up visit after 18 months [R²X = 0.9; R2Y= 1.0; Q²=−0.5].