The Effect of Body Mass Index and Metformin on Matrix Gene Expression in Arthritic Primary Human Chondrocytes

Abstract

Objective

Obesity is a known risk factor for knee osteoarthritis (OA). Diabetes has been associated with progression of OA and metformin is the first-line treatment in type 2 diabetes. The effect of the body mass index (BMI) and metformin on the expression of certain matrix genes in human chondrocytes is unclear. The purpose of this study was to investigate the effect of BMI and metformin on the expression of matrix genes in primary human chondrocytes.

Design

Adult female patients undergoing knee arthroplasty for end-stage OA were enrolled. Primary chondrocytes were cultivated and stimulated with metformin. Matrix gene expression was analyzed using polymerase chain reaction. Clinical data were used in multivariable regression models to assess the influence of BMI and metformin stimulation on gene expression.

Results

A total of 14 patients were analyzed. BMI was a predictor of increased expression in ADAMTS5 (β = −0.11, P = 0.03). Metformin slightly reduced expression in ADAMTS5 (β = 0.34, P = 0.04), HIF-1a (β = 0.39, P = 0.04), IL4 (β = 0.30, P = 0.02), MMP1 (β = 0.47, P < 0.01), and SOX9 (β = 0.37, P = 0.03). The hip-knee-ankle angle and proton pump inhibitors (PPIs) intake were associated with reduced SOX9 expression (β = 0.23, P < 0.01; β = 2.39, P < 0.01). Higher C-reactive protein (CRP) levels were associated with increased MMP1 expression (β = −0.16, P = 0.02).

Conclusion

We found that BMI exerts a destructive effect via induction of ADAMTS5. Metformin reduced the expression of catabolic genes ADAMTS5 and MMP1 and might play a role in disease prevention. Limb malalignment and PPI intake was associated with a reduced expression of SOX9, and higher CRP levels correlated with increased MMP1 expression, indicating a destructive process.

Keywords

metformin osteoarthritis of the knee obesity gene expression primary chondrocytes

Introduction

Osteoarthritis (OA) of the knee is among the most debilitating diseases.¹ In the United States, an estimated 14 million people are suffering from symptomatic knee OA,² with numbers increasing as the population becomes ever older. This leads to a massive economic burden on the health care system. The lifetime cost for patients diagnosed with knee OA was estimated at around $140,300.³

The pathophysiologic mechanisms underlying knee OA are complex as knee OA is a multifactorial disease.^4,5 There are many known risk factors, such as age, gender, body mass index (BMI), trauma, and genetic background.⁶ BMI has been found to alter expression of certain matrix genes such as MMP13 (matrix metallopeptidase 13) or aggrecan.⁷ The relationship with other major players in the pathophysiologic process of OA remains unclear, however.

In addition to damage caused by overweight, recent scientific findings indicate that systemic inflammation plays an important role in onset and progression of the disease.⁴ This is also evidenced by the fact that in addition to OA of the weightbearing joints, obese patients often suffer from OA of the hand as well.⁸ Low-grade systemic inflammation is common in metabolic syndrome.⁹ Diabetes, in particular, has been associated with OA and is a risk factor for disease progression.^10,11 The first-line treatment in diabetes type 2 is metformin. Clinical studies have found that this biguanide derivative might have a protective effect against knee OA.^12-14 In animal models, metformin was found to reduce the expression of the catabolic genes.^15,16 So far the effects of metformin on the expression of cartilage matrix genes in human chondrocytes have not been investigated.

The purpose of this study was therefore to address the following questions: (1) What is the effect of BMI on the expression of certain matrix genes playing a major role in the pathogenesis of knee OA? (2) How can metformin influence this gene expression? (3) How do other clinical parameters and medication relate to gene expression? Our hypothesis was that BMI and metformin have significant effects on these matrix genes. To answer these questions, primary human chondrocytes were cultivated and stimulated with metformin. Interrelations of quantitative polymerase chain reaction (qPCR) data and clinical data were analyzed by a multivariable regression analysis.

Methods and Materials

This study was designed as a cross-sectional clinical/experimental laboratory study and conducted at the Department of Orthopaedics of the Medical University of Graz, Austria, starting in February 2019 until December 2019. Patients were asked for written informed consent. The procedures followed were in accordance with the ethical standards of the responsible committee (Ethics Committee, Medical University of Graz, IRB #31-133) and with the Helsinki Declaration.

Study Population

Female patients undergoing knee arthroplasty for end-stage knee OA at the age of 52 to 82 years were enrolled.

Cartilage Samples and Cell Culture

Femoral cartilage was collected intraoperatively and stored in growth medium at 4 °C for 2 to 4 hours. Cartilage was dissected using a scalpel, washed using phosphate buffered saline (PBS) 1× (Gibco Invitrogen, Carlsbad, CA, USA) and digested using 2 mg/mL collagenase B (Gibco) in growth medium at 36 °C for 24 hours. Cells were then filtered using a 70 µm filter, centrifuged, and seeded in cell culture flasks. For cultivation DMEM/F12 (Dulbecco’s modified Eagle medium F-12) growth medium, containing 10% fetal bovine serum (FBS), 1% penicillin-streptomycin (5.000 U/mL), 1% l-glutamine, 1% insulin-transferrin-selen, 0.01% transforming growth factorβ (TGF-β; 10 ng/µL), and 0.01% fibroblast growth factor (FGF; 10 ng/µL; all Gibco), were used. Cells were incubated at 37 °C in a humidified atmosphere of 5% CO₂ under normoxic conditions. Cell passages 1 and 2 were stored in liquid nitrogen using a 5% dimethyl sulfoxide (DMSO; Sigma-Aldrich, St. Louis, MO, USA) solution. Passage 2 was used for experiments. For transfer and splitting, PBS 1× containing 10% trypsin 2.5% (Gibco) was used.

Cell Viability Assay

In order to find a nontoxic drug concentration for experiments, cell viability assays were performed. Chondrocytes (1 × 10⁴ and 5 × 10³ cells for measurement after 24 and 48 hours respectively) were seeded on 96-well plates and either used as control or treated with metformin in various concentrations (0.5 µM, 0.1 mM, 1 mM, 2 mM). Cell viability was evaluated in 2 patient samples using the CellTiter-Glo Luminescence Cell Viability Assay (Promega, Madison, WI, USA) after 24 and 48 hours according to the manufacturer’s instructions. Background reference values were derived from the culture media. Absorbance values were measured with the Lumistar microplate luminometer (BMC Labtech, Ortenberg, Germany).

Metformin Treatment

Metformin (Sigma-Aldrich, St. Louis, MO, USA) was dissolved in DMSO (Sigma-Aldrich) and stored at a concentration of 100 mM. Cells were stimulated with 1 mM metformin at a confluence of 30% to 50% and incubated for 48 hours. For each patient, a control and a metformin sample (1 mM) was obtained for further analysis.

Gene Expression Analysis by Reverse-Transcription Quantitative Polymerase Chain Reaction (RT-qPCR)

RNA was extracted using the RNeasy Mini kit using spin technology (Qiagen, Hilden, Germany). DNA was digested using RNase-free DNase Set (Qiagen). RNA concentration was determined using spectrophotometry (Nanodrop, ThermoFisher, Waltham, MA, USA). RNA was then analyzed by RNA gel electrophoresis (Bioanalyzer, Agilent, CA, USA). The RNA integrity number (RIN) for all samples was greater than or equal to 9. cDNA was synthesized using 2 µg of extracted RNA using the iScript kit (Bio-Rad Laboratories Inc., Hercules, CA, USA) according to the manufacturer’s recommendations. The following amplification settings were used on a thermocycler (MyCycler, Biorad): Priming at 25 °C for 5 minutes, reverse transcription at 46 °C for 20 min and reverse transcription inactivation at 95 °C for 1 minute. cDNA was stored at −20 °C.

RT-qPCR was performed using SsoAdvanced Universal SYBR Green Supermix (Bio-Rad Laboratories Inc.) according to the manufacturer’s recommendations using technical triplicates. A reaction volume of 10 µL containing 12.5 ng target DNA was used. The following PCR amplification specifications were set on a CFX96 Touch Real-Time PCR Detection System (Bio-Rad): incubation at 95 °C for 30 seconds for polymerase activation and DNA denaturation, followed by 40 amplification cycles of 10 seconds of denaturation at 95 °C and 20 seconds of annealing-elongation at 60 °C. Table 1 shows the primers used for PCR analysis. Results were analyzed using the CFX Maestro Software (Bio-Rad Laboratories Inc., version 1.1). Quantification cycle values (Ct) were exported for statistical analysis. Results with Ct values greater than 32 were excluded from analysis. The arithmetic mean of expression levels of the reference genes RPL and TBP was used for normalization (ΔCt).

Table 1.

Polymerase Chain Reaction (PCR) Primers Used in Quantitative PCR Experiments.^a

Symbol	Name	Catalog Number/Primer Sequence(Forward/Reverse)^b	Source
ADAMTS4	ADAM metallopeptidase with thrombospondin type 1 motif, 4	QT00032949	Qiagen
ADAMTS5	ADAM metallopeptidase with thrombospondin type 1 motif, 5	QT00011088	Qiagen
COL2A1	Collagen, type II, alpha 1	QT00049518	Qiagen
HIF-1a	Hypoxia inducible factor 1, alpha subunit	QT00083664	Qiagen
EPAS1 (HIF2)	Endothelial PAS domain protein 1	QT00069587	Qiagen
IL4	Interleukin 4	5′-ATCTTTGCTGCCTCCAAGAACAC-3′	self-designed
		5′-GTAGAACTGCCGGAGCACAG-3′
IL6	Interleukin 6	QT00083720	Qiagen
CXCL8 (IL8)	Chemokine (C-X-C motif) ligand 8	QT00000322	Qiagen
MMP1	Matrix metallopeptidase 1	5′-CTGTTCAGGGACAGAATGTGCT-3′	Self-designed
		5′-TCGATATGCTTCACAGTTCTAGGG-3′
MMP3	Matrix metallopeptidase 3	5′-TTTTGGCCATCTCTTCCTTCA-3′	Self-designed
		5′-TGTGGATGCCTCTTGGGTATC-3′
MMP13	Matrix metallopeptidase 13	5′-TCCTCTTCTTGAGCTGGACTCATT-3′	Self-designed
		5′-CGCTCTGCAAACTGGAGGTC-3′
SOX9	SRY-box transcription factor 9	5′-CGCCATCTTCAAGGCGCTGC-3′	Self-designed
		5′-CCTGGGATTGCCCCGAGTGC-3′
SREBP2	Sterol regulatory element binding protein	5′-AAGTCTGGCGTTCTGAGGAA-3′	Self-designed
		5′-AGGTCCACCTCATTGTCCAC-3′
TIMP1	TIMP metallopeptidase inhibitor 1	5′-CTTCTGGCATCCTGTTGTTG-3′	Self-designed

Self-designed primers were generated using the online tool Primer BLAST (U.S. National Library of Medicine) and purchased from Sigma/Aldrich.

Qiagen primer sequences are not disclosed by the manufacturer.

Limb Alignment and Joint Space Width

Preoperative full limb standing x-ray images were obtained for every patient. Images were stored and analyzed using the hospital-wide PACS (picture archiving and communications system) (pacsview, version 3.00.123, Institute for Medical Informatics, Statistics and Documentation, Medical University of Graz, Austria). Analysis was performed by 2 investigators (MS, PS), images were read for medial and later joint space width (JSW) and the hip-knee-ankle angle (HKAA) as outlined previously ( Fig. 1 ).¹⁷ Intraclass correlation coefficient (ICC) was found to be excellent (ICC for HKAA 0.964 [0.891-0.988], ICC for medial JSW 0.923 [0.778-0.975], ICC for lateral JSW 0.881 [0.671-0.960]). For analysis the arithmetic mean of the aforementioned values was used.

Figure 1.

The measurement of the hip-knee-ankle angle (HKAA) (left) and joint space width (JSW) (right). The HKAA is defined as the angle between the mechanical axis of the femur and the mechanical axis of the tibia. The mechanical axis of the femur is a line from the center of the femoral head to the femoral reference point at the knee joint. The mechanical axis of the tibia is formed by a lined from the proximal tibial reference point to the mid ankle reference point.¹⁷ The mechanical axes of the femur and tibia are shown on the left image by the black lines. The JSW was measured on the medial and lateral side as indicated by the black lines on the right image.

Clinical Data

Clinical data collected included: demographic data (age), BMI, inflammation markers (leukocyte count, C-reactive protein [CRP] levels), x-ray analysis (JSW, HKAA), and medication. Medication was grouped based on mode of action.

Statistical Analysis

All statistical analyses were performed using the statistical software R (R Foundation for Statistical Computing, Vienna, Austria, version 3.5.2) on GNU/Linux Debian 10 and SPSS 25 (IBM, Armonk, NY, USA) on Windows 10 (Microsoft, Redmond, WA, USA).¹⁸ The following R packages were used: lme4, lmerTest, tidyverse, car.^19-22 Groups were compared using paired Student t test,Wilcoxon rank or Kruskal-Wallis test, where appropriate, after testing for normality using the Kolmogorov-Smirnov test and Q-Q plots. Correlations were assessed using Spearman’s correlation coefficient. Variables showing an association with ΔCt values with a P value <0.1 were used for multivariable analysis. To assess the effect of BMI on gene expression, for each gene of interest a linear regression model was calculated for the control group, using the expression level as outcome and the dependent variables as described above. Similarly, to assess the effect of metformin stimulation and BMI on gene expression, linear mixed models were calculated. A similar approach using regression models for analysis of qPCR data has been described and validated previously and has the advantage of increased flexibility, as it can account for multiple sources of variation compared with the traditional ΔΔCt method.²³ The following modifications were made to the aforementioned protocol: Ct values were averaged, normalized and ΔCt values were then used as the response variable. In mixed models, patient identity was entered as random effect to account for repeated measures. Known risk factors (age, BMI, HKAA) were entered into the models to adjust for them. The final model was selected based on model Akaike information criterion (AIC). Using the β-coefficient of regression models as the ΔΔCt value, the fold increase was calculated as 2⁻^ΔΔCt.24 Model assumptions were assessed using residual versus fitted, normal Q-Q, scale-location and residuals versus leverage plots. A P value <0.05 was assumed to be statistically significant. Due to this study’s exploratory nature, sample size calculation was not performed. A total enrollment of 15 patients was the aim of this study. To asses interrater reliability on x-ray analysis, ICC (ICC 3,1, absolute agreement) was determined. Numerical variables are expressed as mean ± standard deviation. Categorical variables are expressed as count (percentage).

Results

A total of 15 patients were enrolled. Due to technical problems in the PCR experiments, 1 patient had to be excluded from the analysis, so that a total of 14 patients were analyzed. All 14 patients were female, with a mean age of 70.1 ± 10.2 years and a mean BMI of 30.5 ± 6.3 kg/m². Patients were either classified with an American Society of Anesthesiologists (ASA) score of 2 or 3. Three patients were diagnosed with type 2 diabetes, and no patient was insulin dependent or taking metformin. One patient was managed with lifestyle intervention only and not taking any antidiabetic medication, 1 patient was on linagliptin, the other patient was on gliclazide. Table 2 shows the baseline characteristics.

Table 2.

Baseline Characteristics of Patients Enrolled in the Study.^a

	Overall	BMI <25 kg/m²	BMI 25-35 kg/m²	BMI >35 kg/m²
n	14	4	6	4
BMI, kg/m²	30.54 (6.28)	23.49 (1.19)	29.96 (2.67)	38.46 (2.70)
Age, years	70.05 (10.21)	73.16 (12.58)	72.40 (9.55)	63.41 (7.70)
ASA score of 3	7 (50.0)	3 (75.0)	2 (33.3)	2 (50.0)
JSW lateral	4.81 (2.36)	6.27 (1.37)	4.27 (2.90)	4.16 (2.06)
JSW medial	4.25 (1.93)	2.29 (1.65)	5.37 (1.62)	4.52 (1.12)
HKAA	6.68 (4.28)	7.38 (4.35)	6.33 (4.84)	6.50 (4.53)
Leukocyte count, ×10³/L	8.75 (3.17)	7.73 (2.35)	8.95 (4.18)	9.46 (2.58)
CRP, mg/dL	5.07 (3.80)	1.38 (0.26)	6.08 (3.83)	7.25 (3.38)
PPI intake	6 (42.9)	1 (25.0)	3 (50.0)	2 (50.0)
Diabetes mellitus	3 (21.4)	1 (25.0)	1 (16.7)	1 (25.0)

BMI = body mass index; ASA = American Society of Anesthesiologists physical status classification system (all patients were either classified as 2 or 3); JSW = joint space width on the medial or lateral side; HKAA = hip-knee-ankle angle; CRP = C-reactive protein; PPI = proton pump inhibitor.

Continuous variables are displayed as mean (standard deviation), categorical data as count (percentage).

Cell Viability Assay

A metformin concentration of 1 mM (129.16 µg/mL) was found to have no cytotoxic effect in the CellTiter-Glo assay (χ² = 9, P = 0.342 for values after 24 and 48 hours) ( Fig. 2 ). Steady-state metformin plasma levels in human beings range between 1 and 5 µg/mL (7.7-38.7 µmol/L).²⁵ A concentration of 1 mM was found to mitigate loss of cell viability by a decreased production of reactive oxygen species (ROS).¹⁴ We are not aware of any study investigating the metformin concentration in synovial fluid. Thus, based on the positive results of the aforementioned study, and supported by the cell viability assay findings, a final metformin concentration of 1 mM was used for all experiments in order to detect changes in expression, if any at all.

Figure 2.

Results from CellTiter-Glo cell viability assay after 24 hours (top) and 48 hours (bottom). Bars indicate mean cell viability in percentage compared with control. Error bars show the standard deviation. No difference between treatment groups after 24 hours and 48 hours could be observed using a Kruskal-Wallis test.

Association of BMI and Gene Expression

BMI correlated only with ADAMTS5 (r = −0.69, P < 0.01) and MMP1 (r = −0.59, P = 0.03) ( Fig. 3 , Table 3 ) in univariable analysis. Multivariable linear regression was performed for each gene. The following variables were included in the regression models for ADAMTS5 and MMP1: age, BMI, HKAA, CRP level. Linear regression model coefficients are shown in Table 4 . BMI was found to be a significant predictor of gene expression in ADAMTS5 (β = −0.11, P = 0.034, n = 14, 1.079-fold increase for a BMI increase of 1 point) after adjusting for age, BMI, and HKAA. This indicates that patients with a BMI of 30 kg/m² had a 2.14-fold increase in ADAMTS5 expression compared to patients with a BMI of 20 kg/m² in the control group. Considering the exponential nature of PCR replication, this also indicates an exponential relationship of BMI and ADAMTS5. In our study, this exponential relationship was observed in a BMI range of 22 to 40 kg/m². For an increase of 1 point of BMI, there is a 1.08-fold increase, while for 10 points of BMI there is a 2.14-fold increase of ADAMTS5 gene expression relative to reference genes ( Fig. 4 ). BMI was not a significant predictor of MMP1 expression in multivariable regression (β = −0.08, P = 0.148). Similar results were obtained in mixed models, were patients with a BMI of 30 kg/m² had a 1.87-fold increase in ADAMTS5 expression compared to patients with a BMI of 20 kg/m² ( Table 5 ).

Figure 3.

A scatterplot of body mass index (BMI) and ΔCt values of ADAMTS5 (left) and MMP1 (right) with a regression line showing a significant negative correlation. In patients with higher BMI, there is an increased expression of these two genes (a small ΔCt value indicates higher expression levels). r, Spearman’s rank correlation coefficient; P, P value.

Table 3.

Correlation Coefficients of BMI with the ΔCt Values of Genes of Interest Using Spearman’s Rho in the Control Group.^a

	Age		BMI		HKAA		JSW M		CRP
	r	P	r	P	r	P	r	P	r	P
ADAMTS4	−0.35	0.266	0.56	0.063^†	0.02	0.957	0.71	0.012*	0.66	0.022*
ADAMTS5	0.41	0.150	−0.69	0.008*	0.25	0.385	−0.3	0.302	−0.57	0.035*
COL2A1	0.31	0.564	−0.09	0.919	0.12	0.827	0.66	0.175	0.03	1.000
HIF1	−0.36	0.203	0.39	0.170	0.02	0.934	0.27	0.357	0.42	0.137
HIF2	−0.1	0.727	0.04	0.892	−0.21	0.471	0.12	0.682	−0.14	0.627
IL4	0.29	0.501	−0.33	0.428	0.48	0.243	0.12	0.793	0.14	0.752
IL6	−0.42	0.132	0.19	0.512	−0.05	0.863	0.19	0.512	0.04	0.892
IL8	0.15	0.604	−0.51	0.063^†	−0.19	0.505	−0.54	0.047*	−0.55	0.040*
MMP1	0.09	0.762	−0.59	0.030*	−0.1	0.724	−0.33	0.246	−0.49	0.075^†
MMP13	−0.33	0.253	0.15	0.610	−0.37	0.188	0.24	0.404	0.15	0.610
MMP3	−0.02	0.940	−0.06	0.832	0.07	0.822	0.09	0.750	0.01	0.976
SOX9	−0.04	0.904	−0.03	0.928	0.15	0.613	0.11	0.704	0.24	0.400
SREBP	−0.03	0.928	0.06	0.844	0.28	0.331	0.01	0.988	−0.05	0.880
TIMP1	−0.25	0.382	0.02	0.964	0.11	0.707	0.38	0.175	−0.05	0.856

BMI = body mass index; CRP = C-reactive protein; HKAA = hip-knee-ankle angle; r = Spearman’s rank correlation coefficient; JSW M = medial joint space width; JSW L = lateral joint space width.

There is a moderate, significant, negative association of ΔCt values of the 2 extracellular matrix degrading enzymes ADAMTS5 and MMP1 and the BMI indicating that higher BMI correlates with increased ADAMTS5 or MMP1 expression ( Fig. 3 ).

†

P < 0.1. *P < 0.05.

Table 4.

Fixed-Effects Linear Regression Model for ADAMTS5 and MMP1.^a

	β Coefficient (Univariable)	95% CI	P	β Coefficient (Final)	95% CI	P	AIC (Final)
ADAMTS5
BMI	−0.10	−0.19, –0.02	0.018*	−0.11	−0.22, –0.01	0.034*	43.2
HKAA	0.02	−0.14, 0,17	0.826	0.01	−0.16, 0.16	0.952
Age	0.02	−0.04, 0.08	0.519	−0.01	−0.09, 0.06	0.684
CRP	−0.14	−0.29, 0.01	0.063^†	—	—	—
MMP1
BMI	−0.11	−0.20, –0.02	0.018*	−0.08	−0.20, 0.03	0.148	39.0
HKAA	−0.06	−0.22, 0.11	0.451	−0.08	−0.22, 0.06	0.215
Age	−0.01	−0.07, 0.07	0.933	−0.03	−0.09, 0.04	0.340
CRP	−0.19	−0.34, –0.04	0.016*	−0.15	−0.34, 0.03	0.094^†

BMI = body mass index; HKAA = hip-knee-ankle angle; CRO = C-reactive protein; AIC = Akaike information criterion; “—” = variable dropped in the final model.

The results of the linear regression model for ADAMTS5 accounting for known risk factors are shown. BMI decreases ΔCt values and thus increases gene expression. The left column shows the univariable coefficients of the variables included in the final model using backward stepwise selection (accounting for known risk factors for osteoarthritis and variables with an association with a P value <0.1).

†

P < 0.1. *P < 0.05.

Figure 4.

The predicted ΔCt values based on regression models for ADAMTS5 (top left), SOX9 (top right), and MMP1 (bottom left and right), holding all variables except metformin treatment, as well as body mass index (BMI) (top left), hip-knee-ankle angle (HKAA), and proton pump inhibitor (PPI) use (top right) and C-reactive protein (CRP) (bottom left) and HKAA (bottom right), constant. With higher BMI, CRP, or HKAA values, there is a decrease in the respective ΔCt values of ADAMTS5 and MMP1 (top left, bottom right and left) and thus increase in gene expression of the respective gene. Metformin leads to a small increase in ΔCt and thus a decrease in gene expression as depicted in all graphs. PPI intake is associated with a strong increase in ΔCt and thus decrease in SOX9 gene expression (top right).

Table 5.

Linear Mixed Regression Models of Selected Genes Accounting for Confounding Variables and Repeated Measures.^a

	β Coefficient (Univariable)	95% CI	P	β Coefficient (Final)	95% CI	P	AIC (Final)
ADAMTS5
Metformin	0.34	0.02, 0.66	0.043*	0.34	0.02, 0.61	0.043*	76.0
BMI	−0.08	−0.14, –0.012	0.015*	−0.09	−0.16, –0.02	0.020*
HKAA	0.01	−0.11, 0.12	0.919	−0.01	−0.11, 0.10	0.857
Age	0.02	−0.03, 0.06	0.457	−0.01	−0.05, 0.04	0.810
CRP	−0.12	−0.23, –0.02	0.028*	—	—	—
HIF-1a
Metformin	0.39	0.02, 0.75	0.044*	0.39	0.02, 0.75	0.044*	82.7
BMI	−0.01	−0.11, 0.10	0.966	−0.04	−0.15, 0.07	0.455
HKAA	0.01	−0.14, 0.17	0.852	0.08	−0.08, 0.25	0.311
Age	−0.03	−0.1, 0.03	0.276	−0.06	−0.14, 0.01	0.108
IL4
Metformin	0.30	0.06, 0.53	0.022*	0.30	0.06, 0.53	0.022*	30.0
BMI	−0.04	−0.12, 0.04	0.284	−0.04	−0.14, 0.07	0.444
HKAA	0.03	−0.09, 0.15	0.584	0.01	−0.10, 0,13	0.795
Age	0.01	−0.03, 0.06	0.472	<–0.01	−0.06, 0.05	0.997
MMP1
Metformin	0.47	0.17, 0.76	0.005*	0.47	0.17, 0.76	<0.001*	62.9
BMI	−0.11	−0.19, –0.03	0.010*	−0.07	−0.15, 0.01	0.088
HKAA	−0.07	−0.21, 0.07	0.297	−0.11	−0.21, –0.01	0.031*
Age	<0.01	−0.06, 0.06	0.978	−0.02	−0.06, 0.03	0.447
CRP	−0.19	−0.32, –0.06	0.008*	−0.16	−0.29, –0.03	0.019*
SOX9
Metformin	0.37	0.06, 0.68	0.026*	0.37	0.06, 0.68	0.026*	74.3
BMI	−0.02	−0.15, 0.11	0.798	−0.08	−0.18, 0.01	0.070
HKAA	0.06	−0.13, 0.25	0.006*	0.23	0.08, 0.37	0.006*
Age	−0.02	−0.10, 0.06	0.640	−0.05	−0.12, 0.01	0.098
PPI intake	1.60	0.27, 2.93	0.024*	2.39	1.29, 3.49	<0.001*

BMI = body mass index; CRP = C-reactive protein; HKAA = hip-knee-ankle angle; PPI = proton pump inhibitor; AIC = Akaike information criterion; “—” = variable dropped in the final model.

The left column shows the univariable β coefficients of the variables included in the full model (accounting for known risk factors for osteoarthritis and variables with an association with a P value <0.1). The right column shows the coefficients included in the final model after model selection based on AIC. Variables, where no β coefficients for the final model are given. Metformin was found to cause a significant reduction in the expression of ADAMTS5, MMP1, HIF-1a, IL4, and SOX9. For example, in SOX9, there was an increase in the ΔCt value of 0.37 after metformin treatment, indicating a reduction in gene expression.

P < 0.05.

Effect of Metformin Treatment of Chondrocytes on Gene Expression

Metformin had a significant treatment effect on ΔCt values in the following genes compared with the control group: IL4 (11.6 ± 0.6 vs. 11.3 ± 0.6, P = 0.03), MMP1 (7.2 ± 1.2 vs. 6.8 ± 1.1, P < 0.01), and SOX9 (5.1 ± 1.3 vs. 4.7±1.7, P = 0.04) ( Fig. 5 ). After controlling for confounding variables, this effect was significant in the following genes: ADAMTS5 (β = 0.34, P = 0.043, n = 14 per treatment group, 0.79-fold decrease) (controlled for BMI, HKAA, age), HIF-1a (β = 0.39, P = 0.04, n = 14 per treatment group, 0.76-fold decrease) (adjusted for BMI, HKAA, age), IL4 (β = 0.30, P = 0.02, n = 8 per treatment group, 0.81-fold decrease) (accounting for BMI, HKAA, age), MMP1 (β = 0.47, P < 0.01, n = 14 per treatment group, 0.72-fold decrease) (adjusted for BMI, HKAA, age, CRP level), SOX9 (β = 0.37, P = 0.03, n = 14 per treatment group, 0.77-fold decrease) (controlling for BMI, HKAA, age, PPI intake). Table 5 shows linear mixed models and coefficients. This indicates that stimulation with metformin causes a 21% decrease in ADAMTS5, a 24% decrease in HIF-1a, a 19% decrease in IL4, a 23% decrease in SOX9, and a 28% decrease in MMP1 gene expression. Other catabolic markers were not found to be affected by metformin stimulation. This relationship is graphically depicted in Fig. 4 , which shows predicted regression lines for MMP1 by HKAA, and for ADAMTS5 by BMI, holding all other variables constant. Fig. 6 and Table 6 show the fold change in gene expression based on regression results.

Figure 5.

The metformin treatment effect of selected genes of interest. This figure shows the mean ΔCt values in the control versus treatment group. The error bars indicate the standard deviation. Metformin significantly decreased expression of IL4, MMP1, and SOX9 in univariable analysis. “.” P < 0.1. *P < 0.05.

Table 6.

Fold Change in Gene Expression in Select Genes (Based on Mixed Model Regression Analysis).^a

Target	Variable	log2(FC)	95% CI	FC	P
ADAMTS5	Metformin	−0.34	−0.02, –0.61	0.79	0.043*
ADAMTS5	BMI 30 vs. 20 kg/m²	0,9	0.2, −1.6	1.87	0.020*
ADAMTS5	HKAA 10° vs. 0°	0.10	1.1, –1	1.07	0.857
ADAMTS5	Age 75 vs. 65 years	0.10	0.5, –0.4	1.07	0.810
HIF1a	Metformin	−0.39	−0.02, –0.75	0.76	0.044*
HIF1a	BMI 30 vs. 20 kg/m²	0.40	1.5, –0.7	1.32	0.455
HIF1a	HKAA 10° vs. 0°	−0.80	0.8, –2.5	0.57	0.311
HIF1a	Age 75 vs. 65 years	0.60	1.4, –0.1	1.52	0.108
IL4	BMI 30 vs. 20 kg/m²	0.40	1.4, –0.7	1.32	0.444
IL4	Metformin	−0.30	−0.06, –0.53	0.81	0.022*
IL4	HKAA 10° vs. 0°	−0.10	1, –1.3	0.93	0.795
IL4	Age 75 vs. 65 years	0.10	0.6, –0.5	1.07	0.997
MMP1	Metformin	−0.47	−0.17, –0.76	0.72	<0.001*
MMP1	HKAA 10° vs. 0°	1.10	2.1, 0.1	2.14	0.031*
MMP1	BMI 30 vs. 20 kg/m²	0.70	1.4, –0.1	1.62	0.088
MMP1	Age 75 vs. 65 years	0.20	0.6, –0.3	1.15	0.447
MMP1	CRP 10 vs. 0 mg/dL	1.60	2.9, 0.3	3.03	0.019*
SOX9	PPI	−2.39	−1.29, –3.48	0.19	<0.001*
SOX9	Age 75 vs. 65 years	0.50	1.2, –0.1	1.41	0.098
SOX9	BMI 30 vs. 20 kg/m²	0.80	1.8, –0.1	1.74	0.070
SOX9	HKAA 10° vs. 0°	−2.30	−0.8, –3.7	0.20	0.006*
SOX9	Metformin	−0.37	−0.68, –0.06	0.77	0.026*

BMI = body mass index; CRP = C-reactive protein; HKAA = hip-knee-ankle angle; PPI = proton pump inhibitor; FC = fold change.

The first 2 columns show the target gene and treatment group or variable of interest. The next 2 columns show the logarithm of the FC with the 95% confidence interval. Finally, on the last 2 columns we see the FC as well as the P. FC was calculated as FC = 2^−ΔΔCt, ΔΔC_t values were obtained from regression models ( Table 5 ).

P < 0.05.

Figure 6.

The fold change in gene expression in select genes. In this bar chart, we see the deviation from the baseline gene expression in select genes and various treatment groups or variables of interest (0 in the logarithmic scale on the left, 1 in the fold change ratio scale on the right). A bar above baseline indicates an increase in gene expression. Error bars indicate the 95% confidence intervals on the log scale. Fold change was calculated based on regression models ( Tables 5 and 6). For example in ADAMTS5 we see a 1.87-fold increase of gene expression in patients with a body mass index (BMI) of 30 kg/m² compared with those with BMI of 20 kg/m². FC, fold change; log(FC) log2 of fold change. *P < 0.05.

Association of HKAA and JSW and Gene Expression

Medial JSW was positively correlated with ADAMTS4(r = 0.71, P = 0.012), while there was a negative correlation with IL8 (r = −0.54, P = 0.047) in the control group ( Table 3 ). This could not be confirmed by mixed model multivariable regression, however (the variable was dropped in the final model). Instead we found that HKAA was a significant predictor of SOX9 expression (β = 0.23, P < 0.01, n = 14 per treatment group, 0.85-fold decrease for 1° of malalignment) (controlled for age, BMI, and PPI intake) in multivariable mixed model regression. Patients with a deviation of 10° HKAA showed a substantially reduced expression of SOX9 (0.2-fold) ( Tables 5 and 6 , Fig. 6 ).

Association of Inflammation Markers and Gene Expression

CRP showed a significant correlation with ADAMTS4(r = 0.7, P = 0.02), ADAMTS5 (r = −0.6, P = 0.04), and IL8 (r = −0.6, P = 0.04) in the control group ( Table 3 ). In the multivariable mixed model regression, CRP was a significant predictor of MMP1 gene expression (β = −0.16, P = 0.02, n = 14 per treatment group, 1.12-fold increase for 1-point increase in CRP level) (adjusted for BMI, HKAA, and age) ( Tables 5 and 6 , Fig. 6 ). Patients with a CRP level of 10 mg/dL had a 3-fold increase in MMP1 expression compared to patients with a level of 0 mg/dl. CRP was not a significant predictor of ADAMTS4, ADAMTS5, and IL8, however (the variable was dropped in the respective final model).

Association of Comedication

Oral use of PPIs was a significant predictor of reduced SOX9 expression in regression models (β = 2.39, P < 0.01, n = 14 per group, 0.19-fold decrease in PPI intake) (controlled for age, BMI, and HKAA) ( Tables 5 and 6 , Fig. 6 ).

Discussion

The purpose of this study was to assess the effect of BMI and metformin on the expression of important matrix genes in human chondrocytes using RT-qPCR. The hypothesis was that BMI and metformin have significant effects on these matrix genes and thus on the pathogenesis of OA of the knee joint. We found that an increase in BMI influences the expression of matrix degrading gene ADAMTS5, while metformin treatment at the cellular level causes a reduction of the catabolic genes ADAMTS5 and MMP1. Furthermore, we found a reduction of anabolic SOX9 gene expression in chondrocytes in patients with increased HKAA, and patients on PPIs. Finally, higher CRP levels were associated with increased MMP1 expression.

BMI is a known risk factor for knee OA. In addition, BMI is also associated with OA of the hand which indicates that other effects such as systemic inflammation might play a role in pathogenesis.^8,9 In our study, BMI was a significant predictor of the expression of ADAMTS5: Higher BMI values led to a small, exponential increase in gene expression. ADAMTS5, a member of aggrecanases, causes cartilage destruction. In contrast to our findings, a study applying ex vivo physiologic mechanical compression on human cartilage found a change of collagen type II and aggrecan, but no effect on ADAMTS4.²⁶ Another study found a positive association of BMI and TGFβ, aggrecan, MMP-13, IGF1, and TIMP2. As in our study, no effect on type 2 collagen could be found.⁷ Our results show that baseline expression of matrix genes is affected by BMI.

Metformin is the first-line treatment in type 2 diabetes mellitus, a risk factor for OA. The effects of metformin on human cartilage gene expression have not been investigated extensively. The mechanism of action of metformin is complex and not fully understood. It acts via both AMPK-dependent and AMPK-independent mechanisms.²⁷ Findings from clinical studies on metformin treatment in OA are equivocal. In one study using prospective data, metformin was found to reduce cartilage loss.¹² Another study showed that patients with knee OA treated with a combination of meloxicam and metformin had a higher improvement in several clinical scores.²⁸ Another study on patients prescribed metformin, however, could not find an association with the risk of OA.²⁹

In mouse chondrocytes, metformin was able to mitigate cytotoxic effects of IL-1β. Furthermore, metformin also reduced production of reactive oxygen species in that study.¹⁴ In another study, AMPK activation by metformin improved mitochondrial function and reduced oxidative stress.³⁰ In OA mouse model metformin reduced cartilage degradation.³¹

In agreement with these studies, our results showed that expression of catabolic genes can be suppressed after 48 hours of exposure to metformin. We found that metformin slightly decreased expression of catabolic genes ADAMTS5, MMP1. On the other hand, expression of anabolic genes HIF-1a, IL4, and SOX9 was slightly reduced as well.

Our findings are supported by another study that found a reduction of MMP3 and MMP13 in OA mouse knee model after metformin treatment.¹⁵ Also, metformin reduced MMP3 and ADAMTS5 expression in rat nucleus pulposus cell culture.¹⁶ Our results indicate that metformin might reduce expression of these genes and exert a protective effect on cartilage.

SOX9 is a transcription factor and plays an important role as regulator of cartilage formation and chondrocyte differentiation.³² In late-stage OA, SOX9 expression is reduced and its upregulation inhibits IL-1β induced inflammatory response.^33,34 Supporting our findings, metformin reduced SOX9 expression in esophageal cancer cells.³⁵ Contrary to this, metformin was found to increase levels of SOX9 expression in mouse cartilage.¹⁵ How this affects cartilage metabolism and clinical results in humans remains unclear.

HIF-1a is expressed at higher levels in degenerating cartilage than in intact cartilage. Under normal conditions, HIF-1a protein undergoes rapid degradation in proteasomes.³⁶ In hypoxemic conditions, HIF-1a expression is increased and acts cartilage protective.³⁷ We found that metformin caused a decrease in HIF-1a expression under normoxic conditions. This might be an indirect sign of lower inflammation and ROS levels and less oxygen consumption. However, it remains unclear how metformin affects HIF-1a expression under hypoxia when HIF-1a does not undergo rapid degradation. This is in agreement with studies that found reduced HIF-1a levels in various cancer types after treatment with metformin.^38-40

Finally, IL4 exerts anti-inflammatory effects via reduction of inflammatory cytokines and cytokines.⁴¹ In our study, metformin reduced IL4 expression. This confirms studies showing that metformin can reduce IL4 expression in cancer cells.⁴² Likewise, in nondiabetic heart failure patients treated with metformin, IL4 plasma levels were significantly reduced.⁴³ This might be another indirect sign of lower inflammation level and thus lower IL4 expression caused by metformin.

Our results showed a significant negative correlation of CRP levels and ΔCt values of MMP1 in multivariable regression analysis, indicating increased MMP1 expression in patients with higher CRP values. This is supported by a study showing increased MMP1 expression in endothelial cells after incubation with CRP.⁴⁴ At the same time, a systematic review found significantly increased CRP levels in patients with OA.⁴⁵ This again highlights the role of inflammation in OA.

Looking at the association of limb alignment on matrix gene expression, we found that patients with a varus or valgus malalignment had significantly reduced SOX9 gene expression. In animal models of induced OA, there was an increase in subchondral bone thickness, decrease in aggrecan content and increase in MMP3 expression.^46-48 SOX9 plays an important role in cartilage formation and chondrocyte hypertrophy and a study showed a reduction of SOX9 expression in arthritic chondrocytes.^32,49 The authors of that study concluded that the reason for this reduction can only be speculated upon and might be caused by an imbalance in SOX9 signaling pathways. As SOX9 is a potent inhibitor of hypertrophic differentiation, this might eventually lead to increased subchondral bone. We found that patients with a strong varus or valgus malalignment had a reduced SOX9 expression compared to patients with no or little malalignment in this end-stage OA population. It is possible that this observed association is at least partly due to a more severely damaged cartilage in patients with a high HKAA.

We also found that patients taking proton pump inhibitors had a reduced expression of SOX9. A similar effect was found in esophageal cells where PPIs downregulated SOX9 expression and thereby slowed progression of Barrett’s esophagus.⁵⁰ It is known that proton pump inhibitors change bone metabolism, especially in the elderly.⁵¹ The effect of proton pump inhibitors on cartilage, however, remains unclear. Our results are suggestive of a potentially negative effect. However, we cannot exclude confounding effects.

This study has several limitations: First, it was conducted using an exploratory study design, no sample size calculation was performed and only a small sample size was analyzed. This is at least partly counterbalanced by the significant findings. Significant results in underpowered studies overestimate the effect size, however.⁵² While further research is necessary to confirm our findings, we hope that this study can serve to generate new hypotheses and facilitate future research, considering the potential role of metformin in disease prevention. Similarly, the results of the cell viability assay have to be interpreted with caution due to the small sample size. Second, only female patients were included and results might be different in male patients. Third, the concentration of metformin used in our experiments was 20 to 30 times higher than physiologic steady-state plasma concentrations after oral intake. Thus, the findings in this study cannot be applied readily in clinical practice. Fourth, cells were stimulated for 48 hours only and thus long-term effects were not assessed. Fifth, cells were harvested for RNA extraction only after 2 cell passages. While prolonged cell culture of chondrocytes certainly induces cell dedifferentiation, this is a cumulative and linear process from hyaline cartilage toward fibrocartilage.^53,54 In clinical studies, passage 2 chondrocytes haven been used successfully to treat osteochondritis dissecans.⁵⁴ In order to stimulate cells with metformin and obtain enough cell material for further analysis, it was necessary to use chondrocytes of passage 2. Sixth, 3 patients included in this study suffered from diabetes. High glucose levels are known to alter gene expression.⁵⁵ These 3 patients were managed optimally with regard to diabetes, however, and thus any altered effects of diabetes on gene expression seem minimal or very unlikely. Finally, cartilage was collected from random sites intraoperatively, and there might be a site specific gene expression pattern in cartilage.

To summarize, this study is the first to investigate the effects of BMI and metformin on certain important cartilage matrix genes in primary human chondrocytes. We found that BMI is associated with a destructive effect linked to an increased expression of ADAMTS5. On the other hand, metformin stimulation can reduce expression of catabolic genes ADAMTS5 and MMP1. Thus, it might play a role in prevention of OA in obese patients. Limb malalignment was also linked to a reduction in the expression of SOX9. Likewise, intake of proton pump inhibitors was associated with a reduced SOX9 expression. This might be suggestive of destructive process. Higher CRP levels were associated with increased MMP1 expression, highlighting the role of systemic inflammation.

Footnotes

Author Contributions

Conception and design: BL, BSF, PSc, NS, MS, PSa. Analysis and interpretation of the data: PSc, NS, BL, MS, BSF, DG. Drafting of the article: PSc. Critical review of the article: BL, NS, BSF, AL, PSa. Provision of study materials or patients: AL, PSa. Statistical expertise: PSc. Collection and assembly of data: PSc, BL, NS, MS, DG, BSF. Administrative, technical, or logistic support: BL, NS, DG, BSF.

Acknowledgments and Funding

We would like to thank Mag. Gerold Schwantzer, Institute for Medical Informatics, Statistics and Documentation, Medical University of Graz, for his competent and dedicated consultations. The author(s) received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: AL reported grants by Johnson & Johnson, Alphamed, Globus, and Implantec. MS reported financial support by Eli Lilly, Pfizer, Bristol Mayer Squibb, Takeda, AbbVie, Novartis, Roche, MSD, CSL Behring, UCB. BL, BSF, DG, NS, PSa, and PSc reported no conflict of interest.

Ethical Approval

The procedures followed were in accordance with the ethical standards of the responsible committee (Ethics Committee of the Medical University of Graz, Austria, IRB #31-133) on human experimentation (institutional and national) and with the Helsinki Declaration of 1975, as revised in 2000.

Informed Consent

Written informed consent was obtained from all patients before the study.

Trial Registration

Not applicable.

ORCID iD

Paul Schadler

References

Woolf

Pfleger

Burden of major musculoskeletal conditions. Bull World Health Organ. 2003;81(9):646-56.

Deshpande

Katz

Solomon

Yelin

Hunter

Messier

, et al. The number of persons with symptomatic knee osteoarthritis in the United States: impact of race/ethnicity, age, sex, and obesity. Arthritis Care Res (Hoboken). 2016;68(12):1743-50. doi:10.1002/acr.22897

Losina

Paltiel

Weinstein

Yelin

Hunter

Chen

, et al. Lifetime medical costs of knee osteoarthritis management in the United States: impact of extending indications for total knee arthroplasty: costs of knee OA. Arthritis Care Res (Hoboken). 2015;67(2_suppl):203-15. doi:10.1002/acr.22412

Liu-Bryan

Terkeltaub

Emerging regulators of the inflammatory process in osteoarthritis. Nat Rev Rheumatol. 2015;11(1):35-44. doi:10.1038/nrrheum.2014.162

Lane

Brandt

Hawker

Peeva

Schreyer

Tsuji

, et al. OARSI-FDA initiative: defining the disease state of osteoarthritis. Osteoarthritis Cartilage. 2011;19(5):478-82. doi:10.1016/j.joca.2010.09.013

Silverwood

Blagojevic-Bucknall

Jinks

Jordan

Protheroe

Jordan

KP.

Current evidence on risk factors for knee osteoarthritis in older adults: a systematic review and meta-analysis. Osteoarthritis Cartilage. 2015;23(4):507-15. doi:10.1016/j.joca.2014.11.019

Pallu

Francin

Guillaume

Gegout-Pottie

Netter

Mainard

, et al. Obesity affects the chondrocyte responsiveness to leptin in patients with osteoarthritis. Arthritis Res Ther. 2010;12(3):R112. doi:10.1186/ar3048

Yusuf

Nelissen

Ioan-Facsinay

Stojanovic-Susulic

DeGroot

van Osch

, et al. Association between weight or body mass index and hand osteoarthritis: a systematic review. Ann Rheum Dis. 2010;69(4):761-5. doi:10.1136/ard.2008.106930

Sellam

Berenbaum

Is osteoarthritis a metabolic disease?

Joint Bone Spine. 2013;80(6):568-73. doi:10.1016/j.jbspin.2013.09.007

10.

Nieves-Plaza

Castro-Santana

Font

Mayor

Vilá

LM.

Association of hand or knee osteoarthritis with diabetes mellitus in a population of Hispanics from Puerto Rico. J Clin Rheumatol. 2013;19(1):1-6. doi:10.1097/RHU.0b013e31827cd578

11.

Eymard

Parsons

Edwards

Petit-Dop

Reginster

Bruyère

, et al. Diabetes is a risk factor for knee osteoarthritis progression. Osteoarthritis Cartilage. 2015;23(6):851-9. doi:10.1016/j.joca.2015.01.013

12.

Wang

Hussain

Wluka

Lim

Abram

Pelletier

, et al. Association between metformin use and disease progression in obese people with knee osteoarthritis: data from the Osteoarthritis Initiative—a prospective cohort study. Arthritis Res Ther. 2019;21(1):127. doi:10.1186/s13075-019-1915-x

13.

Chung

Lee

Hsieh

Hung

Lin

, et al. Combination COX-2 inhibitor and metformin attenuate rate of joint replacement in osteoarthritis with diabetes: A nationwide, retrospective, matched-cohort study in Taiwan. PLoS One. 2018;13(1):e0191242. doi:10.1371/journal.pone.0191242

14.

Wang

Yang

Zhang

Liu

Yao

Zhang

Protective effects of metformin against osteoarthritis through upregulation of SIRT3-mediated PINK1/Parkin-dependent mitophagy in primary chondrocytes. Biosci Trends. 2019;12(6):605-12. doi:10.5582/bst.2018.01263

15.

Feng

Pan

Zeng

Shao

, et al. Metformin attenuates cartilage degeneration in an experimental osteoarthritis model by regulating AMPK/mTOR. Aging (Albany NY). 2020;12(2_suppl):1087-103. doi:10.18632/aging.102635

16.

Chen

Xia

Pan

Zhou

, et al. Metformin protects against apoptosis and senescence in nucleus pulposus cells and ameliorates disc degeneration in vivo. Cell Death Dis. 2016;7(10):e2441. doi:10.1038/cddis.2016.334

17.

Durandet

Ricci

Saveh

Vanat

Wang

Esat

, et al. Radiographic analysis of lower limb axial alignments. Lect Notes Eng Comput Sci. 2013;2:1354-8.

18.

R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2018. https://www.R-project.org/.

19.

Bates

Mächler

Bolker

Walker

Fitting linear mixed-effects models using lme4. J Stat Softw. 2015;67(1):1-48. doi:10.18637/jss.v067.i01

20.

Kuznetsova

Brockhoff

Christensen

RHB

. lmerTest package: tests in linear mixed effects models. J Stat Softw. 2017;82(13):1-26. doi:10.18637/jss.v082.i13

21.

Wickham

. Tidyverse: easily install and load the “Tidyverse.”. Accessed September 16, 2020. https://CRAN.R-project.org/package=tidyverse

22.

Fox

Weisberg

An R companion to applied regression. 2nd Ed. Sage; 2011. http://socserv.socsci.mcmaster.ca/jfox/Books/Companion

23.

Steibel

Poletto

Coussens

Rosa

GJM

. A powerful and flexible linear mixed model framework for the analysis of relative quantification RT-PCR data. Genomics. 2009;94(2_suppl):146-52. doi:10.1016/j.ygeno.2009.04.008

24.

Livak

Schmittgen

TD.

Analysis of relative gene expression data using real-time quantitative PCR and the 2−ΔΔCT method. Methods. 2001;25(4):402-8. doi:10.1006/meth.2001.1262

25.

Glucophage [package insert]. Sanofi-Aventis Canada Inc; March 2018. Accessed March 1, 2019. http://products.sanofi.ca/en/glucophage.pdf

26.

Dolzani

Assirelli

Pulsatelli

Meliconi

Mariani

Neri

Ex vivo physiological compression of human osteoarthritis cartilage modulates cellular and matrix components. PLoS One. 2019;14(9):e0222947. doi:10.1371/journal.pone.0222947

27.

Rena

Hardie

Pearson

ER.

The mechanisms of action of metformin. Diabetologia. 2017;60(9):1577-85. doi:10.1007/s00125-017-4342-z

28.

Mohammed

Al-Shamma

Jassim

NA.

Evaluation of the clinical use of metformin or pioglitazone in combination with meloxicam in patients with knee osteoarthritis; using knee injury and osteoarthritis outcome score. Iraqi J Pharm Sci. 2014;23(2_suppl):13-23.

29.

Barnett

Jordan

Edwards

van der Windt

DA.

Does metformin protect against osteoarthritis? An electronic health record cohort study. Prim Health Care Res Dev. 2017;18(6):623-8. doi:10.1017/S1463423617000287

30.

Karnewar

Neeli

Panuganti

Kotagiri

Mallappa

Jain

, et al. Metformin regulates mitochondrial biogenesis and senescence through AMPK mediated H3K79 methylation: Relevance in age-associated vascular dysfunction. Biochim Biophys Acta Mol Basis Dis. 2018;1864(4 Part A):1115-28. doi:10.1016/j.bbadis.2018.01.018

31.

Zhang

Liu

Pan

Wang

, et al. Metformin limits osteoarthritis development and progression through activation of AMPK signalling. Ann Rheum Dis. 2020;79(5):635-45. doi:10.1136/annrheumdis-2019-216713

32.

Akiyama

Chaboissier

M-C

Martin

Schedl

de Crombrugghe

The transcription factor Sox9 has essential roles in successive steps of the chondrocyte differentiation pathway and is required for expression of Sox5 and Sox6. Genes Dev. 2002;16(21):2813-28. doi:10.1101/gad.1017802

33.

Lee

GI.

SOX trio decrease in the articular cartilage with the advancement of osteoarthritis. Connect Tissue Res. 2011;52(6):496-502. doi:10.3109/03008207.2011.585409

34.

Ouyang

Wang

Zhu

Fan

Overexpression of SOX9 alleviates the progression of human osteoarthritis in vitro and in vivo. Drug Des Devel Ther. 2019;13:2833-42. doi:10.2147/DDDT.S203974

35.

Honjo

Ajani

Scott

Chen

Skinner

Stroehlein

, et al. Metformin sensitizes chemotherapy by targeting cancer stem cells and the mTOR pathway in esophageal cancer. Int J Oncol. 2014;45(2_suppl):567-74. doi:10.3892/ijo.2014.2450

36.

Maxwell

Wiesener

Chang

G-W

Clifford

Vaux

Cockman

, et al. The tumour suppressor protein VHL targets hypoxia-inducible factors for oxygen-dependent proteolysis. Nature. 1999;399(6733):271-5. doi:10.1038/20459

37.

Yudoh

Nakamura

Masuko-Hongo

Kato

Nishioka

. Catabolic stress induces expression of hypoxia-inducible factor (HIF)-1α in articular chondrocytes: involvement of HIF-1α in the pathogenesis of osteoarthritis. Arthritis Res Ther. 2005;7(4):R904-R914. doi:10.1186/ar1765

38.

Wang

Tang

Sun

Feng

, et al. Suppression of tumor angiogenesis by metformin treatment via a mechanism linked to targeting of HER2/HIF-1α/VEGF secretion axis. Oncotarget. 2015;6(42):44579-92.

39.

Zhou

Chen

Deng

Liu

Liang

, et al. Metformin suppresses hypoxia-induced stabilization of HIF-1α through reprogramming of oxygen metabolism in hepatocellular carcinoma. Oncotarget. 2015;7(1):873-84.

40.

Chen

Feng

Zhang

Bian

Yang

Fang

, et al. Metformin inhibits gastric cancer via the inhibition of HIF1α/PKM2 signaling. Am J Cancer Res. 2015;5(4):1423-34.

41.

Wojdasiewicz

Poniatowski

ŁA

Szukiewicz

The role of inflammatory and anti-inflammatory cytokines in the pathogenesis of osteoarthritis. Mediators Inflamm. 2014;2014:561459. doi:10.1155/2014/561459

42.

Chiang

Chao

Hsu

Chien

Chiu

, et al. Metformin-treated cancer cells modulate macrophage polarization through AMPK-NF-κB signaling. Oncotarget. 2017;8(13):20706-18. doi:10.18632/oncotarget.14982

43.

Cameron

Morrison

Levin

Mohan

Forteath

Beall

, et al. Anti-inflammatory effects of metformin irrespective of diabetes status. Circ Res. 2016;119(5):652-65. doi:10.1161/CIRCRESAHA.116.308445

44.

Montero

Orbe

Varo

Beloqui

Monreal

Rodríguez

, et al. C-reactive protein induces matrix metalloproteinase-1 and -10 in human endothelial cells: implications for clinical and subclinical atherosclerosis. J Am Coll Cardiol. 2006;47(7):1369-78. doi:10.1016/j.jacc.2005.10.070

45.

Jin

Beguerie

Zhang

Blizzard

Otahal

Jones

, et al. Circulating C reactive protein in osteoarthritis: a systematic review and meta-analysis. Ann Rheum Dis. 2015;74(4):703-10. doi:10.1136/annrheumdis-2013-204494

46.

Ziegler

Goebel

Seidel

Cucchiarini

Pape

Madry

Effect of open wedge high tibial osteotomy on the lateral tibiofemoral compartment in sheep. Part III: analysis of the microstructure of the subchondral bone and correlations with the articular cartilage and meniscus. Knee Surg Sports Traumatol Arthrosc. 2015;23(9):2704-14. doi:10.1007/s00167-014-3134-y

47.

Wei

Hjerpe

Brismar

Svensson

Effect of load on articular cartilage matrix and the development of guinea-pig osteoarthritis. Osteoarthritis Cartilage. 2001;9(5):447-53. doi:10.1053/joca.2000.0411

48.

Panula

Lohmander

Rönkkö

Ågren

Helminen

Kiviranta

Elevated levels of synovial fluid PLA₂, strome-lysin (MMP-3) and TIMP in early osteoarthrosis after tibial valgus osteotomy in young beagle dogs. Acta Orthop Scand. 1998;69(2_suppl):152-8. doi:10.3109/17453679809117617

49.

Haag

Gebhard

Aigner

SOX gene expression in human osteoarthritic cartilage. Pathobiology. 2008;75(3):195-9. doi:10.1159/000124980

50.

Huang

Liu

Sun

Fang

J-Y

Wang

Xiong

Omeprazole prevents CDX2 and SOX9 expression by inhibiting hedgehog signaling in Barrett’s esophagus cells. Clin Sci. 2019;133(3):483-95. doi:10.1042/CS20180828

51.

Park

Ahn

Son

Kim

, et al. A proton pump inhibitor’s effect on bone metabolism mediated by osteoclast action in old age: a prospective randomized study. Gut Liver. 2015;9(5):607-14. doi:10.5009/gnl14135

52.

La Caze

Duffull

. Estimating risk from underpowered, but statistically significant, studies: was APPROVe on TARGET? Estimating risk in underpowered studies. J Clin Pharm Ther. 2011;36(6):637-41. doi:10.1111/j.1365-2710.2010.01222.x

53.

Leijten

JCH

Kip

van Blitterswijk

Post

, et al. Gene expression profiling of dedifferentiated human articular chondrocytes in monolayer culture. Osteoarthritis Cartilage. 2013;21(4):599-603. doi:10.1016/j.joca.2013.01.014

54.

Peterson

Minas

Brittberg

Lindahl

Treatment of osteochondritis dissecans of the knee with autologous chondrocyte transplantation: results at two to ten years. J Bone Joint Surg Am. 2003;85:17-24. doi:10.2106/00004623-200300002-00003

55.

Laiguillon

Courties

Houard

Auclair

Sautet

Capeau

, et al. Characterization of diabetic osteoarthritic cartilage and role of high glucose environment on chondrocyte activation: toward pathophysiological delineation of diabetes mellitus-related osteoarthritis. Osteoarthritis Cartilage. 2015;23(9):1513-22. doi:10.1016/j.joca.2015.04.026