Circulating Blood Metabolite Trajectories and Risk of Rheumatoid Arthritis among Military Personnel in the Department of Defense Biorepository

Abstract

Objectives:

We sought to identify metabolic changes potentially related to rheumatoid arthritis (RA) pathogenesis occurring in the blood prior to its diagnosis.

Methods:

In a U.S. military biorepository, serum samples collected at two timepoints prior to a diagnosis of RA were identified. These were matched to controls who did not develop RA by subject age, race, and time between sample collections and RA diagnosis time to stored serum samples. Relative abundances of 380 metabolites were measured using liquid chromatography-tandem mass spectrometry. We determined whether pre-RA case vs. control status predicted metabolite concentration differences and differences over time (trajectories) using linear mixed models, assessing for interactions between time, pre-RA status and metabolite concentrations. We separately examined pre-seropositive RA cases vs. matched controls and adjusted for smoking. Multiple comparison adjustment set the false discovery rate to 0.05.

Results:

291 pre-RA cases (80.8% pre-seropositive RA) were matched to 292 controls, all with two serum samples (2.7±1.6 years; 1.0±0.9 years before RA/matched date). 52.0% were female; 52.8% were White, 26.8% Black, 20.4% other race. Mean age was 31.2 (±8.1) years at earliest blood draw. Fourteen metabolites had statistically significant trajectory differences among pre-RA subjects vs. controls, including sex steroids, amino acid/lipid metabolism and xenobiotics. Results were similar when limited to pre-seropositive RA and after adjusting for smoking.

Conclusions:

In this military case-control study, metabolite concentration trajectory differences in pre-RA cases vs. controls implicated steroidogenesis, lipid/amino acid metabolism, and xenobiotics in RA pathogenesis. Metabolites may have potential as biomarkers and/or therapeutic targets preceding RA diagnosis.

INTRODUCTION

Rheumatoid arthritis (RA) is an inflammatory polyarthritis that destroys synovial joints, causing systemic complications and early mortality. RA affects approximately 1% of the population and two to four times more women than men.^1,2 No medications are curative and treatment, disability and lost productivity are costly.³ Biologic pathways in the initiation and propagation of autoimmunity and inflammation in pre-RA and the transition to clinical synovitis are not fully understood, but abnormalities of autoantibodies, cytokines and chemokines are detectable preceding diagnosis.^4–7 Assays of metabolic changes in biosamples obtained prior to RA may advance understanding of pathogenic events not driven by longstanding disease or therapies and potentially identify novel targets for prevention.

Current metabolomic technologies allow feasible, high-throughput assessment of many metabolites resulting from genomic, transcriptomic, and proteomic variability. Metabolomics provides a profile of biological status reflecting genetic and environmental interactions.⁸ The integration of metabolomics into epidemiologic studies helps to understand molecular pathophysiological processes, define metabolic changes, and discover novel biomarkers and new targets for intervention. In complex diseases, like RA, biological perturbations leading to disease involve multiple pathways that may be reflected in their metabolites. Using multivariable statistics in conjunction with metabolomic data, it is possible to describe patterns of metabolite changes that are highly discriminatory for disease state. Epidemiologic studies are ideally suited for using metabolomic profiling to detect metabolites predictive of disease, informing the mechanisms of disease pathogenesis; metabolomic profiling in large prospective cohorts has led to biomarker discovery for type II diabetes, renal failure, and cardiovascular disease.^9–11

We have previously reported the results of a large untargeted liquid chromatography tandem mass spectrometry (LC-MS/MS) metabolomic study among pre-RA samples and matched controls in two case-control collections nested within large cohorts of women.¹² Within the Nurses’ Health Study (NHS) cohorts, 4-acetomidobutanoate and several short-chain acylcarnitines were lower and spermidine was higher among pre-RA cases compared to matched controls. Homoarginine and N-monomethylarginine (NMMA) were elevated among women less than five years of sero+ RA, and short-chain acylcarnitines were lower among those who developed sero+ RA within less than five years. In an independent pre-RA case-control study from the Department of Defense Serum Biorepository (DoDSR), three short-chain acylcarnitines, associated with sero+ RA within five years, were replicated among females. These prior studies included only women, examined only one blood sample per subject, and 74% were drawn five or more years prior to RA.

We aimed to study changes in untargeted metabolites within samples from two timepoints both within a few years of RA diagnosis, compared to those of matched controls. Our goal was to characterize the trajectory of changes in metabolite concentrations among subjects developing RA, both males and females, compared to controls.

METHODS

Study Population and Blood Sample Collection

The DoDSR was started in 1985 to monitor the health of U.S. military personnel, collecting samples at enlistment, deployment, and every other year of service. Among individuals with banked serum samples, possible RA cases were identified based on ≥2 International Classification of Disease (ICD) codes for RA in the medical record and ≥1 rheumatologist encounter and records were reviewed, as previously described.⁷ RA cases met the 1987 American College of Rheumatology Revised Criteria for the Classification of RA by medical record review or rheumatologist’s diagnosis of RA.¹³ Serum samples were obtained from military personnel in a variety of settings, from medical clinics to field facilities. Blood collection protocols, serum processing, and storage varied; blood samples were sent to the nearest medical facility for processing to serum, transported to the DoDSR facilities and shipped to a central lab for storage at −30°C.¹⁴

We identified all male and female pre-RA cases with two stored serum samples prior to the date of RA diagnosis. Controls were identified and matched to the pre-RA cases based on age at RA diagnosis index date, sex, race, enlistment region, and sample storage duration at each of the two pre-RA time points (matched within one year for each sample). All serum samples and data were anonymized, and this protocol was approved by Partners’ Healthcare, the University of Colorado, and the Walter Reed National Military Medical Center Institutional Review Boards.

Clinical Covariate Data Collection

Men and women were included in this analysis. Subject data, including demographics, smoking status, height and weight, were abstracted in a standardized form from medical record review. Covariates also assessed by medical record review included cigarette smoking (cumulative pack-years) and body mass index (BMI: underweight < 18.5, normal weight 18.5-< 25, overweight 25-< 30 or obese ≥30 kg/m²), but these were not available for all subjects. Sero+ RA status of cases was determined based on the presence at or after RA diagnosis of anti-citrullinated peptide antibody (ACPA) test results by medical records, and direct testing using two versions of an anti-cyclic citrullinated peptide (CCP) assay [CCP2 (Axis-Shield) and CCP3.1 (Inova Diagnostics Inc., San Diego, CA)], and/or positivity of one or more RF isotypes IgA, IgG and IgM (Inova Diagnostics Inc., San Diego, CA). Controls were not tested for seropositivity as above.

Metabolomics Profiling

Serum samples were assessed using two liquid chromatography tandem mass spectrometry metabolomics platforms: hydrophilic interaction liquid chromatography coupled with positive ion mode MS detection to measure amines and polar metabolites (HILIC-Positive) and C18 chromatography coupled with negative ion mode MS detection to measure free fatty acids, bile acids, and metabolites of intermediate polarity (C18-Negative).¹⁵ Briefly, LC-MS systems comprised of Nexera X2 U-HPLC systems (Shimadzu Corp.; Marlborough, MA) coupled to Q Exactive or Exactive Plus orbitrap mass spectrometers (Thermo Fisher Scientific; Waltham, MA) were used to obtain high resolution mass profiling data. Raw LC-MS data were processed using TraceFinder software (version 3.3, Thermo Fisher Scientific; Waltham, MA) and Progenesis QI (Nonlinear Dynamics; Newcastle upon Tyne, UK) for feature detection and signal integration. With the exception of sterols, metabolites were identified by matching to authentic reference standards and samples. Sterols were putatively identified by matching measured masses using METLIN¹⁶.

For quality control, a total of 59 serum samples obtained from 10 RA cases and 20 controls were interspersed randomly among study samples to assess temporal drift in instrument response and batch effects on the metabolomics platforms. Metabolites with mean coefficient of variation (CV) >25%, an intra-class correlation of <40%, poor reproducibility after delays in processing, or low with-in person stability were excluded¹⁰. In metabolites with missingness in ≤ 25% of the subjects, missing values were imputed with half of the minimum value observed per metabolite to avoid excluding metabolites with true missingness patterns (e.g. xenobiotics). Subjects with ≥10% missingness in metabolites were excluded. Metabolite levels were log-transformed and Pareto-scaled within cohort and then pooled. A total of 380 named metabolites were included in these analyses.

Statistical Analyses

We conducted a longitudinal metabolome-wide association study to identify metabolites with trajectories that differed between RA cases and controls. Matched controls were assigned index dates corresponding to the RA diagnosis dates of their matched cases. For each metabolite, longitudinal changes in relative metabolite Z-scores were analyzed using linear mixed effect models (LMEM) with primary effect variables for time, RA status, and the interaction between time × RA status, while specifying subject and match-pair random intercept effects. As our objective was to identify differences in metabolite trajectories over time by pre-RA case-control status, the interaction between time and RA status was specified to capture the extent to which the rate of change in metabolite levels over time to varied by case-control status. Each unit difference in metabolite concentration reflects one standard deviation in the population. Each model produced three sets of β estimates: the first compared metabolite levels in controls at the first vs. second timepoint, the second compared metabolite levels in pre-RA cases vs. controls at baseline, and the third compared the change over time in metabolite level between the first and second timepoint in pre-RA cases vs. controls. The interaction term βs can be interpreted as the change in the standardized level of metabolite per year among pre-RA cases compared to controls.

In secondary analyses, we repeated analyses 1) adjusting each linear regression model for smoking pack-years and 2) including only the pre-RA cases who later developed sero+ RA and their controls. All statistical analyses were conducted in R. Linear mixed effects models were obtained in using the ‘lme4’ package, and p-values were calculated using the Satterthwaite approximation available in the ‘lmerTest’ package. We applied a local false discovery rate (FDR) correction to p-values obtained from the interaction terms in each model.¹⁷ To filter the metabolites with highest evidence to suggest altered trajectories over time by RA diagnosis group, signals were filtered at p<0.005 and FDR q<0.05. Lastly, we investigated steroid hormone metabolite results in sex-stratified analyses.

RESULTS

A total of 1142 serum samples contributed by 291 pre-RA cases and 292 matched controls passed quality control (QC) processing and were included in analyses. Each RA case (80.8% sero+, 51.9% female) had two blood samples selected during the pre-RA phase (1^st mean of 2.7±1.6 years; 2^nd mean of 1.0±0.9 years before clinical RA onset). Matched case-control sets were 52% female; mean age was 31.2 years at first blood draw and 33.9 years at second blood draw and 35.3 years at RA diagnosis (among cases) (Table 1) Smoking history was more common in pre-RA cases than controls (31.3% vs. 25.3%). Being overweight (48.5% vs. 26.0%) or obese (17.9% vs. 8.9%) were more common in pre-RA cases than controls, although there was a high proportion of missing BMI among controls. Among pre-RA cases, 58.1% were sero+ at earliest blood draw, 72.5% at 2^nd pre-RA blood draw, and 80.8% were later diagnosed with sero+ RA (analyzed as sero+ RA). The earlier of the two pre-RA samples were collected up to 12.1 years prior to RA. From the first blood collection, 26.4% did not progress to RA within 3 years; of these, 33 (44%) were seropositive.

Table 1.

Characteristics of the Pre-Rheumatoid Arthritis Cases and Controls in the Department of Defense Serum Biorepository Study Sample

Characteristics	Cases (N=291)	Controls (N=292)	All (N=583)
Age
1st Blood Draw, Mean (SD)	32.6 (8.1)	29.7 (7.9)	31.2 (8.1)
2nd Blood Draw, Mean (SD)	34.3 (8.3)	33.5 (7.5)	33.9 (7.9)
Male, N (%)	140 (48.1)	140 (47.9)	280 (48)
Race, N (%)
White	155 (53.3)	153 (52.4)	308 (52.8)
Black	76 (26.1)	80 (27.4)	156 (26.8)
Other	60 (20.6)	59 (20.2)	119 (20.4)
Smoking Status, N (%)
Ever	91 (31.3)	74 (25.3)	165 (28.3)
Never	191 (65.6)	168 (57.5)	359 (61.6)
Missing	9 (3.1)	50 (17.1)	59 (10.1)
BMI Category, N (%)
Underweight	0 (0)	1 (0.3)	1 (0.2)
Normal	93 (32)	45 (15.4)	138 (23.7)
Overweight	141 (48.5)	76 (26.0)	217 (37.2)
Obese	52 (17.9)	26 (8.9)	78 (13.4)
Missing	5 (1.72)	144 (49.3)	149 (25.6)
RA Features
Age at Diagnosis, Mean (SD)	35.3 (8.21)
Seropositive Status, N (%)
At 1st Blood Draw
Seropositive	165 (58.1)
Seronegative	82 (28.9)
Missing	37 (13.0)
At 2nd Blood Draw
Seropositive	208 (72.5)
Seronegative	52 (18.1)
Missing	27 (9.41)
Ever*
Seropositive	235 (80.8)
Seronegative	56 (19.2)
Years to RA Diagnosis, Mean (SD)
From 1st Blood Draw	2.7 (1.6)
From 2nd Blood Draw	1.0 (0.9)

^*Seropositivity in subsequent analyses based on ever ant-CCP or RF seropositive in medical records or anti-CCP or RF testing on stored samples.

Significant differences in the β estimates in the comparison of all pre-RA case vs. all matched control samples were seen for several metabolites, indicating lower concentrations in the pre-RA cases for steroid hormones, lipids and some xenobiotics. (Table 2) We identified 14 metabolites with evidence of altered trajectories over time in pre-RA cases vs. controls (p<0.005 and FDR q<0.05 in the time*pre-RA case/control status interaction). These fell into four main super class categories: steroids, lipids, amino acids and xenobiotics. Negative interaction effects (β_intx < 0), indicating a more rapid decline in the metabolite over time in the pre-RA cases compared to controls, were seen for several steroids, lipids, and some xenobiotics. A positive β_intx, indicating a significant increase over time in the pre-RA cases compared to controls, was observed for amino acids and naproxen. Of note, while not reaching the level of significance, positive β_intx were found for other xenobiotics, including acetaminophen (β =0.072, p=0.016, q =0.146), acetaminophen glucuronide (β = 0.065, p=0.031, q=0.186), ibuprofen (β = 0.052, p=0.056, q=0.214) and omeprazole (β = 0.0.064, p= 0.018, q= 0.154).

Table 2.

Changes in Relative Metabolites per year in 292 Pre-Rheumatoid Arthritis (RA) Cases and 291 Controls for Metabolites with Differential Trajectories over Time (at p<0.005 and FDR q<0.05)

				Time		Group Pre-RA Cases vs. Controls		Interaction Time*Pre-RA Status
Super Class	Sub Pathway	Name	HMDB ID	β T	p	β G	p	β Intx	p	q
Steroids	Androsterone Glucuronides	±androsterone glucuronide	HMDB0002829	0.017	0.046	−0.359	1.7E-07	−0.074	4.9E-06	1.39E-03
	Pregnenolone Steroids	±17-hydroxypregnenolone sulfate	HMDB0000416	−0.046	1.48E-04	−0.398	2.0E-07	−0.086	3.8E-04	0.020
		pregnenolone sulfate	HMDB0000774	0.001	0.939	−0.390	1.1E-07	−0.064	0.001	0.037
Lipids	Long Chain Fatty Acids	hydroxymyristate	HMDB0002261	−0.016	0.239	−0.349	1.2E-05	−0.089	0.001	0.017
	Fatty Acid Metabolism (Acyl Carnitine)	C2 carnitine	HMDB0000201	0.178	6.27E-43	−0.127	0.067	−0.075	0.002	0.045
		C9 carnitine	HMDB0013288	0.009	0.460	−0.275	0.000	−0.077	0.002	0.039
	Glycerophospholipids	C40:7 phosphatidylethanolamine plasmalogen	HMDB0011394	0.096	2.00E-14	−0.371	0.000	−0.075	0.003	0.049
Amino Acids	Leucine, Isoleucine and Valine Metabolism	N-acetylleucine	HMDB0011756	−0.043	0.002	0.251	0.003	0.115	0.000	0.004
	Tyrosine Metabolism	tyramine	HMDB0000306	−0.036	0.001	0.865	0.000	0.082	0.000	0.016
Xenobiotics	Food/Plant Components	levulinate	HMDB0000720	0.139	6.14E-23	−0.489	8.5E-10	−0.118	1.7E-05	0.002
	Caffeine/Xanthine Metabolites	1,7-dimethyluric acid	HMDB0011103	0.044	0.001	−0.181	0.022	−0.091	0.001	0.028
		theophylline	HMDB0001889	0.033	0.010	−0.163	0.047	−0.083	0.002	0.040
		caffeine	HMDB0001847	0.046	3.82E-04	−0.184	0.023	−0.084	0.001	0.037
	Drugs	naproxen	HMDB0001923	0.022	0.145	0.390	1.6E-06	0.096	0.001	0.038

p: p-value; q: false discovery rate adjusted q-statistic for likelihood ratio test of the time * group effect

C: Carbon

β_T: change in relative metabolite concentration per year in controls; β_G: difference in relative metabolite concentration in pre-RA cases vs. controls; β_Intx: change in β_T per year given pre-RA case status, i.e. difference in rate of change in relative metabolite concentrations per year in pre-RA cases vs. controls

^±Preliminary IDs assigned to these metabolites

Models adjusted for age at blood draw, race, and sex and included subject and match-pair specific random effects

After adjustment for smoking, fewer significant metabolites were identified and no new metabolites were identified at p< 0.05 and FDR q< 0.05 significance; several were now just greater than this threshold, however, including two lipid metabolites and theophylline. (Table 3) For the subgroup of pre-sero+ RA cases (n=229) vs. matched controls over time, after adjustment for smoking, changes in the three steroid hormone metabolites remained significantly associated with pre-RA case status, and no new metabolites were identified at our a priori significance threshold. (Table 4) Sex-stratified analyses revealed decreasing trajectories over time pre-RA for the three significant steroid metabolites in both sexes and a strong decrease in androsterone among men. (Table 5)

Table 3.

Changes in Metabolites per Year in 282 Pre-Rheumatoid Arthritis (RA) Cases vs. 242 Controls for Metabolites with Differential Trajectories, adjusted for smoking

				Time		Pre-RA Cases vs. Controls		Time*Pre-RA Status
Super Class	Sub-Pathway	Name	HMDB	β T	p	β G	p	β Intx	p	q
Steroids	Androsterone Glucuronides	±androsterone glucuronide	HMDB0002829	0.018	0.037	−0.372	3.06E-07	−0.071	2.60E-05	0.005*
	Pregnenolone Steroids	±17-hydroxypregnenolone sulfate	HMDB0000416	−0.040	0.002	−0.411	4.57E-07	−0.090	2.85E-04	0.015*
		pregnenolone sulfate	HMDB0000774	0.004	0.716	−0.410	1.51E-07	−0.069	8.40E-04	0.032*
Lipids	Long Chain Fatty Acids	hydroxymyristate	HMDB0002261	−0.020	0.144	−0.321	1.16E-04	−0.088	1.25E-03	0.040*
	Fatty Acid Metabolism (Acyl Carnitine)	C2 carnitine	HMDB0000201	0.173	8.08E-38	−0.170	0.017	−0.071	0.004	0.075
		C9 carnitine	HMDB0013288	0.008	0.524	−0.263	0.001	−0.075	0.003	0.062
	Glycerophospholipids	C40:7 phosphatidylethanolamine plasmalogen	HMDB0011394	0.092	5.27E-12	−0.394	8.71E-07	−0.067	0.009	0.102
Amino Acid	Leucine, Isoleucine and Valine Metabolism	N-acetylleucine	HMDB0011756	−0.042	0.003	0.250	5.13E-03	0.113	7.67E-05	0.007*
	Tyrosine Metabolism	tyramine	HMDB0000306	−0.040	0.001	0.888	4.88E-25	0.083	2.67E-04	0.014*
Xenobiotics	Food/ Plant Components	levulinate	HMDB0000720	0.136	3.66E-20	−0.524	2.40E-10	−0.117	3.48E-05	0.005*
	Caffeine/Xanthine Metabolism	1,7-dimethyluric acid	HMDB0011103	0.045	0.001	−0.148	0.074	−0.090	0.001	0.037*
		theophylline	HMDB0001889	0.033	0.016	−0.116	0.180	−0.079	0.003	0.072
		caffeine	HMDB0001847	0.045	0.001	−0.154	0.069	−0.085	0.002	0.046*
	Drugs	naproxen	HMDB0001923	0.016	0.315	0.385	7.62E-06	0.100	0.002	0.044*

p: p-value; q: false discovery rate adjusted q-statistic for likelihood ratio test of the time * group effect

C: Carbon,

^*p< 0.05 and FDR q < 0.05

β_T: change in relative metabolite concentration per year in controls; β_G: difference in relative metabolite concentration in pre-RA cases vs. controls; β_Intx: change in β_T per year given pre-RA case status, i.e. difference in rate of change in relative metabolite concentrations per year in pre-RA cases vs. controls

^**p< 0.05 and FDR q < 0.05 after adjustment for smoking

^±Preliminary IDs assigned to these metabolites

Models adjusted for age at blood draw, race, and sex and included subject and match-pair specific random effects

Table 4.

Changes in Metabolites per Year in 229 Pre-Seropositive Rheumatoid Arthritis (RA) Cases and 193 Controls for Metabolites with Differential Trajectories over Time, adjusted for smoking

				Time		Pre-RA Cases vs. Controls		Time*Pre-RA Status
Super Class	Sub-Pathway	Name	HMDB	β T	p	β G	p	β Intx	p	q
Steroids	Androsterone Glucuronides	±androsterone glucuronide	HMDB0002829	0.025	0.015	−0.411	1.12E-06	−0.095	3.44E-06	0.001*
	Pregnenolone Steroids	±17-hydroxypregnenolone sulfate	HMDB0000416	−0.032	0.029	−0.442	3.60E-06	−0.112	1.91E-04	0.025*
		pregnenolone sulfate	HMDB0000774	0.014	0.264	−0.422	3.07E-06	−0.089	4.46E-04	0.039*
Lipids	Long Chain Fatty Acids	hydroxymyristate	HMDB0002261	−0.021	0.192	−0.272	0.003	−0.075	0.018	0.117
	Fatty Acid Metabolism (Acyl Carnitine)	C2 carnitine	HMDB0000201	0.171	4.03E-33	−0.185	0.014	−0.067	0.012	0.105
		C9 carnitine	HMDB0013288	0.014	0.321	−0.267	0.003	−0.079	0.005	0.074
	Glycerophospholipids	C40:7 phosphatidylethanolamine plasmalogen	HMDB0011394	0.104	2.78E-11	−0.415	7.20E-06	−0.056	0.066	0.242
Amino Acids	Leucine, Isoleucine and Valine Metabolism	N-acetylleucine	HMDB0011756	−0.037	0.026	0.231	0.020	0.101	0.002	0.066
	Tyrosine Metabolism	tyramine	HMDB0000306	−0.044	0.001	0.901	3.39E-20	0.080	0.003	0.070
Xenobiotics	Food/Plant Components	levulinate	HMDB0000720	0.147	9.40E-18	−0.507	6.10E-08	−0.103	0.002	0.063
	Caffeine/Xanthine Metabolites	1,7-dimethyluric acid	HMDB0011103	0.058	2.55E-04	−0.238	0.001	−0.101	0.002	0.061
		theophylline	HMDB0001889	0.051	0.001	−0.243	0.001	−0.104	9.48E-04	0.053
		caffeine	HMDB0001847	0.058	2.37E-04	−0.200	0.037	−0.081	0.010	0.100
	Drugs	naproxen	HMDB0001923	0.017	0.360	0.384	1.15E-04	0.112	0.003	0.068

p: p-value; q: false discovery rate adjusted q-statistic for likelihood ratio test of the time * group effect

C: Carbon,

^*p< 0.05 and FDR q < 0.05

β_T: change in relative metabolite concentration per year in controls; β_G: difference in relative metabolite concentration in pre-RA cases vs. controls; β_Intx: change in β_T per year given pre-RA case status, i.e. difference in rate of change in relative metabolite concentrations per year in pre-RA cases vs. controls

^**p< 0.05 and FDR q < 0.05 after adjustment for smoking

^±Preliminary IDs assigned to these metabolites

Models adjusted for age at blood draw, race, and sex and included subject and match-pair specific random effects

Table 5.

Changes in Relative Steroid Metabolites per year in 292 Pre-Rheumatoid Arthritis (RA) Cases and 291 Controls for Metabolites with Differential Trajectories over Time – Stratified by Subject Sex

				Time		Group Pre-RA Cases vs. Controls		Interaction Time*Pre-RA Status
Sub Pathway	Name	HMDB ID	Sex	β T	p	β G	p	β Intx	p
Androsterone Glucuronides	±androsterone glucuronide	HMDB0002829	Men	0.018	0.119	−0.143	0.075	−0.069	3.49E-04
			Women	0.018	0.124	−0.535	5.70E-07	−0.084	0.001
Pregnenolone Steroids	±17-hydroxy-pregnenolone sulfate	HMDB0000416	Men	−0.047	0.017	−0.308	0.002	−0.056	0.087
			Women	−0.049	0.002	−0.447	8.15E-05	−0.119	0.001
	pregnenolone sulfate	HMDB0000774	Men	0.000	0.990	−0.364	1.27E-04	−0.064	0.010
			Women	0.002	0.867	−0.391	3.37E-04	−0.071	0.030

p: p-value

β_T: change in relative metabolite concentration per year in controls; β_G: difference in relative metabolite concentration in pre-RA cases vs. controls; β_Intx: change in β_T per year given pre-RA case status, i.e. difference in rate of change in relative metabolite concentrations per year in pre-RA cases vs. controls

^±Preliminary IDs assigned to these metabolites

Models adjusted for age at blood draw, race, and included subject and match-pair specific random effects

DISCUSSION

Small molecule products or intermediates of metabolism are potential markers for RA development. In this case: control study nested within the DoDSR, we identified metabolites whose concentrations changed significantly during the pre-RA phase, and included steroids, lipids, amino acids, and xenobiotics. (Table 6)

Table 6.

Major Metabolites implicated in Department of Defense Serum Biorepository study of 291 Pre-RA cases and 292 Matched controls, each with banked samples at two points in time prior to RA Diagnosis (or matched Index Date)

Pathway	Metabolites	Pre-RA Cases vs. Controls	Direction of Trajectory	Potential Role in RA Pathogenesis	Refs.
Steroid Biosynthesis	androsterone glucuronide, 17-hydroxypregnenolone sulfate isomer, pregnenolone sulfate	Lower	Decreasing	Androgen to estrogen conversion in inflammatory cells; systemic inflammation suppressing HPA axis and pregnenolone formation	18–24,26
Acyl Carnitines and related lipid metabolites	Hydroxymyristate, C2 carnitine, C9 carnitine, C40:7 PE plasmalogen	Lower	Decreasing	Enhanced cellular oxidation, oxidative stress	12,28–32
Amino Acid Metabolites	N-acetyl leucine, tyramine	Higher	Increasing	Amine acid and polyamine catabolism abnormal in RA	12,34,35
Food/Plant components and Xenobiotics	Levulinate, caffeine/xanthine metabolites (1,7-dimethyluric acid, caffeine, theophylline), naproxen	Levulinate, caffeine and xanthine metabolites: lower; naproxen: higher	Levulinate and caffeine/xanthine metabolites: decreasing; Naproxen: increasing	Levulinate and caffeine/xanthine: either decreased intake or increased catabolism Naproxen: taken for pre-RA symptoms	36–39

Several steroid metabolites concentrations were lower in pre-RA cases than controls and, furthermore, their decline over time was significantly faster in the pre-RA cases than controls. These included androsterone glucuronide (the major circulating human androgen), 17-hydroxypregnenolone sulfate, and pregnenolone sulfate. Androgen deficiencies have been identified in RA and reduced androgens have been found to precede RA in both sexes in another prior RA metabolomics study.^18–20 Androgens may inhibit Th1 responses, induce regulatory B cells, suppress Th17 differentiation, and inhibit interleukin (IL)-1-beta, IL-6, and tumor necrosis factor (TNF) production.^21–23 Androgen may be converted to estrogen in inflammatory cells.²⁴ Systemic inflammation also suppresses the hypothalamic-pituitary-adrenal axis and pregnenolone formation, the rate-limiting step of adrenal steroidogenesis.^25,26 Total and free testosterone, and dehydroepiandrosterone sulfate (DHEAS), levels were not lower pre-RA compared to controls among women in a past NHS study, and current analyses revealed the decrease in androsterone gluconuride pre-RA was particularly strong among men.²⁷ Several metabolites involved in fatty acid metabolism significantly decreased in pre-RA cases vs. controls. Our work and those of others have implicated acylcarnitines as inversely associated with RA risk.^12,28 Acylcarnitines transport acyls into the mitochondrial matrix for β-oxidation; oxidative stress can cause deficiency.²⁹ Patients with RA have been observed to have low circulating acylcarnitine levels, possibly secondary to enhanced cellular oxidation.³⁰ Other lipid metabolites, including C40:7 phosphatidylethanolamine plasmalogen and hydroxymyristate also decreased significantly over time in pre-RA. Plasmalogens, cellular membrane components, may protect other phospholipids from oxidative stress.³¹ In a past study, the plasmalogen proportion of synovial fluid phospholipids in RA patients was lower than in controls.³² Hydroxymyristate, a fatty acid derivative, also fell preceding RA, also suggesting defects in energy utilization.

Elevations in two amino acids were associated with developing RA. Tyramine, an amino acid, is a catecholamine-releasing agent found in animals and plants. It is metabolized by monoamine oxidases in humans.³³ Polyamine catabolism has been reported to be elevated in RA synovial fibroblasts.³⁴ Our previous study of female NHS participants also pointed to polyamine metabolism, involved in aging and oxidative stress¹². N-acetylleucine, a modified amino acid used for vertigo and cerebellar ataxia, is thought to play a role in the mammalian Target of Rapamycin (mTOR) signaling pathway, a critical regulator of T cell proliferation, differentiation and function.³⁵

Five xenobiotics displayed significant changes during pre-RA phase vs. controls. Naproxen, a non-steroidal anti-inflammatory drug (NSAID) often taken for joint pain and swelling, increased significantly among pre-RA subjects compared to controls, an expected proof-of-concept for our analysis; ibuprofen, acetaminophen and omeprazole, also likely taken for early RA symptoms, were also near the significance threshold. Three metabolites (caffeine, theophylline, 1,7-dimethyluric acid) involved in xanthine metabolism declined significantly compared to controls. Individuals developing RA may reduce caffeine intake, although a past study in NHS did not find a relationship between caffeine intake and RA risk.³⁶ These analyses cannot separate the effects of intake from those of catabolism, and, alternatively, caffeine and related xanthine metabolites may be consumed by systemic inflammation. Caffeine is known to downregulate STAT-1 signaling with potential anti-inflammatory effects in RA.³⁷ Theophylline also has anti-inflammatory properties in rats and systemic inflammation may reduce its concentration.³⁸.

The fifth significant xenobiotic, levulinate, was lower in pre-RA cases than controls and declined significantly among pre-RA cases even after adjustment for smoking. Levulinate is found in consumer cosmetic products, herbicides, pharmaceuticals, and in cigarettes to increase nicotine delivery and binding to neural receptors. In animal models, it is catabolized via beta-oxidation to acetyl-CoA, formate and lactate.³⁹ Subjects may have decreased smoking as RA symptoms developed or lower circulating levulinate may reflect impairments in beta-oxidation for energy production.

Although the largest of its kind, sample sizes were limited for subgroup analyses. This RA population differs from the older women in NHS. Men represent approximately 25% of RA cases in the general population. Nonetheless, as a follow-up to earlier NHS studies, we observed similar but stronger patterns of metabolite changes. Smoking data were not available for all subjects. BMI was missing for more controls than cases, likely due to less healthcare system contact. (We attempted to adjust for and stratify by BMI, but due to high missingness, these models did not converge.) Both smoking and obesity have known influences on metabolism and are strongly related to RA risk^40–43. While the current study did not point to metabolic pathways connecting these risk factors to RA, future studies should assess their effects on RA-related metabolites. Other potential confounders such as diet, alcohol, and physical activity were not available. It is not known how NSAID intake may have affected other metabolite concentrations and not all sterol metabolite identities have been fully confirmed. Not all metabolites in each implicated pathway changed significantly over time. Metabolic pathways are complex; some metabolites may increase as others are consumed or feedback loops may lead to suppression when downstream metabolites are over-produced. Measures of systemic inflammation (e.g. C-reactive protein) were not available and relationships between metabolites and specific inflammatory pathways should be addressed in future studies. Finally, assumptions in statistical modeling may have resulted in identifying falsely significant or missing significant metabolites.

Few past studies have used metabolomics to investigate RA pathogenesis. The current study is large and unique as samples were from defined timepoints prior to RA, allowing for discovery of metabolic perturbations not due to treatment that may drive pathogenesis. We matched cases and controls, accounted for multiple comparisons and adjusted for smoking, a strong RA risk factor and potential confounder, analyzed sero+ RA and steroid hormones by sex. Moreover, metabolite change trajectories over time pre-RA onset adds important biologic and mechanistic information that a single point in time analysis does not. Several metabolites and pathways identified were previously implicated in RA pathogenesis or identified in studies in other populations^12,20,28.

Differences in trajectories of several metabolites were identified among patients with pre-RA versus controls. Implicated biochemical pathways included sex steroid, lipid, amino acid and xanthine metabolism. Increasing concentrations of NSAIDs in the time period prior to RA in cases vs. controls provided proof-of-concept. We identified metabolic perturbations before RA diagnosis that could lead to insights into disease pathogenesis and be developed as biomarkers to predict disease or therapeutic targets. Further studies are needed to perform absolute quantifications and to assess whether these metabolites could be biomarkers of RA risk.

KEY MESSAGES:

• Metabolomics using liquid chromatography and tandem mass spectroscopy can detect biological perturbations leading to disease that involve the activation of multiple biologic pathways reflected in the metabolites in these pathways.

• Past studies have identified lower concentrations of sex steroid hormones, in particular androgens, among individuals with RA compared to controls.

• Only a few past metabolomic studies have been conducted using samples from prior to RA disease onset. These have revealed lowered concentrations in sex steroids and lipids, such as carnitines and plasmalogens.

• In this case-control study nested within the U.S. Department of Defense Serum Biorepository, we identified two stored blood samples from 291 individuals who later developed RA (the first 2.7±1.6 years prior to RA onset and the second 1.0±0.9 years prior to RA onset). These were matched by date to two stored serum samples from age-, sex- and race- matched controls from individuals who did not develop RA. We found that the trajectory over time for 14 metabolites changed significantly in subjects preceding diagnosis of RA compared to controls. These findings were robust to adjustment for multiple comparisons and for smoking status, and were similar when limited to cases developing seropositive RA.

• These results implicate changes in steroid hormones, amino acid, lipid, and xanthine metabolism prior to RA onset and likely involved in RA pathogenesis. Not only will these findings lead to insights into RA disease pathogenesis, but they could lead directly to development of biomarkers to predict disease or become new therapeutic targets.