Associations of 25 structural, degradative, and inflammatory candidate genes with lumbar disc desiccation, bulging, and height narrowing

Abstract

Objective

To examine the allelic diversity of structural, inflammatory, and matrix-modifying gene candidates and their association with disc degeneration.

Methods

Subjects were 588 men ages 35–70 years. We investigated associations of single-nucleotide polymorphisms in AGC1 and in 12 collagen, 8 interleukin, and 4 matrix metalloproteinase genes with quantitative magnetic resonance imaging measurements of disc desiccation and disc bulging and height narrowing scores, after controlling for age and suspected risk factors. Analyses were performed using QTDT software. P values were derived from 1,000 permutations, and empirical P values for global significance also were applied.

Results

Twelve of the 99 variants in 25 selected candidate genes provided evidence of association (P < 0.05) with disc signal intensity in the upper and/or lower lumbar regions. Allelic variants of AGC1 (rs1042631; P = 0.001), COL1A1 (rs2075555; P = 0.005), COL9A1 (rs696990; P = 0.00008), and COL11A2 (rs2076311; P = 0.018) genes provided the most significant evidence of association with disc signal intensity. The same variants of AGC1 (P = 0.010) and COL9A1 (P = 0.014), as well as variants in the COL11A1 gene (rs1463035 [P = 0.004]; rs1337185 [P = 0.015]) were also associated with disc bulging, as was AGC1 with disc height narrowing (rs1516797; P = 0.005). In addition, 4 allelic variants in the immunologic candidate genes (rs2071375 in IL1A [P = 0.027]; rs1420100 in IL18RAP [P = 0.005]) were associated with disc signal intensity.

Conclusion

Genetic variants account for interindividual differences in disc matrix synthesis and degradation. The accuracy of the quantitative disc signal intensity measurements we used likely enhanced our ability to observe these associations. Our findings shed light on possible mechanisms of degeneration and support the view that disc degeneration is a polygenetic condition.

Intervertebral disc degeneration is a suspected cause of common back pain (1), but both the etiology and pathogenesis of disc degeneration are poorly understood (2). A 1992 review of relevant scientific literature identified physical demands of occupational and leisure activities as primary risk factors, while the role of genetics was uncertain (3). However, results from later twin studies suggested that heredity plays a major role, accounting for an estimated 34–74% of variance in disc degeneration (4, 5).

Genes coding for the structural components of intervertebral discs are obvious candidates that affect disc degeneration. For example, mutations of major collagen genes, especially those disrupting the formation of the collagen triple helix, are known causes of severe connective tissue disorders (6), while more subtle structural variations in aggrecan and types IX and XI collagen proteins have been implicated in “degenerative disc disease” (6, 7). Most of the previous candidate gene studies for disc degeneration have examined associations with relatively rare polymorphisms in structural matrix proteins, using qualitative assessments of disc degeneration or symptom complaints.

Among the 11 genes that have been investigated in 21 studies previously performed, 9 (AGC1, COL1A1, COL9A2, COL5A1, COL9A3, COL11A2, IL1, IL2, and MMP3) were associated with disc degeneration, pathologic changes, or associated symptoms in at least 1 study (8–17). However, phenotypes vary widely, and the studies included both symptoms assumed to be associated with disc pathology and qualitative assessments of disc degeneration. Furthermore, the qualitative assessments of disc degeneration were typically categorical scores (usually a 4-point scale) based on disc space characteristics and vertebral osteophytes (bony structures). Despite such study limitations, there is reasonably robust evidence of an association between disc degeneration and the genes COL9A2 (8 of 10 studies) and COL9A3 (4 of 8 studies).

Disc degeneration involves dehydration, fragmentation of collagens, and development of annular tears, resulting in disc height reduction. It also alters biomechanical loading patterns, leading to the development of osteophytes and other degenerative changes. All genes involved in matrix turnover and organization, such as collagen, aggrecan, and matrix metalloproteinases are likely involved in the mechanical failure of the extracellular matrix of the disc. Interleukin genes also play a role in stimulating degenerative processes and fibrosis.

Of the 25 biologic candidate genes selected for this study (aggrecan [AGC1], collagen [COL], interleukin [IL], and matrix metalloproteinase [MMP] genes), 14 of them (COL1A2, COL3A1, COL5A1, COL5A2, COL10A1, IL1R1, IL1R2, IL1RL1, IL1RL2, IL18R1, IL18RAP, MMP8, MMP9, and MMP13) have not, to our knowledge, been studied in human disc degeneration.

The main phenotype of interest in disc degeneration was the quantitatively assessed disc signal intensity, in which the T2-weighted disc signal from magnetic resonance imaging (MRI) reflects the water concentration in the disc, since dehydration is the dominant phenomenon in disc degeneration associated with proteoglycan content. In addition, we were interested in the “classic” disc degeneration findings of qualitatively assessed disc height narrowing and bulging (18). Due to the significant differences in the degree of disc degeneration and the possibly different genetic and environmental determinants for the lumbar disc levels, analysis of upper and lower lumbar disc segments separately is warranted (19). The lower lumbar discs represent, on average, more severe stages of disc degeneration (4).

The overall study aims were to investigate the associations of structural, degenerative, and inflammatory candidate genes associated with physiologic and biochemical processes in the connective tissues of the intervertebral discs, using rigorous statistical methods.

SUBJECTS AND METHODS

Study subjects.

The subjects consisted of 588 male monozygotic and dizygotic twins (35–70 years old) from the Twin Spine Study, which aims to investigate the role of heredity and suspected environment determinants for common musculoskeletal problems. The subjects in the Twin Spine Study were selected from the population-based Finnish Twin Cohort and have been shown to be highly representative of the Finnish population (20). All subjects received detailed written information on study procedures and provided informed consent prior to participation. The study design was approved by the ethics committees of both the Department of Public Health at the University of Helsinki and the Faculty of Rehabilitation Medicine at the University of Alberta.

Environmental determinants.

Data on exposure to possible environmental and behavioral risk factors were obtained from an extensive structured interview that determined the subject's detailed work history and leisure activities (Table 1). The structured interview has been described in detail elsewhere (4). A limited analysis of the reliability of measurements of the occupational history variables was possible using a 1-year test–retest interval among the subjects who had not changed jobs after 1 year. Subjects were asked how many hours they spent sitting, the most common amount of weight they lifted, and the frequency of lifting in their current work. The intraclass correlation coefficients (ICC) were 0.74 for sitting time and 0.60 for mean total lifting per day (4) (Table 1). Also, using a 5-year test–retest interval examining the reliability of lifetime exercise history data, the ICC for the test–retest reliability of the summary variable of mean hours of exercise per week was 0.73 (21) (Table 1).

Table 1. Distribution of clinical and magnetic resonance imaging characteristics among 588 men from the Twin Spine Study

	No. of subjects	Mean ± SD	Range
* Based on the most common amount of weight lifted, the positional loading, and the frequency, each occupational and leisure time physical loading activity was categorized as sedentary, moderate degree of material handling and positional loading, or heavy physical loading (for further details, see ref.4).
Age, years	588	50 ± 7.7	35–70
Smoking, pack-years	588	14 ± 17	0–80
Anthropometric measures
Body weight, kg	587	80 ± 11	48–115
Axial L2–L4 vertebral area, pixels	529	7,471 ± 1,149	5,226–11,686
Body weight/axial vertebral area, %	529	1.1 ± 0.17	0.59–1.6
Lifting performance measures
Isokinetic lifting force, N	539	1,024 ± 234	156–1,730
Isokinetic lifting work, J	539	528 ± 184	66–1,191
Physical activity factors*
Average amount of weight lifted after age 20 years, mean kg/week	588	572 ± 1,772	0–24,000
Job heaviness score	588	2.5 ± 0.92	1–4
Twisting and bending at work, years	584	81 ± 99	0–480
Heavy leisure time activities, years	588	2.3 ± 6.6	0–57
Resistance training, years	588	0.063 ± 0.32	0–3.4
Endurance and ball sports, years	588	1.8 ± 2.5	0–25
Disc degeneration measures
Mean disc signal intensity
L1–L4 discs	579	0.34 ± 0.059	0.18–0.58
L4–S1 discs	579	0.31 ± 0.066	0.13–0.60
Mean disc height narrowing score
L1–L4 discs	585	1.7 ± 1.8	0–9
L4–S1 discs	587	2.0 ± 1.6	0–6
Mean disc bulging score
L1–L4 discs	585	1.9 ± 2.0	0–10
L4–S1 discs	587	1.8 ± 1.1	0–6

Assessment of disc degeneration.

Quantitative signal intensity measurements (n = 579) and qualitative disc height narrowing and bulging scores (n = 587) were obtained from MRI scans of the L1–S1 lumbar levels (Table 1). The MRIs were obtained using 1.5T scanners (Magnetom SP 4000 or Magnetom Vision scanner; Siemens, Erlangen, Germany) with a surface coil. Each subject spent 30–45 minutes lying supine immediately prior to MRI scanning in order to minimize the effects of body posture and activity on discs. Disc signal intensity was measured using the midsagittal disc signal, which was adjusted according to the signal intensity of an adjacent cerebrospinal fluid sample extracted from digital MRI data by one of us (VT), using a custom-designed image analysis program (SpEx [“Spine Examiner”]; Twin Spine Study at the University of Alberta, Edmonton, Alberta, Canada). The reliability coefficients (ICC) and error estimates for disc signal intensity and height narrowing were 0.99 and 0.96, respectively. Disc height reduction (relative to neighboring disc heights) and disc bulging phenomena (including herniations) are signs of disc disease and were determined from qualitative evaluations of the films by one of us (KG), using a 4-point scale. Data on basic anthropometric measurements and isokinetic lifting strength measurements (repeatability ICC 0.87) were also recorded (Table 1).

Candidate genes and single-nucleotide polymorphism (SNP) selection.

Based on biologic relevance for the studied phenotypes, a set of 25 candidate genes were chosen for this study: an aggrecan gene (AGC1), 12 collagen genes (COL1A1, COL1A2, COL2A1, COL3A1, COL5A1, COL5A2, COL9A1, COL9A2, COL9A3, COL10A1, COL11A1, and COL11A2), 8 interleukin genes (IL1A1, IL1B1, IL1R1, IL1R2, IL1RL1, IL1RL2, IL18R1, and IL18RAP), and 4 matrix metalloproteinase genes (MMP3, MMP8, MMP9, and MMP13). Next, 158 SNPs with a minimum heterozygosity of 0.2 in the Caucasian population were selected from the Seattle SNP and the Snpper CHIP Bioinformatics Tools databases to cover the selected genes with ∼10-kb SNP spacing. A set of 5 additional SNPs (rs7533552, rs2857401, rs1800012, rs412777, and rs9277933) were selected based on previous association findings. SNP assays were designed for these 163 SNPs using SpectroDesigner software (Sequenom, San Diego, CA).

SNP genotyping.

SNPs were genotyped using the Sequenom MassArray system as recommended by the manufacturer, with additional quality assessment steps described in detail by Silander and collaborators (22). Only 1 individual of the monozygotic twin pairs was genotyped. The genotyping was done in multiplexes of 1–5 SNPs using homogeneous MassExtend assays in 384-well plates. Each 384-well plate contained 8 duplicate samples and 8 nontemplate wells for quality assessment.

Prior to genotyping the study samples, each SNP multiplex was validated by genotyping 81 trio samples (i.e., 27 families with 2 parents and 1 child) to control for Mendelian incompatibilities, using the PedCheck 1.1 program (23). If the genotypes could not be reliably called, if there were unexplainable errors in Mendelian inheritance or between duplicate samples, or if there were genotypes in the nontemplate wells, the SNP was excluded from the study. Furthermore, only SNPs with a call rate of >90% in the study sample were included. Sixty-four of the 163 SNPs were excluded from the study due to technical problems in genotyping or because the SNP did not fit into any multiplex assay. Deviation from Hardy-Weinberg equilibrium was evaluated with Pearson's chi-square test using only unrelated subjects (i.e., only 1 person from a twin pair).

Statistical analysis.

Age-adjusted univariate and multiple regression models were estimated to find the most significant covariates that explained the variation in the phenotypes studied. The most significant covariates of the univariate models were considered first after age until no further significant covariates entered the model. Age was always considered as the first covariate, even when the regression coefficient of age was not significant. Stata statistical software, version 9.1, and survey regression estimation were applied to account for dependency of twins (24).

In association modeling, genotypes with Mendel errors were set as “missing.” Hardy-Weinberg equilibrium of genotype frequencies and bivariate normality of phenotype were assumed. The association analysis was based on the variance components framework, since we had genotypes from family members (twins).

For the variance component model, the variance–covariance matrix was as follows:

• 

where σ is the additive genetic variance, σ is the gene attributable to polygenes, σ is the residual environmental variance, π_ijk is the proportion of alleles shared identically by descent at the marker locus between individuals j and k in family i, and φ_ijk (equal to 1 for the monozygotic twin pairs; 0.5 for the dizygotic twin pairs and parent–offspring pairs) is the kinship coefficient between individuals j and k in family i.

The significance of σ (test of linkage) σ (test of polygenes) was tested, and the most parsimonious variance component model was assumed in modeling of the association (25–29).

The orthogonal model of association, in terms of the means model was as follows:

• 

where b_i and w_ij are orthogonal between- and within-family components g_ij for sibling j in family j; g_ij is the genotype score and is defined as g_ij = m_ij – 1, where m_ij is the marker genotype (i.e., the number of ‘1’ alleles at diallelic locus M, with alleles arbitrarily designated as ‘1’ and ‘2’).

• 

where, n_i is the number of siblings in family i, g_iF is the father's genotype score, and g_iM is the mother's genotype score.

• 

where b_i is the expectation of each g_ij conditional on the family data, and w_ij is the deviation from this expectation for offspring j.

The means model of association can also be extended to components of genetic dominance, as follows:

• 

where bd_i is the between-family component of genetic dominance and wd_ij is the within-family component of genetic dominance.

Association models were adjusted by covariates that were entered in the final multivariate regression models. Because the studied phenotypes are skewed, permutation tests were performed: P values for the within-family component of association were calculated using 1,000 Monte Carlo permutations. In addition, these P values were corrected for multiple testing, and the following empirical single P value thresholds for global significance were provided: ∗ for P = 0.05, ∗∗ for P = 0.01, and ∗∗∗ for P = 0.001.

Hardy-Weinberg equilibrium was tested using the PedStats program (29). Identical-by-descent probabilities were calculated using Simwalk (version 2) (28). Association analyses were performed using QTDT (version 2.4.3) (25, 26).

Since the HapMap haplotype tagging data were not available at the time we selected the SNPs for this study, we first monitored for linkage disequilibrium between the genotyped variants in each gene using the Haploview program (30). Then, haplotypes were estimated for genes that had ≥2 SNPs residing in the same linkage disequilibrium block (r² > 0.95). Haplotypes were formed for the COL3A1 (rs2056156 and rs2203601), COL10A1, (rs3812111 and rs1064583), COL11A1 (rs1337185, 1463035), COL11A2 (rs2072915, rs9277933, and rs2076311), IL1R2 (rs3218984 and rs1008394), IL1RL2 (rs870684, rs1922290, and rs1922295), and IL18R1 (rs2270298 and rs1035130) genes using the Phase2 program (30, 31). A minimum probability of ≥0.60 was required for the individual haplotypes (30, 31). Haplotype association analysis was performed using the QTDT program.

RESULTS

In the multiple regression analyses, body weight/disc area, isokinetic lifting performance, and several physical activity history variables were associated with disc signal intensity, bulging, and/or height narrowing in discs of the upper and/or lower lumbar spine. These variables explained from 4% to 14% of the variance in disc signal intensity (more loading was uniformly beneficial), bulging, and/or height narrowing (Table 2).

Table 2. Statistically significant associations between phenotype measures and covariates, as determined by linear regression analysis*

Covariate	Disc signal		Disc bulging		Disc height narrowing
	L1–L4	L4–S1	L1–L4	L4–S1	L1–L4	L4–S1
* P values (and r^2 values from model with all covariates) were calculated from the model using age and the indicated covariates. † Effect is opposite the effect of age (i.e., effect is beneficial).
Anthropometric measures, body and disc
Body weight/axial vertebral area, %	0.056†	0.051†	0.029
Axial L2–L4 vertebral area
Lifting performance measures
Isokinetic lifting work		0.006†
Isokinetic lifting force						0.066
Physical activity factors
Mean weighted lifting at work			0.054	0.018		0.026
Job heaviness score					0.314
Heavy leisure time activities, years						0.012†
Endurance and ball sports, years
Twisting and bending at work, 1 hour/day	0.004†	0.004†
Smoking, pack-years				0.006
r^2 from model with all covariates	0.136	0.118	0.137	0.040	0.064	0.094

A total of 99 SNPs in the 25 candidate genes passed the preset quality criteria as described in Subjects and Methods, and these were used in the association analysis (Figures 1 and 2). On average, 4 variants were analyzed per gene (varying from 1 SNP each in the IL1A and IL1B genes to 7 SNPs in the IL1R2 gene [see below]), and on average, 42% of the common variation in each gene was covered (r² ≥ 0.5), with coverage varying from no tagging SNPs in COL9A3, COL11A1, MMP9, and IL1A to ≥80% coverage in COL9A2, IL1RL2, and IL1RL1 according to the Tagger software (32).

Figure 1. Linkage disequilibrium structures of the variants of the single aggrecan gene, 12 collagen genes, and 4 matrix metalloproteinase genes analyzed in this study of the association of desiccation, bulging, and height narrowing of discs of the lumbar spine, as determined using Haploview software. Single-nucleotide polymorphisms used to construct the haplotypes are enclosed in boxes. Numbers in the boxes are r² values; shading from white to black indicates increasing r².

Figure 2. Linkage disequilibrium structures of the variants of 6 newly associated interleukin genes analyzed in this study of the association of desiccation, bulging, and height narrowing of discs of the lumbar spine, as determined using Haploview software. Single-nucleotide polymorphisms used to construct the haplotypes are enclosed in boxes. Numbers in the boxes are r² values; shading from white to black indicates increasing r².

By using the QTDT quantitative association analysis approach, 12 of the 99 SNPs provided evidence of association with disc signal intensity, either in the upper or the lower lumbar region or both. Together, allelic variants of 4 structural extracellular matrix proteins were associated with quantitative disc signal intensity: AGC1, COL1A1, COL9A1, and COL11A2 (Table 3). Allelic variants of 4 genes (AGC1, COL9A1, COL9A2, and COL11A1) were associated with the disc bulging score and variants of the AGC1 gene with the disc height narrowing score (Table 3). (Note that in addition to 1,000 permutations, we used empirical single P value thresholds for global significance to control for multiple testing; ∗ for P = 0.05, ∗∗ for P = 0.01, and ∗∗∗ for P = 0.001.)

Table 3. QTDT-based association analysis of the structural candidate genes with disc signal, bulging, and height narrowing*

Gene, rs number	Minor allele frequency	Disc signal		Disc bulging		Disc height narrowing
		L1–L4	L4–S1	L1–L4	L4–S1	L1–L4	L4–S1
* The software program QTDT was used to perform the association analyses with the aggrecan and collagen genes. For some genes, dominant (D) genetic effects of the single-nucleotide polymorphisms were assumed. All P values are based on 1,000 permutations. Only P values that were less than 0.10 are shown. † P value is lower than the corresponding empirical threshold value for global significance (P = 0.01). ‡ P value is lower than the corresponding empirical threshold value for global significance (P = 0.05). § Tests of Hardy-Weinberg equilibrium of genotype frequency yielded P = 0.0013. ¶ P value is lower than the corresponding empirical threshold value for global significance (P = 0.001). # Tests of Hardy-Weinberg equilibrium of genotype frequency yielded P = 0.0286.
AGC1
rs939587	0.28	0.053	0.043
rs4932424	0.28		0.081	0.090		0.040	0.017
rs3825996	0.49			0.025	0.072
rs1042631	0.23	0.001†	0.083		0.010‡
rs1516797	0.34				0.031	0.005‡
COL1A1 (D)
rs1800012	0.12
rs2075555	0.13		0.005‡
rs1007086	0.27		0.049
rs909102	0.14		0.090
COL1A2
rs3763468	0.05
rs388625	0.40				0.064
rs412777	0.42
rs400218	0.35				0.065
rs1034620	0.20
COL2A1
rs1859443	0.18
rs1635529	0.20
rs2276453	0.39
rs917055	0.20		0.073
rs2276458	0.36
rs6823	0.47
COL3A1 (D)
rs1878199	0.29
rs2056156	0.47				0.076
rs2203601	0.47
rs1800255	0.23
COL5A1 (D)
rs4842138	0.12
rs4341231	0.49
rs3128619	0.48
rs10858281	0.45
rs7357740	0.49
COL5A2
rs1983318	0.11
rs13005821	0.34
rs2138374	0.31
rs12693527	0.20
rs1131518§	0.09
COL9A1
rs696990	0.16	0.00008¶			0.014
rs564031	0.40
rs592121	0.32
rs2076816	0.40
rs1200564	0.07
rs997953	0.38				0.077
COL9A2
rs449541	0.34		0.085
rs364281	0.37
rs7533552#	0.20			0.036
COL9A3
rs3891033	0.45
rs1046789	0.27
COL10A1
rs549332	0.47
rs1064583	0.34
rs3812111	0.34
rs568725	0.32
COL11A1 (D)
rs1415359	0.40
rs1337185	0.18			0.015‡
rs1463035	0.18			0.004‡
rs3753841	0.32
COL11A2 (D)
rs2072915	0.24		0.029
rs9277933	0.24	0.051	0.030
rs2076311	0.24	0.070	0.018
rs2855432	0.30
rs2257126	0.36
rs734181	0.18

Of the immunologic candidate genes, 4 allelic variants in 2 genes in the interleukin pathway provided evidence of association with disc signal intensity: IL1A and IL18RAP. No association was observed for any of the immunologic candidate genes and either disc bulging or disc height narrowing (Table 4).

Table 4. Inflammatory and degradative genes associated with disc signal, bulging, and height narrowing*

Gene, rs number	Minor allele frequency	Disc signal		Disc bulging		Disc height narrowing
		L1–L4	L4–S1	L1–L4	L4–S1	L1–L4	L4–S1
* For the IL18RAP gene, dominant (D) genetic effects of the single-nucleotide polymorphisms were assumed. All P values are based on 1,000 permutations. Only P values that were less than 0.10 are shown. † P value is lower than the corresponding empirical threshold value for global significance (P = 0.05).
IL1A
rs2071375	0.35	0.027†	0.082
IL1B
rs1143634	0.26
IL1R1
rs1465325	0.20
rs956730	0.28
rs3917225	0.41
rs2287047	0.25
rs3771200	0.43
IL1R2
rs740044	0.16
rs4141134	0.26
rs719250	0.19
rs3218984	0.41
rs1008394	0.41
IL1RL1
rs1420089	0.13
rs1997466	0.49
rs1041973	0.24
rs12905	0.27
IL1RL2
rs2241132	0.13
rs870684	0.40
rs1922290	0.40
rs1922295	0.40
rs1997502	0.35
rs2302612	0.14
rs1558626	0.46	0.086	0.089
IL18R1
rs2287037	0.39		0.054
rs2270298	0.27
rs1035130	0.27
rs1420096	0.48	0.058
IL18RAP (D)
rs1420106	0.21	0.036
rs1420100	0.48	0.019†	0.005†
rs917997	0.20	0.021†
MMP3
rs645419	0.39
rs646910	0.25
MMP8
rs1940475	0.47
rs2509013	0.38
rs1276283	0.33
MMP9
rs3918241	0.17
rs2664538	0.38
rs20544	0.49
MMP13
rs2252070	0.40
rs3819089	0.11

The association of 13 genotyped variants in COL2A1, IL1RL2, and IL18R1 genes with disc signal intensity and the association of 5 allelic variants of COL2A1 and COL3A1 genes with disc bulging were close to nominal statistical significance (P values varying from 0.054 to 0.089). The analyzed SNPs of 12 of the selected candidate genes (COL5A1, COL5A2, COL9A3, COL10A1, IL1B1, IL1R1, IL1R2, IL1RL1, MMP3, MMP8, MMP9, and MMP13) did not show evidence of association with either the disc signal intensity, disc bulging, or disc height narrowing in this study (Tables 3 and 4).

Haplotype approach.

We further estimated haplotypes for the COL3A1, COL10A1, COL11A1, COL11A2, IL1R2, IL1RL2, and IL18R1 genes, in which ≥2 variants were in high linkage disequilibrium (r² > 0.95) with each other. We were able to estimate haplotypes for at least 98% of the individuals (97.8% for COL3A1, 98.7% for COL10A1, 100% for COL11A1, 99.8% for COL11A2, 99.4% for IL1R2, 99.4% for IL1RL2, and 99.1% for IL18R1).

The haplotype-based analysis provided further evidence of an association with COL3A1 (haplotype frequencies 0.53 for CA, 0.46 for TT, <0.01 for CT, and <0.01 for TA; P = 0.024 for association with disc bulging). The analysis of haplotypes of COL11A1 and COL11A2 did not show significance in excess of that observed with the individual variants, but the allelic haplotype of COL11A2 (haplotype frequencies 0.76 for CTT and 0.24 for AAA) showed association also with disc bulging (P = 0.010) after 1,000 permutations. Hardy-Weinberg equilibrium was P = 0.0033 for COL11A2.

DISCUSSION

In this study, we used quantitative (disc signal intensity) and qualitative (disc bulging and disc height narrowing) phenotypes to estimate disc degeneration. The QTDT-based analysis revealed associations of disc signal intensity with allelic variants of AGC1, COL1A1, COL9A1, and COL11A2 genes encoding for structural proteins. AGC1 and COL9A1 genes, as well as COL9A2 and COL11A1 genes, showed evidence of association with disc bulging, and AGC1 was associated with disc height narrowing. IL1A and IL18RAP genes were associated with disc desiccation (disc signal intensity). Analysis of the individual variants in COL3A1 showed that the associations did not reach statistical significance, but the haplotype approach provided further evidence of association of this gene with disc bulging. The COL11A2 gene also showed further evidence of association with the same phenotype in the haplotype-based analysis.

Although our population sample is among the largest to date to be used in investigations of specific genes and disc degeneration, it is still modest in size for the study of polygenic, multifactorial conditions (32, 33). One other limitation is the low-density sampling across the candidate genes, which did not provide adequate data to detect all true strong associations: additional work is needed to tease-out all genetic contributions, since the study did not cover all the common genetic variations (as indexed by HapMap tagging SNPs) in these genes. Thus, the study did not have the statistical power to exclude genes that did not show association as possible causes of disc degeneration. However, we found some genes associated with human disc degeneration that had not previously been identified.

Among the strengths of the study were the inclusion of the quantitative disc degeneration phenotype of disc signal intensity (desiccation) and the separate examinations of the upper and lower regions of the lumbar spine. Also, comprehensive interview data concerning suspected confounding factors for disc degeneration allowed adjustment for lifetime physical loading history and body anthropometrics. The wide array of 25 candidate genes examined (Table 2) while taking extra care to minimize chance findings was a study strength.

Disc dehydration may be the earliest degenerative sign visible on MRI. When the T2 signal reflects water, and the cerebrospinal fluid is nearly 100% water, the adjusted T2 disc signal intensity provides a good indication of disc desiccation (34). The associations of allelic variants of aggrecan and collagen genes with disc signal intensity observed in this study support the current understanding of the underlying mechanism of disc degeneration. Disc desiccation is thought to result from the breakdown of proteoglycans and collagen network in the nucleus pulposus, the early biochemical changes of disc degeneration.

Disc height narrowing is a sign of advanced degeneration and is associated with bulging and a decrease in disc signal intensity and may be an end point for a variety of pathologic conditions affecting disc degeneration. Thus, one may speculate that disc height narrowing may not provide insights into specific causes of disc degeneration, but instead, may be indirectly associated with most genes involved in disc degeneration. This view is supported by our finding that among 25 candidate genes, only the AGC1 gene, which was strongly associated with disc signal intensity and disc bulging, was associated with disc height narrowing. However, there may also be specific genetic factors predisposing to disc bulging and disc height narrowing.

The type IX collagen molecule is a heterotrimer composed of α1(IX), α2(IX), and α3(IX) chains and apparently functions as a spacer between the collagen fibrils in cartilage. Both COL9A2 and COL9A3 have been shown to be associated with degenerative disc disease in previous studies (6, 7). We also observed that the COL9A2 gene was associated with disc signal intensity, but only at the lower lumbar disc levels, where disc degeneration generally progresses faster than at the upper lumbar disc levels. We did not observe an association of allelic variants of the COL9A3 gene with either disc signal intensity, disc bulging, or disc height narrowing in our study. Interestingly, SNP rs696990, which is located 9.2 kb 5′ of the COL9A1 gene, provided the strongest evidence of association with disc signal intensity in this study. Mutations in the coding region of the COL9A1 gene have been identified in affected members of families with multiple epiphyseal dysplasia and a novel autosomal-recessive form of Stickler syndrome (35,36). In addition, polymorphisms in the COL9A1 gene have recently been reported to be associated with disc degeneration in mice (37).

Our findings of an association with disc signal intensity and the COL1A1, COL3A1, and COL5A1 genes are consistent with the immunolocalization observations reported by Nerlich and collaborators (38), showing that types I, II, and V collagen molecules play a role in disc degeneration and in the degeneratively altered nucleus pulposus. Guehring and collaborators (39) have also shown a significant up-regulation of COL1A2 in compressed discs in an animal model. However, to explain differences in gene associations between the specific disc regions, we must also consider that at different stages of degeneration, different genes may be important and that the influence of different environmental factors and other genes (e.g., vertebral anthropometrics) can vary between discs. These explanations are supported by the findings from the immunolocalization (38) and followup (40) studies.

The observed associations between polymorphisms of the IL1A, IL18RAP, and IL18R1 (P = 0.054) genes and disc signal intensity are supported by the known connections of these genes with inflammatory conditions and with innate immunity (41) (Table 4). Interestingly, SNP rs917997 of the IL18RAP gene, which was associated with disc signal intensity in the present study, was recently shown to be associated with inflammatory bowel disease in 3 independent Dutch cohorts (42). According to the HapMap data, allele A of the rs917997, which was associated with lower average disc signal intensity at the L1–L4 level in our study, perfectly tags the IL1RL1;IL18R1;IL18RAP;SLC9A4 haplotype, which was shown to be associated with both Crohn's disease and ulcerative colitis. SNP rs917997 was further shown to be associated with celiac disease, and carriers of the disease-associated A allele had lower expression of messenger RNA for IL18RAP in whole blood samples (43).

Among the 12 collagen genes we studied, allelic variants of the COL1A2 and COL11A1 genes were associated only with disc bulging: COL1A2 at the lower lumbar level, where the most severe degenerative findings are commonly found (L4–S1), and COL11A1 at the upper lumbar level (L1–L4). Furthermore, Mio and collaborators (13) recently identified an association between an allelic variant of COL11A1 (rs1676486) and disc herniation in a Japanese population. COL11A1 was found to be highly expressed in intervertebral discs, but its expression was decreased in patients with disc herniation. The amount of the transcript containing the disease-associated allele was shown to be decreased because of its decreased stability.

In conclusion, of the 25 selected candidate genes related to disc matrix synthesis and degradation, AGC1 was associated with disc desiccation, bulging, and height narrowing, and COL9A1 and COL9A2 were associated with disc desiccation and bulging. The COL1A1, IL1A, and IL18RAP genes were associated only with disc desiccation and COL11A1 as well as COL3A1 genes with disc bulging. We are not aware of any previous reports of such associations with the COL9A1, COL11A1, IL18R1, and IL18RAP genes in humans. The accuracy of the quantitatively measured phenotype of disc desiccation, using cerebrospinal fluid–adjusted disc signal intensity, likely enhanced our ability to observe related associations. On the other hand, it is likely that we did not have adequate data to detect all true associations because all allelic variations of the genes were not covered by this analysis. Adjustments of the phenotypes excluded possible associations because of some constitutional factors, such as body weight and axial disc area. Overall, the observed associations support the view of disc degeneration as a multifactorial, polygenic condition and shed light on the biochemical mechanisms affecting disc matrix synthesis and degradation that influence disc degeneration. However, the findings need to be replicated in other populations as well as at the functional level.

AUTHOR CONTRIBUTIONS

Dr. Videman had full access to all of the data in the study,and Drs. Saarela and Kaprio had full access to the genotyping and statistical analysis data; these authors take responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Videman, Saarela, Kaprio, Näkki, Peltonen, Battié.

Acquisition of data. Videman, Saarela, Kaprio, Näkki, Gill, Battié.

Analysis and interpretation of data. Videman, Saarela, Kaprio, Näkki, Levälahti, Gill.

Manuscript preparation. Videman, Saarela, Kaprio, Näkki, Levälahti, Gill, Peltonen, Battié.

Statistical analysis. Saarela, Kaprio, Levälahti.

Acknowledgements

The authors wish to thank Brad G. Sinclair for his input in the further development of our MR image analysis software, SpEx; Dr. Laura E. Gibbons and Mr. Kauko Heikkilä for preliminary statistical analyses; and Mr. Aki Salo, Ms Heidi Maunu, and Ms Sanna Kouhia for their contribution to the SNP and gene selection and genotyping.