Do motor control genes contribute to interindividual variability in decreased movement in patients with pain?

The prioritization process generated the candidate gene list shown in Table 1, plus additional genes, not shown, that scored lower than those in Table 1. Our literature search showed that there was virtually no work specifically on the neurochemistry of pain-related motor changes in mammalian models. Therefore, the ratings for "relation to motor control" just reflect the degree of certainty that the molecule is present at key sites in motor control systems, from basal ganglia down to the neuromuscular junction.

Table 2 shows the Roland-Morris Disability Questionnaire items that we included in the motor limitation subscale. Figure 1 shows that patients with higher SF-36 bodily pain magnitude ratings reported greater movement limitation. Pain rating contributed 31% of the variance in movement impairment in patients who chose nonoperative treatment, and 16% of the variance in those who chose surgery. Table 3 shows other non-genetic contributors to variation in movement impairment. For this and subsequent statistical analyses, we did not include pain after observing that a two-factor, gene × pain analysis of covariance for movement impairment yielded many more significant genetic associations than would have occurred by chance, even for lists of genes that were not on our motor candidate list. Our only explanation for this was that because the Roland-Morris movement questions include wording referring to both pain and movement, this might produce unanticipated biases of analyses that include both pain and movement. Therefore we simplified the analysis by recognizing that all patients presented to the study with leg and back pain – an entry criterion – and dropping pain magnitude from the analysis. This step corrected the number of SNP associations to the number expected by chance.

Table 3 shows that greater reduction in movement was associated with lower scores for the SF-36 Vitality subscale (R² = 29%); the choice of surgical treatment (R² = 16%); the patient receiving workers' compensation (R² = 1.5%), the presence of neurological deficits on exam (R² = 4%), and lower scores, indicating greater impairment, on the SF-36 Emotional Role subscale (R² = 12%). A combination of all of the variables in the model explained 44% of the variance in the movement reduction subscale.

Resources permitted us to genotype a set of SNPs from 14 of these genes: 9 of the top priority 15 genes in Table 1, and 5 lower priority genes that were included in the Laboratory of Neurogenetics' array of 130 polymorphic genes of interest to a consensus of neuropsychiatric and addictions researchers. No SNPs in the fourteen genes we tested – the dopamine D2 receptor, GTP cyclohydrolase I, glycine receptor α1 subunit, GABA-A receptor α2 subunit, GABA-A receptor β1 subunit, α-adrenergic 1C, 2A, and 2C receptors, serotonin 1A and 2A receptors, cannabinoid CB-1 receptor, M1 muscarinic receptor, and the tyrosine hydroxylase, and tachykinin precursor-1 molecules – showed a significant association with the movement score after a Bonferroni correction for the multiple candidate genes tested (Additional file 1). The two SNPs in the α-2C adrenergic receptor that were nominally significant in the additive model (without correction for multiple tests, Table 4) made small contributions to the variance in movement scores – approximately 0.6% and 0.9% (partial R²). Neither of these SNPs is known to cause a functional change in molecular function. Haplotype analysis of this gene using all four ADRA2C SNPs genotyped in this study showed a nominally significant association (p = 0.02) with movement scores which loses significance when one corrects for the 22 haplotype blocks examined.

Haplotype analysis of one of the blocks in the GABA-A receptor β1 subunit showed that a haplotype of 11% frequency was associated with less limitation of movement at a significance level (p = 0.0025; Table 5) almost strong enough to correct for testing 22 haplotype blocks.

Power analysis indicated that our study had sufficient power (85%) to detect effect of a single SNP that contributed 5% (partial R² = 0.05) of the movement variance after correcting for Bonferonni error using α = 0.00357 (=0.05 divided by the number of genes tested, 14).

Although muscle stiffness and reduction of movement in pain patients is widely discussed in the literature [31, 32], there is no evidence regarding its molecular basis. Hence we reviewed the motor control research from animal and human studies to compile a list of molecules which appear to be involved in normal motor control and in diseases where muscle tone is altered and movement restricted and muscle stiffness.

We searched the literature for reports of the transmitters, receptors, and signaling molecules mediating control of limb and trunk movements, tone, and posture. We searched PubMed for review articles on motor control mechanisms and for citations of neurotransmitters, receptors, or signaling molecules relevant to motor cortex, brainstem, basal ganglia, cerebellum, locus coeruleus, corticospinal tract, ventral horn, interneuron, Renshaw cells, and neuromuscular junction. We prioritized polymorphisms in the genes for these molecules similar to the criteria proposed by Belfer et al. [19]. Three scores of 0–3 were assigned to each polymorphism according to (a) the strength of evidence supporting involvement of the gene in motor control; (b) population frequency of the polymorphism; and (c) strength of evidence that the polymorphism is associated with an altered clinical or molecular phenotype.

Participants were members of the sciatica group in the Maine Lumbar Spine Study (MLSS [33]). The MLSS was a prospective cohort study conducted by approximately half of Maine's orthopedists and neurosurgeons who actively treated spine disease. Patients were enrolled between 1990 and 1992, and surgical discectomy or non-surgical treatment was chosen based upon clinician judgment and patient preference. Patients completed questionnaires at a baseline assessment, and then at 3, 6, and 12 month follow-up, and then annually through year 10. After completion of the 10 year study, the NIDCR and MLSS investigators developed a collaboration to add a genetic component to the study, under a protocol approved by the NIDCR Institutional Review Board. Patients were invited to contribute a DNA sample. Two hundred ninety of the original 507 enrollees provided DNA and had usable baseline data.

Table 2 lists the 23 items in The Roland-Morris Disability Questionnaire. We selected the 16 items that refer to a pain-related reduction in movement to comprise a "movement impairment subscale." Each "yes" answer contributes one point to the score.

Genomic DNA was extracted from lymphoblastoid cell lines using a salting out protocol [35]. Additional DNA samples were obtained from saliva samples collected using Oragene self-collection kits and the DNA extracted according to the manufacturer's recommendation.

One hundred twenty-one SNPs in the 14 genes studied were selected based on the haplotypes reconstructed by Haploview [36] or SNPHAP [37] with HapMap project genotype data. A SNP selection pipeline based on a double classification tree search algorithm [38] was used to capture the haplotype complexities and the tag SNPs (seleceted with Haploview for haplotype blocks).

We genotyped the DNAs using Illumina GoldenGate chemistry on Sentrix Universal-96 Arrays. The Illumina array used interrogates 1536 SNP simultaneously, using a custom primer assay design (GS-0007064-OPA). 500 ng of DNA per well was genotyped using standard Illumina protocols. Arrays were imaged using an Illumina BeadStation 500 GX and data analyzed using GenCall 6.2.0.4 and GTSreports v5.1.2.0 software (Illumina). Ten percent of DNA samples were run in duplicate in order to obtain an estimate of genotyping reproducibility. The overall error rate was <0.005. Average genotyping completion rate was >0.98. Hardy-Weinberg equilibrium for each marker was tested with Chi-square tests using the R package "genetics" [39] or exact tests [40]. No deviation from the expected Hardy-Weinberg equilibrium values were observed for any of the values analyzed.

Haplotype phases – i.e., how the directly measured SNP alleles were distributed into two chromosomes in each patient – were inferred by the expectation – maximization (EM) algorithm (SAS/Genetics, Cary, North Carolina, USA).

Multiple linear regression was applied to examine the association between movement and each SNP, adjusting for eight other variables: age, sex, SF-36 Vitality, SF-36 Emotional Role, nonsurgical vs. surgical treatment, workman's compensation, marital status, and neurological exam deficits. An additive genetic model was tested by recoding the three SNP genotypes as 0 for the homozygote of the common allele, 1 for the heterozygote, and 2 for homozygote of the uncommon allele. In the haplotype analysis, first haplotype phases were inferred using expectation-maximization (EM) algorithm, where the probability was assigned to each pair of haplotypes which each subject possessed. Then, stepwise regression [41] was applied to test the association between movement score and haplotypes; the linear regression model included all haplotypes with frequency greater than 1% and the eight covariates.