Genotype and the Periventricular Edge in MS

MS lesions are not spread evenly through the brain. They cluster in the periventricular white matter, and how far that pattern extends varies widely from one patient to the next. Sombekke and colleagues set out three reasons to suspect that some of this variation is genetic. Patients tend to relapse in the same regions, and relatives who both have MS show lesions in similar regions. In experimental autoimmune encephalomyelitis, the animal model used to study MS, lesions in the spinal cord and lesions in the brain were shown to be controlled by different quantitative trait loci. And in a subset of this same patient group they had previously found carriership of HLA-DRB1*1501 associated with more lesions in the spinal cord but not with total brain lesion volume, which suggested the allele affects lesion development in one particular region of the central nervous system. They tested that idea by comparing voxelwise lesion probability across genotypes of 69 candidate SNPs in 208 patients, and they call the work exploratory in the title.

The Candidate Set and the Cohort
The SNPs came from a literature search rather than a scan. PubMed and the Genetic Association database were searched for genes and polymorphisms with suggested involvement in MS pathogenesis, prognosis or treatment response; the resulting polymorphisms were confirmed in dbSNP, their frequency in white populations checked against HapMap, and probes and primers designed from SNPper sequences using Primer3. SNPs within the MHC class II region were chosen predominantly on the basis of their described linkage disequilibrium with the HLA-DRB1*1501 allele. That gave 80 validated polymorphisms across 44 genes, from which 11 were dropped, five as monomorphic and six for a minor allele frequency below 5%, leaving 69. The 208 patients were 126 relapsing remitting, 42 secondary progressive and 40 primary progressive, a 19.2% share of primary progressive disease the authors describe as slightly high. They were 37% male, mean age 41.1 years, with a median disease duration of 6.3 years, median EDSS of 3.5 and median relative lesion volume of 12.4 mL, and EDSS was recorded a median of 0.0 months from the MRI.

Getting 208 Brains Into One Frame
Because the patients were sampled from several natural history studies, the imaging came from different acquisition protocols, on a 1T scanner for 77.4% of them and 1.5T for the remaining 22.6%, with 3 to 5 mm sections at an in-plane resolution of 1 × 1 mm². Lesions were identified by an expert reader and outlined with semiautomated seed-growing software using a local thresholding technique, producing binary lesion masks. A common T2 template with 2 × 2 × 2 mm voxels was built by linearly registering every T2 brain image, at 12 degrees of freedom, to the MNI-152 brain image with FLIRT, then averaging and smoothing with a 4-mm full width at half maximum Gaussian kernel. Each individual image was registered to that template the same way, and the resulting matrices applied to the corresponding binary masks, with nearest-neighbor interpolation producing per-voxel lesion presence or absence and registration quality checked by visual inspection. Lesion volumes were computed after registration, so they represent relative volumes and account for differences in head size.

How the Genotype Comparison Was Set Up
Statistical inference ran through the nonparametric Randomise method in FSL 4.0, with a general linear model treating each voxel independently of the others, 5000 permutations and a cluster-forming threshold of pseudo-t = 2, corresponding to a voxelwise P threshold of 0.01. For each SNP the authors made three comparisons, each time setting one genotype against the combination of the other two: homozygote for the frequent allele, homozygote for the rare allele, and heterozygote. Every comparison was tested separately for increased and for decreased lesion probability. Homozygote genotypes with frequencies below 5% in the cohort were folded in with the heterozygotes, because markers with low genotype frequencies can produce unreliable observations in voxelwise analyses even where they matter in complex trait disorders. Two further steps handled the confound the authors expected to matter most. Any genotype producing a significant cluster had its mapping repeated while controlling for total brain lesion volume, and genotype was separately tested against total brain lesion volume while controlling for disease duration.

Eleven Genotypes, Every Cluster Periventricular
Of the 69 SNPs, 11 genotypes across 10 SNPs produced significant clusters of either increased or decreased lesion probability, and every cluster sat periventricularly, abutting the frontal or occipital horn of the lateral ventricles, often asymmetrically. Increased probability came from the heterozygous genotype at three SNPs, rs2227139 in the highly polymorphic MHC class II region on chromosome 6, a region involved in self-versus-nonself immune recognition and consistently shown to have a major effect on susceptibility to MS and other autoimmune disease, rs2076530 in BTNL2 and rs876493 in PNMT, and from the homozygous major allele at two more, rs2107538 in CCL5 and rs9808753 in IFNGR2. Decreased probability came from six SNPs, in BTNL2, CRYAB, NDUFS7, UCP2 and two within FAS. No genotype of any SNP was associated with a decrease in one region and an increase in another, which the authors read as consistency of the observed relations across the brain. BTNL2 rs2076530 was the only SNP with effects for more than one genotype, its AG genotype raising lesion probability in white matter next to the right occipital horn and its GG genotype lowering it next to the left occipital and frontal horns, effects the authors describe as nonconflicting.

What Survives the Lesion-Volume Correction
Once total brain lesion volume entered the model, only rs2227139 retained a significant cluster, with the CT genotype associated with increased local lesion probability compared with the other two genotypes, and that result also held after correction for disease duration. To corroborate it the authors computed voxelwise average lesion frequencies for each of the three genotypes. At the voxel carrying the maximum pseudo-t value in the significant cluster, 4.1, a lesion was present in 29% of the heterozygotes (26 of 89), against 16% of the CC patients (12 of 77) and 2% of the TT patients (1 of 42). Six of the 11 genotypes were significantly associated with total brain lesion volume, three of them still after controlling for disease duration, and the directions matched the mapping results throughout, with genotypes tied to higher total volume also tied to increased local probability. The authors draw the methodological consequence themselves: total brain lesion volume is an important covariate in this kind of analysis, the genetic influence they observe may act partly through it and show up in voxelwise analyses particularly where MS lesion frequencies are high, and future group comparisons should consider correcting for it.

The HLA Null Result, and the Limitations the Authors Set Out
No data are available on rs2227139 and region-specific differences in the brain, and the authors recommend that future work on genetic influence over where lesions form include high-resolution HLA typing, which would allow different alleles to be compared and causative ones detected. Using rs3135388, a SNP predicting the HLA-DRB1*1501 haplotype with very high sensitivity, they confirmed in a larger sample the negative result of an earlier report, finding that allele unrelated to total brain lesion volume and, as a new finding, unrelated to the anatomic distribution of lesions across the brain. They say plainly that they have no explanation for the asymmetry of the clusters, which persisted in post hoc analyses at a lower threshold, that a genetic influence on asymmetric distribution is theoretically unlikely a priori, and that bias from the high number of progressive patients cannot be excluded. Three limitations are named. The field strength was relatively low and therefore less sensitive to lesions than scanners at 3T or above. Linear registration cannot correct for the variability in ventricular and sulcal sizes that follows from atrophy, and that imperfect match may partly explain why clusters appeared only periventricularly and not in areas of lower lesion frequency. They chose it deliberately: with only dual-echo images, nonlinear registration struggles to separate periventricular lesions from ventricular CSF at similar signal intensities, and could shift lesions within the brain, affecting the very thing they set out to measure. And gray matter lesions are absent from the study, since standard MRI mostly misses them, while postmortem work shows they are extensive in MS. What they call for next is independent confirmation, a more focused approach combining mapping with predefined regions of interest to raise statistical power, more homogeneous patient groups, and a longitudinal study of where new lesions develop, which could eventually yield ways of predicting from a patient's genotype whether they are likely to develop lesions in clinically eloquent areas.

Disclaimer: This blog post is based on the cited study and is intended for informational purposes only. It is not intended to provide medical advice. Please consult with a healthcare professional for any health concerns.

Reference:
Sombekke, M. H., Vellinga, M. M., Uitdehaag, B. M. J., Barkhof, F., Polman, C. H., Arteta, D., Tejedor, D., Martinez, A., Crusius, J. B. A., Peña, A. S., Geurts, J. J. G., & Vrenken, H. (2011). Genetic correlations of brain lesion distribution in multiple sclerosis: An exploratory study. American Journal of Neuroradiology, 32(4), 695–703. https://doi.org/10.3174/ajnr.A2352