Updated meta-analysis of SORL1 in Alzheimer's disease: Unmet concerns

16 April 2014

I have read with interest the recent report by Jin and coworkers on the putative role of SORL1 variants in sporadic Alzheimer’s disease [1]. In their report, the authors claim a significant effect of a three-marker haplotype, but refute a second haplotype based on twenty-three case-control studies. This conclusion, it seems, is premature in view of glaring oversights.

Firstly, only a fraction of the total number of cases and controls was included in the meta-analysis. Out of 144 data sets for 11 SNPs listed in Table 2, 24 were dropped without comment in Table 3, e.g., instead of 17 studies available for SNP 23, only 12 were pooled. Which studies were retained, and why others were not, is cryptic. The tests of heterogeneity described in Table 4 refer to a further subset of those studies included in Table 3, and are incomplete for SNPs 4, 8, and 9.

Secondly, information from earlier studies is poorly transcribed. Multiple figures are in error, e.g., sample size [2-5]. Alleles are incorrect for SNP19 which stands for a G/T, not a C/T substitution. For SNPs 5 and 10, the alleles refer to the wrong DNA strand, i.e., the substitutions should read A/G rather than C/T. This, in turn, renders results for SNP23 non-informative. SNP23 is a canonical A/T substitution and can be read either as "A" or as "T" depending on the DNA strand under study. There is no telling which is the minor allele when the remaining SNPs were genotyped on alternating DNA strands. Failure to unequivocally determine SNP23 alleles, finally, casts doubt on association data pertaining to the refuted SNP23-24-25 haplotype. Discrepant results for SNP23 have been noted by others [6].

Thirdly, the interpretation of findings is highly contradictory. The majority of putative “risk alleles” from Tables 3 and 4 are actually protective by the OR shown in these tables. Attempts to define risk alleles “according to the AlzGene database” are not helpful. The database gives no such definition and is a work in progress, i.e., results change as new studies are being added. The authors do not specify the date of access or an AlzGene freeze of reference. Negative association results for the second SORL1 haplotype are best explained by the ambiguity surrounding SNP23 data, imputation-related artefacts introduced by [7], and additional artefacts introduced by sibling structure in a case-control study of twins [8]. The claim of female-specific effects of SORL1 variants is not supported by [7].

On the whole, pending resolution of the above issues, the paper by Jin and coworkers raises more questions than it answers. This is in part due to the practice of carrying data over from earlier meta-analyses [9]. Had the authors consulted the original publications, at least the errors in sample size could have been avoided.

Philipp G. Sand, Department of Psychiatry, University of Regensburg, 93042 Regensburg, Germany

References
[1] Jin C, Liu X, Zhang F, Wu Y, Yuan J, Zhu J, Zhang F, Wang G, Cheng Z (2013) An updated meta-analysis of the association between SORL1 variants and the risk for sporadic Alzheimer's disease. J Alzheimers Dis 37, 429-437.
[2] Cellini E, Tedde A, Bagnoli S, Pradella S, Piacentini S, Sorbi S, Nacmias B (2009) Implication of sex and SORL1 variants in italian patients with Alzheimer disease. Arch Neurol 66, 1260-1266.
[3] Kölsch H, Jessen F, Wiltfang J, Lewczuk P, Dichgans M, Teipel SJ, Kornhuber J, Frölich L, Heuser I, Peters O, Wiese B, Kaduszkiewicz H, van den Bussche H, Hüll M, Kurz A, Rüther E, Henn FA, Maier W (2009) Association of SORL1 gene variants with Alzheimer's disease. Brain Res 1264, 1-6.
[4] Tan EK, Lee J, Chen CP, Teo YY, Zhao Y, Lee WL (2009) SORL1 haplotypes modulate risk of Alzheimer's disease in Chinese. Neurobiol Aging 30, 1048-1051.
[5] Feulner TM, Laws SM, Friedrich P, Wagenpfeil S, Wurst SH, Riehle C, Kuhn KA, Krawczak M, Schreiber S, Nikolaus S, Förstl H, Kurz A, Riemenschneider M (2010) Examination of the current top candidate genes for AD in a genome-wide association study. Mol Psychiatry 15, 756-766.
[6] Lee JH, Cheng R, Honig LS, Vonsattel JP, Clark L, Mayeux R (2008) Association between genetic variants in SORL1 and autopsy-confirmed Alzheimer disease. Neurology 70, 887-889.
[7] Beecham GW, Martin ER, Li YJ, Slifer MA, Gilbert JR, Haines JL, Pericak-Vance MA (2009) Genome-wide association study implicates a chromosome 12 risk locus for late-onset Alzheimer disease. Am J Hum Genet 84, 35-43.
[8] Reynolds CA, Hong MG, Eriksson UK, Blennow K, Johansson B, Malmberg B, Berg S, Gatz M, Pedersen NL, Bennet AM, Prince JA (2010) Sequence variation in SORL1 and dementia risk in Swedes. Neurogenetics 11, 139-142.
[9] Reitz C, Cheng R, Rogaeva E, Lee JH, Tokuhiro S, Zou F, Bettens K, Sleegers K, Tan EK, Kimura R, Shibata N, Arai H, Kamboh MI, Prince JA, Maier W, Riemenschneider M, Owen M, Harold D, Hollingworth P, Cellini E, Sorbi S, Nacmias B, Takeda M, Pericak-Vance MA, Haines JL, Younkin S, Williams J, van Broeckhoven C, Farrer LA, St George-Hyslop PH, Mayeux R (2011) Meta-analysis of the association between variants in SORL1 and Alzheimer disease. Arch Neurol 68, 99-106.

Comments

Response to Sand:
Unmet concerns about an update meta-analysis of the association between SORL1 variants and the risk for sporadic Alzheimer’s disease

Dear Prof. Philipp G. Sand,

We highly appreciate your letter pointing out some shortcomings and mistakes in our original report. The detailed explanations for your concerns are as follows:

In our meta-analysis, we evaluated the association of SORL1 variants with sporadic Alzheimer’s disease (SAD) using four genetic models. Thus, we removed some references in the statistical analysis because we could not obtain detailed genotyping data from the original papers or AlzGene database directly. Based on this principle, for SNP 8, 3 out of 17 articles [1-3] were removed. We also missed two included studies [4,5] in calculating pooled odds ratio in Table 3 (the original article), however, the revised results is not significantly changed (for T versus A: 0.96 (0.92-1.00), p = 0.08; for TT versus AA: 0.93 (0.84-1.03), p = 0.16; TT + TA versus AA: 0.94 (0.88-1.00), p = 0.04; TT versus TA + AA: 1.02 (0.94-1.11), p = 0.64). In addition, the revised tests of heterogeneity between two races are not markedly altered in Table 4 (for SNP 8: I2 = 75%, pheterogeneity = 0.05; for SNP 9: I2 = 0%, pheterogeneity = 0.53). Nevertheless, we apologize for our oversights.

The sample size transcribed in Table 2 is inconsistent with the original references because it was mainly from the previous meta-analysis [6]. However, all statistical data are extracted from the original papers or AlzGene database; the discrepancy of the transcription therefore did not affect the final statistical results. Alleles annotation refers to the AlzGene database in Table 1, and SNP 19 should be the “G/T” allele (“C/T” is a writing error), while we believe that the “T/C’ allele of SNP 10 is consistent with “A/G” allele (AlzGene database). As you mentioned, SNP 23 is a canonical “A/T” substitution. Due to our oversights, we ignored the key discrepancy discussed by Lee et al. [1]. Therefore, we could not adequately draw a conclusion that the SORL1 three-marker haplotype at SNP 23-SNP 24-SNP 25 is not associated with susceptibility to SAD in this meta-analysis. In spite of this, we still suggested that SNP 19 and SNP 24 may be false-positive sites based on the LD analysis (Figure 1 of the original article).

It is well known that the results of genetic association study are usually controversial. Actually, the AlzGene database does not define a certain allele as a “risk allele”. For SNP 5, our results showed that the ‘C’ allele was protective for SAD (Table 3 of the original article). Therefore, the annotation of “risk allele” in our meta-analysis is inappropriate (Tables 3 and 4 of the original article).

We welcome and thank you for your criticism and correction very much. Your serious and rigorous scholarship has deeply affected and impressed us, and benefit us.

Thank you very much.

Chunhui Jin, Ph.D., Department of Geriatrics, Wuxi Mental Health Center of Nanjing Medical University, 156 Qian Rong Road, Wuxi city, Jiangsu Province, China; Tel: +86-510-83250132, Fax: +86-510-83012201, jch1029@126.com.

Zaohuo Cheng, Ph.D., Department of Clinical Psychology, Wuxi Mental Health Center of Nanjing Medical University, 156 Qian Rong Road, Wuxi city, Jiangsu Province, China; Tel: +86-510-83030359, Fax: +86-510-83012201, chengzaohuo@126.com.

References
[1] Lee JH, Cheng R, Honig LS, Vonsattel JP, Clark L, Mayeux R (2008) Association between genetic variants in SORL1 and autopsy-confirmed Alzheimer disease. Neurology 70, 887-889.
[2] Beecham GW, Martin ER, Li YJ, Slifer MA, Gilbert JR, Haines JL, Pericak-Vance MA (2009) Genome-wide association study implicates a chromosome 12 risk locus for late-onset Alzheimer disease. Am J Hum Genet 84, 35-43.
[3] Tan EK, Lee J, Chen CP, Teo YY, Zhao Y, Lee WL (2009) SORL1 haplotypes modulate risk of Alzheimer's disease in Chinese. Neurobiol Aging 30, 1048-1051.
[4] Shibata, N, Ohnuma T, Baba H, Kamboh MI (2008) No association of SORL1 SNPs with Alzheimer’s disease in a Japanese population. Dement Geriatr Cogn Disord 26, 161-164.
[5] Kimura R, Yamamoto M, Morihara T, Akatsu H, Kudo T, Kamino K, Takeda M (2009) SORL1 is genetically associated with Alzheimer disease in a Japanese population. Neurosci Lett 461, 177-180.
[6] Reitz C, Cheng R, Rogaeva E, Lee JH, Tokuhiro S, Zou F, Bettens K, Sleegers K, Tan EK, Kimura R, Shibata N, Arai H, Kamboh MI, Prince JA, Maier W, Riemenschneider M, Owen M, Harold D, Hollingworth P, Cellini E, Sorbi S, Nacmias B, Takeda M, Pericak-Vance MA, Haines JL, Younkin S, Williams J, van Broeckhoven C, Farrer LA, St George-Hyslop PH, Mayeux R (2011) Meta-analysis of the association between variants in SORL1 and Alzheimer disease. Arch Neurol 68, 99-106.