Variation in the uric acid transporter gene (SLC2A9) and memory performance

Abstract

Understanding human cognitive ageing is important to improve the health of an increasing elderly population. Serum uric acid levels have been linked to many ageing illnesses and are also linked to cognitive functioning, though the direction of the association is equivocal. SLC2A9, a urate transporter, influences uric acid levels. This study first tested four SLC2A9 SNPs, previously associated with uric acid levels, in ∼1000 Scots: the Lothian Birth Cohort 1936 (LBC1936). These participants were tested on general cognitive ability at ages 11 and 70. At age 70, they took a battery of diverse cognitive tests. Two replication cohorts were investigated. First, the LBC1921, who were tested on general cognitive ability at age 11. At ages 79 (n = 520), 83 (n = 281) and age 87 (n = 177), they completed cognitive ability test batteries. Second, the Edinburgh Type 2 Diabetes Study (ET2DS) were tested for cognitive abilities aged between 60 and 75 years (n = 1066). All analyses were adjusted for age, gender, body mass index and either childhood cognitive ability test score (LBC) or vocabulary—a measure of prior cognitive ability in ET2DS. Significant associations were detected with SLC2A9 and a general memory factor in LBC1936 and other individual cognitive ability tests (lowest P = 0.0002). The association with logical memory replicated in LBC1921 at all ages (all P < 0.05). These associations were not replicated in ET2DS (all P > 0.1). If the positive associations withstand, then this study could suggest that higher uric acid levels may be associated with increased performance on memory-related tasks.

INTRODUCTION

Certain important human cognitive abilities decline with age (1). This can be debilitating on an individual level, by decreasing quality of life and lowering independence, and at a societal level, with increased healthcare costs (2). Identifying the cerebral basis for people's differences in age-related cognitive decline is a healthcare priority for an increasing elderly population (3). Determining the causes of age-related cognitive changes could reveal the mechanisms of cognitive decline, which ranges from normal cognitive ageing, through mild cognitive impairment, to the dementias. Cognitive ability in youth is the major determinant of cognitive ability in old age (4,5). However, it explains only about half of the variance. Additional causal factors are thought to be genetic, medical, psychological, social and lifestyle factors (3,6). Given its large influence, it is very helpful, when studying cognitive functions in older people, to have a measure of cognitive ability from youth. Two of the three cohorts studied here have this near-unique advantage.

In humans, uric acid is the end product of purine metabolism and is believed to have strong neuroprotective and antioxidant properties (7). Uric acid is also thought to be a key to an evolutionary and behavioural response to starvation and consequently have a role in intellectual development (8). Altered serum uric acid levels have been associated with many illnesses. Low uric acid level is associated with neurodegenerative diseases such as multiple sclerosis, Parkinson's disease and Alzheimer's disease. Conversely, high uric acid level is a risk factor for gout, hypertension, renal disease and cardiovascular disease (as reviewed by Kutzing et al. 9). There are few studies investigating the role of serum uric acid influencing individual differences in normal cognitive ability. Recently, a large study in 1724 participants (age 55 or older) showed that higher uric acid levels were associated with better global cognitive function, executive function and memory function after correcting for cardiovascular risk factors (10). Previous investigations have also proposed the importance of serum uric acid inducing ‘an intelligence and excellence of all-round performance’ but were not definitive studies, hampered by small sample sizes and inconsistent findings between cohorts (11–13). However, the direction of this effect is not conclusive as other studies show the opposite trend that higher uric acid levels is associated with poorer cognitive ability. In a study in 96 adults, aged 65 or older, participants with mildly elevated serum uric acid were more likely to have poor performance on processing speed, verbal memory and working memory tests (14). In a cross-sectional study performed in 1016 older persons, they showed that demented persons had higher serum uric acid levels (15). It is possible that higher uric acid level is associated with better cognitive performance, but this direction of association is ambiguous, and it is not certain which domains of cognitive function are affected. The question of whether factors that influence variation in uric acid levels are relevant to discovering the sources of individual differences in cognitive functions remains open.

Genetic association studies have firmly established SLC2A9 as a gene influencing serum uric acid levels, with gender-specific effects (16–22). SLC2A9 [solute carrier family 2 (facilitated glucose transporter), member 9] is located on chromosome 4p16.1 and spans over 214 kb with 13 coding exons and one non-coding exon. Two transcript variants encoding distinct isoforms have been identified: SLC2A9_L (GLUT9) and SLC2A9_S (GLUT9ΔN) (23). SLC2A9 is predominantly expressed in the kidney and liver and is detectable in brain tissue (23,24) and is also genetically linked with the psychiatric illness, bipolar disorder (25). Importantly, SLC2A9 shows strong uric acid transport activity (16,20) and it is proposed that SLC2A9 directly regulates plasma uric acid (urate) levels by transporting uric acid out of the tubular cell, in a voltage-driven manner (26). Moreover, SLC2A9 physiologically regulates serum uric acid levels in vivo (27).

In the present study, we tested four SNPs in SLC2A9: rs733175 in the promoter region, rs1014290 in intron 3, rs6449213 in intron 4 and rs737267 in intron 7. These were first tested for association with cognitive ageing in a discovery cohort the Lothian Birth Cohort 1936 (LBC1936), and then two replication cohorts of older people, the Lothian Birth Cohort 1921 (LBC1921) and the Edinburgh Type 2 Diabetes Study (ET2DS). In order to address the question of whether uric acid levels are associated with individual differences in cognitive ability, we tested associations between these genetic variants and cognitive phenotypes spanning major domains of cognitive function. Two of these cohorts of older people have validated childhood IQ scores, which are incorporated into the analyses in order to examine cognitive change from early to later life.

RESULTS

Table 1 shows significant association in the LBC1936 between SLC2A9 markers and memory and other cognitive phenotypes; a general memory factor, logical memory, spatial span (a test of non-verbal, spatial learning and memory), verbal paired associates (a test of verbal learning and memory), a general cognitive factor, block design (a test of constructional ability), simple reaction time (RT) mean, and choice RT mean (all P < 0.05). Three associations surpassed the threshold for Bonferroni significance: rs733175 with the general memory factor (β = 0.102, P = 0.0002, increasing effect of the major T allele) and with verbal paired associates (β = 0.097, P = 0.002, increasing effect of the major T allele), and rs6449213 with block design (β = 0.089, P = 0.004, increasing effect of the major T allele). For each SNP, the minor allele is associated with a decrease in cognitive abilities.

Table 1.

Association of SLC2A9 in LBC1936

Cognitive test	rs733175 (C)	rs1014290 (G)	rs6449213 (C)	rs737267 (T)
General factors
G Factor
β	−0.051	−0.037	−0.064	−0.041
P	0.085	0.210	0.033	0.170
G Speed
β	−0.050	−0.041	−0.060	−0.071
P	0.097	0.171	0.050	0.021
G memory
β	−0.102	−0.068	−0.073	−0.072
P	0.0002	0.0118	0.0082	0.0084
Memory phenotypes
Logical memory
β	−0.077	−0.069	−0.072	−0.048
P	0.011	0.024	0.021	0.120
Backward digit span
β	−0.037*	−0.031*	−0.056*	−0.045*
P	0.230	0.300	0.069	0.140
Verbal paired associates
β	−0.097	−0.078	−0.070	−0.059
P	0.002	0.013	0.026	0.060
Letter-number sequence
β	−0.020	−0.001	−0.021	−0.009
P	0.530	0.980	0.500	0.780
Spatial phenotypes
Spatial span
β	−0.050	−0.029	−0.078	−0.035
P	0.110	0.360	0.013	0.260
Block design
β	−0.075	−0.059	−0.089	−0.052
P	0.014	0.051	0.004	0.089
Reasoning phenotypes
MHT at age 70
β	0.000	0.002	−0.017	0.002
P	0.990	0.940	0.570	0.950
Matrix reasoning
β	−0.058	−0.050	−0.057	−0.038
P	0.056	0.099	0.062	0.200
Verbal fluency
β	0.009	0.013	−0.003	−0.015
P	0.761	0.666	0.917	0.625
NART
β	−0.018	−0.002	−0.008	0.003
P	0.540	0.950	0.790	0.930
Speed phenotypes
Digit symbol
β	−0.009	−0.019	−0.043	−0.044
P	0.770	0.540	0.160	0.160
Symbol search
β	−0.038	−0.023	−0.047	−0.036
P	0.220	0.440	0.130	0.240
Simple reaction time mean
β	0.050	0.069	0.076	0.054
P	0.110	0.032	0.017	0.093
Simple reaction time StdDev
β	−0.015	0.043	0.042	0.032
P	0.650	0.180	0.190	0.310
Choice reaction time mean
β	0.036	0.031	0.085	0.058
P	0.260	0.320	0.007	0.069
Choice reaction time StdDev
β	−0.006	0.000	0.024	0.022
P	0.840	0.990	0.440	0.490
Inspection time
β	−0.002	0.018	−0.005	−0.014
P	0.943	0.565	0.877	0.673

Gender, age in days at testing, age 11 MHT score (age residualized) and BMI were included as covariates. Individuals with dementia, gout and/or on allopurinol medication were removed. The beta scores are standardized regression coefficients. The direction of the regression coefficient represents the effect of each extra minor allele where a positive regression coefficient means that the minor allele increases the phenotype mean. The minor allele is listed in the header row. P-values are in italics and significant P-values (<0.05) and corresponding beta values are highlighted in bold. G, General cognitive factor; MHT, Moray House Test; NART, National Adult Reading Test. StdDev, standard deviation.

*A significant SNP × gender interaction in females, where the minor allele of each SNP is associated with a decrease in cognitive abilities and is discussed in the text.

Table 2 shows that SLC2A9 variants are also significantly associated with logical memory in the first replication sample, the LBC1921 (P < 0.05). This significant association to logical memory is apparent in all three waves of testing. There is also an association with the Moray House Test (MHT) at wave 1 (age 79), the simple RT standard deviation at wave 2 (age 83) and G composite and letter-number sequencing at wave 3 (age 87). Again, the minor allele is associated with a decrease in cognitive abilities, except for the MHT at wave 1.

Table 2.

Association of SLC2A9 in LBC1921 wave 1, wave 2 and wave 3

Cognitive test	LBC1921 wave 1 (age 79)				LBC1921 wave 2 (age 83)				LBC1921 wave 3 (age 87)
	rs733175	rs1014290	rs6449213	rs737267	rs733175	rs1014290	rs6449213	rs737267	rs733175	rs1014290	rs6449213	rs737267
G Factor
β	0.003	−0.002	−0.048	−0.052	−0.090	−0.054	−0.039	−0.065	−0.165	−0.156	−0.130	−0.114
P	0.941	0.966	0.244	0.206	0.192	0.437	0.575	0.348	0.029	0.042	0.093	0.138
Memory phenotypes
Logical memory
β	−0.062	−0.039	−0.056	−0.099	−0.183	−0.125	−0.109	−0.159	−0.197	−0.170	−0.125	−0.144
P	0.17	0.38	0.22	0.026	0.0077	0.065	0.12	0.017	0.009	0.026	0.102	0.057
Letter-number sequence
β	—	—	—	—	0.073	0.063	0.001	0.007	−0.069	−0.128	−0.158	−0.117
P	—	—	—	—	0.25	0.33	0.99	0.91	0.36	0.08	0.040	0.12
Reasoning phenotypes
MHT
β	0.114	0.073	0.048	0.010	—	—	—	—	−0.023	0.000	0.027	0.084
P	0.015	0.10	0.30	0.83	—	—	—	—	0.75	1.00	0.71	0.26
Raven's progressive matrices
β	0.087	0.008	0.013	−0.019	0.017	0.022	0.044	0.044	−0.059	−0.054	−0.051	−0.019
P	0.063	0.87	0.77	0.68	0.79	0.74	0.49	0.49	0.45	0.50	0.52	0.80
Verbal fluency
β	0.035	0.004	−0.031	0.000	−0.007	−0.061	−0.030	−0.047	−0.140	−0.149	−0.137	−0.111
P	0.45	0.94	0.51	0.99	0.91	0.35	0.65	0.47	0.08	0.06	0.09	0.16
NART
β	0.003	−0.048	−0.013	−0.033	−0.021	−0.032	0.046	0.033	−0.126	−0.087	−0.019	0.006
P	0.94	0.29	0.77	0.46	0.74	0.62	0.47	0.60	0.09	0.23	0.80	0.93
Speed phenotypes
Digit symbol
β	—	—	—	—	0.100	0.085	0.052	0.074	−0.018	0.030	−0.026	0.033
P	—	—	—	—	0.13	0.19	0.44	0.27	0.82	0.70	0.75	0.68
Simple reaction time mean
β	—	—	—	—	−0.114	−0.095	−0.060	−0.053	−0.011	−0.026	−0.001	−0.013
P	—	—	—	—	0.099	0.167	0.388	0.429	0.89	0.74	0.99	0.87
Simple reaction time StdDev
β	—	—	—	—	−0.042	−0.084	−0.080	−0.131	−0.101	−0.123	−0.094	−0.102
P	—	—	—	—	0.518	0.201	0.212	0.042	0.19	0.11	0.21	0.18
Choice reaction time mean
β	—	—	—	—	−0.023	−0.003	−0.005	−0.031	0.080	0.074	0.065	0.028
P	—	—	—	—	0.38	0.90	0.86	0.23	0.28	0.32	0.38	0.71
Choice reaction time StdDev
β	—	—	—	—	−0.037	−0.012	0.019	0.005	0.087	0.058	0.114	0.054
P	—	—	—	—	0.33	0.75	0.62	0.89	0.24	0.44	0.14	0.48

Individuals with dementia, gout and/or on allopurinol medication were removed. Gender, age 11 MHT score (residualized for age in days), age in days at testing and BMI (wave 1 and 3 only) were included as covariates. The beta scores are standardized regression coefficients. The direction of the regression coefficient represents the effect of each extra minor allele where a positive regression coefficient means that the minor allele increases the phenotype mean. If the test was not performed a ‘—’is indicated. The minor allele is C (rs733175), G (rs1014290), C (rs6449213) and T (rs737267). P-values are in italics and significant P-values (<0.05) and corresponding beta values are highlighted in bold. G, general cognitive factor; MHT, Moray House Test; NART, National Adult Reading Test; StdDev, standard deviation.

Table 3 shows that SLC2A9 SNPs are not associated with cognitive ability at older age in the second replication cohort, the ET2DS (all P > 0.1), as tested by general cognitive and general memory ability factors, together with memory, reasoning and speed phenotypes.

Table 3.

Association of SLC2A9 in ET2DS

Cognitive test	rs733175 (C)	rs1014290 (C)	rs6449213 (C)	rs737267 (T)
G factor
β	0.018	0.017	0.024	0.015
P	0.48	0.51	0.35	0.58
G memory
β	−0.034	−0.038	−0.026	−0.042
P	0.27	0.22	0.41	0.19
Memory phenotypes
Logical memory
β	−0.032	−0.024	−0.016	−0.026
P	0.32	0.45	0.62	0.42
Letter-number sequence
β	−0.015	−0.033	−0.038	−0.039
P	0.64	0.29	0.23	0.22
Faces
β	−0.02	−0.029	−0.008	−0.033
P	0.51	0.36	0.79	0.29
Reasoning phenotypes
Matrix reasoning
β	0.013	0.013	0.016	0.0054
P	0.67	0.67	0.59	0.87
Verbal fluency
β	0.000	0.46	0.023	0.045
P	0.99	0.15	0.47	0.16
Speed phenotypes
Digit symbol coding
β	−0.016	0.0046	0.012	0.008
P	0.59	0.88	0.69	0.79
Trail making
β	0.0098	0.048	0.032	0.039
P	0.75	0.12	0.31	0.22

Individuals with dementia, gout and/or on allopurinol medication were removed. Gender, Mill Hill vocabulary scale as a measure of pre-morbid IQ, age in days at testing and BMI (wave 1 and 3 only) were included as covariates. The beta scores are standardized regression coefficients. The direction of the regression coefficient represents the effect of each extra minor allele where a positive regression coefficient means that the minor allele increases the phenotype mean. P-values are in italics and significant P-values (<0.05) and corresponding beta-values are highlighted in bold. G, general factor.

A meta-analysis showed a significant association between all four SNPs with logical memory in LBC1936, LBC1921 ages 79, 83 and 87, and ET2DS: rs733175 (β = −0.066, P = 0.010), rs1014290 (β = −0.055, P = 0.011), rs6449213 (β = −0.050, P = 0.025) and rs737267 (β = −0.050, P = 0.025). A combined analysis could not be performed because the version of the Wechsler logical memory test differed between cohorts.

As previous studies reported a strong SNP allele-by-sex interaction for serum uric acid levels, such that the effect of the SNP on serum acid levels in females is larger, additive and highly significant (16,18,22), gender by SNP interactions were tested in all cognitive outcomes in LBC1936, and checked for replication in LBC1921 and ET2DS. Significant SNP by gender interactions were noted with digit span backwards in LBC1936 (P < 0.05). The minor allele of each SNP is associated with a decrease in digit span backwards in females only (rs733175 P = 0.017, rs1014290 P = 0.018, rs6449213 P = 0.005 and rs737267 P = 0.017). This cognitive test was not performed in either replication cohort. However, as it tests an aspect of memory we checked for associations with the memory phenotypes in LBC1921 and ET2DS but they did not show sex-specific associations (all P > 0.1).

Reports associating SLC2A9 with uric acid level have differed on their inclusion criteria and covariates incorporated into their analyses; for example, one study did not account for body mass index (BMI) in their analysis (17) and another study included cardiovascular history in their analysis (10). To ensure our results withstand these constraints and are not due to confounding, several checks were performed. Supplementary Material, Table S1 shows that significant associations remained in LBC1936 to SLC2A9 for G memory factor, logical memory, spatial span, verbal paired associates, block design, simple RT mean and choice RT mean (all P < 0.05), when childhood IQ was removed as a covariate. These analyses examine the effect of the SLC2A9 SNPs solely on cognition as it is manifest in old age, not cognitive ability in old age adjusted for cognitive ability in youth (which given an indication of lifelong cognitive change). These associations and the corresponding effect sizes are not as strong as the associations of SLC2A9 with ‘cognitive ageing’ in Table 1. The same result stands when childhood IQ is removed as a covariate in LBC1921. When estimated prior cognitive ability is removed as a covariate in ET2DS, there are still no significant associations with the cognitive abilities listed in Table 3. There is an association with the test used to estimate prior cognitive ability, the Mill Hill Vocabulary Scale, with the minor allele increasing the phenotype mean (rs733175 β = −0.065, P = 0.045; rs1014290 β = −0.058, P = 0.071; rs6449213 β = −0.061, P = 0.056; rs737267 β = −0.076, P = 0.019). However, the equivalent test in the LBC samples is the MHT at age 11 which does not show any significant association with SLC2A9 (Supplementary Material, Table S1).

The significant associations remain after removing BMI as a covariate in LBC1936 and in LBC1921 wave 1 and wave 3, and do not alter the lack of association in ET2DS. The significant associations also remain after excluding individuals with stroke history (LBC1936 n = 53; LBC1921 wave 1 n = 39, wave 2 n = 21, wave 3 n = 14) (all P < 0.05). Including cardiovascular history as a fixed factor in LBC1936 (n = 238), LBC1921 wave 1 (n = 75), wave 2 (n = 48), wave 3 (n = 44), significant associations remained with G Factor, G Memory, logical memory, verbal paired associates, block design, simple RT mean and choice RT mean (all P < 0.05) in LBC1936, and with the MHT (LBC1921 wave 1), logical memory in LBC1921 waves 2 and 3. Including cardiovascular history as a fixed factor in ET2DS (n = 355) did not alter the genetic associations. The results are not shown here but are available from the first author upon request.

DISCUSSION

We examined one large cohort, tested for cognitive ability at age 11 and 70, for association with four SLC2A9 variants and attempted replication in two similar cohorts. Here, we show that markers in SLC2A9 are associated with memory abilities in LBC1936. We show that the SNPs are associated with a general memory factor, which extracts the memory domain from a number of memory-based tests, and we also report the specific subtests, for comparison to LBC1921 where such general factors could not be extracted as not all the subtests were performed. This association with memory abilities is replicated in the logical memory test in LBC1921 at all three different waves, but not in ET2DS. There is a clear association between the minor alleles of the four SLC2A9 variants with lower logical memory in both LBC1936 and LBC1921. This was not replicated in ET2DS. A meta-analysis confirms the association with logical memory to all four SNPs, including all data from the discovery cohort and the two replication cohorts. This association remains after adjusting for possible confounding effects of cardiovascular and cerebrovascular risk factors. There is also an association in females only with backward digit span in LBC1936, with digit symbol in LBC1921 wave 2 and with verbal fluency in LBC1921 wave 3. The minor alleles—rs737267 (T allele), rs6449213 (C allele) and rs1014290 (G allele)—were previously associated with decreased uric acid (16). Thus we show that markers associated with an increase in uric acid are also associated with increased performance in memory-related tasks.

Overall, the results provide some genetic support to the recent report showing that higher serum uric acid levels are associated with better cognitive function later in life (10). Indeed, Euser et al. showed a significant trend (P < 0.05) for global cognition and memory function, indicating better cognition for participants with higher levels of serum uric acid. Our report also supports two smaller studies. One study observed a positive, non-significant correlation between plasma uric acid level and IQ (12), and another study found a positive association between serum uric acid levels and a spatial ability test (mental rotations test) and educational achievement, but this became non-significant on adjusting for height and weight (13).

It is known that there are separate gender-specific effects in cognition, in serum uric acid levels (28) and in the association of SLC2A9 and serum uric acid levels (18). However, no study tested for any specific gender effects between serum uric acid levels and cognition; therefore, we are unable to compare our gender-specific results. It is conceivable that the (as yet unknown) SLC2A9 variant causing an alteration in urate transport activity could cause the same effect of, say, increasing serum uric acid to the same level in both males and females, but this level may have more effect in females, who have a lower threshold than males and consequently effect cognitive ability to varying degrees. Why this is evident in LBC1936 in the working memory task, digit span backwards, and not other memory tasks remains to be clarified. Indeed, the lack of replication of the sex-specific associations in LBC1921 and ET2DS does not suggest that the interaction is robust.

Our report does not concur with another population-based study that hypothesised increased serum uric acid to be associated with lower cognitive function (29) and dementia. Schretlen et al. (29) described an association between higher levels of uric acid and impaired memory function in a sample of community dwelling participants. However, they had a relatively small sample size of 96 individuals and could not exclude the possibility that serum uric acid increased in response to some other pathological process that caused the cognitive impairment. If their results can be attributed to a pathological process that requires stable levels of serum uric acid, then it may be interesting to note that of their wide range of cognitive tests measured, those tapping processing speed, working and verbal memory showed significant associations, which are the same significant cognitive abilities reported in this study. Ruggiero et al. (15) reported an association between serum uric acid and dementia in over 1000 individuals in the InCHIANTI study. However, they could not exclude the possibility of reverse causality that serum uric acid increases in response to dementia. Also, they report an increase in disability score from the middle to the low serum uric acid quartile which fits with our study on normal cognitive ability.

A strength of this study was the ability to control for childhood IQ in two of the phenotypically well-defined cohorts, while simultaneously incorporating extensive health information, including dementia, incidences of gout, medication history and cerebrovascular and cardiovascular risk factors. The main limitation of the study is the failure to replicate the association of SLC2A9 in one of the two replication cohorts (ET2DS). Possible reasons to explain our failure to replicate are 3-fold. First, the original finding in both LBC cohorts was a false positive, such as a chance finding or a spurious result. Also, we cannot rule out that there are unpublished null findings of this association. Second, the finding in the LBC cohorts may be a true positive but the effect size was overestimated (30). If so, then the power to detect the effect sizes in ET2DS found in both LBC1936 and LBC1921 of 0.35–3.88% would be lowered. This is found for other quantitative traits such as height, where whole genome-association studies have shown that each locus explains only a very small proportion of the phenotypic variance (0.3–0.5%) (31). Third, the SLC2A9 association to memory performance may be a real association but the inconsistencies between cohorts may lie in hidden population sub-structure and phenotypic differences that may limit the degree to which the attempted replication is truly a replication. Although the ET2DS is similar to the LBC1936 and LBC1921 in terms of geographical ascertainment, they differ in some other respects; both the LBC cohorts are separately very homogenous cohorts within narrow age range, similar ability and lifestyle; LBC1936 (age ∼70) and LBC1921 (ages 79, 83 and 87) are older cohorts than ET2DS (age range 60–75); ET2DS had a higher BMI (mean 31.46) than LBC1936 (mean 27.58) and LBC1921 (mean 25.5 wave 1 and 25.99 wave 3), see Supplementary Material, TableS2; health status (ET2DS is a diabetes cohort and LBC1936 and LBC1921 are relatively healthy cohorts); childhood IQ measures were not available for ET2DS; and for the directly comparable cognitive ability tests, namely letter-number sequencing, verbal fluency and digit symbol coding, LBC1936 scored higher than ET2DS, see Supplementary Material, Table S2. Age, BMI and disease prevalence (9,32) are known determinants of serum uric acid levels which would also affect the ability to detect relevant genetic associations.

A second limitation is that we have not found the causative variant that links serum uric acid levels with increased performance in memory tasks. Various SLC2A9 non-synonymous mutations and rare variants were recently reported and await functional characterization (16,18). Recently, two loss-of-function heterozygous mutations in SLC2A9 were identified in individuals with renal hypouricaemia. Both of these mutations are located in highly conserved ‘sugar transport proteins signature motifs’ and result in reduced urate transport activity (27). A third limitation is all four markers were not consistently associated with the memory phenotypes between the two cohorts and within the three waves in the replication cohort. This may reflect the different degrees of LD between the SNPs and the alternative causative variant or multiple variants they are tagging which remains to be identified. Supplementary Material, Table S3 shows that the four markers fall into two LD blocks, as defined by solid spine LD (D′ 0.8). Indeed, the four markers are in relatively close LD in LBC1936, all r² > 0.38. Also, slight differences in allele frequencies, coinciding with relatively low power may reduce the ability to detect a significant association in the smaller cohorts, LBC1921 and ET2DS. Indeed, there may be a distant regulatory controlling mechanism related with these associations observed in this study or an interaction with other urate transport related genes. Furthermore, other genes that are implicated in urate transport: SLC22A12, ABCG2, SLC17A3, SLC17A1, SLC22AA11, SLC16A9, GCKR, LRRC16A, PDZK1 (19,33,34) were not included in this study and are also worthy as candidate genes for cognitive ageing or to investigate for interactions with SLC2A9.

To progress from the findings in this study, future work should first measure serum uric acid levels in LBC1936 and confirm the already well-validated association between SLC2A9 and serum uric acid levels. We report SLC2A9 genetic variants that explain 0.35–1.0% (LBC1936) and 0.98–3.88% (LBC1921) (by squaring the standardized beta values in Tables 1 and 2) of the variance in memory phenotypes. This is a smaller effect than 1.7–5.3% of the variance explained in serum uric acid concentrations by SLC2A9 (16). Undoubtedly, an association of high serum uric acid levels with increased memory performance would substantiate the results of this study and allow comparison and comprehension of the effect sizes; high serum uric acid levels as suggested by SLC2A9 genotype is associated with increased memory performance. Second, if serum uric acid level truly contributes to cognitive ageing, then an occurrence of brain pathology, for example the integrity of white matter, could be evident. To date, two studies on 180 elderly individuals have suggested that high serum uric acid levels correlate with increased white matter hyperintensity volume (representing cerebral ischaemia) (14), mediated through poorer working memory, processing speed, fluency and verbal memory (35). This is not the direction of effect that we are reporting here, but it is conceivable that both extremes of the uric acid distribution could cause pathologies detrimental to cognition (e.g. U-shaped relationship).

Indeed, measurement of serum urate in the LBC1936 cohort could elucidate the mechanism linking it to cognitive function. So far, evidence points at a complex relationship of uric acid with disease, where high uric acid levels are a risk factor for gout, hypertension, renal disease and cardiovascular disease, and low levels are associated with neurodegenerative diseases such as multiple sclerosis, Parkinson's disease and Alzheimer's disease (as reviewed by Kutzing et al. 9). Uric acid is thought to have an important role in neuroprotection as it is a natural antioxidant, accounting for the majority of free radical scavenging activity in human blood (7), in particular through an astrocyte-mediated mechanism (36). Indeed, high serum uric acid levels have been proposed to carry an evolutionary advantage (7). However, in a normal human population, the positive effect of the antioxidant and free radical scavenger activity of uric acid to general memory performance remains to be clarified.

This is the first report suggesting a link between cognitive ability in old age, in particular memory performance, to SLC2A9. If proven, our work strengthens the hypothesis that oxidative stress is involved in cognitive processes and provides additional evidence for a protective role for antioxidants such as uric acid. The precise relationship of serum uric acid levels and cognitive ageing remains to be found. This study provides a focus for future research, in particular to first replicate our finding in an independent cohort, with the objective of finding the determinants of cognitive ability differences among older people, and thus providing evidence to help improve health and wellbeing of the elderly.

MATERIALS AND METHODS

Sample

There are 1091 individuals (543 females and 548 males) in the LBC1936. All were born in 1936 and attended school in Scotland in 1947. At an average age of 11 years, they took a valid IQ type test—a version of the MHT no. 12—in the nationwide Scottish Mental Survey 1947 (SMS1947; n = 70 805) (37). At age ∼70 years, the LBC1936 were recruited as surviving and still relatively healthy participants of the SMS1947 who were mostly living in the Edinburgh area (Lothian) of Scotland. They re-sat the same mental test and other cognitive and medical tests, as described elsewhere in detail (38). All participants in the study lived independently in the community and travelled to the Wellcome Trust Clinical Research Facility (WTCRF) at the Western General Hospital, Edinburgh, UK for testing. DNA samples are available for 1078 participants. Fifteen participants were excluded due to possible dementia: 13 scored <24 on the Mini-Mental State Examination (MMSE) (39), one had incomplete MMSE test data and another excluded because of uncertainty over dementia history. Thirty-eight individuals with self-reported incidences of gout or who listed allopurinol on a list of current medications were removed prior to analyses. The final sample used in the analyses was 1016 individuals (524 females and 492 males) with a mean age of 69.5 years (range 67.6–71.3). They had a mean age of 10.9 years (range 10.4–11.4) when tested in the SMS1947.

The first replication sample is the LBC1921; n = 550, 316 females and 234 males. They are surviving participants of the Scottish Mental Survey of 1932 (SMS1932) (40), who were living independently in the Edinburgh area (Lothian) at the time of recruitment. Therefore, at a mean age of 11 years, they took the same mental test that was taken 15 years later by the SMS1947. Details on the recruitment and testing of participants have been described previously (4,41). In wave 1 of re-testing, at a mean age of 79 years, 33 participants were excluded due to possible dementia: nine scored <24 on the MMSE, 19 had incomplete MMSE test data and five reported a history of dementia. Ten participants with a history of gout or on allopurinol medication were removed. In wave 2, at a mean age of 83 years, 18 participants were excluded due to possible dementia: nine scored <24 on the MMSE and nine reported a history of dementia. Fourteen participants with a history of gout or on allopurinol medication were removed in wave 2. In wave 3, at a mean age of 87 years, there were 18 participants excluded due to possible dementia: 15 scored <24 on the MMSE and three reported a history of dementia. Eight participants with a history of gout or on allopurinol medication were removed in wave 3 and five participants had no DNA sample. The final LBC1921 sample used in the wave 1 analyses contained 520 individuals (303 females and 217 males) with a mean age of 79.1 years (range 77.7–80.6). The final sample used in wave 2 analyses contained 281 individuals (158 females and 123 males) with a mean age of 83.3 years (range 82.0–84.6). The final sample used in wave 3 analyses contained 177 individuals (97 females and 80 males) with a mean age of 86.6 years (range 85.7–87.47). The wave 1 individuals had a mean age of 10.9 years (range 10.4–11.5) when tested in 1932.

The second replication sample is the ET2DS which is a cohort of men and women from the Lothian area of Scotland, with type 2 diabetes aged between 60 and 75 years (n = 1066). Recruitment and examination procedures for this cohort have been described previously (44). Individuals with possible dementia (MMSE < 24 or missing; n = 33) and gout (n = 34) were removed from the analysis.

Cognitive tests

LBC1936

A full description of the cognitive tests completed by the LBC1936 is available elsewhere (38). The tests pertinent to this study are described in brief below. The MMSE was used to screen for possible dementia (39). A general measure of cognitive ability with an emphasis on verbal reasoning (MHT no.12) was administered when participants were a mean age of 11 years in the SMS1947 [Scottish Council for Research in Education (SCRE)] (37). The MHT was re-administered at a mean age of almost 70 years for LBC1936, using the same instructions and 45 min time limit that were applied at age 11. Cognitive tests assessing learning, memory and working memory were administered to the LBC1936 at age 70. These include: logical memory (immediate declarative memory), backward digit span (working memory), spatial span (non-verbal, spatial learning and memory) and verbal paired associates (verbal learning and memory) from the Wechsler Memory Scale 3rd Edition (WMS-III^UK) (42); letter-number sequencing (working memory), matrix reasoning (non-verbal reasoning) and block design (constructional ability) from the Wechsler Adult Intelligence Scale 3rd Edition (WAIS-III^UK) (42). The verbal fluency test provided a measure of executive function (43). The National Adult Reading Test (NART) estimated prior cognitive ability (44). The information processing speed battery comprised two psychometric tests from the WAIS-III^UK (digit symbol and symbol search) and two elementary cognitive tasks: simple and four choice RT; and inspection time, which is a psychophysical assessment of the efficiency of the early stages of visual decision making (4,45).

A full description of the cognitive tests completed by the LBC1921 is available elsewhere (4). The tests pertinent to this study are described in brief below. The MMSE was used to screen for possible dementia in all three waves (39). A general measure of cognitive ability with an emphasis on verbal reasoning (MHT no.12) was administered when participants were a mean age of 11 years in the SMS1932 (SCRE) (37,40). The MHT was re-administered at a mean age of over 79 and 87 years for LBC1921, using the same instructions and 45 min time limit that were applied at age 11. Cognitive tests assessing learning, memory and working memory were administered to the LBC1921; at age 79 in wave 1, MHT, Raven's Progressive Matrices, logical memory, verbal fluency and NART; at age 83 in wave 2, Ravens' Progressive Matrices, logical memory, verbal fluency, NART, letter number sequence, digit symbol, simple and choice RT; at age 87 in wave 3, MHT, Raven's Progressive Matrices, logical memory, verbal fluency, letter-number sequencing, NART, digit symbol and simple and choice RT. They are the same cognitive tests as described for LBC1936 with the following exceptions: in LBC1921, the logical memory test was based on an earlier version (46) and Raven's Progressive Matrices tested non-verbal reasoning in LBC1921 only (47).

A detailed explanation of the cognitive tests performed in the ET2DS is available elsewhere and described briefly below (48). The MMSE was used to screen for possible dementia (39). The Verbal Fluency Test was performed to test executive function. Immediate and delayed verbal declarative memory was assessed using one story from the Logical Memory subset of the WMS-III^UK. Non-verbal memory was assessed using the Faces subtest of the WMS-III^UK. As a measure of mental flexibility, the Trail Making Test (Part B) (49) was administered. Subjects completed several subtests of the WAIS-III^UK (42): the Digit Symbol Coding subtest as a measure of speed of information processing, the letter-number sequencing subtest to assess working memory and the matrix reasoning subtest to measure non-verbal reasoning. All subjects were asked to complete a combined version of the Junior and Senior Form A synonyms of the Mill Hill vocabulary scale (50), an indicator of prior (‘best ever’ or pre-morbid) cognitive ability which changes very little with age (51,52).

Genotyping

Genomic DNA was isolated from whole blood by standard procedure at the WTCRF Genetics Core, Western General Hospital, Edinburgh. The SLC2A9 SNPs selected were those most significantly associated with serum uric acid levels in an association study performed in a Croatian population study and replicated in UK (Orkney, Scotland), Croatian and German samples (16). The three markers associated with serum uric acid concentrations in the UK cohort at a stringent Bonferroni-corrected significance threshold (P ≤ 1.7 × 10⁻⁷) were chosen for this study: rs737267, rs1014290 and rs6449213. An SNP in the promoter region, 5′ of SLC2A9 (rs733175) also showed a strong association (P < 1 × 10⁻⁵) to serum uric acid level and was included. The markers were genotyped by Taqman assay (Applied Biosystems, Pleasonton, CA, USA) for LBC1936 and LBC1921 and by a competitive allele-specific PCR system (KASPar) by KBiosciences, Herts, UK for ET2DS.

Statistical analysis

Cognitive phenotypes

The cognition data were prepared by removing outliers in the cognitive variable data with Z scores greater than ±3. The score for logical memory was the logical memory I total recall score plus logical memory II delayed recall total score. The score for verbal paired associates was the verbal paired associates total score plus the recall total score. The simple RT measures (LBC) and trail making Test B scores (ET2DS) were logarithmically transformed to improve their distribution. For LBC1936, a general cognitive ability factor was derived from principal components analysis of the WMS-III subtest (digit-span backwards) and five WAIS-III subtests (matrix reasoning, letter-number sequencing, block design, symbol search and digit symbol), as previously described (53). A general speed factor was separately derived from principal components analysis of speed measures, but excluding the RT standard deviation measures which were strongly correlated with RT mean scores as previously described (53). For LBC1936, a general memory factor was derived from principal components analysis of WMS III—Logical Memory I Total recall score (A + B + B2), WMS III—Logical Memory II Delayed recall total score (A + B), WMS III—Spatial Span Forward, WMS III—Spatial Span Backward, WMS III—Verbal Paired Associates I (List A + B + C + D), WMS III—Verbal Paired Associates II recall total score, WAIS III—Letter-Number Sequencing and WAIS III—Digit Span Backwards. For LBC1921, a general cognitive ability composite score was derived from principal components analysis of Raven's Progressive Matrices, verbal fluency and logical memory (WMS III). For ET2DS, scores from the seven tests were used to obtain a general, fluid cognitive ability factor, g, via a principal components analysis as previously reported (54). A general memory score was derived from principal components analysis of logical memory (WMS III), faces (WMS III) and letter-number sequencing (WAIS-III^UK). Outliers were removed prior to principal components analysis. The sample size, the mean values and standard deviations of the cognitive variables are listed in Supplementary Material, Table S2. This preparation was performed using SPSS version 14.0.

Genotype data

Quality control of the genotyping data was undertaken by removing samples with <75% (three SNPs) genotyping success rate (n = 22) and testing SNPs for Hardy–Weinberg equilibrium using the HW exact SNP tests (all P > 0.05). The genotyping data were of good quality, as the mean genotyping rate in LBC1936 was 95% (range 93.7–96.3%) in 1016 samples and informative, as the mean heterozygosity was 33.5% (range 30–37%). In LBC1921, the mean genotyping rate was 92% (range 91–93%) and 96% (range 94–97%) in ET2DS. The minor allele frequencies of all markers were >0.05. The genotype frequencies were similar (±0.02) to the Orkney, UK cohort in the original study (16). Further characteristics of the SNPs investigated are listed in Supplementary Material, Table S4. These analyses were performed using PLINK version 1.04 (55) and Pedstats (56).

Association analysis

Linear regression analysis, under an additive genetic model, was performed using PLINK version 1.04–1.05 (55). For analysis in LBC1936, the following were included as covariates: gender, BMI, age in days at testing at age 70 and MHT score at age 11 (age residualized). For analysis in LBC1921 wave 1 and 3, gender, BMI, age in days at testing at age 79 or 87 and MHT score at age 11 (age residualized) were included as covariates. For analysis in LBC1921 wave 2, the following were included as covariates: gender, age in days at testing at age 83 and MHT score at age 11 (age residualized). For the analysis in ET2DS, the following covariates—gender, age in days, BMI and Mill Hill vocabulary scale as a measure of pre-morbid IQ—were included. Gender has an effect on serum uric acid levels (28). BMI was included as a covariate to control for the effect of BMI on uric acid levels as BMI is a potential predictor of illnesses associated with high uric acid levels (32). Childhood MHT score (LBC samples) and Mill Hill vocabulary scale (ET2DS) were included as covariates to adjust for prior cognitive ability, thus allowing us to specifically identify associations with cognitive functions in old age while adjusting for cognitive differences in youth. Previous studies include a gender by age in days at testing interaction and this was tested for the cognitive phenotypes, but none were significant (all P > 0.05) so were not included in the model. A meta-analysis was performed using the Comprehensive Meta-analysis package (57). The correlation was used as the standardized beta effect size measure, time points were used to distinguish between wave 1, 2 and 3 LBC1921, and a random effects model was specified to account for systematic differences between LBC1921, LBC1936 and ET2DS.

Power and significance threshold

The parameter value for statistical significance was determined by the Bonferroni method that corrects the critical significance level by the number of tests (n) performed (α = 0.05/n). To find the number of independent cognitive domains in the 16 psychometric tests in LBC1936, principal components analysis extracted three independent domains. On the basis of matrix spectral decomposition, the four SNPs represent 2.87 independent variables (58). The significance threshold thus applied to LBC1936 was P ≤ 0.006. A nominal significance threshold (P < 0.05) was used for the replication cohorts (LBC1921 and ET2DS) as the strict Bonferroni correction was not considered appropriate.

The power to detect an additive effect of a causal variant, in linkage disequilibrium D′ = 1, of a marker with an allele frequency of 0.2, accounting for 1–2% of the variance, at type-1 error rate adjusted for multiple testing (P ≤ 0.006) was 67–96% in LBC1936. This was estimated using the variance component quantitative trait loci association module in the genetic power calculator (59).

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

FUNDING

This work has made use of the resources provided by the Edinburgh Compute and Data Facility (ECDF) (http://www.ecdf.ed.ac.uk/). The ECDF is partially supported by the eDIKT initiative (http://www.edikt.org).The LBC1936 was supported by a programme grant from Research Into Ageing and continues with programme grants from Help the Aged/Research Into Ageing (Disconnected Mind). The LBC1921 data were collected by grants from the Biotechnology and Biological Sciences Research Council (wave 1), a Royal Society-Wolfson Research Merit Award to I.J.D. (wave 2), and the Chief Scientist Office of the Scottish Government (wave 3). The ET2DS is led from the Centre for Population Health Sciences, University of Edinburgh and is funded by the Medical Research Council, the Chief Scientist Office of the Scottish Executive and Pfizer plc. The research was done within the University of Edinburgh Centre for Cognitive Ageing and Cognitive Epidemiology, supported by the BBSRC, EPSRC, ESRC and MRC, as part of the cross-council Lifelong Health and Wellbeing Initiative.

ACKNOWLEDGEMENTS

We thank the LBC1936, LBC1921 and ET2DS cohort members. We thank Michelle Luciano, Veronique Vitart and Lars Penke for genetics advice on this study. We thank the study secretaries Paula Davies and Evelyn Crooks. We thank Janie Corley, Caroline Brett, Caroline Cameron, Alison Pattie, Michelle Taylor and staff of the ET2DS for data collection and data entry. We thank the nurses and staff at the WTCRF where the data were collected. We thank the staff at the Lothian Health Board and the staff at the SCRE Centre, University of Glasgow.

Conflict of Interest statement. None declared.