Schematic of the SNCA gene and patterns of linkage disequilibrium. A, The positions of REP1 and the 5 SNPs significantly associated with PD in tier 1 are shown. Coding and noncoding sequence is indicated by black and light-gray shading, respectively. B, Pairwise linkage disequilibrium between REP1 and these 5 SNPs as measured by r2 (left) and D′ (right) in the combined control sample. REP1 alleles 259 and 261 were grouped together. Ex indicates exon.
The mean plasma α-synuclein levels by rs356219 genotype in patients and control subjects. Error bars indicate 1 SEM. Genotype was significantly associated with transformed plasma α-synuclein level in patients (P = .005) but not in controls (P = .25) in an additive model adjusting for age, sex, and red blood cell and platelet contamination.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Mata IF, Shi M, Agarwal P, et al. SNCA Variant Associated With Parkinson Disease and Plasma α-Synuclein Level. Arch Neurol. 2010;67(11):1350–1356. doi:10.1001/archneurol.2010.279
A functional repeat polymorphism in the SNCA promoter (REP1) conveys susceptibility for Parkinson disease (PD). There is also increasing evidence that single-nucleotide polymorphisms (SNPs) elsewhere in the gene are associated with PD risk.
To further explore the association of common SNCA SNPs with PD susceptibility, to determine whether evidence of allelic heterogeneity exists, and to examine the correlation between PD-associated variants and plasma α-synuclein levels.
Patients and control subjects from the NeuroGenetics Research Consortium.
Main Outcome Measures
We performed a 2-tiered analysis of 1956 patients with PD and 2112 controls from the NeuroGenetics Research Consortium using a comprehensive tag SNP approach. Previously published REP1 genotypes were also included. Plasma α-synuclein was assayed in 86 patients with PD and 78 controls using a highly sensitive Luminex assay.
Five of 15 SNPs genotyped were associated with PD under an additive model in tier 1 (α = .05). Of these, 4 were successfully replicated in tier 2. In the combined sample, the most significant marker was rs356219 (odds ratio, 1.41; 95% confidence interval, 1.28-1.55; P = 1.6 × 10−12), located approximately 9 kilobases downstream from the gene. A regression model containing rs356219 alone best fit the data. The linkage disequilibrium correlation coefficient between this SNP and REP1 was low (r2 = 0.09). The risk-associated C allele of rs356219 was also correlated with higher transformed plasma α-synuclein levels in patients under an adjusted additive model (P = .005).
Our data suggest that 1 or more unidentified functional SNCA variants modify risk for PD and that the effect is larger than and independent of REP1. This variant(s), tagged by rs356219, might act by upregulating SNCA expression in a dose-dependent manner.
The SNCA gene encodes α-synuclein, a small (140 amino acid) protein localized, in part, to presynaptic terminals that modulates vesicle trafficking and neurotransmitter release.1,2 A link between α-synuclein and Parkinson disease (PD) was first demonstrated in 1997 when a missense mutation (A53T) in SNCA was reported to cause autosomal dominant parkinsonism in a large Italian family (the Contursi kindred).3 Shortly thereafter, α-synuclein was shown to be a major component of Lewy bodies, the pathologic hallmark of both familial and sporadic PD.4 Triplication of the gene was later discovered to cause dominant early-onset PD, indicating that overexpression of wild-type synuclein was sufficient to cause disease.5,6 However, missense mutations and multiplications of SNCA proved to be rare, which raised the question of whether common variants with more subtle functional effects might modify susceptibility for PD. Our group and others have reported an association between PD and REP1 (D4S3481), a complex repeat polymorphism located approximately 10 kilobases (kb) upstream from the SNCA translation start site.7-9REP1 is essentially triallelic, and compared with the intermediate-length allele (“261”), the longest allele (“263”) is associated with increased risk and the shortest allele (“259”) with decreased risk for PD. In vitro data suggest that, rather than simply serving as a genetic marker, REP1 alleles might differentially regulate SNCA transcription, possibly through interactions with the DNA-binding protein PARP-1.10,11 Association analyses of SNCA single-nucleotide polymorphisms (SNPs), including 2 small genome-wide association studies (GWASs), yielded mixed results over the past decade.9,12-14 However, 3 larger GWASs15-17 published in 2009 all reported strong association signals from SNPs within SNCA.
Although α-synuclein was initially thought to be neuron specific,18SNCA messenger RNA (mRNA) is highly expressed in erythroid cells, and the protein is detectable in all blood components, including packed red blood cells (RBCs), peripheral blood mononuclear cells (PBMCs), platelets, and plasma.19-21 Whole-blood α-synuclein levels in individuals with SNCA triplications are approximately double those of their mutation-negative relatives and unrelated control subjects.22 Whether a correlation exists between common risk-associated SNCA polymorphisms and peripheral α-synuclein levels is yet unknown. Examining this relationship could provide complementary evidence for use in refining the association signal and in identifying the true risk variants at the SNCA locus.
The objectives of the present study were (1) to further explore the association between common SNCA SNPs and PD susceptibility, (2) to assess whether the signal from any risk-associated SNPs is independent of REP1, and (3) to test for correlation between PD-associated variants and plasma α-synuclein levels.
For the SNCA (OMIM 163890) association analysis, we studied 1956 patients with PD and 2112 controls enrolled through the NeuroGenetics Research Consortium, which includes movement disorder clinics in Albany, New York; Atlanta, Georgia; Portland, Oregon; and Seattle, Washington. The patients had a mean (SD) age at onset (AAO) of 58.7 (11.9) years and a mean (SD) age at enrollment of 67.9 (10.6) years, and 67.8% were male. The controls had a mean (SD) age at enrollment of 67.0 (18.3) years, and 37.4% were male. All patients met United Kingdom Parkinson's Disease Society Brain Bank clinical diagnostic criteria for PD as determined by a movement disorder specialist23 and were consecutively recruited, except that patients who had an AAO younger than 20 years, whose race was not solely classified as white (by self-report), or who carried pathogenic mutations in LRRK2 or PARK2 (homozygotes or compound heterozygotes) were excluded from the sample. Among patients, 22.6% reported a family history of PD in at least 1 first-degree or second-degree relative and were classified as having familial PD for the purpose of this analysis. Controls had no history of parkinsonism and were spouses of patients with PD or were community volunteers.
Participants for the plasma α-synuclein analysis were derived from a separate study on PD biomarkers. The 86 patients with PD all met United Kingdom Parkinson's Disease Society Brain Bank clinical diagnostic criteria and were enrolled at NeuroGenetics Research Consortium clinics in Portland and Seattle; they had a mean (SD) AAO of 56.9 (11.2) years and a mean (SD) age at enrollment of 66.3 (9.4) years, and 76.7% were male. The 78 controls had no evidence of parkinsonism by examination and were enrolled through Alzheimer Disease Research Centers at Oregon Health and Science University, the University of California–San Diego, and the University of Washington School of Medicine; they had a mean (SD) age at enrollment of 65.1 (10.3) years, and 43.6% were male. All blood samples from these subjects were drawn in the morning after an overnight fast, and the plasma was processed and frozen at −70°C within 90 minutes.
The institutional review boards at each participating site approved the study. All subjects gave informed consent.
We divided the sample for the SNCA association analysis into 2 tiers. Tier 1 was composed of 685 patients and 673 controls closely matched for age and sex, while tier 2 comprised the remainder of the sample (1271 patients and 1439 controls). We used the LD-select algorithm, as implemented on the SeattleSNPs Genome Variation Server (http://gvs.gs.washington.edu/GVS/) to choose tag SNPs at the SNCA locus based on data from the International HapMap Project Centre d’Etude du Polymorphisme Humain collection (CEU) population (http://www.hapmap.org). The r2 threshold for bins was 0.80, the minor allele frequency cutoff was 5%, and the region covered included 10 kb of upstream and downstream sequence (136 kb in total). For genotyping in tier 1, we selected 1 tag SNP from each of 13 bins and 2 additional SNPs reported to associate with PD that were not included in the HapMap CEU population (rs2301135 and rs2619363).24 The SNPs associated with PD in tier 1 (α = .05) were then replicated in tier 2.
The SNP genotyping in tier 1 was performed using matrix-assisted laser desorption ionization–time-of-flight mass spectrometry (Sequenom, Inc, San Diego, California). Genotyping for tier 2 (and for SNPs that failed in tier 1) was performed using TaqMan assays on a sequence detection system (ABI 7900HT; Applied Biosystems, Foster City, California). REP1 was genotyped by polymerase chain reaction using fluorescently labeled primers; the amplification products were separated on a genetic analyzer (ABI PRISM 3130) and were analyzed using commercially available software (GeneMapper 4.0) (both from Applied Biosystems).
Plasma α-synuclein levels were measured using a highly sensitive Luminex assay recently developed by our group,25 with minor modifications. Briefly, plasma samples were treated with equal volumes of 2× ristocetin-induced platelet agglutination buffer and then diluted with 0.1% bovine serum albumin–phosphate-buffered saline (pH 7.4) for a final dilution of 1:100 before incubation with capturing antibody–coupled beads. A series of recombinant full-length human α-synuclein (rPeptide, Athens, Georgia) standards was diluted in a serum matrix–mimicking buffer (2% fetal bovine serum–0.1% bovine serum albumin–phosphate-buffered saline [pH 7.4]) and run in parallel. The incubation time with the detection antibody was 4 hours. All samples were analyzed using an available workstation (LiquiChip Luminex 200; Qiagen Inc, Valencia, California).
Because the concentration of α-synuclein in RBCs and platelets is higher than that in plasma,20,21 residual amounts of these 2 blood components could result in spuriously elevated plasma α-synuclein levels. To account for this, we measured markers for RBCs (hemoglobin) and platelets (soluble P-selectin) in plasma. Hemoglobin was assayed using a commercially available kit (Human Hb ELISA Quantitation kit; Bethyl Laboratories, Inc, Montgomery, Texas) and soluble P-selectin was measured using another commercially available kit (Human sP-Selectin/CD62P ELISA Quantitation kit; R&D Systems, Inc, Minneapolis, Minnesota) according to manufacturer instructions.
Each SNP was assessed for Hardy-Weinberg equilibrium in patients and controls (separately) using an exact test. We used logistic regression analysis to test for association between SNCA genotype and PD under an additive model adjusting for sex and age at enrollment (divided into quartiles). Homozygotes for the more common allele were used as the reference. P values were generated by Wald test. Breslow-Day test was used to test the homogeneity of the odds ratios (ORs) for selected SNPs across enrollment sites. For age-stratified analyses, we grouped patients into those with early onset (AAO ≤50 years) or with late onset (AAO >50 years). Controls were assigned to a “younger” group or an “older” group. This was done first by parsing the whole sample into 5-year intervals of age at enrollment, then by determining the proportion of patients in each interval with early-onset disease, and finally by assigning a proportional random sample of controls from each interval of age at enrollment to the younger control group. The remaining controls were assigned to the older control group. Patients with early onset and late onset were compared with younger and older controls, respectively.
We performed stepwise logistic regression analysis using the Akaike information criterion to assess the relative contributions of REP1 and all SNPs replicated in tier 2 to PD risk. REP1 alleles shorter than 259 or longer than 263, which together occur at a frequency of less than 1%, were excluded from the data set.
We used commercially available software (HPlus 3.1; http://qge.fhcrc.org/hplus/index.php) to reconstruct haplotypes from nonphased genotype data and to test for haplotype disease associations before and after adjustment for age and sex. Haplotypes with an estimated frequency of less than 0.01 in patients and controls were excluded from the analysis. We used the coefficient estimates and standard errors calculated by the program to construct ORs and 95% confidence intervals (CIs). Pairwise LD (measured as D′ and the correlation coefficient r2) between markers was calculated using Haploview (http://www.broad.mit.edu/mpg/haploview/).26
The association of plasma α-synuclein with SNCA genotype in patients and controls was examined using linear regression analysis adjusting for age, sex, and plasma levels of hemoglobin and soluble P-selectin. To ensure that the assumptions of linear regression analysis were met, we used the Shapiro-Wilk test to assess for departures from normality and the Breusch-Pagan/Cook-Weisberg test to evaluate for heteroscedasticity of the residuals. The difference in the mean α-synuclein level between patients and controls was evaluated using the t test.
Two of 15 SNPs selected for tier 1 failed initial genotyping by matrix-assisted laser desorption ionization–time-of-flight mass spectrometry, but both were then successfully genotyped by TaqMan. The mean and maximum genotyping failure rates for tier 1 were 0.9% and 2.3%, respectively. One of 13 tag SNPs (rs10002435) and 1 of 2 unbinned SNPs (rs2619363) were out of Hardy-Weinberg equilibrium in controls (P < .001) and were excluded from further analysis. Of the remaining SNPs, 5 were significantly associated with PD in tier 1 (P < .05) (eTable) and were then genotyped in tier 2. After adjusting for age and sex, the association with PD was replicated for 4 of 5 SNPs in tier 2 (Table 1). There was no significant evidence of heterogeneity for any of these SNPs across recruitment sites (P ≥ .28). The most robust association observed in the combined sample (tier 1 plus tier 2) was for rs356219 (OR, 1.41; 95% CI, 1.28-1.55; P = 1.6 × 10−12), which is located approximately 9 kb downstream from the SNCA gene.
We then examined the relative contribution of these 4 SNPs and REP1 to PD risk in our combined sample using stepwise logistic regression analysis. Using a forward selection procedure, the best model (lowest Akaike information criterion) was one containing rs356219 alone. There was no significant improvement in the fit of the model by sequential addition of rs2572324, rs2737029, rs2619364, and REP1 (in that order). A backward elimination procedure yielded similar results.
We constructed haplotypes composed of REP1 and the 4 SNPs replicated in tier 2 (Table 2). To facilitate comparison with previous studies,9,17 we recoded REP1 as a binary variable, grouping the 259 and 261 alleles together. The most common haplotype (No. 1) was used as the reference. Two haplotypes were strongly associated with PD risk (No. 2 [OR, 1.43; P = 4.1 × 10−11] and No. 3 [1.49; P = 3.7 × 10−4]), and another haplotype (No. 5 [1.24; P = .06]) was marginally associated. The only allele shared by all 3 haplotypes was the C allele of rs356219.
Figure 1 shows LD between REP1 and the 5 PD-associated SNPs from tier 1 in the combined control sample (grouping the REP1 259 and 261 alleles together). There were modest to high levels of correlation among all SNPs (r2 range, 0.22-0.68 [Figure 1B, left panel]) but low levels between REP1 and each SNP, including rs356219 (r2 = 0.09). Low correlation between REP1 and rs356219 was also observed when alternate combinations of REP1 alleles were grouped together (r2 <0.01 for 259 and 263 and r2 = 0.03 for 261 and 263). The pattern of LD among these 6 markers in patients was similar to that seen in controls (data not shown).
The association between PD and rs356219 was further explored by performing regression analyses stratified by family history and by AAO (Table 3). A significant effect was seen in familial and sporadic subgroups and in early-onset and late-onset PD.
There was no significant difference in the mean (SD) plasma α-synuclein levels between patients and controls (46.9 [32.6] vs 48.6 [39.9] ng/mL, P = .76). We then examined the relationship between SNCA genotype and plasma α-synuclein separately in patients and in controls. Because our data suggested that the association between SNCA and PD was largely driven by rs356219, we focused only on this SNP. The distribution of plasma α-synuclein by rs356219 genotype deviated significantly from normality in patients (P = .001) and in controls (P < .001).
We used square root transformation in patients and natural log transformation in controls to approximate a normal distribution. There was no significant evidence of heteroscedasticity in the transformed data for patients (P = .70) or for controls (P = .48); thus, the transformed data sets were used for linear regression analysis. In cases, rs356219 was associated with square root plasma α-synuclein level under an additive model after adjusting for age, sex, and RBC and platelet contamination (P = .005). The effect can be visualized by inspection of the raw data, which are plotted in Figure 2. Heterozygous patients had α-synuclein levels that were intermediate between the TT and CC homozygous groups. Although a slight trend in the same direction can be seen in controls, rs356219 was not significantly associated with log plasma α-synuclein level in controls in an adjusted additive model (P = .25).
Our findings suggest that common variation within SNCA modifies susceptibility for PD. The association signal in our data set emanated primarily from a single SNP (rs356219), whose effect was similar in early-onset, late-onset, familial, and sporadic disease. The signal from this SNP was largely independent of REP1, the only putative functional polymorphism identified to date within the SNCA locus. Finally, we demonstrate herein for the first time that a risk-associated SNCA allele correlates with increased levels of α-synuclein protein in vivo in a dose-dependent manner.
Few large studies on SNCA using a comprehensive tag SNP (or comparable) approach have been published among populations of European origin. The largest was a recent GWAS by Simón-Sánchez et al17 of 1713 patients and 3978 controls in which strong association signals were observed from 2 genes, SNCA and MAPT. Several SNPs within SNCA were then successfully replicated in 3361 patients and 4573 controls. The most significant SNP in their combined sample was rs2736990 in intron 4 (OR, 1.23; P = 2.24 × 10−16). This finding is in agreement with our study, as the LD correlation coefficient between rs2736990 and rs356219 (our top SNP) was high (r2 = 0.82 in the HapMap CEU sample). In 2 smaller GWASs of 857 patients and 867 controls15 and of 604 patients and 619 controls,27 no markers met genome-wide significance after correction for multiple testing. However, in the former study,15SNCA was ranked ninth by strength of association among all gene regions (under an additive model), and the most significant SNCA SNP was rs356229 (OR, 1.35; P = 5.5 × 10−5), located approximately 40 kb downstream from the gene. In the latter study,27SNCA showed the highest association of all genes, and the strongest signal came from rs356220 (OR, 1.48; P = 2.7 × 10−6), located approximately 5 kb downstream from the gene. Again, both of these studies were consistent with our results, as the top SNCA SNP from each study was at least moderately well correlated with rs356219 based on HapMap CEU data (r2 = 0.56 for rs356229 and r2 = 0.96 for rs356220).
Results from our stepwise regression analysis, together with the low intermarker correlation observed (r2 = 0.09), indicate that the association signal from rs356219 is likely independent of REP1. This suggests that allelic heterogeneity exists and that 1 or more functional variants, in addition to REP1, act at the SNCA locus to convey risk for PD. The additional risk variant is unlikely to be rs356219 but rather a polymorphism in LD with it. Inspection of the HapMap CEU sample data that we used to select tag SNPs revealed 7 other SNPs within the bin containing rs356219. However, none of these SNPs were obvious candidates to have direct biologic effects. One SNP (rs356165) is in the 3′ untranslated region but not in a microRNA-binding site predicted by the MicroCosm Targets Web resource (http://www.ebi.ac.uk/enright-srv/microcosm/htdocs/targets/v5/), and the other SNPs are located downstream or deep within intron 4. Therefore, the identity of the functional variant(s) in question remains to be determined. Our data are consistent with a recent study14 of 397 patients and 270 controls from Europe in which rs356165 was found to associate with PD independent of REP1, as there was essentially no correlation (r2 = 0.00) between the 2 polymorphisms. In contrast, Simón-Sánchez and colleagues17 examined REP1 in a subset of 1774 samples from their GWAS and concluded that the association signal from REP1 and several SNCA SNPs was not independent and “may be the result of residual LD between these loci.”17(p1309) As evidence, the authors cited an r2 value of 0.37 between REP1 and rs3857059 but did not provide this information for the most significant SNCA SNP in their data set, rs2736990, which was highly correlated with our top SNP (rs356219). Therefore, it is difficult to directly compare our findings with theirs.
We found that the risk-associated C allele of rs356219 was correlated with increased plasma α-synuclein levels among patients with PD in an additive manner (Figure 2). This suggests that the true functional risk variant tagged by this SNP might confer risk by upregulating neuronal SNCA expression, assuming that plasma α-synuclein (derived mainly from RBCs) and brain α-synuclein levels are correlated. Whether this assumption is valid in typical late-onset PD remains to be determined, but carriers of SNCA triplications have a similar 2-fold increase in both whole blood and brain α-synuclein.22 Although a trend for higher plasma α-synuclein levels with increasing copy numbers of the C allele was visible in our control group, this did not reach significance. A possible explanation is that at baseline allele-specific differences in SNCA expression are modest but become accentuated beginning in the preclinical phase of the disease through gene × gene or gene × environment interactions.
Fuchs et al28 examined the relationship between 3 SNCA polymorphisms (REP1, rs2583988, and rs356219) and SNCA mRNA and α-synuclein protein levels in substantia nigra (8 patients and 14 controls), cerebellum (mRNA [5 patients and 5 controls] and protein [17 patients and 24 controls]), and PBMCs (36 patients and 79 controls). They reported that the protective REP1 259/259 genotype was associated with lower PBMC protein levels but did not see an association with the 263 risk allele. REP1 was not correlated with PBMC mRNA or brain mRNA or protein levels. The SNP rs356219 was not associated with PBMC mRNA or protein levels or with brain protein levels. However, the protective TT genotype was associated with higher mRNA levels in cerebellum. In substantia nigra, mRNA was highest in samples of the CT genotype, with lower and similar levels for the TT and CC genotypes. The SNP rs2583988 was not associated with mRNA or protein levels in any of the tissues studied. We found these data difficult to interpret for several reasons, including the seemingly opposite effects of the protective genotypes across tissues, the small sample sizes, and the fact that patients and controls were pooled together in all analyses.
Although investigators in some studies29,30 have proposed that SNCA is divided into 2 or more LD blocks, data from these studies and from our study (Figure 1B, right panel) indicate that there is still substantial LD (measured by D′) between markers at the 5′ and 3′ ends of the gene. Therefore, future work aimed at discovering the true risk variant(s) tagged by rs356219, which could require deep resequencing, should include the entire gene and possibly regions some distance away. The yield from such efforts might be increased by incorporating measurements of plasma α-synuclein. For example, selecting a subset of patients for resequencing from the extremes of the distribution of plasma α-synuclein levels might enrich the sample for chromosomes bearing susceptibility alleles. Defining the full spectrum of SNCA risk variants is an important step in better understanding the molecular mechanisms by which α-synuclein mediates neurodegeneration and ultimately could prove useful for developing PD therapeutics aimed at modifying α-synuclein in vivo.
Correspondence: Cyrus P. Zabetian, MD, MS, Geriatric Research Education and Clinical Center, Veterans Affairs Puget Sound Health Care System, Mailstop S-182, 1660 S Columbian Way, Seattle, WA 98108 (firstname.lastname@example.org).
Accepted for Publication: April 20, 2010.
Author Contributions: Drs Mata and Shi contributed equally to this study. Study concept and design: Mata, Factor, Zhang, Payami, and Zabetian. Acquisition of data: Mata, Shi, Agarwal, Chung, Factor, Galasko, Ginghina, Griffith, Higgins, Kay, Kim, Leverenz, Roberts, Samii, Yearout, Zhang, Payami, and Zabetian. Analysis and interpretation of data: Mata, Edwards, Quinn, Snapinn, Tsuang, Yearout, Zhang, and Zabetian. Drafting of the manuscript: Mata, Zhang, and Zabetian. Critical revision of the manuscript for important intellectual content: Mata, Shi, Agarwal, Chung, Edwards, Factor, Galasko, Ginghina, Griffith, Higgins, Kay, Kim, Leverenz, Quinn, Roberts, Samii, Snapinn, Tsuang, Yearout, and Payami. Statistical analysis: Mata, Edwards, and Snapinn. Obtained funding: Zhang, Payami, and Zabetian. Administrative, technical, and material support: Agarwal, Galasko, Ginghina, Higgins, Roberts, Samii, Yearout, Zhang, and Zabetian. Study supervision: Kim, Leverenz, Zhang, and Zabetian.
Financial Disclosure: Dr Agarwal serves on advisory boards for Ipsen and Merz Pharmaceuticals and has received compensation as a speaker for Boehringer Ingelheim, GlaxoSmithKline, Novartis, and Teva Neuroscience. Dr Factor has received research grants from Ipsen, Schering Plough, and Teva Neuroscience and is a consultant for Allergan, Boehringer Ingelheim, Lundbeck, and UCB. Dr Samii has received compensation as a speaker for Boehringer Ingelheim, Ipsen, and Teva Neuroscience.
Funding/Support: This work was supported by the Michael J. Fox Foundation, by the Parkinson's Disease Foundation, by the American Parkinson Disease Association, by grant 1I01BX000531 from the Department of Veterans Affairs, and by grants P30 AG008017, P42 ES004696, P50 NS062684, R01 AG033398, R01 NS065070, R01 NS036960, and R01 NS057567 from the National Institutes of Health.
Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding agencies.
Online-Only Material: The eTable is available at http://www.archneurol.com.
Additional Contributions: Erica Martinez, BS, and Sydney Thomas, BS, assisted with subject recruitment, and Carolyn Hutter, PhD, and Jia Yin Wan, MS, provided statistical support.
Create a personal account or sign in to: