Demographic data of patients and healthy controls
Demographic data for patients and healthy controls are shown in Table 1. There were significant differences between patients and healthy controls in terms of age, sex, education and body mass index (BMI) (all p < 0.01), which were controlled for in the following analyses.
BDNF serum Levels, allele and genotype frequencies of BDNF rs6265 and rs10835210 in patients and healthy controls
BDNF serum levels in 239 patients and 259 healthy controls were normally distributed (Kolmogorov-Smirnov one-sample test: both p > 0.05). Serum BDNF levels were significantly lower in patients than in healthy controls (8.8 ± 3.7 n/ml vs 12.0 ± 2.3 ng/ml; F = 63.4, df = 1,497, p < 0.001). This difference remained significant after controlling for age, sex, education, and BMI (F = 47.3, df = 5, 490, p < 0.001).
Correlation analysis of all patients showed that BDNF levels were associated with PANSS total score (r = 0.31, df = 250, p < 0.001), positive symptom subscore (r = 0.22, df = 250, p < 0.01), general psychopathology subscore (r = 0.29, df = 250, p < 0.001) and HAMD total score (r = 0.16, df = 250, p < 0.05). Stepwise multiple regression analysis revealed that age (β = 056, t = 4.16, p < 0.001), age of onset (beta=-0.61, t=-4.67, p < 0.001), PANSS total score (beta = 0.04, t = 2.93, p < 0.001) and general psychopathology subscore (beta = 0.06, t = 2.16, p < 0.05) ) were independently correlated with serum BDNF levels.
Genotypes were available for 308 FEDN patients and 422 healthy controls. Allele frequencies, genotype distributions and statistical analyses are shown in Table 1. The x2 goodness-of-fit test showed that all genotype distributions of the two polymorphisms in the BDNF gene in patient and control groups were consistent with Hardy–Weinberg equilibrium (all p > 0.05). rs10835210 genotype and allele frequencies were significantly different between patients and controls (genotype χ2 = 6.03, P = 0.049, allele χ2 = 5.05, P = 0.025). The frequency of the A allele (minor allele) of rs10835210 was higher in patients than in controls. In addition, there were no allelic or genotypic differences in BDNF rs6265 Val66Met (Table 1).
Furthermore, PANSS data were available for 250 patients. The three genotypes differed significantly in PANSS positive symptom subscore for BDNF rs6265 (Val66Met) (F = 6.82, df = 2,247, p < 0.001; Bonferroni corrected p < 0.01). Both Val/Val (23.4 ± 6.7 p < 0.001; effect size = 0.68) and Val/Met (21.6 ± 6.0; p = 0.014; effect size = 0.40) genotypes had significantly higher positive symptom scores than Met/Met (19.2 ± 5.9) genotype. In addition, there was a significant difference in PANSS positive symptom subscore for rs10835210 (CC = 20.1 ± 6.3 vs CA + AA = 23.1 ± 5.8; F = 14.73, df = 1,244, p < 0.001; Bonferroni corrected p < 0.01). Therefore, both BDNF rs6265 and rs10835210 variant alleles may have an effect on positive symptoms of the SCZ patients.
In addition, there was a significant difference in BDNF levels for the rs10835210 genotype (F = 4.59, p < 0.01), showing that patients with the CC genotype had higher BDNF levels than those with the A allele (9.8 ± 4.2ng/ml vs. 7.8 ± 3.3ng/ml). However, there was no significant difference in BDNF levels for the rs6265 genotype (p > 0.05).
Clinical trial of celecoxib: demographic and basic descriptive data
Of 308 FEDN SCZ patients, we randomly selected 90 patients to enter the clinical trial. By random assignment, 46 patients received celecoxib and 44 patients received placebo. Ten patients left before study completion, including 4 patients on celecoxib and 6 on placebo (Fig. 1).
At baseline, there were no significant differences between the two treatment groups in terms of demographic or clinical variables, including HAMD and PANSS scores (Table 2). We present the analysis of study completers with an intention-to-treat analysis showing similar results.
In addition, there was a significant difference in BDNF levels for the rs10835210 genotype (F = 4.59, df = 1,88, p = 0.025), showing that patients with the CC genotype had higher BDNF levels than patients with the A allele (9.8 ± 4.2ng/ml vs 7.8 ± 3.3ng/ml). However, there was no significant difference in BDNF levels for the rs6265 genotypes (p > 0.05).
Treatment outcomes for comparison of celecoxib vs placebo
Table 2 provides the group means for the treatment comparisons for the HAMD and PANSS outcomes. Repeated-measures ANCOVA on the total HAMD score showed a significant group-by-time effect (F = 4.7, df = 1,78, p = 0.03), a significant time effect (F = 39.7, df = 1,78, p < 0.001), and a non- significant group effect (F = 0.73, df = 1,78, p = 0.40). The HAMD total score was significantly lower in the celecoxib group compared to the placebo group at week 12 (F = 10.34, df = 1,78, p = 0.002). This difference in HAMD total score remained significant (F = 9.79, df = 5,74, p = 0.003) after covarying for sex, age, BMI, and duration of illness. The reduction in HAMD total score from baseline to week 12 in the celecoxib and placebo groups were 10.2 ± 9.9 versus 4.7 ± 10.4 (F = 5.79, df = 1,78, p = 0.019, effect size = 0.54) (Fig. 2).
To further assess the effect of treatment, we calculated the percentage of responders. At the end of 12 weeks of treatment, 32 of 42 patients (76.2%) treated with celecoxib had a 50% or greater improvement in their total HAMD score, compared with 14 of 38 patients (36.8%) given placebo (Χ2 = 12.64, df = 1, p < 0.001).
Table 2 also shows significant improvement in PANSS scores between baseline and week 12 (all p < 0.01). A repeated-measures MANOVA of the PANSS total and its 3 subscale scores showed significant group-by-time effects (F = 6.9 ~ 8.0, p < 0.05 ~ 0.006) for the PANSS total score, positive and negative symptom scores, significant time effects (F = 51.5 ~ 484.0, all p < 0.0001) and non-significant group effects (all p > 0.05) for all PANSS total and subscale scores. In addition, at week 12, all PANSS total and subscale scores were significantly lower in the celecoxib group than in the placebo group (F = 8.99 ~ 15.54, p < 0.005 ~ 0.0001, effect size = 0.43 ~ 0.79).
Systemic side effects detected by routine physical and neurological examinations and laboratory tests, as well as side effects assessed using the TESS, AIMS, and SEPS scales, were not significantly different between celecoxib and placebo groups (all p > 0.05). No significant side effects were found with celecoxib treatment.
Effect of BDNF genotypes on treatment outcome for depressive symptoms
Classifying patients with a 50% or greater reduction in their HAMD total score as responders resulted in significantly more celecoxib responders than placebo responders (76.2% vs. 36.8%, Χ2 = 12.64, p < 0.001; odds ratio = 5.41; 95% confidence interval 1.99 ~ 14.71).
Celecoxib and placebo groups were balanced in terms of allele frequency and genotypes for both two BDNF polymorphisms. Logistic regression models of clinical response showed a significant interaction between rs10835210 genotype and treatment improvement, with an adjusted odds ratio of 1.63 (95% confidence interval 1.04 to 2.54; x2 = 4.55, P < 0.05; Table 3). The significant rs10835210 genotype *treatment interaction indicated a greater likelihood of clinical response with celecoxib treatment in the BDNF rs10835210 CC group (celecoxib: 19/23 = 82.6% vs. placebo: 8/22 = 40.9%) than in the A allele carrier group (celecoxib: 13/19 = 68.4% vs. placebo: 6/16 = 43.8%).
BDNF levels as a predictor of treatment outcome
Baseline BDNF levels were associated with improvement in PANSS total score (r = 0.28, df = 1,80, p < 0.05) and positive symptom subscale (r = 0.26, df = 1,80, p < 0.05), but not with improvement in HAMD. Post-treatment BDNF levels did not differ significantly between the two treatment groups (F = 1.64, df = 1,79, p > 0.05).
In addition, the increase in BDNF levels from pre-treatment to post-treatment was significantly associated with baseline PANSS total score (r=-0. 27, p < 0.05) and baseline positive symptom subscale (r=-0.30, p < 0.05), as well as age of onset (r=-0.26, p < 0.05).
Clinical predictors of treatment outcome
Baseline PANSS total score (r = 0.35, df = 1,80, p < 0.01), positive symptoms subscale (r = 0.24, p < 0.05) and general psychopathology subscale (r = 0.42, p < 0.0001) significantly predicted the decrease in HAMD total score during treatment. Regression analyses including BDNF levels, demographics, and clinical assessments revealed that only the baseline PANSS general psychopathology subscale (t=-3.51, p < 0.001) was positively associated with improvement in HAMD total score. Finally, as expected, improvement in HAMD was associated with improvement in PANSS total score (r = 0.55, p < 0.0001), positive symptom subscale (r = 0.43, p < 0.0001) and general psychopathology subscale (r = 0.62, p < 0.0001).