In a preliminary analysis of NSW-BTRC samples, mean PFC iron among schizophrenia cases (16.4 [95%CI 14.6 to 18.2] µmol/g, n = 38) was higher by 29% (95%CI 11–47%) compared to matched controls (12.7 [95%CI 11.2 to 14.2] µmol/g, n = 37, Figure S1a). Following replication in an independent sample set from VBBN and NIMH-HBCC (n = 48/clinical group, mean group difference = 2.15 [95%CI 0.07 to 4.22] µmol/g, Figure S1b), we pooled all samples together. Since analytical samples from the three cohorts were prepared by slightly different extraction protocols that could impact on iron levels, we normalized each cohort by its mean iron level, generating z-scores for each individual sample. In the combined cohort (ncontrol = 85, nscz = 86, Table 1), iron distributions satisfied robust normality assumptions (Figure S2a, Table S4), yet there were some individuals with high values (i.e > 4-5x SD), which led us to use a robust linear regression approach to evaluate mean differences. Iron in schizophrenia cases was higher by 0.58 (95%CI 0.18 to 0.97) standard deviations (SDs) compared with controls (Figs. 1a and S2b, Table S5).
Table 1
Clinical and postmortem characteristics of the Combined Brain Cohort.
|
Control
|
Schizophrenia
|
p-value1
|
N
|
Fraction
|
N
|
Fraction
|
Number of samples
|
85
|
|
86
|
|
|
Brain Bank
|
NSW-BTRC
VBBN
NIMH-HBCC
|
37
18
30
|
0.44
0.21
0.35
|
38
19
29
|
0.44
0.22
0.34
|
|
Sex
|
Female
Male
|
24
61
|
0.28
0.72
|
27
59
|
0.31
0.69
|
0.652
|
Ancestry
|
Caucasian
Non-Caucasian
|
69
16
|
0.81
0.19
|
70
16
|
0.81
0.19
|
0.971
|
BMI2, kg/m2 (mean ± s.d.)
|
29.3 ± 6.1
|
27.9 ± 6.2
|
0.376
|
Smokers2,3
|
No
Yes
|
37
18
|
0.67
0.33
|
15
43
|
0.26
0.74
|
< 0.001
|
Alcohol users2,4
|
No
Yes
|
18
5
|
0.78
0.22
|
17
15
|
0.53
0.47
|
0.058
|
Death circumstances5
|
Natural
Non-natural
|
79
6
|
0.93
0.07
|
55
31
|
0.64
0.36
|
< 0.001
|
Age of death, years (mean ± s.d.) [range]
|
54.4 ± 14.9 [17–85]
|
52.6 ± 16.2 [17–84]
|
0.441
|
pH (mean ± s.d.)
|
6.54 ± 0.29
|
6.49 ± 0.28
|
0.187
|
PMI, hours (mean ± s.d.)
|
33.1 ± 15.3
|
36.5 ± 18.4
|
0.189
|
1Significance of between-group comparison (control vs. schizophrenia) based on independent-samples t-test (continuous variables) or Pearson-chi2 (categorical variables). |
2Data for BMI, smoking and alcohol use were available for a sizeable subset of individuals (Table S1). |
3Smokers were defined as current smokers, heavy ex-smokers or individuals with a positive postmortem toxicology essay for nicotine. |
4Alcohol users were defined as those with a history of drinking an average of ≥ 20 g ethanol/day. |
5As no suicide cases have been documented in the control group, for ‘death circumstances’ to be a meaningful covariate, deaths by accidents and suicides were pooled together into a ‘Non-natural’ category. |
NSW-BTRC, New South Wales Brain Tissue Resource Centre; VBBN, Victoria Brain Bank Network; NIMH-HBCC, National Institute for Mental Health Human Brain Collection Core; BMI, body mass index; PMI, post-mortem interval. |
Following the notion that mean and variance measures provide complementary insights, examining within-group heterogeneity is increasingly emphasized in schizophrenia research56. We identified markedly increased PFC iron heterogeneity in schizophrenia patients, manifested as a wider distribution (Figure S3a), with a variance more than double of that in controls (Figure S3b, Table S6). Reflecting clinical heterogeneity, gene-environment interactions, secondary disease factors, or any combination of the above, this finding is consistent with recent reports indicating that relative to healthy controls, people with schizophrenia display increased interindividual differences in cortical structure57, 58 and function56.
Controlling for diagnosis, neither demographic (age and mode of death, sex, and ethnicity) nor sample-quality (pH and PMI) variables were associated with PFC iron (Table S7). The effect of diagnosis on iron displayed only minimal changes when controlling for these covariates individually or simultaneously (Table S7) or following regression-adjusted propensity-score matching (Tables S8-S9 and Figures S4-S5), consistent with the effect on iron reflecting the disease itself. For sizeable subsets of individuals (one to two thirds of entire cohort), data relating to smoking habits, alcohol use and body-mass-index (BMI) were available (Table S1). As expected59, 60, smoking and alcohol use were more prevalent among schizophrenia cases (Table 1), yet consistent with a previous report38, these habits were not associated with PFC iron, and when included as covariates, the effect of diagnosis remained significant (Tables S10-S11). BMI did not differ between the matched groups analysed here (Table 1), and iron discriminated cases from controls when controlling for BMI (Table S12).
Based on limited animal data indicating that chronic treatment with typical neuroleptics altered the BBB and facilitated brain iron uptake32, 33, a small iron elevation previously noted among schizophrenia patients was intuitively ascribed to neuroleptic treatment, rather than to the pathophysiology of the disease itself61. While a causal link between antipsychotics and iron elevation cannot be unequivocally addressed in postmortem tissue, based on data available for a subset of cases in our cohort (n = 43, Table S2), PFC iron levels were not correlated with mean antipsychotic daily dose or cumulative lifetime exposure (Figure S6b, Table S13). These results remained similar after adjusting for age-of-onset and treatment duration (Table S14). Moreover, while animal data indicated that atypical neuroleptics could have less profound effects on brain iron compared to typical agents32, we did not find an effect of neuroleptic (a)typicality on PFC iron levels (Table S15). Consistent with a previous report that indicated no correlation between a neuroleptic-free period before death and brain iron62, we did not find an association between presence of antipsychotic drugs (post-mortem toxicology assay) and PFC iron (Table S16). Finally, while lithium pharmacotherapy was recently reported to increase iron levels in the substantia nigra and hippocampus of teenagers who were at high risk of psychosis63, the paucity of patients in our sample with positive lithium toxicology (n = 4) rendered this drug an unlikely confounder. Taken together, our data are consistent with the notion that in schizophrenia, PFC iron elevation represents a primary disease-related, rather than a secondary drug-induced, abnormality.
Like iron, copper is a redox-active transition metal with myriad roles in brain function, including monoamine metabolism, mitochondrial activity, and myelination64. Reduced copper activity results in schizophrenia-like behavioral impairments65, yet copper dyshomeostasis in schizophrenia brain tissue has also rarely been studied. PFC copper, quantified in nearly 80% of our samples (Table S1), was normally distributed (Table S17). In contrast to a recent study reporting copper deficiency in midbrain specimens from schizophrenia cases66, in the PFC we observed no such difference (Figure S7a, Tables S18-S19). Moreover, brain iron is thought to be modulated by ceruloplasmin, a multi-copper ferroxidase67, and consistent with correlational data in rodents68, a positive effect of copper on iron levels was evident across both control individuals and schizophrenia cases (Figure S7b, Table S20). As iron elevation in the schizophrenia PFC remained significant throughout the physiological range of copper levels (Figure S7b), our data do not demonstrate an overt copper perturbation in schizophrenia PFC and are consistent with a primary disturbance of iron in this disorder.
To appreciate the relevance of iron perturbation in schizophrenia, we decided to quantify ferritin, a highly conserved iron-storage protein whose expression is driven by iron. In grey matter, ferritin is expressed in all cells to store and detoxify labile cytoplasmic Fe2+, rendering it chemically inactive and redox-insensitive40. Quantified by western blots (Figure S8a, antibody recognizes both heavy and light chains), PFC ferritin normally distributed across individuals (Table S21). We observed lower ferritin in tissue from schizophrenia cases (-0.45 [95%CI -0.82 to -0.08] SDs, Fig. 1b, Figure S8b, Table S22). This difference remained prominent following adjustment for relevant covariates (Table S23). As an elevated iron-to-ferritin ratio has been associated with iron toxicity and accelerated aging69, we quantified individual iron-to-ferritin ratios in our sample, and noted that the magnitude by which this ratio was elevated among schizophrenia cases (0.62 [95%CI 0.22 to 1.02] SDs, Fig. 1c, Figure S8c, Table S24) was larger than the group-differences observed for iron and ferritin alone.
In PFC tissues from control individuals, increasing age of death was associated with higher iron levels. As the rate of iron accumulation decreased with age, this relationship was best estimated using a linear-logarithmic model (t83 = 2.39, p = 0.019, Figs. 2a and S9a, Table S25). This observation is consistent with findings from postmortem52 and imaging studies38, 39, 70–72, in which the rate of age-dependent iron accumulation in the PFC slowed after the third decade of life. Notably, among schizophrenia cases PFC iron remained stable across age (t84 = 0.26, p = 0.794, Fig. 2a, Figure S9a, Table S26).
To further probe how differential trajectories of age-dependent iron accumulation affect schizophrenia risk, we performed a series of logistic regression analyses (Figure S9b). Using sequential age cutoffs, we examined the odds ratio (OR) of a schizophrenia diagnosis attributed to a 1 SD increase in covariate-adjusted iron (Table S27). As the OR was maximal when the analysis included only individuals younger than 35 (Fig. 2b), this age was selected as a cutoff for generating younger (age < 35) and older (age ≥ 35) subcohorts. While in both cohorts the probability of having been diagnosed with schizophrenia increased as PFC iron rose (Fig. 2c, Table S28), the effect of postmortem iron in predicting whether an individual had been previously diagnosed with schizophrenia was larger among individuals younger than 35 (OR = 3.72 [95%CI 1.16 to 11.9]) compared to the older individuals (OR = 1.49 [95%CI 1.14 to 1.95]). Mirroring the prediction models, between-group comparison revealed that the difference in covariate-adjusted iron between cases and controls was very large in the young subcohort (1.53 [95%CI 0.55 to 2.51] SDs), yet marginal in the older subcohort (0.46 [95%CI -0.02 to 0.94] SDs, Fig. 2d, Table S29). Taken together, our data are consistent with an accelerated iron accumulation in the developing schizophrenia prefrontal cortex, culminating in a pathological iron elevation whose peak occurs during the typical age of schizophrenia onset73.
Following an expected iron-ferritin co-expression pattern74, among both control individuals and schizophrenia cases higher levels of PFC iron were mirrored by higher ferritin (t83 = 2.65, p = 0.010 and t84 = 2.74, p = 0.007, respectively, Figs. 3a and S10a). Likely reflecting the capacity of this storage protein to continue incorporating iron at higher iron-to-ferritin ratios75, in both groups we observed a decline in the rate of ferritin accumulation upon increasing iron levels, best estimated using linear-logarithmic models (Tables S30-S31). At iron corresponding to the control group's mean, our model predicted that ferritin among cases would be lower by 0.79 (95% CI 0.31 to 1.27) SDs (Table S32).
To probe if differential patterns of iron-dependent ferritin accumulation affected disease risk, we performed a series of logistic regression analyses using sequential iron cutoffs, each examining the OR of a schizophrenia diagnosis attributed to a 1 SD increase in covariate-adjusted ferritin (Figure S10b, Table S33). As the OR was farthest from 1 when the analysis included only individuals with iron levels below the control group's mean (Fig. 3b), this value was selected as a cutoff for generating low-iron and high-iron subcohorts. While in the low-iron cohort the probability of having been diagnosed with schizophrenia markedly decreased as PFC ferritin rose (OR = 0.284 [95%CI 0.139 to 0.578]), in the high-iron subcohort ferritin did not predict an individual’s disease status (Fig. 3c, Table S34). Mirroring the prediction models, the difference in covariate-adjusted ferritin between cases and controls was large in the low-iron subcohort (-0.96 [95%CI -1.46 to -0.47] SDs), yet minimal (-0.17 [95%CI -0.67 to 0.33] SDs) in the high-iron subcohort (Fig. 3d, Table S35). Extending our previous findings indicating that elevated iron in young adulthood predicted disease status, these data reveal that even in the context of low prefrontal iron, a deficiency in iron's protective storage protein markedly increased the risk for a schizophrenia diagnosis.
To gauge the proportion of schizophrenia cases that could be attributed to perturbed iron biology, we quantified the discriminatory performance of iron, ferritin, and iron-to-ferritin ratio, focusing on relevant subcohorts (Figs. 2–3). As expected, while PFC iron could predict disease status in the entire cohort (Figure S11a), its discriminatory performance among individuals younger than 35 was higher (AUC = 0.837 [95%CI 0.633 to 0.964], Fig. 4ai). Similarly, although ferritin only marginally predicted disease status in the entire cohort (Figure S11b), its discriminatory performance among individuals with low iron improved (AUC = 0.766 [95%CI 0.639 to 0.866], Fig. 4aii). Although iron-to-ferritin ratio was more efficient at discriminating cases from controls when focusing on younger individuals (Figure S12), it was not superior to iron alone in this age group (Fig. 2b). However, across the entire cohort, the discrimination offered by the iron-to-ferritin ratio (AUC = 0.674 [95%CI 0.586 to 0.754], Fig. 4aiii) was higher than that of iron or ferritin alone (Figure S11). Having focused on the discriminatory performance of iron and ferritin in specific subcohorts (Fig. 4ai − ii), by deriving optimal cutoffs (Fig. 4b, Table S36) we could assess all three predictors simultaneously (Table S37). Such a model was superior to two-predictor models (Table S38), consistent with iron, ferritin and their ratio capturing (partially) distinct pathophysiological components. 69% of schizophrenia cases were classified as having perturbed iron biology while in 72% of healthy individuals iron biology was intact (Fig. 4c, Table S39). Overall, classification based on prefrontal iron biology was 71% (95%CI 63 to 77%) accurate in predicting schizophrenia (Fig. 4c).