Iron, zinc and copper in the Alzheimer's disease brain: a quantitative meta-analysis. Some insight on the influence of citation bias on scientific opinion

Abstract

Dysfunctional homeostasis of transition metals is believed to play a role in the pathogenesis of Alzheimer's disease (AD). Although questioned by some, brain copper, zinc, and particularly iron overload are widely accepted features of AD which have led to the hypothesis that oxidative stress generated from aberrant homeostasis of these transition metals might be a pathogenic mechanism behind AD. This meta-analysis compiled and critically assessed available quantitative data on brain iron, zinc and copper levels in AD patients compared to aged controls. The results were very heterogeneous. A series of heavily cited articles from one laboratory reported a large increase in iron in AD neocortex compared to age-matched controls (p<0.0001) while seven laboratories failed to reproduce these findings reporting no significant difference between the groups (p=0.76). A more than three-fold citation bias was found to favor outlier studies reporting increases in iron and this bias was particularly prominent among narrative review articles. Additionally, while zinc was not significantly changed in the neocortex (p=0.29), copper was significantly depleted in AD (p=0.0003). In light of these findings, it will be important to re-evaluate the hypothesis that transition metal overload accounts for oxidative injury noted in AD.

Figure 1. Meta-data of studies reporting the effect of Alzheimer’s disease on neocortical iron, zinc and copper concentrations

Data from hippocampus, frontal, temporal and parietal lobes were combined to assess neocortical metal levels. Studies from the University of Kentucky (U of K) are indicated by red parentheses for iron data. The mean effect size indicated by red vertical lines includes data reported by U of K (p=0.13). The black vertical line indicates the meta-effect size for iron when this data source is excluded. There is no significant increase in neocortical iron in Alzheimer’s brain: effect size = −0.05, 95%CI −0.34-0.25; n= 206 control, 251 AD (p=0.76). There is a trend toward an increase in neocortical zinc, although the dataset is significantly heterogenous: effect size = 0.26, 95% CI −0.22-0.75; n= 166 control, 118 AD (p=0.29). Copper levels are significantly depleted in Alzheimer’s disease neocortex: effect size = −0.59, 95% CI −0.87- −0.31; n= 123 controls, 115 AD (p<0.0003). This correlates to a reduction of about 0.4 μg Cu/g tissue, or a 13.8% reduction.

Figure 2. Laboratory of origin and not analytical technique appears to be the source of heterogeneity in measurements of neocortical iron in Alzheimer’s disease

Subgroups analysis found that the University of Kentucky results indicate a significant positive effect size. Data from other groups using INAA for analysis found no significant change and groups using non-INAA techniques also found no significant change. The results reported by the University of Kentucky are different from the results obtained from groups whether they used INAA for analysis (p<0.0001) or non-INAA techniques (p<0.0001).

Figure 3. No evidence of publication bias is present in the dataset

Funnel plot illustrates symmetric distribution of the studies and exclusion of the results from the University of Kentucky does not alter this analysis. The long-term impact factor of journals publishing studies which analyzed iron content of Alzheimer’s disease brain was not statistically different between those that argue that iron levels change and those that argue that it does not.

Figure 4. Significant citation bias is present favoring studies which found an increase in brain iron

There was a significant increase in both the number (p<0.05) and frequency (p<0.01) of citations favoring studies which found a significant increase in brain iron in AD. A more than three-fold increase in the number and frequency of citation (p<0.0001 for each) was observed favoring articles which interpreted their data as supporting the hypothesis that iron is increased compared to those that do not. In some cases the authors’ interpretation appeared to differ from the data – in these cases the interpretation may have been based on data from a particular brain region or other non-quantitative findings or the data may have appeared significant prior to conversion to the standard measure (μg Fe/g wet weight tissue) which was necessary for inclusion in this study.

Figure 5. The divergent effects of quality measures and citation bias lead to different interpretations of the dataset

When all available studies were included in the analysis, a small positive effect size was observed. Introduction of quality-based exclusion criteria adjusted the effect size to −0.08 (green vertical arrow). Finally, when all available studies were included and weighted by the number of citations received (instead of conventional weighting by inverse variance), the apparent effect size was increased to 0.53 (red arrow). It should be noted that even this exaggerated effect size correlated to less than 3 g/g increase in brain iron (about 5% higher than control).

Figure 6. Citation map illustrating citation bias in review literature describing the role of iron in Alzheimer’s disease

Of 115 review articles reviewed, only 59 cited quantitative literature and only one (in white square) argued that iron levels were not altered in Alzheimer’s disease. Of the 94 citations shown in the map above, only 6 referenced negative literature and of those 3 misrepresented the findings of those studies to suggest they demonstrated an increase in brain iron.