Meta-analysis of the Glial Marker TSPO in Psychosis Revisited: Reconciling Inconclusive Findings of Patient–Control Differences

To the Editor:

Drugs targeting the immune system are a promising new approach for treatment of schizophrenia. In search of methods for patient stratification, great effort has been invested into positron emission tomography studies using radioligands that bind to the 18-kDa translocator protein (TSPO), a glial cell marker, yielding seemingly inconclusive results (1–12). To address this issue, we previously conducted an individual participant data meta-analysis, pooling data from all 5 published studies at the time that employed second-generation TSPO radioligands that show superior signal-to-noise ratio compared with the first TSPO radioligand, (R)-[¹¹C]PK11195 (13). Contrary to the originally expected hypothesis of an upregulation of TSPO, we observed strong evidence of lower total distribution volume (V_T)—an index of radioligand binding to TSPO—in psychosis patients compared with healthy control subjects (14).

Concurrently, an additional study was published using the radioligand [¹⁸F]PBR111 in patients with schizophrenia, showing an interaction effect between patient status and age (11). This study was subsequently included in a new meta-analysis by Marques et al., which showed no difference in V_T between patients and control subjects in studies using second-generation TSPO radioligands (15). However, this meta-analysis only made use of summary statistics, such that the results are not directly comparable to Plavén-Sigray et al. (14). Since then, data collection has been completed for a further study examining first-episode psychosis patients and control subjects using [¹¹C]PBR28 (16). Here, we aim to reconcile the differences in conclusions from TSPO positron emission tomography studies in psychosis, at the levels of both individual studies and meta-analyses. To this end, we conducted a new individual participant data meta-analysis, making use of all of the now-available data obtained using second-generation TSPO radioligands, using a preregistered analysis plan agreed upon by representatives of all included studies (see https://github.com/pontusps/TSPO_psychosis). The primary objective was to re-evaluate the hypotheses of 1) higher, 2) lower, or 3) no difference in radioligand binding between patients and healthy control subjects. Secondary objectives were to assess the effects of antipsychotic medication on TSPO levels, as well as relationships between TSPO and disease duration, symptom measures, and age-group interaction effects.

The pooled individual participant data amounted to 99 patients and 109 healthy control subjects. The analysis approach was based on our previous study, whose main outcomes were V_T values from the frontal cortex, temporal cortex, and hippocampus. Only data without partial volume correction were used, to allow for more homogeneous pooling. A linear mixed-effects model was used to examine patient–control subject differences in V_T, while accounting for the nested structure of data (varying intercepts and slopes for all included studies), with genotype as covariate. Age, sex, medication, duration of illness, and symptom severity were all examined as covariates. We also specified an additional model including a group-by-age interaction effect to test the hypothesis of Ottoy et al. (11). Alongside p values, Bayes factors (BFs) were applied to examine the relative support for the hypotheses of higher, lower, or no change of TSPO levels in patients compared with healthy control subjects [see (14) for specification of priors].

In all three regions, BFs showed moderate to strong support (all BFs > 5) for lower V_T in patients as compared with no difference (all p < .005) and strong support (all BFs > 10) for lower V_T compared with higher V_T in patients (Figure 1). There was no effect of antipsychotic medication on TSPO (all p > .6), and no association was observed between V_T and disease duration or symptom levels (all p > .5). We found no age-by-group interaction effect on V_T in any of the examined regions (all p > .6) (for full results, see https://github.com/pontusps/TSPO_psychosis).

Figure 1.

Standardized differences in V_T between patients with psychosis and HC subjects. The top panel shows the TSPO V_T values of all individual patients and HC subjects, subdivided into HAB and MAB groups, respectively. The data have been standardized within genotype and study with a mean of 0 and an SD of 1, in order to allow for visualization of compiled data obtained using different radioligands. The lower panel shows the estimated standardized difference in V_T using a Bayesian linear effects model, with study as random effect. The black circle denotes the posterior mean, and the thick line denotes the 95% credible interval of the estimated random slopes (study specific effects); these are also presented in text next to the plots. The cross denotes the patient–control subject mean difference in raw data (together with its 95% confidence interval) without performing linear mixed-effects modeling. Hence, the difference between the dot and the cross displays the model shrinkage toward the mean. HAB, high-affinity binder; HC, healthy control; MAB, mixed-affinity binder; TSPO, 18-kDa translocator protein; V_T, total distribution volume.

While the magnitude of the observed differences is reduced compared with those observed in Plavén-Sigray et al. (14), the conclusions remain unchanged. The result of no elevation of TSPO in second-generation radioligand studies is in agreement with the meta-analysis by Marques et al. (15). Instead, the current meta-analysis found patients to have lower V_T compared to control subjects. This could be explained by an increased sensitivity in our analysis, since the individual participant data approach allows for control of potential confounding effects, such as age, sex, or medication, and estimation of the effect sizes using the data directly, rather than relying on the specific summary statistics reported in the original publications (17). We chose not to include patient–control subject comparisons using (R)-[¹¹C]PK11195 because the outcome measures used in these studies show low reliability and consistency (18–20), likely owing to low specific binding of this radioligand (13,21). Hence, based on the best available evidence so far, we conclude that TSPO is lower in patients with psychotic disorders as compared with healthy control subjects.

The presence of lower TSPO in patients does not necessarily abrogate the hypothesis of proinflammatory activation in psychosis. While TSPO has been shown to be sensitive in systemic inflammation paradigms (22,23), TSPO is not specific to microglia (24,25), and in vitro data have shown a lack of correspondence between TSPO and proinflammatory activation (26,27), although conflicting data exist (28,29). For schizophrenia specifically, the lack of a relationship between TSPO and cell- and activation-specific markers was recently confirmed in a postmortem study (30). In addition, body mass index was recently reported to influence TSPO values in a large sample of healthy control subjects (31). This could be a potential confounder for the present analysis, but it could also offer a future line of investigation into disease mechanisms and risk factors underlying psychotic disorders.

Although we observed no associations between symptom severity or duration of illness and V_T, it is noteworthy that the three studies displaying the strongest effects (8,10,16) included patients in an early phase of the disorder with a majority being unmedicated or drug naïve. On the opposite side of the meta-analytical estimate is the study by Ottoy et al. (11), which included a sample of patients with longer duration of illness and expressing more severe symptoms (32). To further examine this pattern, longitudinal studies monitoring the in vivo expression of immune markers in patients from the onset throughout different stages of the disorder are warranted.