Cytoplasmic Rbfox1 isoform is enriched in PVIs in human PFC.
To investigate if cytoplasmic Rbfox1 isoform levels are altered in PVIs in schizophrenia, we first examined whether Rbfox1 and its isoforms are enriched in these neurons in human PFC. To assess the enrichment of Rbfox1, we compared the levels of Rbfox1 protein across PVIs, calbindin-expressing interneurons (CBIs), and calretinin-expressing interneurons (CRIs), the three major interneuron subtypes in human PFC37 (Fig. 1A-C). Total Rbfox1 protein levels, defined as the mean level of Rbfox1 immunoreactivity in the entire cell body of each interneuron, significantly differed (F2,6=7.1, p = 0.026) across interneuron subtypes (Fig. 1D). Total Rbfox1 protein levels were 4.0-fold greater in PVIs and 3.0-fold greater in CBIs relative to CRIs. Post hoc analyses revealed that total Rbfox1 protein levels were significantly greater in PVIs than in CRIs but did not differ between CBIs and CRIs. These findings are unlikely to reflect any interneuron differences in susceptibility of Rbfox1 immunoreactivity to postmortem effects as a similar pattern of Rbfox1 protein levels across three interneuron types was also present in the PFC of perfused macaque monkeys (Supplementary Fig. 1).
Next, we examined whether the protein levels of cytoplasmic and nuclear Rbfox1 isoforms in PVIs differ from CBIs or CRIs. Cytoplasmic Rbfox1 isoform levels, defined as mean levels of Rbfox1 immunoreactivity in the cytoplasm, significantly differed (F2,6=11.7, p = 0.008) across interneuron subtypes (Fig. 1E). Cytoplasmic Rbfox1 levels were 9.6-fold greater in PVIs and 5.0-fold greater in CBIs relative to CRIs. Post hoc analyses revealed that cytoplasmic Rbfox1 protein levels were significantly greater in PVIs relative to CRIs but did not differ between CBIs and CRIs. Similarly, nuclear Rbfox1 isoform levels, defined as mean levels of Rbfox1 immunoreactivity in the nucleus, significantly differed (F2,6=6.8, p = 0.029) across interneuron subtypes (Fig. 1F). Nuclear Rbfox1 isoform levels were 3.0-fold greater in PVIs and 2.6-fold greater in CBIs relative to CRIs. Post hoc analyses revealed that nuclear Rbfox1 isoform levels were significantly greater in PVIs relative to CRIs but did not differ between CBIs and CRIs.
To compare the magnitude of the shift in splicing between cytoplasmic and nuclear isoforms of Rbfox1 across interneuron subtypes, we assessed the cytoplasmic-to-nuclear (C/N) ratio of Rbfox1 isoform levels. The C/N ratio of Rbfox1 isoform levels significantly differed (F2,6=13.7, p = 0.006) across interneuron subtypes (Fig. 1G). The Rbfox1 C/N ratio was 1.8-fold and 3.3-fold greater in PVIs relative to CBIs and CRIs, respectively. Post hoc analyses revealed that Rbfox1 C/N ratio was significantly greater in PVIs relative to both CBIs and CRIs. Consistent with this finding, the distribution of the C/N ratio across individual interneurons was shifted to the right in PVIs relative to both CBIs and CRIs. (Fig. 1H-J). These findings suggest that Rbfox1 is highly expressed in PVIs in human PFC and that the splicing of Rbfox1 is shifted to produce relatively more cytoplasmic than nuclear isoform in PVIs compared to other major interneuron subtypes.
Cytoplasmic Rbfox1 protein levels in PVIs are lower in schizophrenia.
To investigate whether schizophrenia is associated with altered levels of cytoplasmic Rbfox1 in PVIs, we quantified the protein levels of cytoplasmic Rbfox1 in PVIs from the PFC of 20 pairs of schizophrenia and unaffected comparison subjects. Mean levels of cytoplasmic Rbfox1 immunoreactivity were significantly (F1,19=17.2, p < 0.001) 29% lower in PVIs from individuals with schizophrenia relative to unaffected comparison subjects (Fig. 2A). In addition to lower levels of cytoplasmic Rbfox1, the mean levels of nuclear Rbfox1 (F1,19=19.9, p < 0.001) and total Rbfox1 (F1,19=19.9, p < 0.001) immunoreactivities were also lower in PVIs in schizophrenia subjects (Fig. 2B, C). However, the Rbfox1 C/N ratio did not significantly differ between subject groups (Fig. 2D). Moreover, the levels of cytoplasmic and nuclear Rbfox1 isoforms were significantly positively correlated across both unaffected comparison and schizophrenia subjects (Fig. 2E) and these correlations were also evident across individual PVIs (Fig. 2F). These findings suggest that the levels of cytoplasmic and nuclear Rbfox1 isoforms are co-regulated and that schizophrenia is associated with lower levels of both isoforms in PVIs without evidence of shift in splicing.
Next, we investigated whether lower cytoplasmic Rbfox1 levels in PVIs in schizophrenia might be due to methodological confounds or diagnosis-associated co-occurring factors. Consistent with our prior study using the sample subject cohort30, the mean level of PV immunoreactivity was significantly (F1,17=24.4, p < 0.001) 37% lower in schizophrenia (Fig. 3A), validating that our sampling method robustly captured PVIs that are affected in the illness. The number and the surface area of sampled PVIs did not differ between subject groups (Fig. 3B, C), suggesting that our findings are not due to methodological confounds associated with sampling or masking of PVIs, respectively. Finally, the levels of cytoplasmic (Fig. 3D), nuclear or total Rbfox1 (Supplementary Fig. 2) did not significantly differ as a function of a diagnosis of schizoaffective disorder; the presence of a substance use disorder at time of death; use of nicotine, antidepressants, benzodiazepines, or valproic acid at the time of death; or death by suicide. Together, these findings support our hypothesis that schizophrenia is associated with lower cytoplasmic Rbfox1 levels in PVIs and suggest that this alteration is not due to methodological confounds or diagnosis-associated co-occurring factors.
Lower cytoplasmic Rbfox1 protein levels predict lower mRNA levels of Vamp1 in PVIs.
Cytoplasmic Rbfox1 has been shown to enhance the mRNA stability of its major target Vamp118. To investigate if lower cytoplasmic Rbfox1 protein levels predict lower Vamp1 mRNA levels in PVIs in schizophrenia, we first explored whether Vamp1 is highly expressed in these neurons in human PFC. Vamp1 mRNA levels were highest in layer 4 of human PFC where PVIs are most abundant37 (Fig. 4A). In this layer, the density of Vamp1 mRNA grains was two-fold greater in PVIs relative to non-PV neurons (Fig. 4B), consistent with previous mouse studies reporting that Vamp1 is enriched in PVIs18,38.
Next, we explored if we could co-detect Vamp1 mRNA and Rbfox1 protein in the same PVIs in human PFC. Using a novel approach that combines multi-label in situ hybridization and immunohistochemistry39, we were able to detect the presence of Rbfox1 immunoreactivity and Vamp1 mRNA grains in the same PVIs in fresh-frozen human PFC sections (Fig. 4C). To validate that this approach can reliably sample affected PVIs in fresh-frozen PFC sections of schizophrenia, we selected the 10 subject pairs with the largest difference in cytoplasmic Rbfox1 protein levels in fixed PFC sections. Then, we quantified cytoplasmic Rbfox1 protein levels and the density of PV mRNA grains in PVIs in fresh-frozen PFC sections from these 10 pairs. The mean levels of cytoplasmic Rbfox1 immunoreactivity (F1,9=5.2, P = 0.048; Fig. 4D) and the mean density of PV mRNA grains (F1,9=4.4, P = 0.065; Fig. 4E) were each 14% lower in PVIs in schizophrenia, respectively. These findings demonstrate that our approach robustly detects deficits in both mRNA and protein levels in affected PVIs in schizophrenia using fresh-frozen PFC sections.
Finally, we investigated the relationship between the levels of cytoplasmic Rbfox1 protein and Vamp1 mRNA in PVIs in schizophrenia. The mean density of Vamp1 mRNA grains was significantly (F1,9=5.9, P = 0.038) 20% lower in PVIs in PFC of schizophrenia relative to comparison subjects (Fig. 4F). Furthermore, the levels of cytoplasmic Rbfox1 immunoreactivity were significantly positively correlated with the density of Vamp mRNA grains across individual PVIs in both unaffected comparison and schizophrenia subjects (Fig. 4G). Together, these findings demonstrate that Vamp1 mRNA levels are lower in PVIs in schizophrenia and suggest that this alteration could be due to deficits in cytoplasmic Rbfox1 levels in these neurons.
Lower release probability of GABA from PVIs reduces gamma power in a computational model network.
Vamp1 regulates the release probability of neurotransmitters and the loss of Vamp1 levels in PVIs reduces the release probability of GABA from these neurons18,28,29. To explore the functional consequence of Rbfox1-Vamp1 alterations in schizophrenia, we simulated the effect of reduced release probability of GABA from PVIs (RPI◊E) on gamma power in a model network of pyramidal neurons and PVIs (Fig. 5A).
Maximum gamma power occurred at RPI◊E=1 (Fig. 5B). Gamma power then sharply decreased with RPI◊E<1 and reached a stable nadir at RPI◊E≤0.7, demonstrating that reducing RPI◊E disrupts the generation of gamma oscillations in the model network. To further investigate the network properties affected by lower RPI◊E, we assessed its effect on network activity, measured by the firing rates of pyramidal and PVIs, and its effect on network synchrony, measured by the coefficient of variation of the interspike interval (CVISI) of pyramidal neurons. Decreasing RPI◊E from 1 to 0.7 had minimal effect on neuronal firing rates (Fig. 5C), but steeply increased CVISI (Fig. 5D). These findings suggest that lower RPI◊E reduces gamma power primarily by disrupting network synchrony (i.e., higher CVISI) while minimally affecting network activity (Fig. 5E).
In addition to lower release probability, the strength of PVI-mediated inhibition is also thought to be lower in schizophrenia based on prior studies reporting lower levels of the GABA synthesizing enzyme GAD67 in PVIs10,40,41 and lower levels of GABAA1 receptor α1 subunit (GABAA1α1) mRNA in pyramidal neurons42,43. To investigate how deficits in the release probability and the strength of inhibitory drive from PVIs together influence PFC gamma power in schizophrenia, we simulated the effect of the interaction between RPI◊E and PV-mediated inhibitory strength (GI◊E) on gamma power. To simulate the effect of this interaction, GI◊E (Fig. 5F; green line) or RPI◊E (Fig. 5F; blue line) was each reduced incrementally from its value that produced maximum gamma power (i.e., GI◊E=0.7 and RPI◊E=1, corresponding to 100% parameter weight in Fig. 5F). Next, the magnitude of the reduction in gamma power from each parameter was summed to predict the expected reduction in gamma power from lowering both parameters together (Fig. 5F; red line). These values were then compared to the actual gamma power reduction simulated by lowering GI◊E and RPI◊E concurrently (Fig. 5F; orange line). Our simulation showed that the actual magnitude of reduction in gamma power was greater than the expected reduction until GI◊E and RPI◊E were lowered by 15%, after which the magnitude of gamma power reached a stable nadir. Thus, these findings suggest that lower GI◊E and lower RPI◊E synergistically interact to reduce gamma power non-linearly in the model network.