Decoding frontotemporal and cell type-specific vulnerabilities to neuropsychiatric disorders and psychoactive drugs

Article activity feed

1. 

   Review Commons

   Dec 5, 2023

   Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Reply to the reviewers

   Reviewer #1

   We thank this Reviewer for the time spent assessing our manuscript, and for suggesting approaches to strengthen the robustness of the differences (e.g., TL vs FL) reported in our results. We have carefully addressed each point raised by this and other reviewers, providing new analyses and data - see list below. Indeed, these analyses combined helped us to make our main results reproducible, corroborating the main findings and refining the message of the manuscript.

   New analyses/data added:

   1. *Effect of batch due to different lanes - comparison of DEGs (TL/FL) obtained when samples in different lanes are tested individually (new Figure S15). *

   2. Ef…

   Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Reply to the reviewers

   Reviewer #1

   We thank this Reviewer for the time spent assessing our manuscript, and for suggesting approaches to strengthen the robustness of the differences (e.g., TL vs FL) reported in our results. We have carefully addressed each point raised by this and other reviewers, providing new analyses and data - see list below. Indeed, these analyses combined helped us to make our main results reproducible, corroborating the main findings and refining the message of the manuscript.

   New analyses/data added:

   1. *Effect of batch due to different lanes - comparison of DEGs (TL/FL) obtained when samples in different lanes are tested individually (new Figure S15). *

   2. Effect of batch correction on our results - comparison of the DEGs (TL/FL) obtained with and without batch removal (new Figure S15).

   3. Sensitivity of our enrichment results for GWAS significance – we performed the enrichment of GWAS genes using different GWAS thresholds, 10-6, 10-7, 5x10-8 (new Figure S14).

   4. Expression analysis of GRIN2A and SLC12A5 in Allen Brain Atlas data and qPCR results of GRIN2A and SLC12A5 in patients with frontal and temporal lobe traumatic injury (new Figure S12, Table S3).

   5. Comparison of the DEGs (TL/FL) with DEGs (autism/Ctrl) obtained from single cell RNA seq (new Figure S16, Table S7).

   6. Comparison of the results using the GWAS genes derived from Trubetskoy et al. with our gene lists (new Figure S17).

   7. Description of the data quality (Figure S2) Major points:

   8. * The main limitation of the work is the small starting sample size. The authors studied 1 frontal lobe sample and 2 temporal lobe samples. Although this information was in Table S3 it would be good to include upfront in the Methods. snRNA-seq was generated on the 10x platform. It would be helpful to know if the 10x step and sequencing was performed as one batch, or as individual batches. Similarly, were the sample libraries all sequenced on the same lane, or different lanes. The authors do not state in the Methods how many nuclei they were targeting and this should be included. Sample pre-processing was well described and standard.* We now provide additional details about the sequencing step (nuclei, sample pre-processing, etc.) in the revised manuscript (see Methods, and text below). The potential batch effect of the lane is discussed and addressed in the next point.

   ‘10X Genomics uses a microfluidic system for cell sorting. Cells and enzymes, combined with Gel Beads, enter the oil phase to form GEMs. The resulting sample libraries were sequenced on separate lanes. To enhance sequencing depth, the primary target number of nuclei for the two samples from TL is set at 10,000, considering an RNA integrity number (RIN) of 6.5. In contrast, for the sample from FL, the target is set at 20,000 nuclei due to a higher RIN of 8.1.’

   * In relation to Batch correction - as with any batch correction method, it is unclear whether the correction is adjusting for biological differences or technical. Since this is a study of the differences between FL and TL, would it not be more appropriate not to correct for batch, particularly as the samples were analysed individually - particularly if batch effects were carefully controlled for in the initial study design. The authors should test whether the results are robust to batch correction or not.*

   Since the samples are sequenced by different lanes of 10X platform we can’t exclude potential batch effects. To address this, we corrected the batch by CCA (canonical correlation analysis) which enhanced the clustering and the UMAP visualization, which is now less affected by batch-specific variations.

   Moreover, in an attempt to account for the sample size limitation, we employed 3 approaches to confirm the main transcriptional differences between the 2 regions, and that these are “robust” to batch correction, as is shown in new Figure S15: (1) Comparison of the gene expression differences (2 TL vs 1FL) with and without removing batches (new Figure S15. a, c); (2) The results obtained by comparing the differences between each individual TL sample (processed in different lanes) and FL sample are contrasted with the results after batch removal (new__ Figure S15. b, d__); (3) To confirm a limited effect of lane, we provide analysis of the expression similarity of three samples which demonstrates, consistently for each major cell type and neuronal sub-types, a strong correlation between the two TL samples (form different lanes) as compared with FL (new Figure S15. e).

   As shown in panel a, c, below, the majority of DEGs (2TL vs FL) identified with batch effects largely overlap with the DEGs (2TL vs FL) without considering batch effects for both major cell types and neuronal sub-types. In panel b, d, we show that the majority of DEGs with batch correction (2TL vs FL) overlap with the individual DEGs found in each TL vs FL comparison. In panel e, we show that the transcriptomic profiles of 2 TL exhibit higher similarity compared with the sample from FL. Overall, based on these analyses we concluded that our results are robust to batch correction.

   In addition, we highlight that, differently from other tissues, it is very difficult to obtain the “fresh” human samples of brain cortex, which most likely provides different transcriptome information than the more commonly used post-mortem brain samples. These analyses offered another evidence supporting the differences between TL and FL, which complement (and align with) the comparative analyses using the data from Allen Brain Atlas (see Figure S9, original results).

   Figure S15. Comparison of biological (gene expression) differences in each major cell type and neuron-subtype between the 2 regions with and without batch effect removal. ____a, c. Comparison of the DEGs (2 TL vs 1FL) with and without removing batches (a, up-regulated in TL; c, up-regulated in FL). b, d. Comparison of the DEGs (2TL vs FL) following the removal of batch effects with the DEGs calculated by individual TL vs FL samples. __e. __Expression correlation between each sample (without batch correction for lane), showing higher transcriptional similarity within the same tissue type than across tissues, consistently in major cell-types and neuronal subtypes.

   3.Differential gene expression analyses between the FL and TL was undertaken using edgeR. It is unclear if this was performed on aggregated counts or not - i.e., sum of counts per gene per cell type. If it was, then with such a small sample size (1 frontal lobe and 2 temporal lobe samples), it is unclear how well edgeR will perform. Similarly, if the DE analysis was performed using individual gene per cell counts, then there is a type 2 error risk due to pseudoreplication. It is reassuring that the primary results were replicated in a second dataset. Moreover, the downstream analyses (functional enrichment analysis, heritability enrichment analysis etc) are designed to cope with noisy data so I'm happy with the broad conclusions.

   We acknowledge the reviewer’s point, and here we specify that edgeR performs differential expression analysis at the level of individual genes across individual cells, and we performed DE analysis for each cell type. We and others consider edgeR a robust tool for analyzing RNA-Seq data; edgeR has been extensively benchmarked alongside other widely used statistical methods, e.g., edgeR-LRT and edgeR-QLF which showed high performance1. Another study about different tools for differential expression in single cell data demonstrated that edgeR (and others) has usually higher precision, larger than 0.9, yielding lower false positive2. Therefore, based on previous formal assessments showing the robustness of edgeR, we select this approach for DE analysis.

   Moreover, it has been previously documented that edgeR can be used also to analyses small samples due to several inherent features. First, edgeR uses an empirical Bayes framework to estimate dispersion, which is a measure of the biological variability in gene expression. This approach uses information across genes, helping to stabilize the variance estimates even when sample sizes are small. This makes edgeR more robust in cases with a limited number of replicates. Second, edgeR accounts for overdispersion, which can effectively handle small sample sizes and provide more accurate statistical tests. In the revised manuscript, we now discuss the advantages of edgeR in Methods, in particular for edgeR performance on small sample size in single cell RNA seq.

   It is unclear if this was performed on aggregated counts or not - i.e., sum of counts per gene per cell type

   We specify that edgeR performs differential expression analysis at the level of individual genes across individual cells, and we performed DE analysis for each cell type. This is now indicated in Methods.

   __*4.To calculate the enrichment of "genetic risk" associated with psychiatric disorders, the authors used a hypergeometric test for the overlap between cell type specific genes and the GWAS variant-mapped genes for each disease, which is widely used to evaluate the enrichment of genetic risk genes. To identified GWAS variant mapped genes the authors used a GWAS SNP threshold of To test the sensitivity of the enrichment analysis, we selected the GWAS genes with each threshold respectively: 10-6, 10-7, 5x10-8. The new results are largely consistent with those obtained using a P-value of 10-5. Susceptibility genes for neuropsychiatric disorders are enriched for expression in neuronal cell types for each P-value. With respect to neuronal subtypes, we found stronger enrichment in INH than in EX sub-clusters, with INH PVALB, SST and EX L5 being the neuronal sub-clusters mostly enriched for expression of GWAS genes (new Figure S14).

   Figure S14 Cell type for expression of neuropsychiatric disorder associated GWAS genes with each threshold respectively: 10-6, 10-7, 5x10-8. a-c. adjusted P-value of enrichment in each 7 major cell type. d-f. adjusted P-value of enrichment in each neuron subtype.

   Moreover, the Reviewer suggests using an alternative tool, FUMA, which requires the whole set of SNP GWAS associations. While these can be available for single diseases and GWAS data (assuming the authors made all data available, and assuming one obtains approval by the consortia managing the GWAS data), unfortunately these SNPs data are not available for several diseases in the NHGRI-EBI GWAS catalog, which provides only SNPs with a max P=10-5. Since in our study we wanted to consider GWAS data from 7 neuropsychiatric diseases, we pragmatically opted for obtaining data from NHGRI-EBI GWAS catalog rather than seeking GWAS SNP data from individual studies.

   We also acknowledge the limitations for the variant to gene mapping (revised Discussion, page 17, line 17), and we also highlight that several other studies rely on the variant to gene mapping from NHGRI-EBI GWAS catalog for enrichment analyses3-5. There are also studies that investigate the enrichment of mapped genes (from NHGRI-EBI GWAS catalog) in different cell types using the hypergeometric test 6-7, as we do in our study. Therefore, the methods used in our manuscript are consistent with approaches adopted in previously published studies. Perhaps more importantly, in the revised manuscript, we replicated the main GWAS enticement results (e.g., in INH neurons and in PVLAB from the temporal lobe) in the Brain Allen Atlas datasets, which shows that, despite these limitations of variant to gene mapping, our main enrichment results are replicable. We discussed these limitations in our paper (see Discussion, page 17, line 6).

   However, where individual genes are mentioned then the authors may wish to confirm the results from edgeR for a few selected genes with a second technique such as qPCR. For example, GRIN2A and SLC12A5.

   To address this point, first, we check the expression of the 2 genes using the data from Allen Brain Atlas data, which show significantly high expression in TL (new Figure S12. b, and below). In addition, we carried out new qPCR analysis, and found the mRNA expression levels of GRIN2A and SLC2A5 in patients with traumatic brain injury in the temporal lobe region were higher than those in patients with frontal lobe injury (new Figure S12. c).

   Figure S12. b. Expression level of GRIN2A and SLC12A5 in 2 regions using Brain Allen Atlas. ***P-value-ΔΔCt method. Significance was determined through T-test (two-tailed). qPCR for each TL or FL sample was repeated 3 times.

   * *

   Reviewer #2

   We thank this Reviewer for the time spent evaluating our manuscript. In the revised manuscript we have now included several new analyses and data that allowed us to replicate and strengthen our main findings, and especially we considered the psychoactive drug target genes using the whole psychoactive drugs DB. We believe these new data helped us to refine the message and overall improve reproducibility of the main findings presented. We have carefully addressed each point raised by this and other reviewers, by providing revisions and explanations, and adding new data to our manuscript, as follows:

   New analyses/data added:

   1. *Effect of batch due to different lanes - comparison of DEGs (TL/FL) obtained when samples in different lanes are tested individually (new Figure S15). *
   2. Effect of batch correction on our results - comparison of the DEGs (TL/FL) obtained with and without batch removal (new Figure S15).
   3. Sensitivity of our enrichment results for GWAS significance – we performed the enrichment of GWAS genes using different GWAS thresholds, 10-6, 10-7, 5x10-8 (new Figure S14).
   4. Expression analysis of GRIN2A and SLC12A5 in Allen Brain Atlas data and qPCR results of GRIN2A and SLC12A5 in patients with frontal and temporal lobe traumatic injury (new Figure S12, Table S3).
   5. Comparison of the DEGs (TL/FL) with DEGs (autism/Ctrl) obtained from single cell RNA seq (new Figure S16, Table S7).
   6. Comparison of the results using the GWAS genes derived from Trubetskoy et al. with our gene lists (new Figure S17).
   7. Description of the data quality (Figure S2) 1.The manuscript is unfortunately lacking (supplemental) figures showing the preprocessing, batch effect correction, and cell type annotation of single nucleus RNAseq data. Although this part is described in the methods in detail, it is hard to judge if these parts were done properly if data is not shown in any of the figures. Regarding the batch effect correction, it reads as if the batch effects have been removed for both brain regions separately. This potentially introduces a bias between brain regions that hugely questions the later performed analysis of differential expression analysis in FL vs TL. In any case, this analysis is not convincing since it has been performed on n=3 vs. n=3 samples and is thus tremendously underpowered.

   We thank the reviewer for the suggestions. First, we added the cell type annotation process for the major cell type by showing the expression of known markers in Figure S2. f. To show the validity of our cell classification, we calculated the significance of overlap with major cell type markers derived from known study in __Figure S2. e. We also provide the distribution of nUMI, nGenes, percentage of mitochondrial genes after quality control in Figure S2. b __to show the large number of cells contributing to the overall quality and depth of the scRNA-seq dataset despite the small number of individual samples.

   Figure S2. Description of snRNA-seq data. b. Distribution of nUMI, nGenes, percentage of mitochondrial genes after QC. e. Significance of overlap with major cell type markers derived from known study. f. Expression of known markers for each cell type.

   Since the samples are sequenced by different lanes of 10X platform, therefore, we can’t exclude potential batch effects. To account for this potential batch effect, we corrected the batch by doing CCA (canonical correlation analysis) which enhanced the clustering and the UMAP visualization more biologically meaningful and less driven by batch-specific variations.

   Moreover, in an attempt to account for the sample size limitation, we employed 3 approaches to confirm the main transcriptional differences between the 2 regions, and that these are “robust” to batch correction, as is shown in new Figure S15 (see next page): (1) Comparison of the gene expression differences (2 TL vs 1FL) with and without removing batches (new Figure S15. a, c); (2) The results obtained by comparing the differences between each individual TL sample (processed in different lanes) and FL sample are contrasted with the results after batch removal (new Figure S15. b, d); (3) To confirm a limited effect of lane, analysis of the expression similarity of three samples demonstrates, consistently for each major cell type and neuronal sub-types, a strong correlation between the two TL samples (form different lanes) as compared with FL (new Figure S15. e).

   As shown in panel a, c, below, the majority of DEGs (2TL vs FL) identified with batch effects largely overlap with the DEGs (2TL vs FL) without considering batch effects for both major cell types and neuronal sub-types. In panel b, d, we show that the majority of DEGs with batch correction (2TL vs FL) overlap with the individual DEGs found in each TL vs FL comparison. In panel e, we identified that the transcriptomic of 2 TL exhibit higher similarity compared with the sample from FL.

   Overall, based on these analyses we concluded that the results are robust to batch correction.

   Figure S15. Comparison of biological (gene expression) differences in each major cell type and neuron-subtype between the 2 regions with and without batch effect removal. a, c. Comparison of the DEGs (2 TL vs 1FL) with and without removing batches (a, up-regulated in TL; c, up-regulated in FL). b, d. Comparison of the DEGs (2TL vs FL) following the removal of batch effects with the DEGs calculated by individual TL vs FL samples. __e. __Expression correlation between each sample (without batch correction for lane), showing higher transcriptional similarity within the same tissue type than across tissues, consistently in major cell-types and neuronal subtypes.

   In addition, we highlight that, differently from other tissues, it is very difficult to obtain the “fresh” human samples of brain cortex, which most likely provides different transcriptome information than the more commonly used post-mortem brain samples. These analyses offered another evidence supporting the differences between TL and FL, which complement (and align with) the comparative analyses using the data from Allen Brain Atlas (Figure S9, original results).

   *2.Furthermore, the way that the authors treat GWAS data for disease does not seem to follow best practices. For schizophrenia, last year the largest GWAS so far was published (Trubetskoy et al, Nature, 2022) with very careful prioritization of genes. The authors should re-analyze their data using the gene list from this paper (and similar from other disorders) rather than the gene list that they came up with using their approach. The approach to select genes from different GWAS introduced seems highly arbitrary and leaves the reader unsure about statistical rigor. *

   We have carefully considered the suggestion regarding the treatment of GWAS data, particularly with respect to the gene list derived from the recent schizophrenia GWAS by Trubetskoy et al. (Nature, 2022). In this paper, the author mainly identified 120 genes (106 protein-coding) that are likely to underpin associations with schizophrenia which implicate fundamental processes related to neuronal function including synaptic organization, differentiation and transmission.

   With respect to our study, first, we found there is significant overlap between prioritized genes in Trubetskoy et al’ study and GWAS genes included in our study. We showed the P value for overlap significance below, and listed the 27 genes. Among the prioritized genes, GRIN2A is also identified to be important in neuropsychiatric disorder, which is also confirmed to differ between the 2 regions and dysregulated in disease brain.

   Enrichment of genes obtained from the prioritized schizophrenia-associated genes in Trubetskoy et al. Significant overlap (P=0.013, hypergeometric test) between schizophrenia-associated genes (120 prioritized genes from Trubetskoy et al.) and our GWAS genes (from GWAS catalogue).

   Second, we conducted a supplementary analysis focused on the 120 genes prioritized by Trubetskoy et al, as shown below. We found the 120 prioritized genes in this paper are significantly enriched in excitatory and inhibitory neurons (panel b, below), aligning with our main findings conducted by schizophrenia related genes in our previous GWAS gene lists. Within the neuronal subcluster, we found a significant enrichment in L4, LAMP5 and PVALB cells (panel c); L4 and PVALB are largely consistent with our previous results (shown in Figure 3. c). Furthermore, we also found the 120 schizophrenia-associated genes are highly significantly enriched in DEGs (TL/FL) in VIP and PVALB subtypes (panel d).

   b-c. Enrichment of 120 prioritized schizophrenia-associated genes in major cell types and neuronal subtypes. d. For each cell type, the enrichment of 120 genes is calculated with respect to the set of DEGs (TL/FL). Approach used for enrichment analysis is hypergeometric test (significance level, P-valueThese results suggest that while new gene lists from larger GWAS studies (e.g., Trubetskoy et al) come up regularly, the lists of GWAS genes prioritized in our enrichment analysis has some overlap with the newest GWAS. We agree that including more (larger) GWAS studies will strengthen the manuscript, but based on the analyses above, we believe our GWAS enrichment results are robust. In the revised manuscript, the new analysis including the detailed comparison with schizophrenia GWAS by Trubetskoy et al. (Nature, 2022) are reported in new Figure S17.

   To improve on the GWAS enrichment analysis, we carried out additional sensitivity analyses to support our GWAS enticement results. We selected additional thresholds to evaluate the robustness of our results to the choice of gene lists to test the sensitivity of the enrichment analysis, we selected the thresholds: 10-6, 10-7, 5x10-8. The new results are largely consistent with those obtained using P-value of 10-5. Susceptibility genes for neuropsychiatric disorders are enriched for expression in neuronal cell types for each P-value. With respect to neuronal subtypes, we found stronger enrichment in INH than in EX sub-clusters, with INH PVALB, SST and EX L5 being the neuronal sub-clusters mostly enriched for expression of GWAS genes. These results are reported in new Figure S14.

   Figure S14. Enrichment of cell type expression of neuropsychiatric disorder-associated GWAS genes for different GWAS-thresholds. a-c. Adjusted P-value of enrichment in each 7 major cell type. d-f. Adjusted P-value of enrichment in each neuron subtype.

   3.Similarly, the choice of data set for disease-related differentially expressed genes is unclear as much larger (two orders of magnitude) published data sets exist for many of the disorders. For three of those DEG analyses performed on bulk RNAseq data, for the remaining two the DEG list of papers is used directly -making a comparison complicated. One would have to run DEG analysis in a standardized way for all 5 datasets/ disorders. It would be good to also indicate the respective sample size in Fig. 5a. (On a different note, the OCD publication is Piantadosi et al. 2021, not Sean C.et.al..) In addition, the authors matched brain regions to their regions of interest (frontal and temporal lobe) as shown in Fig. 5a. Still, they vary across disorders, which makes it hard to compare their findings across disorders and does not allow for a general statement about frontal vs. temporal lobe. ____To generalize for any of those psychiatric disorders I would recommend including more RNA-seq studies of the same disorder. Nowadays there are getting more and more case-control single nuclei studies on such disorders published. The authors could also include those by transforming them to pseudo bulk datasets and running their DEG analysis with edgeR as documented.

   We acknowledge there might be a bias introduced by using the DEGs from the original paper directly. In addition, there is a general limitation affecting all bulk-RNA studies in complex tissues with different anatomical structures (e.g., kidney, brain, etc.), which form a great part of the publicly available data sources. In brain research, it is also more difficult to collect fresh human brain samples from patients with psychiatric disorders, which poses additional tissue availability constraints. Despite these limitations, we argue that bulk-RNA studies in anatomically complex tissues, and the DEGs reported therein, can be useful for GWAS enrichment analysis and not all DEGs are due to spurious or artificial signals. Furthermore, due to the lower sequencing depth inherent in single-cell RNA sequencing compared to bulk RNA sequencing, we set up to contrast our findings with results found by bulk-RNA seq.

   We agree with the Reviewer that “*One would have to run DEG analysis in a standardized way for all 5 datasets/ disorders”, *however this approach assumes that the raw data are directly available and/or that the authors are keen to share the raw data. Both these assumptions are – unfortunately – not valid in many cases. (In several instances, we did contact authors to have access to raw data, with no success). Furthermore, when a commonly shared gene set in the DE genes is identified when using “heterogenous DE gene lists”, this might suggest a strongest convergence, or a convergence that is “robust” despite the differences between the heterogeneous DE lists (from authors or newly generated by us). Therefore, despite the limitations, our approach was motivated by practical considerations.

   In addition, the brain region differences can be more prevalent and have a larger impact for specific psychiatric disorders. In our manuscript, for MDD we specially looked at only the BA8/9 which come from dorsolateral prefrontal cortex. Regarding OCD, BP, and MDD, several studies showed that there are no significant functional differences clinically observed between the orbitofrontal cortex and dorsolateral prefrontal cortex (Schoenbaum G, Setlow B. Integrating orbitofrontal cortex into prefrontal theory: common processing themes across species and subdivisions. Learning & Memory, 20018. Golkar A, Lonsdorf T B, Olsson A, et al. Distinct contributions of the dorsolateral prefrontal and orbitofrontal cortex during emotion regulation. PloS one, 20129). In the case of ASD, Brodmann area 41, 42, 22 refers to a subdivision of the cytoarchitecturally defined temporal region of cerebral cortex, exhibiting similar functionality to the temporal gyrus. Therefore, ASD and SCHI may arise from specific regions within the temporal lobe, while OCD, MDD, and BP may be associated with regions within the frontal lobe.

   To address the Reviewer’s point more directly - we carried out additional analyses to investigate the effect of this factor on our main results. One of our aims was to understand how regional gene expression differences (TL/FL) in PVALB neurons are associated with gene dysregulation in the brain of neuropsychiatric disease patients. We have now extended these analyses to a separate dataset, and tested whether the dysregulated genes in neuropsychiatric disease are expressed mainly in TL and FL using single cell data from Brain Allen Atlas (4 patients, each with 6 brain regions profiled). The new results are shown in new Figure S11 b-f (and reported in the next page).

   Briefly, we found that the percentage of dysregulated genes in SCHI, BP, OCD, and MDD that are expressed in MTG (SCHI: 75%, BP: 81%, OCD: 68%, MDD: 71%) and CgG (SCHI: 77%, BP: 80%, OCD: 60%, MDD: 77%) is higher compared with those in all other regions included in Brain Allen Atlas dataset. The percentage of ASD dysregulated genes expressed in the 6 regions from Brain Allen Atlas are quite similar. This analysis suggests that, despite the potential impact of heterogeneity of regions, the DEGs in psychiatric conditions are typically expressed at higher level in MTG (TL) and CgG (FL) compared with other regions, therefore highlighting the potential role of these two regions in psychiatric conditions. Therefore, we believe that despite the heterogeneity of regions included in the published RNA-seq studies, the strongest signal of enrichment for DEGs is detected consistently in TL and FL, i.e., in the 2 brain regions where the DEGs are also most highly expressed compared with other regions. These new data, reported in a new Figure S11 of the revised manuscript, provide additional evidence to support our main conclusions.

   Due to the difficulties obtaining the human sample of psychiatric disorders causing limited public data resource, we found one study about molecular changes of ASD revealed by single cell RNA seq coming from Velmeshev et al. Science. 2019; 364(6441):685-689 (PMID: 31097668), including 22 ASD samples and 19 control samples. We compared the DEGs (TL/FL) with the DEGs (ASD/Ctrl), and report the results in new Figure S16. Briefly, the results show that except LAMP5, Endo, and L4, ASD-associated dysregulated genes significantly overlap with DEGs between FL and TL in several cell types, especially in VIP and astrocytes. While PVALB is not the most apparent cluster reflecting regional differences contributing to ASD, we found a moderate association (R2 =0.11, P=0.04) between changes in TL/FL and those in ASD/Ctrl brain. These findings suggest that gene expression differences between the 2 regions may contribute to ASD disorder, providing additional evidence to support our main conclusions.

   Figure S16. Overlap of genes dysregulated in ASD and genes differentially expressed between TL and FL in each major cell type and neural subtype. Venn diagram plots in a-m showing the number of overlapped genes. Dot plot in each panel shows the relationship between the log2FC(TL/FL) [our study] and log2FC(ASD/Ctrl) [Velmeshev et al. Science. 2019 study]. Significance of the overlap: *0.001-0.01, **0.0001-0.001, ***0.00001-0.0001, ****4.For cell type enrichment of disease signal based on GWAS signal several carefully controlled studies exist using more sophisticated statistical methods (Skene et al., Nature Genetics, 2018, Bryois et al., et al. Nature Genetics 2020, MJ Zhang et al Nature Genetics 2022 to mention a few). I applaud that the authors aim to go beyond this basic characterization but I think it is worrisome that by using less sophisticated (and importantly less controlled) statistical and genetic approaches they reach a different signal -and then they go on and analyze this signal. It is potentially interesting they reach a different conclusion, but they need to provide a careful statistical analysis to explain how the chosen method is superior or at least different to previous efforts.

   The Reviewer suggests the use of alternative approaches to link GWAS variants to genes, like MAGMA, LDSC, FUMA to improve the gene mapping from GWAS signals, and are better than the gene mapping based on proximity alone. While these approaches can provide some advantages, most of these methods do require the whole set of SNP GWAS associations, including non-significant associations. While these can be available for single diseases and specific GWAS data (assuming the authors made all data available, and assuming one obtains approval by the consortia managing the GWAS data) these SNPs are not available for several diseases in the NHGRI-EBI GWAS catalog, which provides only SNPs with a max P=10-5. Since in our study we considered GWAS data from 7 neuropsychiatric diseases, we (pragmatically) opted for obtaining data from NHGRI-EBI GWAS catalog rather than seeking GWAS SNP data from individual studies.

   We now acknowledge the limitations for the variant to gene mapping (revised Discussion, page 17, line 17), and we also report that several other studies rely on the variant to gene mapping from NHGRI-EBI GWAS catalog for enrichment analyses4-6. There are also studies that investigate the enrichment of mapped genes (from NHGRI-EBI GWAS catalog) in different cell types using the hypergeometric test 7-8, as we do in our study. Perhaps more importantly, in the revised manuscript, we replicated the main GWAS enticement results (e.g., in INH neurons and in PVLAB from the temporal lobe) in the Brain Allen Atlas datasets, which shows that, despite these limitations of variant to gene mapping, our main enrichment results are replicable.

   (Other comments)

   - only n=3, ~45 000 cells making it hard to generalize

   - no supplementary figures for the methods (i.e. preprocessing, cell type annotations), thus hard to judge if done properly if they do not show any data - much higher level of transparency needed

   - The methods part is not clear, in general, it is only descriptive, with no equations

   - Unconvincing determination of DEGs for each disorder

   - DEGs and pathways based on n1=1 vs n2=2 feals handwavy

   - DEG analysis and cell type annotation are mixed up and it is unclear how DEGs were determined

   While we acknowledge the limitation of sample size in our study, we also emphasize again the challenges in of availability of fresh human sample, which provide more transcriptomic information than postern sample. Despite the small number of individual samples, the large number of cells (~45,000) contributes to the overall quality and depth of the scRNA-seq dataset. Hence, our study provides a foundational perspective on the gene expression between the frontal lobe (FL) and temporal lobe (TL), and valuable data source for further investigations.

   With respect to the additional description of the data processing and cell annotation process, in the revised manuscript we now elucidate the cell type annotation process by showing the expression of some known markers in new Figure S2. f, the significance of overlap with major cell type markers derived from known study in new Figure S2. e, the distribution of nUMI, nGenes, percentage of mitochondrial genes after quality control in new __Figure S2. b. __

   To strengthen the differential gene expression analysis, we replicated our main findings through SMART RNA-seq from Brain Allen Atlas including the DEGs identified in our study (Figure S9).

   More technical details are provided in the revised manuscript, as detailed below:

   In the revised Methods section – (1) Differential expression analysis in FL vs TL and pathway enrichment analysis, we added more details about how the DEGs are identified and how this is robust to batch correction. (2) Replication analyses in human Brain Allen Atlas, we provide more details about how we replicated the DEGs using Allen Brain Atlas dataset. (3) Enrichment of neuropsychiatric disease GWAS genes in brain cell clusters, we now added more methodological details about the enrichment analysis.

   __ __

   Reviewer #3

   We thank the Reviewer for his/her overall positive comments. In the revised manuscript we have now included several new analyses requested by this and other reviewers (see list below), which allowed us to replicate and strengthen our main findings. We also add details of the method used in this paper. We believe these new analyses and data helped us to improve reproducibility and strengthen the main findings presented in our manuscript.

   New analyses/data added:

   1. *Effect of batch due to different lanes - comparison of DEGs (TL/FL) obtained when samples in different lanes are tested individually (new Figure S15). *
   2. Effect of batch correction on our results - comparison of the DEGs (TL/FL) obtained with and without batch removal (new Figure S15).
   3. Sensitivity of our enrichment results for GWAS significance – we performed the enrichment of GWAS genes using different GWAS thresholds, 10-6, 10-7, 5x10-8 (new Figure S14).
   4. Expression analysis of GRIN2A and SLC12A5 in Allen Brain Atlas data and qPCR results of GRIN2A and SLC12A5 in patients with frontal and temporal lobe traumatic injury (new Figure S12, Table S3).
   5. Comparison of the DEGs (TL/FL) with DEGs (autism/Ctrl) obtained from single cell RNA seq (new Figure S16, Table S7).
   6. Comparison of the results using the GWAS genes derived from Trubetskoy et al. with our gene lists (new Figure S17).
   7. Description of the data quality (Figure S2) 1.The authors integrated the brain snRNA-seq data with GWAS data to annotate the cell type specific expression, which is one of the key points for this analysis, however a more detailed description of the method is lacking.

   We have made changes to the text to improve and clarify this aspect. In the revised Methods section, we now specify: “To calculate the enrichment of genetic risk associated with psychiatric disorders, we used a hypergeometric test for the overlap between cell type specific genes (DEGs between one cell with other cell types, log2FC>0.5, adjusted.P __2.The authors found a set of genes which is associated with psychiatric disorders and specific cell types, for example inhibitory neurons are the most vulnerable cell type to genetic susceptibility through their analysis. The correlation of each cell type and each psychiatric disorders can be discussed.*__

   We thank the Reviewer for this suggestion; we have now added more details discussing the relationship between other cell types with psychiatric disorders other than PVALB-neuron in this part – see Discussion in the revised manuscript, where we added: “Astrocyte, OPC are also associated with psychiatric disorders, and play essential roles in maintaining brain homeostasis, regulating synaptic transmission, and supporting neuronal function. Astrocytes also contribute to maintaining the integrity of the blood-brain barrier (BBB) and interact closely with neurons. Disruptions in this communication impact neural circuitry, which is relevant to many psychiatric disorders. OPCs generate oligodendrocytes, producing myelin crucial for signal conduction and brain structural integrity, which potentially impacts brain connectivity and communication between brain regions.* Among neuronal subtypes, our data suggest that disruption of specific biological process in PVALB, SST and L5 neurons may contribute to neuropsychiatric disorders. PVALB cells are believed to activate pyramidal neurons only if the signal from excitatory neurons is sufficient and optimize the signaling in both EX and INH72. SST neurons gate excitatory input onto pyramidal neurons within cortical microcircuits, mainly coming from L5 layer of excitatory neuron** which is involve in motor control, decision-making, and information transfer between the cortex and subcortical structures73. These signaling processes, when dysregulated, have been implicated in psychiatric diseases74.** The relationship between psychiatric disorders and other layers of the cerebral cortex is still under investigation.* *L2-3 neurons handle local processing, relevant to conditions like schizophrenia and autism. L6 neurons in thalamocortical circuits are crucial for sensory processing and information relay, involving sensory perception abnormalities.” *

   3.The authors have found a group of interesting genes, such as GRIN2A, DGKI, and SHISA9 and confirmed them with the Allen Brain Atlases. Experimental validation would be helpful to confirm such findings.

   In our manuscript, we emphasized that GRIN2A and SLC12A5 (both implicated in schizophrenia and bipolar disorder) were significantly upregulated in TL PVALB neurons and in psychiatric disease patients’ brain. To address this point, first, we check the expression of the 2 genes using the data from Allen Brain Atlas data, which showed significantly high expression in TL (new Figure S12. b). By means of new qPCR analysis in primary TL/FL samples, we found the mRNA expression levels of GRIN2A and SLC2A5 in patients with traumatic brain injury in the temporal lobe region were higher than those in patients with frontal lobe injury (new Figure S12. c).

   Figure S12. b. Expression level of GRIN2A and SLC12A5 in 2 regions using Brain Allen Atlas. ***P-value-ΔΔCt method. Significance was determined through T-test (two-tailed). qPCR for each TL or FL sample was repeated 3 times.

   Lastly, we want to highlight that since we believe in “Data Democratization” and sharing our data resources, upon publication, we will make all our data (including the single cell in “fresh” (surgically resected) brain tissue samples) and corresponding detailed results available to the scientific community.

   We believe our study (which is focused on psychiatric diseases) will prompt other groups to use our single cell data and to dig deep into the role of temporal and frontal lobes in other neurogenerative diseases.

   __ __

   References

   1. Squair, J.W., Gautier, M., Kathe, C. et al. Confronting false discoveries in single-cell differential expression. Nat Commun 12, 5692 (2021).
   2. Wang, T., Li, B., Nelson, C.E. et al. Comparative analysis of differential gene expression analysis tools for single-cell RNA sequencing data. BMC Bioinformatics 20, 40 (2019).
   3. Bhattacherjee A, Djekidel MN, Chen R, Chen W, Tuesta LM, Zhang Y. Cell type-specific transcriptional programs in mouse prefrontal cortex during adolescence and addiction. Nat Commun. 2019 Sep 13;10(1):4169.
   4. Grubman A, Chew G, Ouyang JF, Sun G, Choo XY, McLean C, Simmons RK, Buckberry S, Vargas-Landin DB, Poppe D, Pflueger J, Lister R, Rackham OJL, Petretto E, Polo JM. A single-cell atlas of entorhinal cortex from individuals with Alzheimer's disease reveals cell-type-specific gene expression regulation. Nat Neurosci. 2019 Dec;22(12):2087-2097
   5. Przytycki, P.F., Pollard, K.S. CellWalker integrates single-cell and bulk data to resolve regulatory elements across cell types in complex tissues. Genome Biol 22, 61 (2021).
   6. Swindell, William R., et al. "RNA-Seq analysis of IL-1B and IL-36 responses in epidermal keratinocytes identifies a shared MyD88-dependent gene signature." Frontiers in immunology 9 (2018): 80.
   7. Geirsdottir, Laufey, Eyal David, Hadas Keren-Shaul, Assaf Weiner, Stefan Cornelius Bohlen, Jana Neuber, Adam Balic et al. "Cross-species single-cell analysis reveals divergence of the primate microglia program." Cell 179, no. 7 (2019): 1609-1622.
   8. Schoenbaum G, Setlow B. Integrating orbitofrontal cortex into prefrontal theory: common processing themes across species and subdivisions[J]. Learning & Memory, 2001, 8(3): 134-147.
   9. Golkar A, Lonsdorf T B, Olsson A, et al. Distinct contributions of the dorsolateral prefrontal and orbitofrontal cortex during emotion regulation[J]. PloS one, 2012, 7(11): e48107

   Read the original source
2. 

   Review Commons

   Dec 5, 2023

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #3

   Evidence, reproducibility and clarity

   In the manuscript "Decoding frontotemporal and cell type-specific vulnerabilities to neuropsychiatric disorders and psychoactive drugs", the authors integrated brain with no history of neuropsychiatric disorder snRNA-seq data with public GWAS data among 7 psychiatric disorders to explore the heterogeneity between temporal lobe (TL) and frontal lobe (FL), the genetic risk factors and potential drug responsible genes. Multiple bioinformatics technics have been used in the manuscript. The authors found critical pathways and key genes that are related to the psychiatric disorders and GWAS genes enriched cells such as PVALB cells, which …

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #3

   Evidence, reproducibility and clarity

   In the manuscript "Decoding frontotemporal and cell type-specific vulnerabilities to neuropsychiatric disorders and psychoactive drugs", the authors integrated brain with no history of neuropsychiatric disorder snRNA-seq data with public GWAS data among 7 psychiatric disorders to explore the heterogeneity between temporal lobe (TL) and frontal lobe (FL), the genetic risk factors and potential drug responsible genes. Multiple bioinformatics technics have been used in the manuscript. The authors found critical pathways and key genes that are related to the psychiatric disorders and GWAS genes enriched cells such as PVALB cells, which can help the understanding in the field. Overall, the manuscript is well written and organized, but there are some issues need to be addressed.

   1. The authors integrated the brain snRNA-seq data with GWAS data to annotate the cell type specific expression, which is one of the key points for this analysis, however a more detailed description of the method is lacking.
   2. The authors found a set of genes which is associated with psychiatric disorders and specific cell types, for example inhibitory neurons are the most vulnerable cell type to genetic susceptibility through their analysis. The correlation of each cell type and each psychiatric disorders can be discussed.
   3. The authors have found a group of interesting genes, such as GRIN2A, DGKI, and SHISA9 and confirmed them with the Allen Brain Atlases. Experimental validation would be helpful to confirm such findings.

   Significance

   Strength: this manuscript is strong in bioinformatics analysis. Limitation: wet-lab validation of some of the findings would be helpful.

   Read the original source
3. 

   Review Commons

   Dec 5, 2023

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #2

   Evidence, reproducibility and clarity

   Paper review: Decoding frontotemporal and cell type-specific vulnerabilities to neuropsychiatric disorders and psychoactive drugs

   In their manuscript with the title Decoding frontotemporal and cell type-specific vulnerabilities to neuropsychiatric disorders and psychoactive drugs, the authors describe their work on integrating snRNAseq data from "fresh" human frontal and temporal lobe of three healthy donors with genetic risk factors of 7 psychiatric disorders, bulk RNAseq data from healthy and disease human cortex/ DEG lists from previous studies for 5of the psychiatric disorders, and gene targets for commonly prescribed …

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #2

   Evidence, reproducibility and clarity

   Paper review: Decoding frontotemporal and cell type-specific vulnerabilities to neuropsychiatric disorders and psychoactive drugs

   In their manuscript with the title Decoding frontotemporal and cell type-specific vulnerabilities to neuropsychiatric disorders and psychoactive drugs, the authors describe their work on integrating snRNAseq data from "fresh" human frontal and temporal lobe of three healthy donors with genetic risk factors of 7 psychiatric disorders, bulk RNAseq data from healthy and disease human cortex/ DEG lists from previous studies for 5of the psychiatric disorders, and gene targets for commonly prescribed psychoactive drugs. The authors claim that PVALB neurons in the temporal lobe are most vulnerable to genetic risk factors and even more to psychoactive drugs for psychiatric diseases and suggest GRIN2A and SLC12A5 as the genes that most contribute to their vulnerability.

   According to my overall impression, the paper has major problems in terms of quality, clarity, and statistical power. I do not recommend publishing this manuscript in its current form.

   The manuscript is unfortunately lacking (supplemental) figures showing the preprocessing, batch effect correction, and cell type annotation of single nucleus RNAseq data. Although this part is described in the methods in detail, it is hard to judge if these parts were done properly if data is not shown in any of the figures. Regarding the batch effect correction, it reads as if the batch effects have been removed for both brain regions separately. This potentially introduces a bias between brain regions that hugely questions the later performed analysis of differential expression analysis in FL vs TL. In any case, this analysis is not convincing since it has been performed on n=3 vs. n=3 samples and is thus tremendously underpowered.

   Furthermore, the way that the authors treat GWAS data for disease does not seem to follow best practices. For schizophrenia, last year the largest GWAS so far was published (Trubetskoy et al, Nature, 2022) with very careful prioritization of genes. The authors should re-analyze their data using the gene list from this paper (and similar from other disorders) rather than the gene list that they came up with using their approach. The approach to select genes from different GWAS introduced seems highly arbitrary and leaves the reader unsure about statistical rigor. Similarly, the choice of data set for disease-related differentially expressed genes is unclear as much larger (two orders of magnitude) published data sets exist for many of the disorders. For three of those DEG analyses performed on bulk RNAseq data, for the remaining two the DEG list of papers is used directly -making a comparison complicated. One would have to run DEG analysis in a standardized way for all 5 datasets/ disorders. It would be good to also indicate the respective sample size in Fig. 5a. (On a different note, the OCD publication is Piantadosi et al. 2021, not Sean C.et.al..) In addition, the authors matched brain regions to their regions of interest (frontal and temporal lobe) as shown in Fig. 5a. Still, they vary across disorders, which makes it hard to compare their findings across disorders and does not allow for a general statement about frontal vs. temporal lobe. To generalize for any of those psychiatric disorders I would recommend including more RNAseq studies of the same disorder. Nowadays there are getting more and more case-control single nuclei studies on such disorders published. The authors could also include those by transforming them to pseudo bulk datasets and running their DEG analysis with edgeR as documented. For cell type enrichment of disease signal based on GWAS signal several carefully controlled studies exist using more sophisticated statistical methods (Skene et al., Nature Genetics, 2018, Bryois et al., et al. Nature Genetics 2020, MJ Zhang et al Nature Genetics 2022 to mention a few). I applaud that the authors aim to go beyond this basic characterization but I think it is worrisome that by using less sophisticated (and importantly less controlled) statistical and genetic approaches they reach a different signal -and then they go on and analyze this signal. It is potentially interesting that they reach a different conclusion, but they need to provide a careful statistical analysis to explain how the chosen method is superior or at least different to previous efforts.

   Plus:

   • Flash-frozen human tissue with little post-mortem delay
   • TL and FL comparison: interesting
   • Multiple comparison corrections
   • Replication analysis included The drug target genes angle is interesting

   Minus:

   • only n=3, ~45 000 cells making it hard to generalize
   • no supplementary figures for the methods (i.e. preprocessing, cell type annotations), thus hard to judge if done properly if they do not show any data - much higher level of transparency needed
   • The methods part is not clear, in general, it is only descriptive, with no equations
   • Unconvincing determination of DEGs for each disorder
   • DEGs and pathways based on n1=1 vs n2=2 feals handwavy
   • DEG analysis and cell type annotation are mixed up and it is unclear how DEGs were determined

   Unclear:

   • Is the background dataset used for enrichment of genetic risk calculation different for each region and cell type? If so? How is this a fair comparison?
   • Which subset of GWAS genes is used for Gene co-expression networks

   Significance

   Mainly weaknesses

   Advancement provided by the study remains modest due to low confidence in the findings. Potentially interesting approach but needs to utilize state-of-the-art methodology and data sets. A typical audience would be journals targeting molecular/biological psychiatry

   Read the original source
4. 

   Review Commons

   Dec 5, 2023

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #1

   Evidence, reproducibility and clarity

   The authors integrated snRNA-seq analysis with genetic risk and drug-specific signatures to investigate brain regional differences in risk for neuropsychiatric disease and drug response. To replicate the main findings, the authors also analyzed single nuclei data from the Brain Allen Atlas. Overall, the manuscript is very well written. The methods are comprehensive, clear and well-wrtten which is very welcome. The authors have undertaken a large number of bioinformatic investigations using appropriate methodology and careful design.
   The main limitation of the work is the small starting sample size. The authors studied 1 frontal lobe …

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #1

   Evidence, reproducibility and clarity

   The authors integrated snRNA-seq analysis with genetic risk and drug-specific signatures to investigate brain regional differences in risk for neuropsychiatric disease and drug response. To replicate the main findings, the authors also analyzed single nuclei data from the Brain Allen Atlas. Overall, the manuscript is very well written. The methods are comprehensive, clear and well-wrtten which is very welcome. The authors have undertaken a large number of bioinformatic investigations using appropriate methodology and careful design.
   The main limitation of the work is the small starting sample size. The authors studied 1 frontal lobe sample and 2 temporal lobe samples. Atlhough this information was in Table S3 it would be good to include upfront in the Methods. snRNA-seq was generated on the 10x platform. It would be helpful to know if the 10x step and sequencing was performed as one batch, or as individual batches. Similarly, were the sample libraries all sequenced on the same lane, or different lanes. The authors do not state in the Methods how many nuclei they were targeting and this should be included. Sample pre-processing was well described and standard. In relation to Batch correction - as with any batch correction method, it is unclear whether the correction is adjusting for biological differences or technical. Since this is a study of the differences between FL and TL, would it not be more appropriate not to correct for batch, particularly as the samples were analysed individually - particularly if batch effects were carefully controlled for in the initial study design. The authors should test whether the results are robust to batch correction or not. Differential gene expression analyses between the FL and TL was undertaken using edgeR. It is unclear if this was performed on aggregated counts or not - i.e., sum of counts per gene per cell type. If it was, then with such a small sample size (1 frontal lobe and 2 temporal lobe samples), it is unclear how well edgeR will perform. Similarly, if the DE analysis was performed using individual gene per cell counts, then there is a type 2 error risk due to pseudoreplication. It is reassuring that the primary results were replicated in a second dataset. Moreover, the downstream analyses (functional enrichment analysis, heritability enrichment analysis etc) are designed to cope with noisy data so I'm happy with the broad conclusions. However, where individual genes are mentioned then the authors may wish to confirm the results from edgeR for a few selected genes with a second technique such as qPCR. For example, GRIN2A and SLC12A5. To calculate the enrichment of "genetic risk" associated with psychiatric disorders, the authors used a hypergeometric test for the overlap between cell type specific genes and the GWAS variant-mapped genes for each disease, which is widely used to evaluate the enrichment of genetic risk genes. To identified GWAS variant mapped genes the authors used a GWAS SNP threshold of <10-5, and mapped SNPs to genes using the GWAS DB. The background set of genes is appropriate as is the statistical method. Given the small sample size however, I think it would be helpful to see a sensitivity analysis of the results that (a) uses different GWAS thresholds e.g., 10-6, 10-7, 5x10-8 and (b) uses an alternative SNP to gene mapping tool such as FUMA. Overall, whilst well written, the manuscript as a whole feels overly long. I think it could be improved by a more stringent focus on the most important biological and translational findings.

   Significance

   Overall, the manuscript is very well written. The methods are comprehensive, clear and well-wrtten which is very welcome. The authors have undertaken a large number of bioinformatic investigations using appropriate methodology and careful design.

   Read the original source
5. 

   Version published to 10.1101/2022.11.30.518630v1 on bioRxiv

   Dec 1, 2022