Abstract
Frontotemporal dementia (FTD) is a heterogeneous neurodegenerative disorder characterized by neuronal loss in the frontal and temporal lobes. Despite progress in understanding which genes are associated with the aetiology of FTD (C9orf72, GRN and MAPT), the biological basis of how mutations in these genes lead to cell loss in specific cortical regions remains unclear. In this work we combined gene expression data for 16,912 genes from the Allen Institute for Brain Science atlas with brain maps of gray matter atrophy in symptomatic C9orf72, GRN and MAPT carriers obtained from the Genetic FTD Initiative study. A set of 405 and 250 genes showed significant positive and negative correlation, respectively, with atrophy patterns in all three maps. The gene set with increased expression in spared cortical regions, i.e., signaling regional resilience to atrophy, is enriched for neuronal genes, while the gene set with increased expression in atrophied regions, i.e., signaling regional vulnerability, is enriched for astrocyte genes. Notably, these results extend earlier findings from proteomic analyses in the same cortical regions of interest comparing healthy controls and patients with FTD. Thus, our analysis indicates that cortical regions showing the most severe atrophy in genetic FTD are those with the highest astrocyte density in healthy subjects. Therefore, astrocytes may play a more active role in the onset of neurodegeneration in FTD than previously assumed, e.g., through emergence of neurotoxic (A1) astrocytes.
Abbreviated summary Altmann et al. investigated the concordance between spatial cortical gene expression in healthy subjects and atrophy patterns in genetic frontotemporal dementia. They found that gene expression of astrocyte-related genes was higher in regions with atrophy. Thus, suggesting a more active role of astrocytes in the onset of neurodegeneration.
Introduction
Frontotemporal dementia (FTD) is a heterogeneous neurodegenerative disorder characterized by neuronal loss in the frontal and temporal lobes, with clinical symptoms including behavioral, language and motor deficits (Seelaar et al., 2011). Around 30% of FTD is familial, most commonly caused by autosomal dominant genetic mutations in one of three genes: progranulin (GRN), microtubule-associated protein tau (MAPT) or chromosome 9 open reading frame 72 (C9orf72) (Rohrer et al., 2009). Despite progress in understanding the pathophysiological basis of genetic FTD, the biological basis of how mutations in these genes leads to cell loss in specific cortical regions and subsequently to specific clinical phenotypes is unclear.
An alternative approach to elucidate the molecular biology of autosomal dominant FTD is to study the gene expression profiles of brain regions which are atrophic in symptomatic mutation carriers. This approach has been enabled by publicly available data from the Allen Institute for Brain Science (AIBS) which features post-mortem high-resolution brain-wide gene expression data from cognitively normal individuals (Hawrylycz et al., 2012). In recent years the AIBS atlas has been successfully integrated with brain maps obtained from case-control studies. For instance, one study investigated the link between gene expression and both regional patterns of atrophy and amyloid deposition, finding a positive correlation of APP gene expression and amyloid (Grothe et al., 2018). A transcriptional analysis of cortical regions vulnerable to cortical thinning in the common epilepsies implicated microglia (Altmann et al., 2018). In the broader FTD context, Rittman et al. (2016) studied the expression of MAPT in the context of Parkinson’s disease and progressive supranuclear palsy.
In this work we combine gene expression data from the AIBS atlas with brain maps of gray matter atrophy in symptomatic C9orf72, GRN and MAPT mutation carriers from the Genetic FTD Initiative (GENFI) compared with non-carriers (Cash et al., 2017). The aim of this study was to investigate the molecular basis of the atrophy pattern in mutation carriers. We firstly investigated the spatial overlap between gray matter atrophy in each of the three genetic groups and the gene expression of the corresponding gene. Secondly, we aimed to identify which genes showed a high spatial correspondence between their expression throughout the brain and the atrophy pattern in each genetic group. We hypothesized that these genes or groups of genes may implicate molecular processes or brain cell types that explain why these regions are particularly vulnerable in FTD.
Methods
Allen Human Brain Atlas Data
We used the brain-wide microarray gene expression data generated by the Allen Institute for Brain Science (AIBS) (downloaded from http://human.brain-map.org/) (Hawrylycz et al., 2012). The dataset consists of a total of 3,702 microarray samples from six donors (one female). Each sample comprised 58,692 gene probes and provides coordinates in MNI152 space. The gene expression data have been normalized and corrected for batch effects by the Allen Institute (‘TECHNICAL WHITE PAPER: MICROARRAY DATA NORMALIZATION’, n.d.). We first restricted the set of samples to cortical regions based on the provided slab type (‘cortex’), retaining only samples with a maximal distance of 2mm to a cortical region of interest (ROI) obtained by a parcellation (Cardoso et al., 2015) of the study template used in Cash et al. (2017) leaving 1,654 microarray samples in total. Next, as previously described (Richiardi et al., 2015) we reannotated all microarray probe sequences with gene names using Re-Annotator (Arloth et al., 2015). We excluded probes that sampled more than one gene (N=6,434), were mapped to intergenic regions (N=91) or could not be mapped to any genomic region (N=1,569), leaving 50,598 probes covering 19,980 unique genes. Furthermore, we removed probes that were marked as expressed in less than 300 of the 1,654 cortical samples (N=12,082). Thus, the analysis was carried out using 37,031 microarray probes covering 16,912 distinct genes. Further, since the majority of the samples were obtained from the left hemisphere, and the high correlation between right and left hemisphere gene expression (Hawrylycz et al., 2012), we attributed all right hemisphere samples to the left hemisphere by mirroring the MNI coordinate at the x-axis.
Image data preparation
In order to quantify the amount of atrophy in carriers of FTD mutations, we used results from a voxel-based morphometry (VBM) analysis of the GENFI dataset (Cash et al., 2017). In particular, in this analysis we used the maps showing the voxel-wise t-statistic (t-maps) comparing symptomatic mutation carriers (MAPT: N=10; GRN: N=12; C9orf72: N=25) to non-carriers (N=144) (Figure 1; top). Here, higher t-scores signify more atrophy in the symptomatic group analysis. A mean bias corrected image from all the normalized T1 images in the GENFI study served as a study template. This template was warped into MNI space using the non-rigid registration based on fast free form deformation implemented in NiftyReg (version 10.4.15) (Modat et al., 2010). The obtained transformation was then applied to each of the three t-maps. For each MNI coordinate of the eligible cortical gene expression samples we located the corresponding voxel in the t-map and extracted a 3×3×3 voxel cube of t-values centered on that MNI coordinate, then the t-score corresponding to a gene expression sample was computed as the mean of the non-zero values in the 3×3×3 cube. This was done in order to accommodate uncertainties in registration. This procedure was carried out for each of the three t-maps and resulted in a 1,654 by three matrix, i.e., each gene expression sample was linked to three t-scores from the VBM analysis (one for each FTD gene).
Association analysis
The overall analysis is depicted in Figure 1. We analyzed the association between atrophy and gene expression in a non-parametric fashion. For a given atrophy map and a given probe, we computed the Spearman (or rank) correlation (ρ) between the local t-score and the gene expression level separately for each of the six donors. We computed separate P-values for positive and negative correlation using the cor.test function in R with one-sided hypotheses, respectively. Next, we combined the six p-values for positive correlations into a single meta p-value (P+) using the sum of Z scores method. The process was repeated for the six p-values for negative correlations (P-). On purpose, we did not use a weighted approach in order to avoid over-emphasizing the impact of donors with more gene expression samples. This procedure was carried out for each for the 37,031 probes and each of the three genetic group atrophy maps. P-values in each of the six resulting lists were corrected for multiple testing using the method by Holm (Holm, 1979). Significantly positively correlated genes (i.e., higher gene expression is linked to higher atrophy) were those where any probe targeting the gene reached a holm-corrected p-value < 0.05 for positive correlations (P+Holm < 0.05); likewise significantly negatively correlated genes (i.e., higher gene expression is linked to lower atrophy) were required to have a holm-corrected p-value < 0.05 for negative correlations (P-Holm < 0.05) for any of the probes targeting that gene. We also created two overlap lists, one containing the overlap of genes in the three positive lists, the other the overlap of genes in all three negative lists. In the following we refer to these two lists as consensus lists. The entire analysis was repeated using a linear mixed effects models (lme4 package in R (Bates et al., 2015)) for testing the association between gene expression and atrophy. In this analysis the local VBM t-score and donor were the fixed effect and the random effect, respectively. Due to high agreement between the results produced by the two association approaches, results based on the Spearman correlation analysis will be reported, since it is the approach with the least model assumptions.
Overrepresentation analysis
In order to identify cellular pathways or cellular processes and cell type signature genes that may be enriched in the significant gene lists, we conducted an overrepresentation analysis. We obtained the following gene sets and gene ontologies from the MSigDB database version 6.1 (date accessed 11/23/2017): KEGG (N=186 pathways), REACTOME (N=674 pathways) and Gene Ontology (N=5,917 ontologies). We used Fisher’s exact test to compute the p-value for overrepresentation of genes in a given set. All tests were carried out using the 16,912 cortex expressed genes as the background set. For each of the six gene lists (i.e., two per FTD gene) we corrected the p-values using the FDR across all 6,777 pathways and ontologies.
In addition to enrichment analysis for GO terms and pathways we used marker gene lists for six brain cell-types obtained from RNA sequencing of purified human cells (Zhang et al., 2016) to determine if the expressed genes implicate a specific class of brain cell types.
Expression Weighted Cell-Type Enrichment
In an additional analysis we sought to identify potential brain cell types that were implicated by all three FTD genes. To this end we conducted Expression Weighted Cell-type Enrichment (EWCE) analysis (Skene and Grant, 2016) on the two consensus lists using a recently published dataset of brain single-cell sequencing data in the mouse brain that identified 265 different cell types (www.mousebrain.org) (Zeisel et al., 2018). From this dataset we removed 76 cell types that were not directly brain related (e.g., cell belonging to enteric nervous system or the spinal cord), leaving 189 different cell-type signatures. Each of the cell types is also attributed with a high-level annotation (astrocytes, ependymal, immune, neurons, oligos, vascular). In brief, from the single cell mouse dataset we used only genes that had a unique human homolog (1-to-1 mapping). Then, we analyzed the two consensus lists separately for high-level cell type enrichment using EWCE with correction for gene length and GC content. P-values are based on 100,000 permutations and enrichment P-values were corrected for multiple testing using the FDR method. We used the available R package for EWCE.
Results
We tested 16,912 genes (from 37,031 microarray probes) for their association with atrophy pattern across the cortex in genetic FTD using two different approaches (Figure 1). The Spearman rank correlation approach showed a high agreement with the LME-based analysis: the Pearson’s correlation coefficient between the two sets of -log10 p-values for each FTD gene ranged from 0.925 to 0.959. The numbers of significant probes (PHolm<0.05) and genes with their direction for each of the three FTD genes are listed in Table 1. The association results per probe are available as supplementary material (Dataset S1).
C9orf72
The strongest association between C9orf72 and the atrophy pattern in symptomatic C9orf72 repeat extension carriers was measured with microarray probe A_23_P405873, which showed a negative Spearman correlation ρ=-0.0973 (Puncor=0.00076; Figure 2).
The most negatively associated gene was NEFH (neurofilament heavy; represented by probe CUST_463_PI416408490; ρ=-0.35; P=2.53e-56; Figure 2; Dataset S1). Other negatively correlated genes included those encoding synaptic (SYT2, VAMP1) and ion channel proteins (KCNA1, SCN4B) (Table 2). Top-ranked gene sets based on the significantly negatively correlated genes include genes related to mitochondria (GO_MITOCHONDRIAL_PART; OR=2.73; PFDR=2.32e-24) and the respiratory chain (GO_RESPIRATORY_CHAIN; OR=9.78; PFDR=1.11e-17; Dataset S2). Notably, KEGG pathways for neurodegenerative disorders were highly enriched (KEGG_PARKINSONS_DISEASE OR=7.03 PFDR=2.60e-17; KEGG_HUNTINGTONS_DISEASE OR=4.81 PFDR=3.69e-15; KEGG_ALZHEIMERS_DISEASE OR=5.50 PFDR=4.67e-17). Among brain cell types, there was a strong enrichment for neuronal genes (OR=2.57; P=2.83e-37; Dataset S3).
Among the most significantly positively correlated genes were ion channel related genes such as KCNG1 (ρ=0.31; P=2.92e-45), SCN9A (ρ=0.32; P=8.42e-43) and KCTD4 (ρ=0.32; P=1.46e-44). Top-ranked GO terms for positively correlated genes included cell-cell signaling (GO_CELL_CELL_SIGNALLING, OR=2.31, PFDR=2.43e-10; Dataset S2). Among brain cell types, there was a strong enrichment for genes associated with mature astrocytes (OR=4.92; P=2.2e-70) as well as neuronal genes (OR=2.05; P=4.43e-20; Dataset S3).
GRN
The strongest association between GRN expression and the atrophy pattern in symptomatic GRN mutation carriers was measured with microarray probe A_23_P49708, which showed a positive Spearman correlation ρ=0.0587 (Puncor=0.0013; Figure 2).
Negatively associated genes include those encoding synaptic proteins (SLC17A, Table 2). For the negatively correlated genes, none of the tested gene sets showed statistically significant enrichment after multiple testing correction (Dataset S2). However, among brain cell types, there was a strong enrichment for neuronal genes (OR=2.77; P=8.61e-22; Dataset S3).
Positively associated genes included those encoding proteins involved in the immune response (CD6, WFDC1, SPON2). Top-ranked GO terms (Dataset S2) for positively correlated genes are related to tissue development, (GO_TISSUE_DEVELOPMENT OR=2.2 PFDR=2.58e-12), the extracellular matrix (GO_EXTRACELLULAR_MATRIX OR=3.28 PFDR=1.4e-10) and response to wounding (GO_RESPONSE_TO_WOUNDING OR=2.51 PFDR=2.10e-07). Again, genes related to mature astrocytes showed the strongest enrichment (OR=3.34; P=8.23e-28), followed by endothelial cells (OR=2.17; P=2.59e-06) and weak enrichment for neuron related genes (OR=1.28; P=0.0076; Dataset S3).
MAPT
The strongest association between MAPT expression and the atrophy pattern in symptomatic MAPT mutation carriers was measured with microarray probe CUST_449_PI416408490, which showed a positive Spearman correlation ρ=0.0831 (Puncor=1.96e-07; Figure 2).
Among the most significantly negatively correlated genes were those encoding ion channels (SCN1B, SCN1A, SLC24A2, Table 2). As in the case of C9orf72, the significantly negatively correlated genes showed enrichment for mitochondria (GO_MITOCHONDRIAL_PART; OR=1.69; PFDR=6.06e-09) and cellular respiration (GO_CELLULAR_RESPIRATION; OR=3.11; PFDR=1.07e-07; Dataset S2). Notably, neuron related genes were strongly enriched (OR=1.85; P=1.59e-26) as well as genes associated with neurodegenerative disorders in KEGG (KEGG_PARKINSON OR=3.19 PFDR=1.23e-06; KEGG_ALZHEIMERS_DISEASE OR=2.49 PFDR=1.97e-05; KEGG_HUNTINGTONS_DISEASE OR=2.36 PFDR=3.17e-05).
Significantly positively associated genes included those encoding ion channels (KCTD4, SCN9A). Significantly positively correlated genes are enriched for cell movement (GO_MOVEMENT_OF_CELL_OR_SUBCELLULAR_COMPONENT; OR=1.63, PFDR=4.66e-09) and neuron projection (GO_NEURON_PROJECTION_DEVELOPMENT; OR=1.90; PFDR=5.34e-08). Furthermore, the positively correlated genes were enriched for genes related to mature astrocytes (OR=4.66, P=1.43e-90), microglia (OR=1.88, P=2.63e-16) and for neurons (OR=1.48, P=1.66e-10; Dataset S3).
Cell-type analysis
From the gene lists obtained for each of the FTD gene atrophy maps we created two consensus lists: one comprising 405 genes that were significantly positively correlated with atrophy in all three FTD genes and one list comprising the 250 genes that were significantly negatively correlated in all three maps (Figure 1). Using these lists, we aimed to identify a common theme underlying the atrophy in the three causative genes. In particular, we used EWCE paired with high resolution cell specific murine gene expression profiles to identify specific cell-types that are enriched in genes that showed significant correlations with atrophy in all three FTD maps. EWCE was executed twice, once using 189 cell-type annotations and once using high-level annotations. The high-level analysis EWCE showed a strong enrichment for astrocyte marker genes (Figure 3; Z-score=5.75; P<0.00001) and a borderline enrichment for ependymal marker genes (Z-score=2.76; P=0.004) among genes with positive correlation to atrophy severity. Genes negatively correlated with atrophy were enriched for neuronal marker genes (Z-score=6.58; P<0.00001). These results were confirmed using cell-type marker genes derived from RNA sequencing of purified human cells (Zhang et al., 2016) (Figure 3; right column).
Discussion
We investigated the gene expression correlates of the cortical regions specifically atrophic in the three main genetic causes of FTD (MAPT, C9orf72, GRN). Overall, the analysis showed that the most atrophic cortical regions in symptomatic mutation carriers have a markedly different gene expression profile in the six cognitively normal subjects from the AIBS gene expression dataset. The type of gene expression profiling used in AIBS is based on measuring bulk expression of tissue samples, i.e., a group of diverse cell types is sampled at once and the resulting expression profile represents the group average of this set of cells and their states. Thus, in this type of analysis, genes positively correlated with atrophy indicate potential cellular processes and cell types that promote atrophy in genetic FTD. Conversely, genes that are negatively correlated with atrophy indicate potential cellular processes and cell types that confer resilience to disease-related neurodegeneration. Whilst there is no association with expression of the gene itself (i.e., GRN, MAPT and C9orf72) in each form of genetic FTD, our analysis revealed that groups of genes commonly associated with astrocytes showed higher expression levels in regions with more atrophy and genes commonly associated with neurons showed higher expression levels in the relatively spared regions.
Astrocytes are the most abundant cell-type in the human CNS and they carry out a plethora of functions including biochemical support for the blood-brain barrier-forming endothelial cells, trophic support for neurons, regulation of extracellular ion balance and participation in repair and scarring processes of the brain following injuries. Astrocytes reacting to injuries in the central nervous system, reactive astrocytes, are characterized by expression of glial fibrillary acidic protein (GFAP). Depending on the context, such reactive GFAP+-astrocytes can be neurotoxic (Liddelow and Barres, 2015; Liddelow et al., 2017) or neuroprotective (Anderson et al., 2016). There are already multiple lines of evidence linking astrocyte (dys-)function to neurodegeneration (Rodríguez et al., 2009; Phatnani and Maniatis, 2015; Sofroniew, 2015) and to FTD in particular. For instance, histopathological studies in FTD have shown that severity of astrocytosis and astrocytic apoptosis correlated with the degree of neuronal loss as well as with the stage of the disease, while at the same time neuronal apoptosis was rare (Broe et al., 2004). In addition, astrocyte reactivity appears to be region specific in that higher numbers of reactive (GFAP+) astrocytes were found in the frontal and temporal cortices of FTD patients compared to controls (Martinac et al., 2001). These observations extend to the CSF where levels of GFAP were increased in various neurodegenerative disorders compared to cognitively normal adults with the highest levels in FTD patients (Ishiki et al., 2016). Martinac et al. (2001) found that degrading astrocytes were inversely correlated with cerebral blood flow in FTD. However, more importantly, astrocytes derived from induced pluripotent stem cells of patients with mutations in MAPT were found to demonstrate increased vulnerability to oxidative stress and exhibit disease-associated gene-expression changes (Hallmann et al., 2017). Co-culture experiments of such modified FTD astrocytes with previously healthy neurons led, among other things, to increased oxidative stress in these neurons.
Hence, taken together astrocyte reactivity and activation of GFAP co-occur with disease onset and disease progression in FTD, but could astrocytes dysfunction alone be the initiator for neurodegeneration in FTD? The answer may lie partially in the lysosomes of astrocytes: studies have demonstrated that neuronal cell-derived proteins such as α-synuclein can be transferred to close-by astrocytes via endocytosis and that these proteins are destined for degradation in the astrocytes’ lysosome (Lee et al., 2010). Likewise, astrocytes are known to bind and degrade extracellular amyloid-β, a key player in Alzheimer’s disease (Wyss-Coray et al., 2003; Koistinaho et al., 2004). Recent work established a strong link of both Parkinson’s disease and FTD with lysosomal storage disorders (LSDs) (Deng et al., 2015; Burbulla et al., 2017; Evers et al., 2017). LSDs are a large group of rare inherited metabolic disorders with defective lysosome function resulting in faulty degradation and recycling of cellular constituents. Intriguingly, mutations in the familial Parkinson’s disease gene GBA and the familial FTD gene GRN cause the LSDs Gaucher Disease and neuronal ceroid lipofuscinosis (NCL), respectively (Ward et al., 2017). Mounting evidence suggests that astrocyte dysfunction alone is sufficient to trigger neurodegeneration in LSDs (Rama Rao and Kielian, 2016). For instance, in a mouse model astrocyte-specific deletion of Sumf1 in vivo induced severe lysosomal storage dysfunction in these astrocytes, which in turn was sufficient to induce degeneration of cortical neurons in vivo (Di Malta et al., 2012). More importantly, though, one of the three FTD genes, GRN, is also known to cause NCL (Ward et al., 2017) and recent work posits that lysosomal dysfunction is a central disease process in GRN-associated FTD (Ward et al., 2017). Indeed, GRN is increasingly associated with regulating the formation and function of the lysosome (Kao et al., 2017). In addition, the transcription of GRN is co-regulated with other lysosomal genes (Belcastro et al., 2011).
In addition to detecting higher levels of astrocyte-related genes in regions with neurodegeneration in FTD, we found that genes associated with neurons are more enriched in brain regions that are spared in FTD. Moreover, for the MAPT and C9orf72 atrophy maps we additionally noted enrichment for genes that are associated with mitochondria, particularly cellular respiration, in regions that are not affected by atrophy. These results confirm earlier results from an unbiased proteomic screen of tissue samples where the modules related to synapse (M1), mitochondrion (M3) and neuron differentiation (M8) showed negative correlations with clinicopathological traits in FTD, i.e., these three modules were consistently negatively correlated with FTD pathology (Umoh et al., 2018). Consistent with our results, there was a positive correlation of astrocyte specific modules (M5: Extracellular matrix and M6: Response to biotic stimulus) with clinicopathological traits. While Umoh et al. (2018) interpreted the negative (and positive) correlations in part with a disease-related shift in cell population in the sampled ROIs, our results extend this observation to regional cell type densities since the gene expression samples were obtained from six cognitively normal subjects.
Lastly, in the GRN group only, there was a positive correlation with a number of genes involved in the immune response (CD6, WFDC1, SPON2), i.e., these were associated with the GRN-associated FTD pattern of atrophy. This is consistent with previous work showing that inflammation and microglial activation have an aetiological role in GRN-associated FTD (Bossù et al., 2011; Martens et al., 2012).
In summary, our analysis indicates that cortical regions showing the most severe atrophy in genetic FTD are those with the highest astrocyte density in healthy subjects. Therefore, astrocytes may have a more active role in the onset of neurodegeneration in FTD than previously assumed. This fits with recent findings of neurotoxic potential of astrocytes (Liddelow et al., 2017). We hypothesize that the distinct regional atrophy pattern in genetic FTD may be driven by regions with naturally increased astrocyte density where these universal astrocyte neurotoxic effects come to bear with higher frequency. Thus, neurodegeneration may be the result of the toxic combination of increased potential for lysosomal storage in astrocytes caused by FTD mutations and age-related increase in neurotoxic (A1) and senescent astrocytes, which lost many normal astrocytic functions.
Funding
AA holds a Medical Research Council eMedLab Medical Bioinformatics Career Development Fellowship. This work was supported by the Medical Research Council (grant number MR/L016311/1). The Dementia Research Centre is supported by Alzheimer’s Research UK, Brain Research Trust, and The Wolfson Foundation. This work was supported by the NIHR Queen Square Dementia Biomedical Research Unit, the NIHR UCL/H Biomedical Research Centre and the Leonard Wolfson Experimental Neurology Centre (LWENC) Clinical Research Facility as well as an Alzheimer’s Society grant (AS-PG-16-007). JDR is supported by an Medical Research Council Clinician Scientist Fellowship (MR/M008525/1) and has received funding from the NIHR Rare Disease Translational Research Collaboration (BRC149/NS/MH). This work was also supported by the Medical Research Council UK GENFI grant (MR/M023664/1). M.R. holds an Medical Research Council Clinician Scientist Fellowship (grant number MR/N008324/1). R.H.R. was supported through the award of a Leonard Wolfson Doctoral Training Fellowship in Neurodegeneration. This work was supported by Italian Ministry of Health (CoEN015 and Ricerca Corrente). Several authors of this publication (JvS, MS, RV, AD, MO, JR) are members of the European Reference Network for Rare Neurological Diseases - Project ID No 739510. This project was supported, in part, via the European Union’s Horizon 2020 research and innovation program grant 779257 “Solve-RD” (to M.S.).
Supplementary information
Dataset S1: Results from the spatial correlation analysis. This table shows for every eligible probe, the mapped gene name, the Spearman (or rank) correlation with the three separate t-maps, P+ values for positive correlation and P- values for negative correlation.
Dataset S2: Enrichments for pathways. Each sheet in this table shows the results of the enrichment analysis for the six gene lists (genes positively and negatively correlated with atrophy, respectively). Columns represent the pathway name, the number of overlapping genes between list and pathway, the expected number of overlapping genes, the odds ratio (OR) and the overrepresentation p-value (including FDR-correction).
Dataset S3: Enrichments for brain six cell-types. Enrichment analysis testing gene lists with significant positive or negative correlation for cell-type marker genes. Human cell-type gene lists for neurons (N), microglia (MG), mature astrocytes (MA), endothelial cells (EC), oligodendrocytes (OLG) and oligodendrocyte precursor cells (OPC) were obtained from Zhang et al. (2016).
Appendix
List of GENFI consortium authors
Caroline Greaves BSc1, Georgia Peakman MSc1, Rachelle Shafei MRCP1, Emily Todd Mres1, Martin N. Rossor MD FRCP1, Jason D. Warren PhD FRACP1, Nick C. Fox MD FRCP1,2, Henrik Zetterberg2, Rita Guerreiro PhD3, Jose Bras PhD3, Jennifer Nicholas PhD4, Simon Mead PhD5, Lize Jiskoot PhD6, Lieke Meeter MD6, Jessica Panman MSc6, Janne Papma PhD6, Rick van Minkelen PhD7, Yolanda Pijnenburg PhD8, Myriam Barandiaran PhD9,10, Begoña Indakoetxea MD9,10, Alazne Gabilondo MD10, Mikel Tainta MD10, Maria de Arriba BSc10, Ana Gorostidi PhD10, Miren Zulaica BSc10, Jorge Villanua MD PhD11 Zigor Diaz12, Sergi Borrego-Ecija MD13, Jaume Olives MSc13, Albert Lladó PhD13, Mircea Balasa PhD13, Anna Antonell PhD13, Nuria Bargallo PhD14, Enrico Premi MD15, Maura Cosseddu MPsych15, Stefano Gazzina MD15, Alessandro Padovani MD PhD15, Roberto Gasparotti MD16, Silvana Archetti MBiolSci17, Sandra Black MD19, Sara Mitchell MD19, Ekaterina Rogaeva PhD20, Morris Freedman MD21, Ron Keren MD22, David Tang-Wai MD23, Linn Öijerstedt MD24, Christin Andersson PhD25, Vesna Jelic MD26, Hakan Thonberg MD27, Andrea Arighi MD28,29, Chiara Fenoglio PhD28,29, Elio Scarpini MD28,29, Giorgio Fumagalli MD28,29,30, Thomas Cope MRCP31, Carolyn Timberlake BSc31, Timothy Rittman MRCP31, Christen Shoesmith MD32, Robart Bartha PhD33,34, Rosa Rademakers PhD35, Carlo Wilke MD36,37, Hans-Otto Karnarth MD38, Benjamin Bender MD39, Rose Bruffaerts MD PhD40, Philip Vandamme MD PhD41, Mathieu Vandenbulcke MD PhD42,43, Catarina B. Ferreira MSc44, Gabriel Miltenberger PhD45, Carolina Maruta MPsych PhD46, Ana Verdelho MD PhD47, Sónia Afonso BSc48, Ricardo Taipa MD PhD49, Paola Caroppo MD PhD50, Giuseppe Di Fede MD PhD50, Giorgio Giaccone MD50, Sara Prioni PsyD50, Veronica Redaelli MD50, Giacomina Rossi MSc50, Pietro Tiraboschi MD50, Diana Duro NPsych51, Maria Rosario Almeida PhD51, Miguel Castelo-Branco MD PhD51, Maria João Leitão BSc52, Miguel Tabuas-Pereira MD53, Beatriz Santiago MD53, Serge Gauthier MD56, Pedro Rosa-Neto MD PhD57, Michele Veldsman PhD58, Toby Flanagan BSc60, Catharina Prix MD61, Tobias Hoegen MD61, Elisabeth Wlasich Mag. rer. nat.61, Sandra Loosli MD61, Sonja Schonecker MD61, Elisa Semler Dr.hum.biol Dipl. Psych62, Sarah Anderl-Straub Dr.hum.biol Dipl.Psych62, Luisa Benussi PhD63, Giuliano Binetti MD63, Michela Pievani PhD63, Gemma Lombardi MD64, Benedetta Nacmias PhD64, Camilla Ferrari64, Valentina Bessi64, Cristina Polito65.
Affiliations
1Dementia Research Centre, Department of Neurodegenerative Disease, UCL Queen Square Institute of Neurology, London, UK; 2Dementia Research Institute, Department of Neurodegenerative Disease, UCL Institute of Neurology, Queen Square, London, UK; 3Center for Neurodegenerative Science, Van Andel Research Institute, Grand Rapids, Michigan, USA.4Department of Medical Statistics, London School of Hygiene and Tropical Medicine, London, UK; 5MRC Prion Unit, Department of Neurodegenerative Disease, UCL Institute of Neurology, Queen Square, London, UK; 6Department of Neurology, Erasmus Medical Centre, Rotterdam, Netherlands; 7Department of Clinical Genetics, Erasmus Medical Centre, Rotterdam, Netherlands; 8Amsterdam University Medical Centre, Amsterdam VUmc, Amsterdam, Netherlands; 9Cognitive Disorders Unit, Department of Neurology, Donostia University Hospital, San Sebastian, Gipuzkoa, Spain; 10Neuroscience Area, Biodonostia Health Research Institute, San Sebastian, Gipuzkoa, Spain; 11OSATEK, University of Donostia, San Sebastian, Gipuzkoa, Spain; 12CITA Alzheimer, San Sebastian, Gipuzkoa, Spain; 13Alzheimer’s disease and Other Cognitive Disorders Unit, Neurology Service, Hospital Clínic, Barcelona, Spain; 14Imaging Diagnostic Center, Hospital Clínic, Barcelona, Spain; 15Centre for Neurodegenerative Disorders, Neurology Unit, Department of Clinical and Experimental Sciences, University of Brescia, Brescia, Italy; 16Neuroradiology Unit, University of Brescia, Brescia, Italy; 17Biotechnology Laboratory, Department of Diagnostics, Spedali Civili Hospital, Brescia, Italy; 18Clinique Interdisciplinaire de Mémoire Département des Sciences Neurologiques Université Laval Québec, Quebec, Canada; 19Sunnybrook Health Sciences Centre, Sunnybrook Research Institute, University of Toronto, Toronto, Canada; 20Tanz Centre for Research in Neurodegenerative Diseases, University of Toronto, Toronto, Canada; 21Baycrest Health Sciences, Rotman Research Institute, University of Toronto, Toronto, Canada; 22The University Health Network, Toronto Rehabilitation Institute, Toronto, Canada; 23The University Health Network, Krembil Research Institute, Toronto, Canada; 24Department of Geriatric Medicine, Karolinska University Hospital-Huddinge, Stockholm, Sweden; 25Department of Clinical Neuroscience, Karolinska Institutet, Stockholm, Sweden; 26Division of Clinical Geriatrics, Karolinska Institutet, Stockholm, Sweden; 27Center for Alzheimer Research, Divison of Neurogeriatrics, Karolinska Institutet, Stockholm, Sweden; 28Fondazione IRCCS Ca’ Granda Ospedale Maggiore Policlinico, Neurodegenerative Diseases Unit, Milan, Italy; 29University of Milan, Centro Dino Ferrari, Milan, Italy; 30Department of Neurosciences, Psychology, Drug Research and Child Health (NEUROFARBA), University of Florence, Florence, Italy; 31Department of Clinical Neurosciences, University of Cambridge, Cambridge, UK; 32Department of Clinical Neurological Sciences, University of Western Ontario, London, Ontario Canada; 33Department of Medical Biophysics, The University of Western Ontario, London, Ontario, Canada; 34Centre for Functional and Metabolic Mapping, Robarts Research Institute, The University of Western Ontario, London, Ontario, Canada; 35Department of Neuroscience, Mayo Clinic, Jacksonville, Florida, USA; 36Department of Neurodegenerative Diseases, Hertie-Institute for Clinical Brain Research and Center of Neurology, University of Tübingen, Tübingen, Germany; 37Center for Neurodegenerative Diseases (DZNE), Tübingen, Germany; 38Division of Neuropsychology, Hertie-Institute for Clinical Brain Research and Center of Neurology, University of Tübingen, Tübingen, Germany; 39Department of Diagnostic and Interventional Neuroradiology, University of Tübingen, Tübingen, Germany; 40Laboratory for Cognitive Neurology, Department of Neurosciences, KU Leuven, Leuven, Belgium; 41Neurology Service, University Hospitals Leuven, Belgium, Laboratory for Neurobiology, VIB-KU Leuven Centre for Brain Research, Leuven, Belgium; 42Geriatric Psychiatry Service, University Hospitals Leuven, Belgium; 43Neuropsychiatry, Department of Neurosciences, KU Leuven, Leuven, Belgium; 44Laboratory of Neurosciences, Institute of Molecular Medicine, Faculty of Medicine, University of Lisbon, Lisbon, Portugal; 45Faculty of Medicine, University of Lisbon, Lisbon, Portugal; 46Laboratory of Language Research, Centro de Estudos Egas Moniz, Faculty of Medicine, University of Lisbon, Lisbon, Portugal; 47Department of Neurosciences and Mental Health, Centro Hospitalar Lisboa Norte -Hospital de Santa Maria & Faculty of Medicine, University of Lisbon, Lisbon, Portugal; 48Instituto Ciencias Nucleares Aplicadas a Saude, Universidade de Coimbra, Coimbra, Portugal; 49Neuropathology Unit and Department of Neurology, Centro Hospitalar do Porto -Hospital de Santo António, Oporto, Portugal; 50Fondazione IRCCS Istituto Neurologico Carlo Besta, Milano, Italy; 51Faculty of Medicine, University of Coimbra, Coimbra, Portugal; 52Centre of Neurosciences and Cell biology, Universidade de Coimbra, Coimbra, Portugal; 53Neurology Department, Centro Hospitalar e Universitario de Coimbra, Coimbra, Portugal; 54Department of Psychiatry, McGill University Health Centre, McGill University, Montreal, Québec, Canada; 55McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University, Montreal, Québec, Canada; 56Alzheimer Disease Research Unit, McGill Centre for Studies in Aging, Department of Neurology & Neurosurgery, McGill University, Montreal, Québec, Canada; 57Translational Neuroimaging Laboratory, McGill Centre for Studies in Aging, McGill University, Montreal, Québec, Canada; 58Nuffield Department of Clinical Neurosciences, Medical Sciences Division, University of Oxford, Oxford, UK; 59Faculty of Medical and Human Sciences, Institute of Brain, Behaviour and Mental Health, University of Manchester, Manchester, UK; 60Faculty of Biology, Medicine and Health, Division of Neuroscience and Experimental Psychology, University of Manchester, Manchester, UK; 61Neurologische Klinik, Ludwig-Maximilians-Universität München, Munich, Germany; 62Department of Neurology, University of Ulm, Ulm; 63Instituto di Ricovero e Cura a Carattere Scientifico Istituto Centro San Giovanni di Dio Fatebenefratelli, Brescia, Italy; 64Department of Neuroscience, Psychology, Drug Research, and Child Health, University of Florence, Florence, Italy. 65Department of Biomedical, Experimental and Clinical Sciences “Mario Serio”, Nuclear Medicine Unit, University of Florence, Florence, Italy.
Abbreviations
- AIBS
- Allen Institute for Brain Science
- C9orf72
- Chromosome 9 open reading frame 72 gene
- EWCE
- Expression Weighted Cell-type Enrichment
- FTD
- Frontotemporal dementia
- GENFI
- Genetic FTD Initiative
- GFAP
- glial fibrillary acidic protein
- GO
- Gene Ontology
- GRN
- Progrenulin gene
- LME
- Linear Mixed Effects
- LSD
- Lysosomal Storage Disorder
- MAPT
- tau gene
- MNI
- Montreal Neurological Institute and Hospital
- NCL
- neuronal ceroid lipofuscinosis
- OR
- Odds Ratio
- VBM
- Voxel Based Morphometry