Effect of progranulin gene status on global gene expression

Genetic screening of our FTLD-U cases revealed 7 cases containing progranulin gene abnormalities (GRN+). Of these, four possessed truncation mutations previously associated with disease and believed to be pathogenic, while the remaining three cases exhibited new variants of unknown significance (VUS, Table ​Table4).4). Preliminary studies suggest that at least some of these VUS are pathogenic, and this avenue of investigation is currently ongoing. Using Spearman correlation coefficients, we clustered all frontal cortex and hippocampal samples into a hierarchical dendrogram to look for any naturally arising subsets of samples. We found that almost all GRN+ samples (9/11), regardless of region of origin, clustered together into one branch without interspersion of GRN− cases of FTLD-U or control samples (Fig. ​(Fig.3A).3A). This clustering result was robust to statistical methodology, seen in dendrograms produced using Pearson correlation coefficients as well.

Open in a separate window

Figure 3.

Cluster analysis and principal components analysis of samples from histopathologically affected brain regions show a distinct expression signature for GRN+ FTLD-U. (A) Hierarchical dendrogram produced by clustering of expression levels for 39 brain samples from frontal cortex or hippocampus (columns) and 22 277 transcripts (rows), using Spearman correlation coefficients. With two exceptions, all 11 GRN+ FTLD-U samples (yellow) fell into one branch (bold), while samples from GRN− FLTD-U (red) and normal controls (blue) were admixed. Brain region for each sample is shown as well, with orange denoting frontal cortex samples and teal denoting hippocampal samples. Heatmap tiles show standardized expression levels of individual genes with red denoting high expression, grey denoting medium expression and blue denoting low expression levels. (B) Principal components analysis of frontal and hippocampal samples using 22 277 transcripts also revealed a distinct expression signature for GRN+ FTLD-U. GRN+ FTLD-U samples (yellow) were characterized by a low score along the first principal component (X-axis) and a high score along the second principal component (Y-axis), separating them from GRN− FTLD-U samples (red) and normal samples (blue). Arrows indicate the two samples that were more distant in global gene expression by cluster analysis. Within GRN+ FTLD-U cases, samples bearing truncation mutations (squares) and samples bearing variants of unknown significance (triangles) were admixed. R493X=predicted protein change for indicated truncation mutant. R418X=predicted protein change for indicated truncation mutant.

Principal component analysis was performed as a second methodology for looking in an unbiased manner at overall data structure (Fig. ​(Fig.3B).3B). Within histopathologically affected regions of brain, GRN+ FTLD-U cases again grouped together, defined by a low score along the first principal component (X-axis) and a high score along the second principal component (Y-axis).

Based on the observation from unbiased cluster and principal component analyses that GRN+ FTLD-U cases have a molecular expression signature which is distinct from GRN− FTLD-U cases and control subjects, we performed subgroup analyses of our gene expression data. Comparing 6 GRN+ frontal cortex samples to 8 normal frontal cortex samples, we found 1311 probe sets representing 1131 genes with differential expression at a P-value of 0.001, with an FDR of 1.9% (Table ​(Table2).2). Of these, 579 probe sets were increased in disease, and 732 were decreased in disease. After imposing a minimum fold change cutoff of 2, 772 probe sets representing 657 genes were still found to be differentially expressed at a P-value of 0.001. Of these, 284 probe sets were increased in disease, and 488 were decreased in disease (Supplementary Material, Table S3); the top 20 transcripts showing differential expression are summarized in Table ​Table5.5. In contrast, comparing 10 GRN- frontal cortex samples to 8 normal frontal cortex samples, we found only 154 probe sets representing 147 genes showing differential expression at a P-value of 0.001 (FDR 13.8%, Table ​Table2).2). In fact, when GRN+ and GRN− frontal cortex samples were compared directly, more genes were found to be differentially expressed (208 probe sets representing 214 genes—several probes recognize more than one gene) between these molecular subtypes of disease than between the GRN− cases and normals (Supplementary Material, Table S4).

The finding that there are many more dysregulated genes among the GRN+ cases is striking given the fact that the smaller number of GRN+ cases offers less statistical power. The large number of dysregulated genes common to this subset of cases corroborates evidence from our cluster and principal component analyses that the GRN+ cases have a distinct expression signature.

As with the full dataset, cerebellar samples exhibited many fewer dysregulated genes (Table ​(Table22).

Quantitative real-time QRT–PCR validation of gene changes detected on microarrays

To corroborate gene changes detected on our microarrays, we performed QRT–PCR for a subset of upregulated and downregulated genes. For all genes tested, microarray and QRT–PCR results were consistent in both GRN+ and GRN− FTLD-U cases (Fig. ​(Fig.4).4). We used ß-actin as our reference gene for normalizing gene expression levels based on microarray data showing stable expression; comparable results were obtained using cyclophilin A as an alternate reference gene (data not shown). In general, relative to QRT–PCR, microarray results tended to underestimate the magnitude of gene changes. This has been observed in other studies as well (30,32), and likely reflects the greater amount of noise around microarray measurements, with their subsequent need for normalization.

Open in a separate window

Figure 4.

Validation of microarray results by quantitative reverse transcription real-time PCR (QRT–PCR). Relative expression levels of genes are indicated by the fold changes in expression level for FTLD-U samples with progranulin gene abnormalities (GRN+), and FTLD-U samples without progranulin gene abnormalities (GRN−). For QRT–PCR, average fold changes and standard deviations of three cases in each group are shown. PCR reactions were performed in duplicate. GLRB, glycine receptor beta subunit; GRM5, metabotropic glutamate receptor 5; HDAC1, histone deacetylase 1; HSPA2, heat shock protein 2 (70 kD); HTR2A, serotonin receptor 2A; ID4, inhibitor of DNA binding 4; NOTCH2, Notch homolog (Drosophila) 2.

A distinct expression signature for GRN+ FTLD-U

By cluster analysis, principal component analysis and biological pathway analysis, we found that GRN+ FTLD-U cases had a distinct expression signature. While there have been suggestions of clinical and pathological differences in some FTLD-U cases with GRN mutations (36,38,39), the overall heterogeneity of presentations makes it difficult or nearly impossible to make distinctions on clinical or histopathological grounds alone (10,20,21). The ability to detect distinct molecular phenotypes, however, may be important in identifying patient groups more responsive to targeted therapies.

It is important to note that cluster and principal component analyses revealing the distinct expression signature for GRN+ FTLD-U cases were performed in an unbiased manner; we did not pre-specify subgroups but rather evaluated the overall pattern of similarity in global gene expression among all samples. The fact that the GRN truncation mutants (four cases) formed a group with the three cases carrying GRN gene VUS was somewhat surprising. This finding has implications for the pathogenicity of these variants, which are under investigation now.

Notably, 2 of the 11 GRN+ samples were more distant in global gene expression from the other 9 samples (Fig. ​(Fig.3B,3B, labeled arrows). Of these, one (Fig. ​(Fig.3A,3A, arrow) was particularly distant and remained so regardless of clustering methodology. This frontal cortex sample came from a case bearing the truncation mutation R493X. The corresponding hippocampal sample from this case did cluster with the other GRN+ cases. On reviewing this case, we noted significant alpha-synuclein pathology in addition to TDP-43 pathology in the frontal cortex sample, which may explain the status of this case as an outlier. The R493X mutation has previously been found in multiple FTLD-U families (10). Frontotemporal dementia and primary progressive aphasia have been the associated clinical diagnoses; pathologically, cases bearing this mutation have been characterized by ubiquitin-positive intranuclear inclusions, consistent with a Sampathu Type 3 classification (10,19). Coexisting alpha-synuclein pathology has not been previously described; thus, along with the cases bearing GRN gene VUS, this case is also being investigated further. The other sample clustering separately from the majority of the GRN+ FTLD-U cases is a frontal cortex sample also bearing a truncation mutation (R418X). This sample, however, appeared more related to the other GRN+ FTLD-U samples, with distance decreasing with the use of a parametric correlation coefficient. The corresponding hippocampal sample from this case fell into the main GRN+ FTLD-U cluster.

We considered various possibilities to explain our finding of a distinct GRN+ FTLD-U expression signature. First, this finding could be the spurious result of different treatment conditions for these samples. However, the 56 brain samples used in this study were collected in random order, with the operator blinded to disease and gene status. Processing of all samples after collection was performed on two consecutive days, using the same batch of microarray platforms, and the GRN+ samples were divided between the 2 days. Second, this finding could be the result of differing RNA quality in this subset. However, review of RNA quality parameters for these selected cases did not reveal any obvious differences from the rest of the samples. Third, while all cases in this study bearing GRN gene abnormalities exhibited Sampathu Type 3 histopathology (19), not all of the GRN− FTLD-U cases were Type 3, raising the possibility that the non-Type 3 cases drove this apparent difference in global gene expression. We therefore re-analyzed our data excluding the four GRN− FTLD-U cases that were not Sampathu Type 3. Excluding these cases did not materially change results (Supplementary Material, Fig. S1). Thus, we believe that the distinct global expression signature observed for GRN+ FTLD-U cases in this study carries biological significance.

Comparison to previous FTLD and ALS gene expression studies

While our study corroborated the findings of a previous FTLD gene expression study (40) that synapse-related genes are downregulated and cytoskeleton-associated genes are upregulated, both our experimental design and overall findings exhibited several important differences from the previous study. Because we compared affected and unaffected brain regions and also cases with and without GRN mutations, we were able to detect significant differences in expression levels in frontal cortex versus cerebellum, and in GRN+ cases versus GRN- cases relative to each other and controls. In addition, compared to the previous study, we found many more genes, involving many more biological pathways, to be dysregulated in both our FTLD-U overall group and in our GRN+ subset (Supplementary Material, Table S2). Fluorescent signal was present for TARDBP and GRN gene expression in all of our samples, which was not the case in the previous study (40). Finally, we found approximately equal numbers of up- and down-regulated genes for FTLD-U overall, as well as our GRN+ and GRN- subsets. In contrast, the previous study found almost all of the genes showing significant differential expression to be downregulated in disease (40).

The differing results of these two studies could reflect the fact that our study included many more samples, with greater pathological homogeneity, both of which increase statistical power.

In addition, we compared our results to a study of gene expression changes in sporadic ALS motor cortex (31) and found many more genes with differential expression (Fig. ​(Fig.2).2). In sporadic ALS, TDP-43-positive inclusions have been found in lower motor neurons (16) as well as motor cortex (37,41). These regional differences in TDP-43 histopathology and gene expression changes in FTLD-U and ALS are worthy of further examination.

Genes implicated in previous neurodegenerative disease studies

Several of the individual genes found to be dysregulated in our study have been implicated in previous studies of neurodegenerative disease. The gene found to be most downregulated in our GRN+ FTLD-U frontal cortex samples was the 68 kD neurofilament (NF) light polypeptide (NEFL), which has been associated with motor neuron disease on pathological, mouse model and human genetic grounds. Pathologically, aberrant NF accumulation has long been considered a hallmark of ALS (42). In addition, mice overexpressing NEFL (43) and mice carrying a point mutation in NEFL (44) show selective dysfunction and degeneration of spinal motor neurons. More recently, it has been argued that TDP-43 acts physiologically as a stabilizer of NEFL mRNA, with loss of this normal function involved in the pathology of FTLD-U (45). Because NFs are abundant structural components of neurons, the apparent loss of NF mRNA from postmortem samples of brain or spinal cord might be attributable simply to neuronal loss. Arguing against this, however, is the fact that we found NEFL mRNA expression to be downregulated only in our GRN+ cases. The GRN− FTLD-U cases, which have as much neuronal loss, did not show any differences in levels of NEFL mRNA expression.

GRN mutations are associated with FTLD-U, but we did not find dysregulated GRN mRNA expression in either our GRN+ or GRN− subsets. We considered several possible explanations. First, this result could be an artifact of combining GRN truncation mutants (four cases) with cases containing GRN gene VUS (three cases). We therefore evaluated these two categories separately but were still unable to find a statistically significant difference in GRN gene expression between either group and normal controls. Second, although total mRNA and protein levels of progranulin have been shown to be decreased in FTLD-U cases with GRN mutations (8,9), most of these studies have focused on mRNA isolated from lymphoblastoid cells rather than brain. In addition, sequencing of cDNA from brain and lymphoblastoid cells has demonstrated preferential expression of the normal allele (8,9), but this does not exclude the possibility that there is a compensatory increase in expression of the normal allele in the affected organ (brain), resulting in little difference in total mRNA levels. Third, in one case of documented decreased total mRNA expression in brain (46), GRN mRNA levels were decreased by ~20% in frontal cortex and ~10% in cerebellum; by microarray standards, such changes are relatively small and may have been below the detection level of our experimental design. Finally, the P-values for differential expression for the three GRN probe sets on our microarray platform were 0.006, 0.048 and 0.005. In an effort to detect the most robust gene changes, we may have set our statistical threshold too stringently (at P < 0.001) to identify this gene as significantly dysregulated. Of note, we are systematically evaluating progranulin transcript levels in different regions of human brain with QRT–PCR now and will report these findings separately.

TDP-43, the major protein found in the characteristic inclusions of FTLD-U and ALS, has been the subject of intense study recently. Initial genetic studies of TDP-43, encoded by the gene TARDBP, have not revealed obvious mutations within FTLD-U patients (47); mRNA expression studies are lacking. In our microarray analysis, TARDBP mRNA levels were not significantly changed among either GRN+ or GRN− cases, compared to controls. There was a suggestion (Fig. ​(Fig.4)4) of slightly higher mRNA levels in the three GRN+ cases selected for QRT–PCR testing, but this observation remains preliminary.

Novel biological pathways in GRN+ FTLD-U

The present study of RNA expression profiles in human FTLD-U brain provides evidence for the commonality of some biological pathways in multiple neurodegenerative diseases. This accords with previous work implicating mechanisms such as oxidative stress (48), mitochondrial dysfunction (49), aberrant protein folding and accumulation (50), synaptic dysfunction (51) and others in several neurodegenerative diseases. Indeed, some have argued that neurodegenerative disorders are linked not just by final common pathways of cell dysfunction and death but by an actual network of interacting proteins in which abnormalities in one or a few proteins may cause dysfunction of the entire network (52,53). One recent study constructed an interacting proteomic map, with nodes representing various proteins implicated in AD, PD, ALS, HD, dentatorubropallidoluysian atrophy and prion disease (34); this map was then used to identify KEGG biological pathways enriched in disease-associated proteins. Of the 8 KEGG pathways identified in this manner, 3 (axon guidance, focal adhesion, regulation of actin cytoskeleton) were identified in our study as well, with an additional 2 (adherens junction, P = 0.07, Wnt signaling pathway, P = 0.08) falling just short of our P-value cutoff for significance.

Although these common pathways are important in understanding neurodegeneration as a whole, biological pathways specific to a particular disease may be key to developing targeted therapies. To that end, our study identified two biological pathways which may be specific to GRN+ FTLD-U. We found the categories of TGF-ß signaling and the cell cycle to be significantly enriched in genes upregulated in GRN+ FTLD-U. These pathways did not emerge from human brain gene expression studies of HD (35) or AD (29), general proteomics studies of neurodegeneration (34) or from analyses of our GRN− FTLD-U subset of cases.

Limitations of postmortem brain studies

One limitation of studies using postmortem tissue is that observed changes in gene expression may reflect either primary events in pathogenesis or secondary responses such as inflammation, cell death and neuronal loss. In addition, we used a regional sampling approach which by definition involves sampling of mixed populations of glial and neuronal cells. We chose this approach to minimize the amount of RNA degradation, which is greater with microdissection techniques. However, in doing so, we considered the possibility that primary pathophysiological changes might be masked by secondary changes of less biological importance.

Two lines of evidence from our present study support the idea that most of the genes and pathways identified have biological relevance. First, we found a distinct molecular phenotype for the GRN+ cases, which is unlikely to be due simply to non-specific inflammation and cell death, since the GRN− cases were clinically and histopathologically similar and handled the same way. Similarly, the presence of a distinct GRN+ FTLD-U expression signature made it unlikely that our findings are simply the result of neuronal cell loss, causing a more ‘glial’ signature in diseased brains, since both GRN+ and GRN− cases exhibited comparable amounts of neuronal cell loss. Second, many microarray studies have found genes involved in stress or immune response and apoptosis to be upregulated in their datasets (29,31), raising the issue of whether microarrays on postmortem tissue simply reflect the end-stage events of vulnerable cells dying. While categories of response to stress and apoptosis did appear in our biological pathway analysis, they were much less significantly enriched and did not meet our stringent statistical criterion for inclusion.

RNA extraction and hybridization

Frozen tissue samples were homogenized for 30–60 s using a PowerGen 35 (Fisher Scientific). RNA was isolated using TRIzol (Invitrogen) and RNeasy Mini columns (Qiagen), according to standard manufacturer protocols. All subsequent steps were conducted as described in the Affymetrix GeneChip Expression Analysis Technical Manual (Affymetrix). Briefly, 3 µg of total RNA from each sample were used to prepare biotinylated fragmented cRNA with products from Affymetrix according to manufacturer protocols. Microarrays (Affymetrix Human Genome U133A) assessing expression of 22 277 transcripts were hybridized for 16 h at 45°C. Chips were washed and stained, and hybridization signals were collected with a GeneChip 3000 7G scanner (Affymetrix).

Microarray quality control

Measures were taken both prior to microarray hybridization and after hybridization to ensure quality control. After RNA isolation, RNA purity and integrity were assessed by spectrophotometric measurement of 260/280 nm OD ratios and by capillary electrophoresis on an Agilent 2100 Bioanalyzer. All samples used for further analysis had a 260/280 ratio of >1.9, with sharp ribosomal RNA peaks (Supplementary Material, Fig. S2), virtually undetectable low molecular weight RNA, and RIN numbers >8 (55). After microarray hybridization, 3′/5′ ratios of GAPDH were calculated, and only chips with ratios <3 were retained. Finally, we used a quality assessment algorithm based on weights and residuals from robust regression models of gene expression to identify outlier chips (56), which were excluded from analysis. Four samples were eliminated from analysis after all quality control measures; of these, two were identified by all three methods, implying that initial RNA quality may be the most important factor.

Microarray data analysis

Statistical analyses of gene expression were performed using open source R software packages available from http://www.bioconductor.org (57). Fluorescence intensity measures of gene expression were normalized and quantified by robust multi-array analysis (58), using the affy package (59). To identify transcripts differentially expressed between disease (all FTLD-U, GRN+ subset, GRN− subset) and controls for each brain region, we used the limma package (60), which adopts a Bayesian approach. Our model corrected for gender and age. FDR were calculated based on nominal P-values as described (35) and by using a class permutation test to adjust for co-expressed genes. R-scripts for these analyses are available on request. In addition, principal components analysis and hierarchical cluster analysis using Spearman and Pearson correlation coefficients was performed using Partek software, version 6.3 Copyright © 2007 (Partek, Inc., St Louis, MO, USA).

Biological pathway analysis

We used the DAVID 2007 program available at http://david.abcc.ncifcrf.gov/ (61) to assign differentially expressed genes to biological pathways and processes annotated in the GO and KEGG databases, available at http://www.geneontology.org (62) and http://www.genome.jp/kegg/ (63), respectively. DAVID uses a Fisher’s exact test to find overrepresentation of differentially expressed genes within each biological pathway represented in the GO and KEGG databases. In addition, because of the hierarchical organization of GO categories, simple lists tend to result in similar, redundant categories. DAVID curates these GO categories into related groups based on this hierarchical structure, allowing for a more biologically meaningful analysis. We retained those pathways with a P-value of 0.05 or less (or, in the case of GO categories, functional groups with a median P-value of 0.05 or less) as statistically significant.

We are grateful to the Penn Microarray Facility and Dr Don Baldwin for advice and technical assistance with array samples. We thank Dr Bruce Miller for recruiting patients who were studied. Finally, we thank our patients and their families whose generosity has made this work possible.

Conflict of Interest statement: None declared.