Generation and characterization of sporadic AD-iPSCs
Dermal fibroblasts were isolated from an 82-year-old woman diagnosed with final stage AD. The cell line was named NFH-46, and lack of AD-related mutations, such as APP, PSEN1 and PSEN2 [1,5], was confirmed by direct sequencing analysis (Additional file 1). HLA haplotype analysis in the AD donor did not reveal any association of HLA alleles to Morbus Alzheimer. The HLA-alleles HLA-A*01:01,*03:01; B*08,*35, C*04:01,*07:01, DRB1*03:01,*11:01 were found in NFH-46. However, the Alzheimer-related HLA-alleles HLA-A*02, HLA-B*07 and HLA-C*07:02 could not be detected.
AD-iPSCs were generated by retroviral transduction using the classical Yamanaka cocktail [27], which includes the four transcription factors OCT4, KLF4, SOX2, and c-MYC, as demonstrated previously [28]. In a single reprogramming experiment several colonies exhibiting hESC-like morphologies were identified and manually picked for expansion and characterization. Two iPSC lines, AD-iPS5 and AD-iPS26B, were successfully established from this reprogramming experiment and characterized with respect to pluripotency-associated properties. Both lines exhibited hESC-like morphologies (Figure 1), telomerase activity (Additional file 2), alkaline phosphatase (AP) activity (Additional file 3a), expression of pluripotency-associated markers NANOG, SSEA4, TRA-1-60, and TRA-1-81 (Figure 2), expression of pluripotency-associated genes such as NANOG, POU5F1, SOX2, LIN28, TDGF1, DPPA4, FGF4, GDF3, LEFTY1, LEFTY2 (Additional file 4) and the genetic fingerprinting pattern of the parental NFH-46 fibroblasts (Additional file 3b).
Finally, the transcriptomes of the AD-iPSC lines are similar to hESCs (H1 and H9) and to iPS lines previously generated from control NFH-2 fibroblasts [28] (Additional file 5).
The ability to differentiate into almost all tissue types as a hallmark of human pluripotent stem cells was analyzed employing embryoid bodies (EBs) based differentiation in-vitro and teratoma formation in-vivo. The AD-iPSC lines were able to differentiate in-vitro into all three embryonic germ layers, as detected by the expression of marker proteins specific for ectoderm (b-TUBULIN III and NESTIN), for mesoderm (Smooth Muscle Actin (SMA) and T/Brachyury), and endoderm (Alpha feto protein (AFP) and SOX17) (Additional file 6).
Finally, both AD-iPSC lines successfully generated teratomas (Additional file 7). For AD-iPS5, the presence of known endoderm-associated structures appeared unclear. However, this must not necessarily imply an impairment of this line towards endoderm differentiation in-vivo, since the teratoma assay itself is not standardized [29]. Moreover, in the in-vitro differentiated cells from AD-iPS5, SOX17 and AFP, both protein markers representative of endoderm, could be detected. Thus, we consider AD-iPS5 to be pluripotent.
Chromosomal analysis of AD-iPS5 revealed the loss of a gonosome, probably the X chromosome, because during mitosis they revealed a normal female karyotype. This is in agreement with our previous study showing that iPSC lines generated from old donors are more likely to contain chromosomal aberrations [28]. In nine mitoses, a very small supernumerary marker chromosome (sSMC) was found besides the monosomy X (Additional file 8). It is unlikely that the presence of a small supernumaray maker chromosome has an effect on AD-iPS5 as sSMCs are a common phenomenon in human. The karyotypes of the second iPSC line AD-iPS26B and the parental cells NFH-46 were normal (Additional file 8).
Generation of neuronal cells from AD-iPSCs (AD-iPS neurons)
We derived neuronal cells from AD-iPSCs in one experiment to address the potential of these to reflect neuropathological features found in neuronal cells of sporadic AD patients. As a control, neuronal cells were derived from the female hESC line H9 in one differentiation experiment. The neuronal cells were generated following a recently published protocol, which requires the exposure to TGF-β receptor (SB431542) and MEK1/2 (PD0325901) inhibitors [30]. AD-iPSC lines (AD-iPS5 and AD-iPS26B) and H9 were successfully differentiated into neuronal cells. The efficiency of differentiation varied, as AD-iPS5 showed more pronounced neuronal differentiation than AD-iPS26B. All induced neuronal cells were positive for neuronal cell markers PAX6, NESTIN, and b-TUBULIN III as shown in Figure 3. Most of the neuronal marker genes in the heatmap shown in Figure 4a are expressed in a similar manner in AD-iPSC neurons and H9 neurons, hence confirming a neuronal differentiation of comparable quality across all used pluripotent cell lines.
Expression of neuronal marker genes in AD-iPS neurons and H9 neurons
The heatmap in Figure 4a shows the expression pattern of pre-synaptic and post-synaptic genes as well as markers of distinct subtypes of neural progenitors and mature neuronal cell types in AD-iPS neurons and H9 neurons of one differentiation experiment conducted. Neuronal markers FABP7, HES5, SOX2, PROM1 and ASCL1 are expressed in AD-iPS5 and AD-iPS26B neurons, however, FABP7 was not detected in H9 neurons. GALC, a marker of oligodendrocyte progenitor cells is expressed in all samples but lower in AD-iPS26B and H9 neurons. MAP2, a marker of neuronal dendrites is expressed in AD-iPS5 neurons but not in neuronal cells of AD-iPS26B and H9. Moreover, markers of retinal ganglion cells (POU4F2), dopaminergic neurons (TH) and glutamatergic neurons (SLC17A6) are expressed in AD-iPS5 neurons (Figure 4a). In addition, neural genes such as NeuN (HRNBP3 or FOX1), GFAP and GAD1, GAD2 (GABA-ergic genes) on the one hand and specific AD-related neuronal genes such as CALBINDIN1 and 2 (also known as calretinin) as well as SST (somatostatin or SRIF) on the other were analyzed. SRIF-positive interneurons are inhibitory neurons which express GAD1 and/or GAD2 as well as CALB2 [31] and are the most affected subtypes of neurons in AD [32]. To confirm the array-derived heatmap data we analyzed relative gene expression by real-time PCR (Figure 4b) using the samples of the same neuronal differentiation experiment which were hybridized for transcriptome analysis. By matching the gene expression to adult brain RNA, again the H9-derived neuronal cells did not express subtype-specific neuronal genes. The astrocyte-specific gene GFAP was barely detected in all neural cells compared to the mRNA level of the adult and AD brain (Figure 4b). The neuronal cells from AD-iPS5 and AD-iPS26B were positive for SYNAPSIN I, vGLUT2 (SLC17A6) and GAD2 and have the same transcript level as the AD brain for CALB2 and GAD1 (Figure 4b).
Proof-of-principle drug discovery using sporadic AD-iPSC derived neuronal cells
In addition to gene expression based analysis of Alzheimer-related genes we evaluated the possible medical relevance of our neuronal cell model in terms of drug discovery and selection of an appropriate therapy for sporadic AD. For this purpose, we subjected the induced neuronal cells to treatment with the γ-secretase-inhibitor Compound E (CE). The experiment was carried out once. Two distinct concentrations were employed: low 10 nM and high 100 nM. After one week of treatment, cells were lysed directly and the protein expression levels of p-tau, tau, p-GSK3B and GSK3B were investigated isolating samples from one well of one inhibitor treatment experiment conducted. Neuronal cells derived from both AD-iPSC lines (AD-iPS5 and AD-iPS26B) exhibited the expression of tau and p-tau, which were undetectable in the parental fibroblasts (NFH-46) (Figure 5). The results were confirmed using two antibodies, one recognizing only p-tau and the other binding to both the phosphorylated and non-phosphorylated forms of tau (Figure 5). Drug treatment did not result in any reduction of p-tau in AD-iPS5 derived neuronal cells. On the other hand, we observed a significant reduction of p-tau and tau expression in neuronal cells differentiated from AD-iPS26B compared to untreated cells following high doses of CE (Figure 5). The expression of both p-GSK3B and GSK3B was significantly higher in neuronal cells compared to parental fibroblasts NFH-46 (Figure 5). However, no evident change in their expression could be identified following CE treatment. Neuronal cells were identified based on the expression of NESTIN and b-TUBULIN III, however, expression was also detected in their parental fibroblast cells, thus confirming previous observations in fibroblasts [33].
Differential gene expression associated with Alzheimer-related pathways and biological processes in AD-iPSC neurons compared to H9 neurons
Using microarray based gene expression analysis we looked at the changes in the biological processes within the AD-iPS neuronal cells compared to H9 neurons as control. The hybridized samples were isolated from one neuronal differentiation experiment. Processes related to WNT signaling pathway and the alanine, aspartate and glutamate metabolism, in the case of AD-iPS5 neurons as well as the lysosome pathway and glutathion metabolism in the case of AD-iPS26B appeared to be up-regulated compared to H9 neurons. Pathways related to Alzheimer’s disease, Huntington’s disease, Parkinson’s disease and the proteasome were down-regulated in both AD-iPS neurons compared to H9 neurons (Additional files 9 and 10). AD-iPS5 and AD-iPS26B neurons showed up-regulated gene expression for biological processes such as neuronal fate commitment, neuron maturation, response to oxygen radical and/or response to reactive oxygen species (Additional file 9). In contrast to that, the UPS, apoptosis, and oxidative phosphorylation emerged as down-regulated biological processes (Additional file 10). Overall, these data suggest that AD neuronal cells exhibit alterations in key signaling pathways related to cell death, anabolism and catabolism in comparison to the healthy control.
AD-iPSC neurons show a distinct gene expression pattern of Alzheimer-associated genes of genome wide association studies compared to H9 neurons
To further analyze the reflection of Alzheimer-specific gene expression patterns in our iPSC-based model system we performed data mining to extract disease relevant gene expression using an Alzheimer gene list recently published by the European Alzheimer’s Disease Initiative (EADI) [34]. The cluster analysis in Figure 6 showed that AD-iPS5 neurons and AD-iPS26 neurons were more similar to each other than to H9 neurons. Basically there are six gene clusters: (i) a cluster of genes expressed in all experiments such as APOE and APP, (ii) a cluster of genes expressed in no experiment such as CASS4 and CR1, (iii) a cluster of genes expressed in both AD-iPSC neuron experiments but not in the H9 neuron experiment such as PTK2B and PICALM, (iv) a singleton cluster of HLA-DRB5 not expressed in both AD-iPS neuron experiments but expressed in the H9 neurons, (v) a singleton cluster of MEF2C not expressed in AD-iPS26B experiments but expressed in AD-iPS5 and the H9 neurons and (vi) a cluster of genes expressed in AD-iPS26B neurons but not in AD-iPS5 and H9 neurons containing genes such as SLC24A4 and ABCA7.
AD-iPSC neurons show down-regulation of genes involved in Alzheimer’s, Huntington’s and Parkinson’s disease compared to H9 neurons
Analysis of differences in the iPS-derived neurons when compared to the annotations Alzheimer’s disease, Parkinson’s disease and Huntington’s disease revealed that most genes down-regulated in AD-iPS5 vs. H9 neurons (Figure 7) and AD-iPS26B vs. H9 neurons (Figure 8) were common to all three neural disorders. Exclusively associated with Alzheimer’s disease were 16 genes in AD-iPS5 vs. H9 neurons and 10 genes in in AD-iPS26B vs. H9 neurons. These genes were APP, APOE, PSENEN, CDK5, HSD17B10, TNFRSF1A, PPP3CB, PPP3CC, CHP, GAPDH, CAPN2, CAPN1, ATP2A2, GSK3B, CALM3 and CALM2 in the experiment AD-iPS5 vs. H9 neurons and APP, CDK5, HSD17B10, CHP, GAPDH, NAE1, ATP2A2, GSK3B, CALM3 and CALM2 in the experiment AD-iPS26B vs. H9 neurons.
Brain allocation of Alzheimer-specific genes down-regulated in AD-iPSC neurons compared to H9 neurons
The expression in different brain regions of the Alzheimer-exclusive genes that were found to be down-regulated in AD-iPSC neurons of one differentiation experiment were investigated using the GNF/Atlas organism part. The expression of the largest set of genes was allocated to pons with 12% in the case of AD-iPS5 neurons for the genes APOE, APP, ATP2A2, CALM2, CALM3, CAPN2, CDK5, GAPDH, GSK3B and PPP3CB (Figure 4c) and 13% in AD-iPS26B neurons for the genes APP, ATP2A2, CALM2, CALM3, CDK5, GAPDH and GSK3B (Figure 4d). This was followed by globus pallidus with 9% in AD-iPS5 neurons for the genes APP, ATP2A2, CALM2, CALM3, CDK5, GAPDH and PPP3CB and 9% in AD-iPS26B neurons for the genes APP, ATP2A2, CALM2, CALM3, CDK5 and GAPDH. The percentages for medulla oblongata, prefrontal cortex and amygdala were found to be 5–8% (Figure 4c and d).
An Alzheimer-relevant functional protein association network can be built using an Alzheimer-specific gene set down-regulated in AD-iPSC neurons compared to H9 neurons
To further specify the reflected Alzheimer-related phenotype in our iPSC-based neuronal disease model we constructed protein association networks by means of the gene expression data generated from one differentiation experiment. Therefore, protein-interaction networks were generated using genes annotated with AD pathway that are down-regulated in AD-iPS5 neurons vs. H9 neurons, in AD-iPS26B neurons vs. H9 neurons as well as the overlap of both datasets. We successfully modeled the association of Alzheimer-related proteins within our cellular system in both AD-iPS neuronal differentiation experiments through the subsequent comparison to non-AD embryonic stem cell line H9 neuronal cells and further network construction by applying STRINGv9. The generated networks in Figures 9, 10 and 11 depict associations between proteins in a color code. The color of the line between proteins represents the following evidence categories: neighborhood in the genome (dark green line), gene fusion (red line), co-occurrence across genomes (dark blue line), co-expression (black line), experimental/biochemical data (purple line), association in curated databases (light blue line) and co-mentioning in PubMed abstracts/textmining evidence (light green line).
The protein association networks built from genes down-regulated in AD-iPS5 vs. H9 neurons in Figure 9 contains 18 more Alzheimer-related proteins than the AD-iPS26B vs. H9 neurons network in Figure 10. These proteins are NDUFB10, NDUFA9, NDUFB8, NDUFB9, ATP5G1, CAPN2, UQCRQ, NDUFA1, CAPN1, NDUFS7, SDHB, TNFRSF1A, CASP3, APOE, SDHD, PPP3CB, PPP3CC and PSENEN (Figure 9). However, the two proteins NDUFA6 and NAE1 are only part of a network built from the AD-iPS26B vs. H9 neurons dataset (Figure 10). Interestingly, both interaction networks contain APP and GSK3Β as well as CDK5 and HSD17B10. In the AD-iPS5 vs. H9 neurons and not in the AD-iPS26B vs. H9 neurons network APP is depicted to be associated with CASP3 and APOE by experimental evidence and textmining evidence as well as with PSENEN, however, only via textmining evidence (Figure 9). In addition, only the AD-iPS26B vs. H9 neurons network depicts the association of APP and NAE1 through co-expression, database, experimental and textmining evidence (Figure 10). Furthermore, only in the AD-iPS5 vs. H9 neurons and not in the AD-iPS26B vs. H9 neurons network functional associations between CASP3 and CDK5, CAPN2, TNFRSF1A, CALM3, APOE and COX4I1 that are proven by textmining evidence could be found. TNFRSF1A which is only a part of the network in Figure 9 is associated with CDK5, CALM3 by textmining evidence and with CASP3 with additional database evidence. Exclusively in this protein association network APOE, which plays a major role in Alzheimer pathogenesis, is associated with CASP3, CALM3, CALM2 and GAPDH by textmining evidence whereas CAPN2 is associated with CASP3, CALM2 and CALM3 by textminig evidence. Surprisingly, no associations could be found for CAPN1 in this network. Additional proteins only part of the AD-iPS5 vs. H9 neurons network are PPP3CC and PPP3CB both of which are associated with each other by co-occurrence, database, experimental and textmining evidence. They are associated with CALM2 and CALM3 with experimental and textmining evidence. PSENEN, a subunit of the γ-secretase complex, occurs only in this protein association network and is associated with UQCRH via textmining and co-expression evidence. Additional proteins exclusively part of the AD-iPS5 vs. H9 neurons network are NDUFS7, NDUFA9, NDUFB8, NDUFB10, NDUFA1, NDUFB9, UQCRQ, ATP5G1, SDHB and SDHD. These are part of a complex protein association network-cluster mainly consisting of proteins involved in oxidative phosphorylation (Figure 9). Interestingly, NDUFA6 is only part of the protein association network based on the genes in the AD-iPS26B vs. H9 neurons dataset (Figure 10).
The network built from the genes overlapping between the AD-iPS5 vs. H9 neurons and AD-iPS26B vs. H9 neurons datasets in Figure 11, shows associations with experimental and textmining evidence of APP with HSD17B10, GAPDH, CDK5, GSK3B and SNCA. In addition, database evidence to prove the association of APP, SNCA and GAPDH could be found. Associations with textmining evidence between CDK5 with CALM2 and CALM3 as well as between GAPDH and ATP2A, CALM2, CALM3, SDHA, ATP5B and ATP5J could be found by our method. Furthermore, the interaction of GAPDH with ATP5B and ATP5J is associated with co-expression evidence in this network. GAPDH is associated with CHP via experimental evidence. Additional Alzheimer-related genes are interconnected to a complex protein association network cluster similar to Figures 9 and 10. consisting of proteins like NDUFB3, ATP5E, NDUFB5, UQCRC, NDUFB6, NDUFB7, ATP5B, NDUFAB1, NDUFB2 that are mainly involved in oxidative phosphorylation or in the electron transport chain in mitochondria. In our Alzheimer-related protein association network in Figure 11 we found experimental evidence association between SNCA and NDUFB6 as well as the co-expression evidence interactions of HSD17B10 with NDUFV2, NDUFB7, UQCRH and ATP5J. These associations connect Alzheimer-specific APP, GSK3B, CDK5, CALM2, CALM3 and ATP2A2 with proteins involved in Alzheimer related failure of the function of mitochondrial processes of the respiratory chain of the protein association cluster. A further association to proteins involved in oxidative phorphorylation are the interactions of GAPDH with ATP5J, ATP5B and SDHA (Figure 11).
A subset of UPS-related genes is down-regulated in AD-iPSC neurons compared to H9 neurons
The cluster analysis of UPS-associated genes assembled both AD-iPS neurons datasets into a cluster separated from the H9 neurons. Genes were divided into three clusters (i) 36 genes which have lower gene expression values in both AD-iPS neuron compared to H9 neurons, among them PSMC1, PSMA5, NEDD8. (ii) 2 genes characterized by higher expression in both AD-iPS neuron compared to H9 neurons: PSMD5, PSMB9. (iii) 25 genes the expression of which varies between the three samples. The last cluster is subdivided into (i) 5 genes with high expression values in H9 neurons and AD-iPS5 neurons but not in AD-iPS26B neurons - PSME2, PSMD8, (ii) 9 genes with high expression values in H9 neurons and AD-iPS26B neurons but not in AD-iPS5 neurons - PSMD14, PSMD4, (iii) 5 genes with a low gene expression values in H9 neurons and AD-iPS26B neurons and a high gene expression in AD-iPS-5 neurons - PSMD4, SUGT1, (iv) 6 genes with low gene expression values in H9 neurons and AD-iPS5 neurons and a high gene expression values in AD-26B neurons - PSME1, PSMB10 (Figure 12).