New lessons learned from disease modeling with induced Pluripotent Stem Cells

Abstract

Cellular reprogramming and generation of induced pluripotent stem cells (iPSCs) from adult cell types has enabled the creation of patient-specific stem cells for use in disease modeling. To date, many iPSC lines have been generated from a variety of disorders, which have then been differentiated into disease-relevant cell types. When a disease-specific phenotype is detectable in such differentiated cells, the reprogramming technology provides a new opportunity to identify aberrant disease-associated pathways and drugs that can block them. Here, we highlight recent progress as well as limitations in the use of iPSCs to recapitulate disease phenotypes and to screen for therapeutics in vitro.

Introduction

Understanding the molecular basis of many diseases have been hampered by the lack of appropriate in vitro cell culture models that accurately reflect the disease phenotypes. The availability of such models is also crucial for the discovery and development of therapeutics. However, primary human cells and especially disease-bearing ones are difficult to obtain and propagate in culture for extended periods of time. Since their discovery by Takahashi and Yamanaka and application to the human context, induced pluripotent stem cells (iPSCs) have emerged as a way to generate disease-specific cell types [1–5]. With this technology, somatic cells isolated from patients (in most cases fibroblasts from skin biopsies), can be reprogrammed into a pluripotent state by overexpressing four transcription factors (Oct4, Sox2, Klf4 and c-Myc)[1]. The resulting iPSCs can be expanded in culture virtually indefinitely, and then be differentiated into cell types of interest.

The first disease-specific iPSCs were derived from patients with familial amyotrophic lateral sclerosis (ALS) and a number of genetic diseases with either Mendelian or complex inheritance [6,7]. Since then an ever-growing number of disease-specific iPSCs are being generated from primary cell samples from patients afflicted with a variety of genetically inherited and sporadic diseases affecting virtually every organ system (Table 1). In general, patient-specific iPSCs are then differentiated to the key cell types that are affected in the disease in question. The next step is the identification of disease-associated phenotypes that are readily detectable by molecular and/or cellular assays. Once a robust assay is identified, larger scale screens can be undertaken to discover key disease pathways and therapeutics (Figure 1).

Table 1. Summary of published human disease-specific iPSCs.

Disease	Molecular defect	Disease phenotype observed / References
ADA-severe combined immunodeficiency	Mutations in ADA	Not determined [7].
Adrenoleukodystrophy, X-linked	Mutation in ABCD1	Very long chain fatty acid accumulation in oligodendrocytes [50].
Alzheimer's disease, familial	Mutation in PS1 or PS2, Duplication of APP	Increased amyloid β42 secretion[51]. Higher levels of the pathological markers amyloid-β(1–40), phospho-tau and active glycogen synthase kinase-3β [52].
Alzheimer's disease, sporadic	Unknown	Higher levels of the pathological markers in one out of two patient-derived lines [52].
Amyotrophic lateral sclerosis (ALS), familial	Mutation in SOD1 or VAPB	Not determined [6]. VAPB protein levels are reduced in ALS8-derived motor neurons [53].
Angelman syndrome	Loss of maternal UBE3A allele	Reduced UBE3A expression upon neuronal differentiation [28].
Atypical Werner syndrome	Mutation in LMNA	Alterations in the nuclear membrane, slow proliferation, senescence [54].
α1-antitrypsin deficiency	Mutation in α1-antitrypsin	Loss of α1-antitrypsin expression [10][55].
Becker muscular dystrophy	Mutation in DMD	Not determined [7].
β-thalassaemia	Deletion in β-globin gene	Not determined [56] [57].
Chronic granulomatous, X-linked	CYBB deficiency	Neutrophils differentiated from X-CGD iPSCs lack ROS production [44].
Crigler–Najjar syndrome	Mutation in UGT1A1	Not determined [58].
Cystic fibrosis	Mutations in CFTR	Not determined [55].
Diabetes, Type1	Multifactorial; unknown	Not determined [7][59].
Down syndrome	Trisomy 21	Not determined [7].
Dyskeratosis congenita	Mutations in DKC1, TERT or TCAB1	Increased expression of telomerase components [30]. Progressive telomere shortening, loss of self-renewal [31].
Duchenne muscular dystrophy	Mutation in DMD	Not determined [7]. Loss of dystrophin expression in iPSC-derived muscle tissue [13].
Dystrophic epidermolysis bullosa	Mutations in COL7A1	Lack of expression of type VII collagen [60][61].
Emanuel Syndrome	Partial trisomy 11;22	Not determined [62]
Fanconi Anemia	FANCA or FANCC deficiency	Corrected loss of FANCA function [20][21].
Familial dysautonomia	Mutation in IKBKAP	Defects in neural crest differentiation migration [17].
Familial hypercholesterolaemia	Mutation in gene encoding LDL receptor	Reduced ability to incorporate LDL in iPSC-derived hepatoctyes [10].
Fragile X syndrome	Trinucleotide (CGG) expansion, silencing of FMR1	Not determined [27][29].
Friedreich's ataxia (FRDA)	Trinucleotide GAA repeat expansion in FXN	Reduced FXN mRNA expression [63].
Gaucher's disease	Mutation in β-glucocerebrosidase	Impaired lysosomal protein degradation, accumulation of α-synuclein, and neurotoxicity [64].
Glycogen storage disease type 1A	Deficiency in glucose-6-phosphate	Hyperaccumulation of glycogen [10] [58].
Gyrate atrophy	Mutation in OAT	Not determined [65].
Hereditary tyrosinaemia type 1	Mutation in FAH	Not determined [58].
Huntington's disease	Trinucleotide expansion in huntingtin gene	Not determined [7][66].
Hutchinson–Gilford progeria syndrome	Mutation in LMNA	Increased senescence, progerin accumulation, DNA damage, nuclear abnormalities [67][68][54].
Inherited dilated cardiomyopathy	Mutation in LMNA	Increased senescence and apoptosis in iPSC-derived fibroblasts, nuclear abnormalities [54].
LEOPARD Syndrome	Mutation in PTPN11	Increased cardiomyocyte size, decreased MAPK signaling [69].
Lesch–Nyhan syndrome (carrier)	Heterozygous for HPRT1	Not determined [7].
Long QT syndromes 1 and 2	Mutations in KCNQ1 or KCNH2	Increased cardiomyocyte depolarization [11][18]
MPS type I (Hurler syndrome)	IDUA deficiency	Not determined [70].
MPS type IIIB	α-N-acetylglucosaminidase deficiency	Defects in storage vesicles and Golgi apparatus [71].
Osteogenesis imperfect	Mutations in COL1A1 or COL1A2	Not determined [72]
Parkinson's disease, sporadic	Unknown	Not determined [7][15][16][73].
Parkinson's disease, familial	Mutations in LRRK2 or PINK1 or triplication of SNCA	Sensitivity to oxidative stress in LRRK2-mutant neurons[74]. Impaired mitochondrial function in PINK1-mutant dopaminergic neurons[75]. Increased α-synuclein protein in neurons[76].
Patau syndrome	Trisomy 13	Not determined [62].
Polycythaemia vera	Mutation in JAK2	Enhanced erythropoiesis [77]
Pompe disease	Mutations in GAA	High levels of glycogen and defective cellular respiration in iPSC-derived cardiomycyte-like cells [78].
Prader-Willi syndrome	Paternal deletion of 15q11-q13	Imprint disorder [79][28].
Progressive familial hereditary cholestasis	Unknown	Not determined [58].
Retinitis pigmentosa	Mutations in RP1, RP9, PRPH2 or RHO	Decreased numbers of differentiated rod cells [80].
Rett syndrome	Mutation in MECP2	Decreased synapse number, reduced spine density [9][81][82].
Schizophrenia	Unknown	Decreased neuronal connectivity, neurite number, and PSD95 and glutamate receptor expression [19].
Scleroderma	Unknown	Not determined [55].
Shwachman–Bodian– Diamond syndrome	Mutations in SBDS	Not determined [7].
Sickle cell anaemia	Mutation in β-globin gene	Not determined [55][46].
Spinal muscular atrophy (SMA)	Mutations in SMN1	Loss of SMN gene expression, reduced size and number of motor neurons [8].
Spinocerebellar ataxia type 3	Trinucleotide expansion in MJD1 gene	Neuron-specific formation of SDS-insoluble aggregates [14].
Timothy Syndrome	Mutations in CACNA1C	Increased cardiomyocyte depolarization [12].
Turner Syndrome	Monosomy X	Not determined [62].
Wilson's disease	Mutations in ATP7B	Defective copper transport in iPSC-derived hepatocyte-like cells [83].
Warkany Syndrome 2	Trisomy 8	Not determined [62].

Figure 1. Disease modeling and drug discovery using patient-derived iPSCs.

Generation of an iPSC-based disease model starts with cells isolated from patients, usually by a skin punch biopsy. Upon reprogramming, several iPSC clones are selected, expanded and characterized. High-quality iPSCs are then differentiated into mature cell types exhibiting a disease-specific phenotype that is readily detected by cellular and/or molecular assays. High-throughput screens based on such assays can be carried out to discover therapeutics that reverse the disease phenotypes. Hits from these screens are candidates for lead optimization by medicinal chemistry, and then further preclinical studies.

In this review we will highlight some of the salient themes that emerge from studies reported to date and point out some of the promises and limitations of using iPSCs in disease modeling.

Identification of disease relevant cellular phenotypes

For iPSC-based disease modeling to be meaningful, a relevant cellular or molecular phenotype must be observable in either the derived iPSCs or their differentiated progeny. Monogenic diseases affecting a specific cell type that can efficiently be derived from pluripotent stem cells are the best candidates for this approach. Some examples include neurological diseases such as spinal muscular atrophy (SMA) and Rett syndrome, metabolic diseases such as α1-antitrypsin deficiency, familial hypercholesterolemia and glycogen storage disease type 1A, and cardiovascular diseases such as Timothy syndrome and Type 1 and 2 Long QT syndrome. Upon differentiation into disease-relevant cell types, most of these iPSCs manifest an observable phenotype. SMA-specific iPSCs displayed progressive loss of motor neurons upon differentiation, which may be similar to the developmental loss of this cell type during disease progression[8]. Neurons derived from Rett syndrome-specific iPSCs had reduced spine density, smaller soma size and altered calcium signaling when compared to controls[9]. For the metabolic diseases mentioned above, differentiation of iPSCs into hepatocyte-like cells resulted in several phenotypes including aggregation of misfolded α₁-antitrypsin in the endoplasmic reticulum, deficient LDL receptor-mediated cholesterol uptake, and elevated lipid and glycogen accumulation[10]. In the modeling of cardiovascular diseases, disease-specific phenotypes manifest themselves in the differentiated cardiomyocytes such as the prolonged action potentials of Long QT and Timothy syndrome-specific cells [11,12]. In certain cases, the disease phenotype could also be observed upon spontaneous differentiation of iPSCs in teratoma assays, as recently shown for the absence of dystrophin expression in muscle-like tissues derived from Duchenne muscular dystrophy-specific iPSCs [13].

A recent study by Brustle and colleagues highlights some of the advantages of iPSCs-based disease modeling. Spinocerebellar ataxia (also called Machado-Joseph disease) is a late-onset neurodegenerative disease caused by the expansion of polyglutamine (CAG) repeats in the MJD1 gene. Upon proteolytic cleavage, the MJD1 gene product ataxin3 (ATXN3) is thought to cause SDS-insoluble protein aggregates which then have a critical role in neurodegeneration. To explore the role of neuron specific proteolysis in initiating the disease process, Koch et al. first derived iPSCs from MJD patients and showed that the expanded MJD allele was expressed in both the pluripotent cells and their differentiated neuron-like progeny[14]. They then showed that glutamate-induced excitation of these differentiated cells initiates Ca²⁺ - dependent proteolysis of ATXN3, which is followed by aggregate formation. Interestingly, this observed phenotype could be abrogated by the inhibition of Ca²⁺ -dependent calpain proteases. This cell model manifest a neuron-specific phenotype; aggregate formation was not observed in the iPSCs, fibroblasts or glia, thus providing an explanation for the neuron-specific manifestation of the disease phenotype[14]. Although the molecular mechanisms linking microaggregate formation to late-stage neurodegeneration remains elusive, this study shows how aspects of a late-onset disease can also be modeled in a cell-type specific manner.

Unfortunately, there have been numerous studies in which disease-specific iPSCs were generated but no phenotype has been assessed or observed (Table 1). For some diseases, this is most likely due to lack of suitable differentiation protocols to the affected cell type and/or specific assays to detect the molecular defect. For example, modeling hematological disorders such as sickle-cell anemia or β-Thalassemia, and ADA-SCID has been hampered by a lack of pure in vitro derivation protocols for the specific blood lineages most relevant to these conditions (beta-globin expressing adult red blood cells, and T cells, respectively). In another example, iPSCs generated from sporadic Parkinson's disease patients could in fact be efficiently differentiated into dopaminergic neurons, but the resulting cells did not show significant differences in various in vitro assays, when compared to control iPSC-derived cells [15,16]. It remains to be seen whether other late-onset diseases that have complex environmental and genetic causes will yield detectable in vitro phenotypes.

Disease-specific iPSC-derived cells as discovery and screening platforms

When a robust disease phenotype is identified in patient-specific iPSCs or their differentiated progeny, such cells can then be used to discover new therapeutic compounds or test existing candidates. One of the first demonstrations that disease-specific iPSCs can be used to test therapeutic compounds came from iPSCs generated from Familial dysautonomia (FD), a rare genetic disorder affecting sensory and autonomic neurons due to mutations in the I-κ-B kinase complex-associated protein (IKBKAP) gene. These mutations result in a tissue-specific splicing defect that lowers the levels of IKAP protein. Lee et al. derived iPSCs from a FD patient which they then differentiated into neural precursors cells that exhibited three FD-associated phenotypes: defective IKBKAP splicing, decreased rate of neurogenesis, and reduced migration[17]. Such iPSC-derived cells were then used to test a number of candidate compounds for their ability to affect any of these phenotypes. One compound, kinetin, could partially reverse the aberrant splicing and the neurogenic differentiation and migration defects[17].

A growing number of disease-specific IPSC-derived cells are being used to test candidate small molecules. For example, motor neuron survival defects observed in SMA-specific cells were partially abrogated by the application of two compounds, valproic acid and tobramycin, which increased the level of SMN protein in the patient-derived iPSCs [8]. In Rett-iPSC derived neurons, IGF1 treatment increased synapse formation[9]. In Long QT syndromes, cardiomyocyte-specific defects could be attenuated by beta-blockers and several ion channel blockers [11,18]. The antipsychotic loxapine has been shown to ameliorate the diminished neuronal connectivity and decreased neurite number phenotype of Schizophrenia-iPSC-derived neurons[19]. All in all, these examples show that the stage is set for the use of disease-specific iPSC-derived cells to be used in larger screens aimed to discover new therapeutic agents that will block the disease-associated phenotypes. Given the disease-relevance of human iPSC-based drug screens, there is considerable hope for therapeutic development.

Limitations of iPSCs

As enumerated above, iPSC-based disease modeling offers significant potential in drug discovery. However, as it currently stands, there are also limitations to this technology, which may preclude its use for some types of diseases. Certain genetic lesions inhibit or may even preclude the derivation of iPSCs from patients by interfering with the reprogramming process itself. A case in point comes from the attempt to model Fanconi anaemia (FA), the most common genetic bone marrow failure syndrome that is caused by recessive autosomal or X-linked mutation in one of 13 genes in the FA pathway. This pathway plays an important role in DNA repair, and it was recently shown that reprogramming leads to activation of the FA pathway, increased DNA double strand breaks and senescence [20]. Initial attempts to reprogram FA fibroblasts were unsuccessful unless the genetic defect was first corrected with a viral vector [21]. Therefore it was concluded that restoration of the FA pathway is a prerequisite for iPSC generation from FA patients [21]. Subsequent work has shown that it is possible to reprogram FA fibroblasts, albeit at a very low efficiency [20]. Nevertheless, these studies suggest that in cases where the diseases caused by mutations that affect DNA repair and senescence pathways, it may be difficult to generate iPSCs from patient cells. It should be noted that a number of small molecules have been shown to improve reprogramming efficiency of human fibroblasts and could therefore prove useful in such cases [22–26].

Another example of the limitation of iPSC-based disease modeling comes from studies of Fragile X syndrome, the most common form of inherited mental retardation. Fragile X is caused by silencing of the FMR1 gene due to CGG triplet expansion in its 5′UTR, which then results in aberrant DNA methylation and accumulation of repressive histone marks. Urbach et al. successfully derived iPSC lines from Fragile X patients, and observed that the FMR1 gene remained inactive and was thus refractory to factor-based epigenetic reprogramming [27]. This contrasted to the active FMR1 gene expression and down-regulation following differentiation in embryonic stem cell (ESC) lines derived from embryos affected by Fragile X [27]. Thus, in this context, Fragile X-iPSCs could not model the silencing of the FMR1 gene that occurs during development, but ESC lines could. There is also evidence indicating that other epigenetic disorders such as imprinting defects are not reset by somatic cell reprogramming [28]. Sheridan et al. recently derived additional Fragile X iPSCs and observed clonal variation among the iPSC lines with respect to CGG-repeat length, CpG methylation and expression levels of FMR1 suggesting that iPSCs may not faithfully reproduce these phenotypes present in the original fibroblasts [29]. Furthermore, in one case, the patient derived fibroblast line was a heterogeneous mixture of normal and expanded-repeat cells, and this line yielded individual iPSC clones bearing either the normal or expanded-repeat lengths, suggesting that reprogramming can capture single target cells with variable properties [29]. This point highlights the clone-to-clone variability that can confound and complicate disease models while also being potentially useful as a platform for line-to-line comparison from specific individuals.

Whether clonal variation among different iPSC lines from patients influences the observed disease phenotype remains an important concern. Agarwal et al. and Batista et al. both reprogrammed fibroblasts from patients afflicted with dyskeratosis congenita (DC), a premature aging syndrome caused by mutations in telomerase components [30,31]. While both groups observed increases in expression of multiple members of the core telomerase machinery in the reprogrammed iPS cells, there was clonal variation with respect to maintenance of telomere length. Agarwal et al. observed that some iPSCs recovered enough telomerase activity to regrow telomeres, whereas in the study by Batista et al. DC-specific iPSCs continued to show telomere decay [32]. Given these different outcomes, one must remain wary of the clonal variation that results from the technical inefficiency and infidelity of reprogramming when iPSC-based disease models are generated. In general, the number and location of viral integrations, heterogeneity in the starting somatic cell populations, passage numbers and culture conditions as well as genetic alterations introduced during reprogramming are all likely to contribute to the clonal variability among iPSC lines[33][34]. In addition, there is growing evidence that epigenetic remodeling can be incomplete, thus influencing the consistency of disease phenotypes [35-39]. Therefore, multiple clones from multiple patients will most likely need to assessed before a truly representative and faithfully reprogrammed iPSC-based disease model is established.

One potential way to overcome the variability among disease-specific iPSC lines, especially with respect to monogenic diseases, is to generate genetically defined sibling cell lines. Engineered zinc finger nucleases (ZFNs) and transcription activator-like effector nucleases (TALENs) have emerged as important tools to achieve “genome editing” in pluripotent stem cells [40-42]. This technology can be used to correct the disease-specific mutations by gene targeting or generate de novo disease-specific mutations in wild-type lines. This approach has recently been to used to model susceptibility variants for familial Parkinson's disease by genetically modifying single base pairs in the alpha-synuclein gene[43]. Genome editing has also been employed for X-linked chronic granulomatous disease where a wild-type gene has been inserted into a ZFN-induced break induced at a potential “safe harbor” locus in iPSCs and for sickle-cell anemia in which case correction of mutations in the beta-globin gene has been achieved by either ZFNs or by homologous recombination [44-46]. Such panels of isogenic disease and control cell lines that differ only at precise disease-associated loci are going to be useful in studying the effects of the mutations in question.

Modeling complex diseases using iPSCs will be especially challenging. First of all, robust differentiation protocols for disease-relevant cell types may be lacking. Modeling of diabetes would be greatly facilitated by the discovery of reliable differentiation methods for generating pancreatic β-cells in vitro from iPSCs. Although progress has been made on this front, routine production of β-cells from diabetic patients to be used in genetic and drug screens are still some time away [47]. In addition, certain disease-associated phenotypes might arise from non-cell autonomous interactions between two different cell types. In such cases, it will be essential to differentiate the iPSCs into multiple different cell lineages and to generate in vitro co-culture assays. For example, Di Giorgio et al. have shown that in a minority of ALS cases resulting from SOD1 mutations, neurodegeneration may be due to the non-cell autonomous effect of the glial cells on the motor neurons [48]. Finally, for late-onset diseases that have large environmental components, it will be important to identify ways by which disease latency can be shortened or environmental conditions recreated by the applications of exogenous stresses [49].

Conclusions

The growing number of studies reporting the generation of disease-specific iPSCs is a testament to the potential of this technology in drug and pathway discovery. However, it is also becoming increasingly apparent that iPSC-based disease modeling will have to incorporate more reliable protocols for directed differentiation into mature cell types in order to realize its full potential. Furthermore, more robust disease phenotypes, especially for complex and common diseases, will have to be detected and assayed. Despite the progress described above, it should be noted that all the published studies to date have tested the effects of previously reported candidate compounds on disease-specific iPSCs. As such, these studies constitute only proof-of-principle demonstrations for larger higher throughput screens to discover novel agents that can potentially reverse the disease phenotypes observed in iPSCs-derived cells. The full benefit of the iPSC platform for disease modeling and drug development, although promising, remains to be proven.