Hargis and Blalock have analyzed publicly available transcriptome data from humans and rodents and from purported transgenic rodent models of Alzheimer’s disease to assess whether the transgenic rodents are, in fact, modelling the disease. They found good concordance between changes in gene expression seen for human AD brains and between humans and rodents in terms of changes in gene expression during aging. However, there was little concordance between human AD and transgenic rodent models of AD. To quote from their conclusions:
“Among five transgenic mouse models, three (J20, Tg2576, 3xFAD) show poor concordance with each other as well as with human AD. Two (5xFAD and CK-p25) show moderate agreement with one another and with human AD, although this agreement is primarily centered on upregulated genes.”
Hargis and Blalock’s findings may possibly be regarded as unsurprising when we reflect on our slow progress in understanding of the mechanisms underlying AD and the history of failure of translation of candidate AD drugs from rodents to humans. However, as someone currently attempting to create animal models of AD, this paper makes plain to me that the mechanistic assumptions underlying the transgenic rodent models are very likely wrong. The development of the models has largely been guided by whether or not they show the Aβ plaque and tau tangle cellular pathologies that are written into the definition of the disease. To obtain those pathologies we have created genetically bizarre models with multiple transgenes containing multiple mutations and driven by exogenous promoters. Clearly Hargis and Blalock’s analysis shows us that this approach is misguided. (By extension, it also undermines the idea that the plaque and tangle pathologies are actually central to Alzheimer’s disease.) The AD research field needs to return to first principles in genetics and to analyze in detail models that are as similar to the genetic state of human AD as possible. This means, for familial AD models, analyzing single, heterozygous fAD-like mutations in endogenous fAD orthologous genes. Only by analyzing single genetic differences in detail can we properly interpret the effects of those differences. Even single-transgene models can represent multiple simultaneous genetic differences (due to exogenous promoters, multiple sequence differences between the inserted and endogenous genes, insertion site effects, etc.) that confound functional interpretation.
Hargis and Blalock’s study has very significant implications for researchers currently testing hypotheses or developing drugs using transgenic rodent AD models and should be kept in mind when interpreting the research results from the thousands of papers that have previously exploited these models.
Thanks to Dr. Lardelli for the kind comments on our recent publication. We would also like to thank the careful researchers who have shared such well-annotated and thorough transcriptome studies, and the NIH for providing the supporting data repository infrastructure. The parent work on which these analyses were performed clearly required a great deal of effort and technical prowess from the originating research labs.
The highly transparent data we used was provided at the resolution of the individual subject/observation. This resource likely represents an ideal test bed for evaluating animal model translatability in a variety of human conditions beyond Alzheimer's, and clearly can be used to assess reproducibility, so long as we, as a research community, are willing to fund, perform, and publish replication studies. This is in keeping with the NIH’s recently formalized transparency/reproducibility/ rigor requirements and the well-publicized “omics” problem (e.g., McCarthy et al., 2008; Heller et al., 2014; Birney et al., 2016). Other “big data” technologies also may benefit from similar attention (e.g., Eklund et al., 2016).
This publication grew out of a scientific shopping trip. We were writing a proposal (not funded!) to study an Alzheimer’s disease intervention. Because our lab focuses on normal aging, we weren’t as familiar with animal models of Alzheimer’s disease. We solicited advice from various researchers who do have extensive experience with these animals as to which model to use, but the experts’ suggestions conflicted with one another. So we decided to perform a less-biased assessment using transcriptional profiles. We had observed that there appeared to be reasonable agreement in normal brain aging between rodent and human hippocampal transcriptomes, but this assessment was in no way formalized. Thus, we developed an analysis approach to assess similarity across transcriptional profiles in the context of what would be expected by chance across thousands of measures. We were able to confirm that rodent brain aging is similar to that seen in the human at the transcriptional level. However, our results comparing human AD transcriptomes to those of mouse models were frustratingly poor, suggesting that druggable targets we had identified based on human transcriptional studies would not be well modeled at the intervention or outcome levels in these mice.
In addition to Dr. Lardelli’s clear concerns, a couple of additional caveats may be worth considering regarding this lack of agreement. First, the mouse models are typically not aged. However, as mentioned above, there is a consistent aging transcriptome effect. Of course, there are also consistent effects on behavioral, electrophysiological, sleep architecture, hormonal, and neuroinflammatory (among others) parameters across multiple species with aging. So the lack of an aged background upon which to overlay genetic insults may represent a fundamental model-to-human dissimilarity in AD research. Additionally, the human studies we used to define the transcriptional profile were based on late-onset idiopathic AD, while the mice were developed (with the exception of the CK-p25) based on familial AD mutations. Unfortunately, we were not able to identify any early onset familial human AD transcriptional profiles in public databases. If someone is aware of such a transcriptional profile, or knows of available brain tissue for such profiles to be measured, please let me know. It is critical at this point to at least determine if human LOAD and human EOAD may represent frankly different diseases, albeit with common outcomes.
References:
Birney E, Smith GD, Greally JM.
Epigenome-wide Association Studies and the Interpretation of Disease -Omics.
PLoS Genet. 2016 Jun;12(6):e1006105. Epub 2016 Jun 23
PubMed.
Eklund A, Nichols TE, Knutsson H.
Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates.
Proc Natl Acad Sci U S A. 2016 Jul 12;113(28):7900-5. Epub 2016 Jun 28
PubMed.
Heller R, Bogomolov M, Benjamini Y.
Deciding whether follow-up studies have replicated findings in a preliminary large-scale omics study.
Proc Natl Acad Sci U S A. 2014 Nov 18;111(46):16262-7. Epub 2014 Nov 3
PubMed.
McCarthy MI, Abecasis GR, Cardon LR, Goldstein DB, Little J, Ioannidis JP, Hirschhorn JN.
Genome-wide association studies for complex traits: consensus, uncertainty and challenges.
Nat Rev Genet. 2008 May;9(5):356-69.
PubMed.
Comments
The University of Adelaide
Hargis and Blalock have analyzed publicly available transcriptome data from humans and rodents and from purported transgenic rodent models of Alzheimer’s disease to assess whether the transgenic rodents are, in fact, modelling the disease. They found good concordance between changes in gene expression seen for human AD brains and between humans and rodents in terms of changes in gene expression during aging. However, there was little concordance between human AD and transgenic rodent models of AD. To quote from their conclusions:
“Among five transgenic mouse models, three (J20, Tg2576, 3xFAD) show poor concordance with each other as well as with human AD. Two (5xFAD and CK-p25) show moderate agreement with one another and with human AD, although this agreement is primarily centered on upregulated genes.”
Hargis and Blalock’s findings may possibly be regarded as unsurprising when we reflect on our slow progress in understanding of the mechanisms underlying AD and the history of failure of translation of candidate AD drugs from rodents to humans. However, as someone currently attempting to create animal models of AD, this paper makes plain to me that the mechanistic assumptions underlying the transgenic rodent models are very likely wrong. The development of the models has largely been guided by whether or not they show the Aβ plaque and tau tangle cellular pathologies that are written into the definition of the disease. To obtain those pathologies we have created genetically bizarre models with multiple transgenes containing multiple mutations and driven by exogenous promoters. Clearly Hargis and Blalock’s analysis shows us that this approach is misguided. (By extension, it also undermines the idea that the plaque and tangle pathologies are actually central to Alzheimer’s disease.) The AD research field needs to return to first principles in genetics and to analyze in detail models that are as similar to the genetic state of human AD as possible. This means, for familial AD models, analyzing single, heterozygous fAD-like mutations in endogenous fAD orthologous genes. Only by analyzing single genetic differences in detail can we properly interpret the effects of those differences. Even single-transgene models can represent multiple simultaneous genetic differences (due to exogenous promoters, multiple sequence differences between the inserted and endogenous genes, insertion site effects, etc.) that confound functional interpretation.
Hargis and Blalock’s study has very significant implications for researchers currently testing hypotheses or developing drugs using transgenic rodent AD models and should be kept in mind when interpreting the research results from the thousands of papers that have previously exploited these models.
University of Kentucky
Thanks to Dr. Lardelli for the kind comments on our recent publication. We would also like to thank the careful researchers who have shared such well-annotated and thorough transcriptome studies, and the NIH for providing the supporting data repository infrastructure. The parent work on which these analyses were performed clearly required a great deal of effort and technical prowess from the originating research labs.
The highly transparent data we used was provided at the resolution of the individual subject/observation. This resource likely represents an ideal test bed for evaluating animal model translatability in a variety of human conditions beyond Alzheimer's, and clearly can be used to assess reproducibility, so long as we, as a research community, are willing to fund, perform, and publish replication studies. This is in keeping with the NIH’s recently formalized transparency/reproducibility/ rigor requirements and the well-publicized “omics” problem (e.g., McCarthy et al., 2008; Heller et al., 2014; Birney et al., 2016). Other “big data” technologies also may benefit from similar attention (e.g., Eklund et al., 2016).
This publication grew out of a scientific shopping trip. We were writing a proposal (not funded!) to study an Alzheimer’s disease intervention. Because our lab focuses on normal aging, we weren’t as familiar with animal models of Alzheimer’s disease. We solicited advice from various researchers who do have extensive experience with these animals as to which model to use, but the experts’ suggestions conflicted with one another. So we decided to perform a less-biased assessment using transcriptional profiles. We had observed that there appeared to be reasonable agreement in normal brain aging between rodent and human hippocampal transcriptomes, but this assessment was in no way formalized. Thus, we developed an analysis approach to assess similarity across transcriptional profiles in the context of what would be expected by chance across thousands of measures. We were able to confirm that rodent brain aging is similar to that seen in the human at the transcriptional level. However, our results comparing human AD transcriptomes to those of mouse models were frustratingly poor, suggesting that druggable targets we had identified based on human transcriptional studies would not be well modeled at the intervention or outcome levels in these mice.
In addition to Dr. Lardelli’s clear concerns, a couple of additional caveats may be worth considering regarding this lack of agreement. First, the mouse models are typically not aged. However, as mentioned above, there is a consistent aging transcriptome effect. Of course, there are also consistent effects on behavioral, electrophysiological, sleep architecture, hormonal, and neuroinflammatory (among others) parameters across multiple species with aging. So the lack of an aged background upon which to overlay genetic insults may represent a fundamental model-to-human dissimilarity in AD research. Additionally, the human studies we used to define the transcriptional profile were based on late-onset idiopathic AD, while the mice were developed (with the exception of the CK-p25) based on familial AD mutations. Unfortunately, we were not able to identify any early onset familial human AD transcriptional profiles in public databases. If someone is aware of such a transcriptional profile, or knows of available brain tissue for such profiles to be measured, please let me know. It is critical at this point to at least determine if human LOAD and human EOAD may represent frankly different diseases, albeit with common outcomes.
References:
Birney E, Smith GD, Greally JM. Epigenome-wide Association Studies and the Interpretation of Disease -Omics. PLoS Genet. 2016 Jun;12(6):e1006105. Epub 2016 Jun 23 PubMed.
Eklund A, Nichols TE, Knutsson H. Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates. Proc Natl Acad Sci U S A. 2016 Jul 12;113(28):7900-5. Epub 2016 Jun 28 PubMed.
Heller R, Bogomolov M, Benjamini Y. Deciding whether follow-up studies have replicated findings in a preliminary large-scale omics study. Proc Natl Acad Sci U S A. 2014 Nov 18;111(46):16262-7. Epub 2014 Nov 3 PubMed.
McCarthy MI, Abecasis GR, Cardon LR, Goldstein DB, Little J, Ioannidis JP, Hirschhorn JN. Genome-wide association studies for complex traits: consensus, uncertainty and challenges. Nat Rev Genet. 2008 May;9(5):356-69. PubMed.
Make a Comment
To make a comment you must login or register.