Mouse models rarely mimic the transcriptome of human neurodegenerative diseases: A systematic bioinformatics-based critique of preclinical models.

European Journal of Pharmacology ∎ (∎∎∎∎) ∎∎∎–∎∎∎

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Q1 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

Contents lists available at ScienceDirect

European Journal of Pharmacology journal homepage: www.elsevier.com/locate/ejphar

Pharmacogenomics

Mouse models rarely mimic the transcriptome of human neurodegenerative diseases: A systematic bioinformatics-based critique of preclinical models Terry C. Burns n, Matthew D. Li, Swapnil Mehta, Ahmed J. Awad, Alexander A. Morgan Department of Neurosurgery, Institute for Stem Cell Biology and Regenerative Medicine, Stanford University School of Medicine, 300 Pasteur Drive, Stanford, CA 94305, USA

art ic l e i nf o

a b s t r a c t

Article history: Received 10 February 2015 Received in revised form 12 March 2015 Accepted 12 March 2015

Translational research for neurodegenerative disease depends intimately upon animal models. Unfortunately, promising therapies developed using mouse models mostly fail in clinical trials, highlighting uncertainty about how well mouse models mimic human neurodegenerative disease at the molecular level. We compared the transcriptional signature of neurodegeneration in mouse models of Alzheimer's disease (AD), Parkinson's disease (PD), Huntington's disease (HD) and amyotrophic lateral sclerosis (ALS) to human disease. In contrast to aging, which demonstrated a conserved transcriptome between humans and mice, only 3 of 19 animal models showed significant enrichment for gene sets comprising the most dysregulated up- and down-regulated human genes. Spearman's correlation analysis revealed even healthy human aging to be more closely related to human neurodegeneration than any mouse model of AD, PD, ALS or HD. Remarkably, mouse models frequently upregulated stress response genes that were consistently downregulated in human diseases. Among potential alternate models of neurodegeneration, mouse prion disease outperformed all other disease-specific models. Even among the best available animal models, conserved differences between mouse and human transcriptomes were found across multiple animal model versus human disease comparisons, surprisingly, even including aging. Relative to mouse models, mouse disease signatures demonstrated consistent trends toward preserved mitochondrial function protein catabolism, DNA repair responses, and chromatin maintenance. These findings suggest a more complex and multifactorial pathophysiology in human neurodegeneration than is captured through standard animal models, and suggest that even among conserved physiological processes such as aging, mice are less prone to exhibit neurodegeneration-like changes. This work may help explain the poor track record of mouse-based translational therapies for neurodegeneration and provides a path forward to critically evaluate and improve animal models of human disease. & 2015 Published by Elsevier B.V.

Keywords: Huntington's disease Parkinson's disease Alzheimer's disease Amyotrophic lateral sclerosis Neurodegeneration Bioinformatics Gene set enrichment

1. Introduction Animal models provide a critical platform upon which translational efforts for treating human neurodegenerative diseases are built. While there is no substitute for studying true human biology, animal models provide opportunities for experimentation that are often impossible in human patients. Transgenic animals carrying human mutations provide an opportunity to understand mechanisms underlying human disease pathogenesis. Moreover, animal models routinely serve as gatekeepers to putative therapies being considered for clinical trials (Burns and Verfaillie, in press). Unprecedented progress has been made in the past two decades based in part on animal models of Huntington's disease (HD), amyotrophic

n

Corresponding author. Tel.: þ 1 612 812 7223. E-mail addresses: [email protected], [email protected] (T.C. Burns).

lateral sclerosis (ALS), Parkinson's disease (PD) and Alzheimer's disease (AD). With this new understanding, hundreds of pharmaceutical agents that have shown promise in preclinical animal models of neurodegenerative disease have progressed to clinical trials. Unfortunately, almost none have proven effective in humans. This stark reality has prompted a thoughtful re-evaluation of the role of mouse models of neurodegeneration and neuroinflammation (Cavanaugh et al., 2014; Doody et al., 2014; Gladstone et al., 2002; O’Collins et al., 2006; Panza et al., 2014; Scott et al., 2008). Comparison of transcriptome data between experiments is made challenging by the wide variety of methodologies employed for comprehensive transcriptome analysis. Even identical protocols routinely yield incomparable results between experiments due to batch effects. Comparisons between species are made even more challenging by a lack of clear homologs for many genes, speciesspecific differences in the function of certain genes and timing of cellular responses to stimuli, as well as a lack of standardized

http://dx.doi.org/10.1016/j.ejphar.2015.03.021 0014-2999/& 2015 Published by Elsevier B.V.

Please cite this article as: Burns, T.C., et al., Mouse models rarely mimic the transcriptome of human neurodegenerative diseases: A systematic bioinformatics-based critique of preclinical models. Eur J Pharmacol (2015), http://dx.doi.org/10.1016/j.ejphar.2015.03.021i

67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87

T.C. Burns et al. / European Journal of Pharmacology ∎ (∎∎∎∎) ∎∎∎–∎∎∎

2

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

methodological approaches to inter-species comparisons. Nevertheless, continually improved annotations and the development of robust bioinformatics techniques now permit meaningful comparisons of transcriptional responses between species. We recently took advantage of the rapidly expanding inventory of transcriptional profiles for human neurodegenerative diseases to perform a meta-analysis focusing on AD, PD, HD, and ALS (Li et al., 2014). This work identified a common neurodegenerative disease module that is shared across human neurodegenerative diseases. Given that transcriptome data additionally exist for several animal models of neurodegenerative diseases, we sought here to address the following questions: 1) Can transcriptional responses between species be meaningfully compared, using the relatively conserved aging process as a positive control? 2) Are the transcriptional signatures of human neurodegenerative diseases appropriately reflected in animal models? 3) Do variables such as disease stage or brain region analyzed substantially confound the results of comparative analyses? 4) Could alternate mouse models exist that more closely mimic human neurodegeneration than standard models of AD, PD, ALS and HD? 5) Do reproducible differences exist between human neurodegeneration and mouse models across multiple diseases?

2. Materials and methods 2.1. Human neurodegenerative disease meta-signatures We previously performed a meta-analysis of human neurodegenerative diseases (Li et al., 2014). We used the individual ranked gene lists for each disease generated in the course of our prior analysis as a baseline against which to compare the transcriptome of the corresponding animal models. The total ranked gene lists for each disease are provided in Supplemental Table S1. In the “discovery” portion of our prior meta-analysis, the effect sizes for each individual AD, PD, HD and ALS data set were then combined to determine the pooled effect size for human neurodegeneration using the random effects inverse-variance technique. The resultant ranked gene lists are provided in Supplemental Table S1.

ALS; and (2) the microarray platform had accessible probe-to-gene mapping annotations available for use through the GSEA Java application v2.0.14. Data sets with o3 mice per group were excluded if another sufficient data set was available for that disease/model at a comparable or later time point. When multiple time points were available, we included only the latest available time point, with the exception of E-GEOD-31372, E-GEOD-4390, and E-MEXP-453, for which each of the 2–3 available time points were evaluated in parallel. All samples included were from fresh whole tissue, with the exception of E-MEXP-453, a study of motor neurons purified by single-cell laser capture. When data for multiple brain regions were available, we selected one or two brain regions most relevant to the disease pathophysiology. The ArrayExpress identifiers (e.g. E-GEOD, E-MEXP) for each study evaluated are included in tables where relevant throughout the manuscript. Table 1 provides the complete list of data sets utilized for this study. These were compared to the human neurodegenerative disease meta-analysis ranked gene lists, as provided in Supplemental Table S1. 2.4. Enrichment for human disease signatures in animal models Gene set enrichment analysis is a well-established technique for comparing genomic responses between independent samples, diseases and species (Yu et al., 2011). We used Gene Set Enrichment Analysis (GSEA) (Subramanian et al., 2005) to evaluate for the enrichment of (1) human neurodegenerative disease signatures in human neurodegenerative disease ranked gene lists; (2) human aging signatures in mouse aging samples; and (3) human neurodegenerative disease signatures in mouse models of neurodegeneration. Additionally, to evaluate for alterations in enrichment for functional gene modules and cell-type signatures, sets of genes found in the literature to be co-regulated in healthy (Oldham et al., 2008) and AD brain samples (Zhang et al., 2013) were included, along with gene sets representing the most highly differentially expressed genes in specific prospectively isolated CNS cell types as well as gliosis and microglial activation. GSEA was performed using default settings, including 1000 permutations based on gene set. The Pre-rank tool was used for GSEA based upon ranked gene lists. Analysis was based upon probe set IDs from the processed public data sets, or gene symbol for ranked gene lists. False discovery rates (FDR) o0.05 were considered significant. Gene sets employed for our analysis are provided in Supplemental Table S2. 2.5. Spearman's rank correlation

2.2. Human aging signatures We previously identified 3 human aging brain data sets each including at least 30 patients: E-GEOD 30272, E-GEOD 11882, and EGEOD 1572 (Li et al., 2014). For each gene in the neurodegeneration data sets, the Kendall tau coefficient between the log2 transformed gene signal intensity and age was determined using the “Kendall” R package. The resulting ranked gene lists are provided in Supplemental Table S1. 2.3. Animal models of neurodegeneration We searched the public data repository ArrayExpress (November 2014) for gene expression microarray data sets from mouse models of neurodegenerative disease using search terms “neurodegeneration,” “Alzheimer,” “Parkinson,” “Huntington,” and “amyotrophic lateral sclerosis.” Additionally, mouse models identified from review of relevant literature were included as well as select studies of mouse brain aging. Available processed data sets were included if they met the following criteria: (1) samples were from mouse CNS tissue samples, including any region of forebrain for AD and HD, midbrain or forebrain samples for PD, and spinal cord samples for

Spearman's rank correlation analysis of the mouse and human gene expression datasets was performed using R/Bioconductor. For every ranked gene list, genes with duplicate entries were omitted (i.e., probe sets mapping to multiple genes). Using the 1315 genes that were in common across all 51 lists, we calculated Spearman's rank correlation for each pairwise comparison between ranked gene lists. Heat maps featuring hierarchical clustering (Euclidean distance) of Spearman's rank correlations were generated using the pheatmap R package. 2.6. Comparison of mouse and human neurodegenerative diseases signatures GSEA is based in part upon the generation of a ranked gene list with the most highly upregulated genes at the top and the most highly downregulated genes at the bottom of the ranked gene list. The order of genes in the list thereby summarizes the transcriptome for given condition. In order to determine which genes were differentially regulated in response to human disease versus mouse models of disease, we compared the percentile rank position of each gene in the ranked gene lists for mouse models and human diseases. By subtracting the human rank percentile from the mouse rank


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

3

Table 1 List of datasets. Data sets employed in this study are accessible through ArrayExpress using the E-GEOD identification numbers provided. When available, citation information is also provided, in addition to details regarding the model signature evaluated. In all cases, data are compared to age-matched wild-type or untreated mice from the same study. Additionally, human meta-analysis ranked gene list signatures of human AD, PD, ALS and HD are provided in Supplemental Table S1, based upon data from our prior meta-analysis. ( Li et al., 2014) CX, cortex; HP, hippocampus; ST, striatum; Br, brain – not otherwise specified; WB, whole brain; VTA, ventral tegmental area; SN, substantia nigra; MB, midbrain; SC, spinal cord; na, no corresponding publication identified for the data set. Accession #

Model type

Age/timepoint

Location

Reference

Alzheimer's models E-GEOD-52022 E-GEOD-36981 E-GEOD-53480 E-GEOD-15128 E-GEOD-31372 E-GEOD-36237 E-GEOD-23847 E-GEOD-56772

Tg6799 3xTg Tg4510 Tg2576 TgCRND8 Tg2576 Dicer KO Tg4510

8m 14m 4m 17m 70, 80, 150d 5m unk 6m

HP HP HP HP Cx HP CX HP

Noh et al. (2014) na Polito et al. (2014) Pereson et al. (2009) na na Hebert et al. (2010) na

Parkinson's models E-GEOD-6041 E-GEOD-19534 E-GEOD-17542 E-GEOD-13033 E-GEOD-4758 E-GEOD-7707

PINK1 KO SNCA KO MPTP HtrA2 KO aSyn MPTP

18m 12m 10d 1m 4m 14w

ST ST SN, VTA Br SN MB, ST

Gispert et al. (2015) Kurz et al. (2010) Phani et al. (2010) Moisoi et al. (2009) na na

Huntington's models E-GEOD-3621 E-GEOD-7958 E-GEOD-9038 E-GEOD-9375 E-GEOD-9857 E-GEOD-10202 E-GEOD-29681 E-GEOD-38219 E-GEOD-44306 E-GEOD-62210

R6/1 Q92 HdhQ111 Hdh4/Q80 R6/2 CHL2 Q150 R6/2 R6/2 N171-HD82Q R6/2

27w 18m 3–10w 12m 12w 22m 12w 15w 20w 8w

FB ST ST ST ST ST ST CX HP ST

Hodges et al. (2008) Kuhn et al. (2007) Fossale et al. (2011) Kuhn et al. (2007) Kuhn et al. (2007) Kuhn et al. (2007) Labbadia et al. (2011) Mielcarek et al. (2013) Valor et al. (2013) Kurosawa et al. (2015)

ALS models E-MEXP-453 E-GEOD-50642 E-GEOD-4390 E-GEOD-3343

SOD1G93A MNs SOD1G93A SOD1G93A SOD1G93A

60, 90, 120d 80d 75, 110d 10w

SC SC SC SC

Perrin et al. (2005) De Oliveira et al. (2013) Lukas et al. (2006) na

Alternate models E-GEOD-39621 E-GEOD-23182

Niemann–Pick PRN/LPS

80–84d 18w þ6 h

HP HP

Alam et al. (2012) Lunnon et al. (2011)

Mouse aging E-GEOD-11291 E-GEOD-3253 E-GEOD-29075 E-GEOD-45044 E-GEOD-45043

Aging Aging Aging Aging Aging

30v5m 24v6m 18v3.5m 18v2m 18v2m

CX WB HP VTA SN

Barger et al. (2008) Godbout et al. (2005) Kohman et al. (2011) Gao et al. (2013) Gao et al. (2013)

Human aging E-GEOD-11882 E-GEOD-1572 E-GEOD-30272

h Aging h Aging h Aging

20–99y 26–106y 20–78y

HP CX CX

Berchtold et al. (2008) Lu et al. (2004) Colantuoni et al. (2011)

percentile, we obtained the “rank difference.” For example, if gene X was upregulated in mice and downregulated in humans, being 500 in a list of 10,000 genes (5th percentile) in mice, and 9500 in a list of 10,000 genes in humans (95th percentile), the rank difference score for that gene would be 90 (5 minus 95). All genes present in the mouse and human lists for a given species response comparison therefore earned a rank difference score between þ100 and 100 with the genes most differentially upregulated in the mouse versus human transcriptome comparison having the highest score and genes most differentially downregulated in the mouse versus human transcriptome comparison having the lowest score. We used this approach to compare mouse and human transcriptomes for multiple mouse models of neurodegeneration. The resulting “rank difference” lists are provided in Supplemental Table S3. These were then evaluated via GSEA using (1) curated gene sets from the Broad Institute's Molecular Signature Database (MSigDB; C2 collection)

(Subramanian et al., 2005); (2) gene sets generated by taking the top 250 genes from other rank difference gene lists; (3) mouse and human disease signatures; and (4) custom gene sets as detailed above. Additional functional analysis of these gene signatures was performed based on the top and bottom 250 genes of the rank difference using the Enrichr web tool (amp.pharm.mssm.edu/Enrichr) (Chen et al., 2013).

3. Results 3.1. Similarities between different human neurodegenerative diseases We recently performed a meta-analysis of brain gene expression data in human neurodegenerative diseases, and identified a


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132

4

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 Q5 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66


Fig. 1. Overview of data processing using gene set enrichment analysis (GSEA). Top Right: Ranked gene lists (from most up- to most-downregulated) were generated to summarize the transcriptome of mouse models of neurodegeneration as well as human neurodegeneration. The top and bottom 250 genes from the ranked gene list were used as gene sets for GSEA. A sample enrichment plot (top right) demonstrates positive enrichment for the human Alzheimer's UP gene set (H_AD_UP) in the human Huntington's disease ranked gene list. Black lines (“hits”) indicate the locations of genes from the H_AD_UP gene set, within in the ranked gene list for human Huntington's disease (NES 3.4, as shown in top left cell of Table (circled; illustration from Table 4D). Color coded heatmaps used throughout the manuscript are based upon normalized Enrichment score (NES) from GSEA. FDR q-values not show here, but also provided in tables throughout the manuscript. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

common neurodegeneration gene module shared among patients with AD, PD, HD, and ALS (Li et al., 2014). Using the same data sets employed for the “discovery” portion of our meta-analysis, we generated 4 ranked gene lists summarizing the transcriptional changes of AD, PD, HD and ALS, with each list containing the rankordered list of genes by fold-change for each disease (Fig. 1; Supplemental Table S1). The top and bottom 250 genes from each list were then input as “gene sets” in Gene Set Enrichment Analysis (GSEA) to evaluate the degree of enrichment of these transcriptional changes for each disease within the other ranked gene lists (Table 2). AD, PD, and HD and ALS gene lists all showed significant enrichment for the up- and down-regulated gene sets, respectively. Considering gene sets comprising both up- and down-regulated genes for each of 4 disease meta-signatures, the degree of enrichment between diseases was evaluated in 26 total disease versus disease comparisons, of which 24 showed highly significant enrichment at FDRo 0.00005, These findings are consistent with significantly overlapping transcriptional signatures across human neurodegenerative diseases. These individual disease signatures, in concert with the overall human neurodegeneration meta-analysis transcriptome, therefore provided a robust, reproducible and internally consistent benchmark against which to evaluate mouse neurodegenerative disease signatures.

3.2. Similarities between human and mouse aging brains Given that sporadic neurodegeneration does not naturally occur in mice and must be induced deterministically via some genetic anomaly or environmental stressor, we first wished to evaluate aging as a “positive control” to determine the consistency of mouse and human brain responses to stressors that are as analogous as possible. Indeed, well-established aging-related functional and histological changes occur in both mice and humans, including decreased memory, progressive atrophy, neuroinflammation, and impaired neurogenesis. To determine the correlation between mouse and human transcriptional changes during the nonmanipulated state of natural aging, we generated ranked gene lists from 3 independent aging brain data sets for humans (Berchtold et al., 2008; Colantuoni et al., 2011; Lu et al., 2004). By then taking 6 gene sets comprising the top 250 most highly up- and downregulated genes for each data set and performing GSEA on the other human aging data sets, we found all human aging data sets to be significantly enriched for each other, demonstrating the reproducibility of the human aging brain transcriptome (Table 3). Enrichment for these 6 gene sets was then evaluated in 3 independent mouse aging brain data sets (Barger et al., 2008; Godbout et al., 2005; Kohman et al., 2011). Significant enrichment (FDRo0.05) was


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

5

Table 2 Robust enrichment of CNS transcriptome signatures by GSEA across diseases and between species. TAGSEA comparing human neurodegenerative diseases. Gene sets comprising the top (UP) and bottom (DOWN) 250 genes from the ranked gene lists of human neurodegenerative diseases, as well as the up- and down-regulated components of the common neurodegeneration module, are significantly enriched in the ranked gene lists of other neurodegenerative diseases. Values shown in the upper panel are normalized enrichment score (NES), based on gene set enrichment analysis (GSEA). NES values are color coded by strength of enrichment: Red indicates positive enrichment and Blue indicates negative enrichment, as indicated by negative NES values. Corresponding FDR q-values are shown in the lower panel and color coded based on degree of significance with darker shading indicating more significant results. Bold in both panels indicates significant enrichment at FDRo 0.05. The color coding and annotations are used for all tables. Disease

AD

HD

PD

ALS

Σ ND

Tissue

CNS

CNS

CNS

CNS

CNS

Age Model Data source

n/a Human AD Meta-Analysis

n/a Human HD Meta-Analysis

n/a Human PD Meta-Analysis

n/a Human ALS Meta-Analysis

n/a Human AD+HD+PD+ALS Meta-Analysis

Gene sets NES

H_AD_UP H_HD_UP H_PD_UP H_ALS_UP COMM NEURODEGEN MOD UP H_AD_DOWN H_HD_DOWN H_PD_DOWN H_ALS_DOWN COMM NEURODEGEN MOD DOWN

5.35 3.82 3.31 2.88 3.73 -4.87 -3.27 -2.83 -2.29 -3.63

3.40 4.88 3.54 2.38 3.17 -3.21 -4.84 -3.43 -2.13 -3.88

3.07 3.66 4.98 2.53 3.53 -2.90 -3.60 -5.16 -1.08 -4.05

2.66 1.95 1.18 5.31 3.52 -2.52 -2.64 -2.04 -4.62 -3.16

4.27 4.47 4.27 3.85 4.09 -3.59 -3.97 -3.78 -2.96 -4.10

FDR q-val

H_AD_UP H_HD_UP H_PD_UP H_ALS_UP COMM NEURODEGEN MOD UP H_AD_DOWN H_HD_DOWN H_PD_DOWN H_ALS_DOWN COMM NEURODEGEN MOD DOWN

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.37 0.00

0.00 0.00 0.18 0.00 0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

found in 14 of 18 comparisons, with changes occurring in the same direction between mice and humans in all cases (Table 3). Notably, enrichment for the human aging gene sets was strongest in the oldest mouse transcriptome (6/6 significant in a study comparing 30-month-old to 5-month-old mice E-GEOD-11291, and the weakest in the least aged mouse transcriptome (3/6 significant in a study comparing 18-month-old mice to 3-month-old mice). We thus concluded that the transcriptome of even a relatively “mild” human neuropathological state such as aging could be detected in aged mouse brains using this methodology. 3.3. Inconsistent gene expression patterns in mouse models of neurodegeneration We next turned our attention to mouse models of neurodegeneration. We identified several publically available complete transcriptome data sets for mouse models of PD, AD, ALS and HD, details and citations for which are provided in Table 1. In order to meet inclusion criteria, each data set provided gene array data for both diseased mice and age-matched controls. Using these mouse data sets, we performed GSEA, to evaluate the degree of enrichment of mouse models for human neurodegenerative disease signatures, including the top 250 up- and down-regulated genes for each human neurodegenerative disease metasignature, as well as the up- and down-regulated portions of the human common neurodegeneration module (Fig. 1). Results were highly variable and are summarized by disease below, and in Tables 4A–4D. A comprehensive integrated summary of all gene set enrichment analysis performed throughout the manuscript is additionally provided in Supplemental Table S4. 3.3.1. Alzheimer's disease Overall, the best correlation was found in AD Tg4510 mice, with similar results in two independent data sets (Table 4A). Appropriate significant enrichment for all 4 human neurodegeneration data sets

was also seen in Dicer KO mice. Interestingly, both Tg4510 data sets actually showed the highest enrichment for genes upregulated in human ALS, rather than AD as should have been expected. Among downregulated gene sets, both Tg4510 and dicer KO mice showed higher concordance with human HD than human AD. These findings suggest that although overlapping gene expression changes occur in Tg4510 mice, Dicer KO mice, and human AD, some of the unique characteristic features of human AD may not be optimally represented in these mouse models. None of the other seven mouse AD data sets evaluated showed significant enrichment for genes upregulated in AD; though two others showed significant enrichment for the upregulated genes of the human common neurodegeneration module; none of the remaining seven mouse AD data sets showed significant downregulation of the genes downregulated with human neurodegeneration.

3.3.2. Amyotrophic lateral sclerosis All of the available mouse model data sets for ALS were derived from the SOD1-G93A mouse (Table 4B). Of the 7 data sets evaluated, 3 represented a time-series of gene expression in purified motor neurons rather than spinal cord whole tissue. Of the remaining 4 data sets, only one (110 days) showed significant positive- and negative-enrichment for the up- and downregulated human ALS signatures, respectively. The remaining 3 data sets (70–80 days) each demonstrated correlations reaching significance (FDR o0.05) for either the up- or down-regulated human ALS gene set, but not both. These findings suggested that tissue harvested from end-stage but not early stage SOD1-G93A mice is robustly enriched for the human ALS signature—a signature, which itself is derived from end-stage tissue. Motorneurons from SOD1-G93A mice were positively enriched for genes upregulated in 4/4 human neurodegenerative diseases at 90 and 120 days. Interestingly, at 60 days, they were significantly enriched for genes upregulated in human AD, ALS and PD, but enrichment for human ALS-upregulated genes missed significance at this early


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132


6

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

Table 3 GSEA comparing human and mouse aging data sets. Gene sets comprising the top (UP) and bottom (DOWN) 250 genes from the ranked gene lists of human aging studies are significantly enriched in the ranked gene lists of mouse brain aging. Human aging

Mouse aging

Disease

Aging

Aging

Aging

Aging

Aging

Tissue

HP

CX

CX

CX

WB

HP

Age

20-99y

26-106y

20-78y

30v5m

24v6m

18v3m

Aging

Model

Human

Human

Human

WT mouse

WT mouse

WT mouse

Data source Gene sets

E-GEOD-11882

E-GEOD-1572

E-GEOD-30272

E-GEOD-11291

E-GEOD-3253

E-GEOD-29075

NES

H_AGING_E-GEOD-11882_UP H_AGING_E-GEOD_1572_UP H_AGING_E-GEOD_30272_UP H_AGING_E-GEOD_11882_DOWN H_AGING_E-GEOD_1572_DOWN H_AGING_30272_DOWN

4.47 2.50 2.42 -3.85 -2.06 -2.03

4.16 5.15 2.46 -4.12 -5.03 -2.80

3.57 2.88 5.23 -3.72 -3.00 -5.04

2.37 2.22 1.66 -1.48 -1.84 -1.72

2.22 1.85 1.61 -1.79 -1.54 -1.45

1.23 1.62 1.19 -1.60 -1.29 -1.53

FDR q-val

H_AGING_E-GEOD-11882_UP H_AGING_E-GEOD_1572_UP H_AGING_E-GEOD_30272_UP H_AGING_E-GEOD_11882_DOWN H_AGING_E-GEOD_1572_DOWN H_AGING_30272_DOWN

0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.03 0.00 0.01

0.00 0.00 0.02 0.01 0.04 0.06

0.19 0.01 0.21 0.03 0.12 0.04

Table 4A Variable enrichment for human neurodegenerative disease signatures in mouse models of neurodegeneration. (A): Alzheimer’s disease. Two data sets utilizing the Tg4510 mouse, as well as Dicer knockout mice both show strong enrichment for genes upregulated in human neurodegenerative diseases, as well as variably significant negative enrichment for genes downregulated in human Alzheimer’s disease. Other animal models show variable enrichment patterns. Of note, TgCRND8 mice show increasingly aberrant positive enrichment for genes downregulated in human disease with increasing age from 70 to 150 days.

Table 4B Variable enrichment for human neurodegenerative disease signatures in mouse models of neurodegeneration. (B): ALS. The SOD1-G93A model of ALS shows strong positive enrichment for human neurodegeneration at 110 days. At earlier time points (70–80 days), the signature is less consistent in 3 independent studies. Motorneurons (MNs) from these mice show strong enrichment for the upregulated component of neurodegeneration from 60 to 120 days with increasing significance of enrichment over time, but show positive enrichment for genes downregulated in the human ALS whole tissue signature.


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

7

Table 4C Variable enrichment for human neurodegenerative disease signatures in mouse models of neurodegeneration. (C): Parkinson’s Disease. Striatal samples at 14 weeks after MPTP show generally appropriate enrichment for human degeneration, though specific enrichment for the human Parkinson’s disease upregulated genes misses significance. The signature of all other data sets, including 14 week midbrain MPTP samples demonstrate variable responses that fail to follow the pattern of human disease.

Table 4D Variable enrichment for human neurodegenerative disease signatures in mouse models of neurodegeneration. (D): Huntington’s Disease. No mouse models analyzed replicated the transcriptional response of human Huntington’s disease. At least a trend toward upregulation of the downregulated portion of the human neurodegeneration module was present in 7 of 10 datasets. Results for Q92 mice were virtually the inverse of those seen in human samples.

time point. Genes downregulated in human ALS actually moved in the opposite direction, being significantly upregulated in SOD1G93A motorneurons, at 90 and 120 days even though they were significantly downregulated in SOD1-G93A whole spinal cord tissue at 110 days, suggesting that the gene expression changes within individual cell types may not be easily inferred by evaluation of whole tissue transcriptome data.

3.3.3. Parkinson's disease None of the PD models showed significant enrichment for the top 250 genes upregulated in human PD (Table 4C), and in fact half of the data sets showed a tendency toward downregulation of the human PD-upregulated gene set. Nevertheless, striatum harvested 14 weeks after MPTP administration was significantly enriched for genes upregulated in each of the other 3 neurodegenerative diseases even though enrichment for human PD missed significance. Moreover, appropriate negative enrichment was seen in this data set for the downregulated human PD gene set, making the overall gene expression pattern analogous to the human condition. By contrast, both the

Pink1 and SCNA knockout (KO) mice demonstrated significant upregulation of genes that normally go down in human PD, as did midbrain harvested 14 weeks after MPTP. Alpha-synuclein mutant mice showed a trend toward movement of both human PD up- and down-regulated genes in directions opposite to those observed in human PD.

3.3.4. Huntington's disease Among mouse models of HD none showed both appropriate positive and negative enrichment for human HD up- and downregulated genes, respectively (Table 4D). Only 2 of 10 data sets, N171HD82Q hippocampus and HdhQ111þ /þ striatum, showed significant enrichment for the human HD upregulated gene set. However, both of these HD models also showed significant up-regulation of genes normally downregulated in human neurodegeneration. Two striatal datasets using R6/2 mice showed appropriate downregulation of genes from the human HD downregulated gene set, showed upregulation of genes normally downregulated in human PD and ALS and the common neurodegeneration module. This contrasts with the


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132

8

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66


Fig. 2. Hierarchical clustering and heat map of Spearman's correlations between data sets. There were 51 data sets evaluated throughout the study, including human metaanalysis signatures of HD, PD, ALS, AD, 3 human forebrain aging datasets, 3 mouse forebrain aging data sets, mouse aging in VTA and SN, as well as mouse models of AD, HD, PD, and ALS, and prion and/or LPS and NP, were compared based upon 1315 genes commonly identified across all data sets. All human data sets, including aging and neurodegeneration clustered together, separately from all mouse data sets. Tg4510 and SOD1-G93A, together with prion disease mice clustered most closely with the human CNS data sets, in agreement with findings obtained using GSEA. Hierarchical clustering dendrogram is based upon Euclidean distance of the Spearman's rank correlations.

pattern for human HD wherein downregulated gene sets for all neurodegenerative diseases are strongly downregulated (Table 4D). Indeed, this tendency of certain mouse models to upregulate genes from the downregulated portion of the human neurodegeneration module was observed in multiple data sets for AD, PD, and HD, as well as in ALS MNs, though was most prominent in HD, where 7 of 10 data sets showed at least a trend toward up-regulation of these human-downregulated genes. One of these mice, Q92, additionally showed significant downregulation of all human up-regulated disease gene sets, yielding a transcriptional picture opposite to that of neurodegeneration. This mouse still clustered with other Huntington's mice in hierarchical clustering analysis based on Spearman's correlation analysis (Fig. 2, Supplemental Table S5).

3.4. Alternate mouse models of neurodegeneration Given the relatively small number of identified animal models that appropriately replicated the human signature without aberrant up-regulation of human disease-downregulated genes, we asked if other mouse models may exist that better capture the essence of the neurodegenerative signature than most currently available models. Given that human neurodegenerative diseases are characterized in part by chronic neuroinflammation and lysosomal dysfunction (Usenovic and Krainc, 2012), we evaluated the transcriptome of mice acutely treated with lipopolysaccharide (LPS) and/or chronically infected with prion protein, as well as a mouse model of Niemann–Pick disease which involves lysosomal


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

9

Table 5 Alternate models of neurodegeneration. The transcriptional responses of Niemann–Pick mice as well as chronically prion-infected mice with or without additional LPS administration, replicate the transcriptional responses of human neurodegeneration, and those of the best ALS, PD and AD mice. Mice acutely treated with LPS demonstrate significant enrichment for genes upregulated in human neurodegenerative diseases, but non-significant and inconsistent downregulation of human downregulated genes.

dysfunction. Of these, all demonstrated appropriate positive enrichment for genes upregulated in human disease (Table 5). The acute LPS model generally showed no significant change in genes down-regulated in human neurodegeneration though these were significantly downregulated in the prion models, and to a lesser extent, the Niemann–Pick models. Although GSEA can be used to detect enrichment for defined gene sets, other statistical methods can more quantitatively compare complete data sets based on overall similarities and differences. To evaluate the relative similarities between all evaluated human and mouse transcriptomes, we performed Spearman's correlation analysis and hierarchical clustering analysis. (Fig. 2 and Supplemental Table S5). Both Tg4510 data sets, and 110d SOD1-G93A clustered most closely to human neurodegeneration and human aging in the hierarchical clustering dendrogram, followed by prion and prionþ LPS models (Fig. 2). Specific evaluation of R correlation values for each human disease metasignature revealed that prionþ LPS, followed by LPS were the most similar available animal models to each human neurodegenerative disease, in some cases with correlation scores surpassing those of certain human aging samples (Supplemental Table S6). These were followed by Tg4510, Dicer KO and Niemann–Pick mice, in strong agreement with results from GSEA.

3.5. Gene expression changes in animal models during disease progression Given that the most established mouse models showed poor correlation with human neurodegeneration, we wondered if these may consistently represent the transcriptomes of early stage disease, as suggested in the SOD1-93A spinal cord tissue samples, which showed marginal correlation with human ALS at 70–80 days, but robust correlation at 110 days. As such, we evaluated TgCRND8 AD mice, at 70, 80 and 150 days. Although a trend was seen toward down-regulation of the human AD up-gene set at 70 days, this switched to a trend towards up-regulation of these genes by 150 days. Further, while only 1 of 22 curated human upregulated gene sets was significantly upregulated at 70 days, 14 of 22 were significantly upregulated at 150 days (Table 3A and Supplemental Table S2). These findings supported the idea that positive correlation with the human neurodegeneration signature may be maximized at later time points. However, the opposite situation occurred for the human down-regulated gene sets, which actually became increasingly up-regulated, opposite to human

neurodegeneration, by 150 days (Table 4A and Supplemental Table S4). Taken together, these data suggested that in contrast to aging, wherein mouse models generally follow the transcriptional changes seen in humans, only very few mouse models of neurodegeneration successfully replicate the human neurodegenerative disease transcriptome. Although the best correlation with the human neurodegenerative disease signature may be obtained in end-stage disease models, the frequently observed “erroneous” pattern of mice up-regulating genes that are downregulated in human disease cannot merely be attributed to evaluating transcriptional changes prematurely during the disease course. 3.6. Region- and cell-type-specific signatures of mouse neurodegeneration In some animal models, we found robust correlation between results obtained from multiple independent datasets. Tg4510 hippocampus data from E-GEOD-53480 and E-GEOD-56772 clustered together (Fig. 2) and yielded a very similar pattern of gene set enrichment (Table 4A and Supplemental Table S4A). Similarly, 3 independent striatal data sets for R6/2 mice (E-GEOD-62210, E-GEOD-29681, and E-GEOD-9857), showed similar patterns of gene set enrichment; though these contrasted with the one available R6/2 cortex data set (E-GEOD-38219; Table 4D), suggesting that transcriptional responses to disease may vary in a region-specific manner. Likewise, 14 week MPTP mice (E-GEOD-7707) yielded contrasting signatures in striatum versus midbrain (Table 4C). Few studies to date have addressed the differential response of individual cell types to neurodegeneration. In one study, however, we noted that the transcriptional responses of genes normally going down in the whole tissue spinal cord, whole tissue of end stage SOD1-G93A mice were actually upregulated in purified motorneurons from these same SOD1-G93A mice (Table 4A, Supplemental Table S4). In order to first determine whether any particular cell type signature is specifically enriched in neurodegeneration, we generated ranked gene lists for cell-type-specific identities (Zhang et al., 2014) and responses. Genes positively and negatively enriched in human neurodegeneration generally followed the same pattern in the transcriptome of purified mouse microglia, astrocytes and endothelium (Supplemental Table S4 rows BD–BF), and was opposite to that of purified neurons (row BJ). These data suggested that the transcriptome of human neurodegeneration could be explained at least in part by a smaller number of neurons relative to astrocytes, microglia and endothelial cells. However, a simple change in cell


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132

10

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66


Table 6 Aging SN and aging VTA responses mimic those of “good” and “aberrant” mouse models of neurodegeneration, respectively. The pattern of gene set enrichment in the substantia nigra (SN), which is susceptible to neurodegeneration, generally follows that of human neurodegenerative disease, human and mouse aging, and the “best” animal models of neurodegeneration. Responses of the ventral tegmental area (VTA), which is relatively refractory to neurodegeneration, generally follow those of the several aberrant animal models that show upregulation of genes normally downregulated with neurodegeneration.

numbers could not explain the transcriptional changes in several mouse models, wherein genes upregulated in neurons and downregulated in human neurodegeneration, are actually upregulated without the relative addition of neurons to the mouse brain. Indeed, given that that these same neuron-associated genes that decline in whole tissue samples are even more highly enriched in purified motorneurons of SOD1-G93A mice at 90 and 120 days suggested they may be involved in the neuronal response to stress in an attempt to preserve neuronal integrity—defenses that may progressively decline in the human neurodegenerative condition, as previously suggested (Li et al., 2014). To further evaluate this hypothesis, we examined the responses of 2 mouse brain regions: ventral tegmental area (VTA) and substantia nigra (SN) to the physiological stress of normal aging. Dopaminergic neurons in the VTA neurons are known to be uniquely resilient to stress, in contrast to substantia nigra (SN) neurons, which selectively degenerate in response to stresses, ranging from toxins to head trauma, viral infection and aging, leading to Parkinsonism. Indeed, while the aging SN showed a similar transcriptional response to other human aging, mouse aging, and human neurodegeneration samples, as well as a small subset of the “best” animal models of neurodegeneration, the aging VTA demonstrated the same up-regulation of human diseasedownregulated genes as seen in other “aberrant” mouse models (Table 6). In order to understand the differences between aging responses between VTA and SN, we used the rank difference methodology to determine which genes are most strongly upregulated in VTA aging response, as compared to the SN aging response. We used GSEA for GO terms to functionally evaluate the ranked list of genes ordered from most—to least-strongly upregulated in the VTA relative to SN aging signatures. We found 59 GO terms to be significantly enriched at FDRo0.05, including multiple terms for ubiquitin and catabolic processes, as well as endosome transport, vesicle transport, Golgi apparatus, RNA processing, DNA repair and response to stress (Supplemental Table S9). Collectively, these findings suggested that the response of various brain regions to physiologic stresses ranging from aging to toxin exposure and genetic mutations may vary by region; that signatures of resilience (e.g., genes upregulated in VTA during aging) are enriched in several mouse models, relative to human neurodegeneration), and that such responses may be particularly relevant to neuronal responses to stress, though that cell-type-specific responses may not be reliably interpreted from analysis of whole tissue samples. 3.7. Analysis via GSEA Given that our GSEA revealed human neurodegenerative disease signatures to be significantly enriched in each other but poorly or inconsistently enriched in most mouse models, we sought to devise

strategies to decipher the functional differences between mice and humans. Weighted gene correlation network analysis (WGCNA) has previously been employed as a method to identify transcriptional “modules” of co-regulated genes that may share functional importance, define constituent cell types within the whole tissue analyzed, or define regulatory networks. WGCNA has previously been used to evaluate healthy (Oldham et al., 2008) and AD-affected (Zhang et al., 2013) human brains. In healthy brains, WGCNA revealed modules that appeared to define the transcriptome of individual cell types such as neurons and glial subtypes (Oldham et al., 2008), whereas when AD brains were included in WGCNA, an immune- and microglia-specific module was found to correlate strongly with late-onset AD (Zhang et al., 2013). We performed GSEA using gene sets comprising modules defined by these previously published studies, as well as gene sets defining the specific transcriptome of healthy or reactive CNS cell types (Li et al., in press; Zamanian et al., 2012; Zhang et al., 2014). Thirty-three gene sets identified in a preliminary screen to be highly enriched at FDRo0.001 (NES41.9) in the neurodegenerative disease metaanalysis ranked gene list were then used to evaluate the full collection of human and mouse aging and neurodegenerative disease datasets compiled throughout the study (Supplemental Table S4, rows 26–58). Excellent concordance for most gene sets was seen across human neurodegenerative diseases. Indeed, several modules originally defined based upon AD patients proved to be even more highly enriched in other neurodegenerative diseases, validating the co-regulated rather than merely AD-associated nature of these modules. Among human neurodegenerative diseases, positive enrichment was also seen for gene sets and modules related to inflammation, astrocytosis and microglial activation, and negative enrichment was seen for those related to neuronal identity, synaptic function, electron transport chain and vesicle trafficking. Largely concordant results were found between human neurodegenerative diseases, human and mouse aging, Tg4510 AD mice, Dicer KO mice and 110 day-old SOD1-G93A mice, as well as Niemann–Pick, Prion and LPS mice. For other mice, several gene sets normally downregulated in human disease were frequently upregulated, in agreement with findings noted in Tables 4A–D. Of note, several of these modules are of mixed or incompletely characterized function and the extent to which they truly represent individual unique cell types as opposed to coordinately regulated signaling pathways across multiple cell types requires further evaluation. For example, overlapping modules were identified via WGCNA of human brain samples by Oldham et al., including M16 (“blue”) and M18 (“yellow”) in multiple brain regions, including cortex (“A”) and caudate nucleus (“C”). Functional analysis of these modules by Oldham et al. suggested M16 (blue) to encode neuronal genes, though the function of M18 (yellow) remained uncertain. Each of these 4 gene sets are significantly enriched in VTA aging,


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

11

Table 7 Rank difference analysis reveals a conserved signature of mouse versus human response to CNS degeneration. Rank difference analysis was performed by ranking genes from most differentially upregulated in mouse disease versus human disease, to most differentially downregulated in mouse disease versus human disease. The mouse and human ranked gene lists compared to generate the rank difference lists are shown in list (mouse) and list 2 (human). Each of the top 7 lists (box) representing aging, NP and prion disease, as well as the best available models for each of AD, PD, HD and ALS, were then evaluated via GSEA for enrichment for the top (UP; more mouse-like) and bottom (DOWN; more human-like) 250 genes from each of the other ranked gene lists. Strong enrichment across independently generated ranked gene lists indicates that mouse and human responses to aging and neurodegeneration are different, and that this difference is conserved across multiple neuropathological conditions.

ALS motorneurons at 60, 90 and 120 days, as well as, to varying degrees, many of the “aberrant” mouse models. Although these gene sets were generally enriched in healthy mouse neurons (Supplemental Table S4, column BJ), 3 of these 4 modules are were also significantly enriched in reactive mouse astrocytes 1 day after stroke, suggesting a potential role in response to acute stress. We selected one of these gene sets,“M16A(ctx)Blue-Neuron” for functional analysis using Enrichr and found that 6 of the top 10 enriched GO biological process terms in this module pertained to protein catabolism, protein localization to organelles, or ubiquitin (Supplemental Table S10). Collectively, these data further supported the idea that several “aberrant” mouse models are capable of chronically up-regulating stress response genes whose expression typically declines during the course of human neurodegeneration. 3.8. Comparative signatures of mouse versus human responses Finally, given that both aging as well as the “best” animal models showed appropriate positive and negative enrichment for most upand down-regulated gene sets, respectively, we asked what overarching differences may exist between even the best available mouse models and their human counterparts. As such, we performed rank difference analysis, based on the difference in gene rank position between the human and mouse responses. Rank difference lists were thus generated comparing each of the best animal models to the actual human disease. We also compared mouse aging to human aging response, and compared the Niemann–Pick and prion models to the human neurodegeneration meta-analysis ranked gene list. Using GSEA to evaluate enrichment of the top and bottom 250 genes from each rank difference list to the others, we found strikingly robust enrichment of each mouse versus human response in each of the other lists—both in terms of UP (more mouse-like) and DOWN (more human-like) genes (Table 7). This provided evidence that reproducible differences are present between transcriptional responses in human disease versus mouse models across a variety of human disease and animal models, remarkably, including aging. GSEA for gene sets representing human disease up- and downregulated WGCNA-derived and cell type-specific gene sets in these ranked gene lists generally revealed that even among these “best” mouse models, gene sets normally downregulated in human disease, such as the M16 and M18 modules, were generally more highly expressed in mice than humans (Supplemental Table S7). To obtain a detailed and unbiased understanding of these reproducible

differences, we evaluated each rank difference list using the C2 “curated” gene sets from MsigDB (Subramanian et al., 2005) which contains 44000 gene sets, including selected signatures from published experimental data as well as the KEGG (Ogata et al., 1999) and REACTOME (Croft et al., 2011) gene set collections. Over 500 gene sets were significantly enriched at FDRo0.05 in at least 1 of the rank difference comparisons (Supplemental Table S8), the 50 most consistently mouse-like and human-like of which are shown in Tables 8A and B, respectively. Since HD mice were the most “aberrant” of the 7 mice evaluated, gene sets were ranked based on the average scores of the other 6 mice. Interestingly, however, HD followed the same general pattern, with generally higher scores for most genes for the same gene sets. Strong concordance between the mouse:human comparisons was supported by the fact that of the top 118 terms, the response was in the same direction for each mouse:human comparison, including aging. Generally, the rank difference lists enriched for the smallest number of terms were Aging (mouse vs. human aging comparison) and prion (mouse prion vs. human neurodegeneration metaanalysis comparison), suggesting that these models most strongly resemble the respective human genetic responses. That gene sets representing those genes found to be downregulated in a microarray study of AD versus control hippocampal tissue (Blalock et al., 2004) “BLALOCK_ALZHEIMERS_DISEASE_DN” as well as genes downregulated in aged versus young human cortex (Lu et al., 2004) “LU_AGING_BRAIN_DN” were in the top 3 gene sets supported the robust nature of our methodology. Common themes among the top mouseenriched gene sets included mitochondrion, electron transport, and oxidative phosphorylation-related gene sets, consistent with a generally more profound bioenergetic failure in mice. P53 dependent and independent DNA damage responses, ubiquitin, proteasome, antigen presentation, wnt signaling, and regulation of proliferation also featured prominently (Table 8A and Supplemental Table S8). Similar results were obtained via functional analysis using Enrichr (data not shown). Among these results, hub proteins repeatedly predicted to regulate genes enriched in animal models included multiple histone deacetylases as well as nuclear receptor corepressor 1, which works with HDACs to maintain chromatin structure and epigenetic state (Delcuve et al., 2012). Though findings among the most human-like (“DOWN”) gene sets were less consistently significant across diseases, several inflammation-related gene sets were significantly more highly enriched in human diseases, including STAT3 and NFKB targets,


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132

12

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66


Table 8 Functional differences between the transcriptional responses of mice and humans. (A): Genes upregulated in mouse models relative to human disease. Gene set enrichment analysis for gene sets in the Broad Institute’s Molecular Signature Database (MSigDB) curated “C2” collection. A robust signature of consistent differences between human neurodegenerative diseases and their respective mouse models was seen, with most differences seen in NP, AD, ALS and HD mice compared to human disease, and fewest differences with aging and in prion disease mice compared to human neurodegeneration. (B): Genes upregulated in human disease relative to human disease.

responses to TGFB1, and pathways involving NKT and IL23 and complement, in addition to AD, aging brain, and various cancerrelated gene sets (Table 8B and Supplemental Table S8).

differences compared to human neurodegeneration. Although the reasons for the poor transcriptional performance of mouse models varied, the unifying theme was the failure of mouse models to exhibit the variety and severity of diverse defects observed in human neurodegeneration.

4. Discussion 4.1. Transcriptional comparison of mouse models and human disease We here demonstrate that most available mouse models of neurodegenerative disease fail to recapitulate the salient transcriptional alterations of human neurodegeneration and that even the best available models show significant and reproducible

Transcriptional differences between mice and humans have recently come under increasing scrutiny, yet varying conclusions have been reached. Seok et al. (2013) evaluated human and mouse


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66

leukocyte transcriptional changes in response to various inflammatory stimuli and found very poor concordance of genomic responses mice. In fact, scarcely more than half of genes moved in the same direction between mice and humans. However, reanalysis of the very same data sets by Takao and Miyakawa (2015) yielded opposite conclusions, with highly significant agreement between mouse and human transcriptional data. These conflicting data raise concerns not only about the relevance of animal models, but also about how best to analyze and interpret comparative analyses between species. Seok et al. evaluated genes significantly differentially expressed in humans and, after evaluating multiple time points in mice and humans, compared the maximum fold change of expression for each gene using Pearson correlation. By contrast, Takao selected genes that were significantly differentially expressed in both humans AND mice, and performed Spearman correlation analysis at the single time point showing best concordance of responses between humans and mice. While the pros and cons of these varying analysis strategies can be debated, the great potential for differing methodologies to fundamentally alter conclusions must be carefully considered. Acknowledging these challenges, we sought to perform an unbiased analysis between human neurodegeneration and mouse neurodegenerative diseases, using two independent methodologies— Spearman's correlation, based on all genes detected in common across the 51 data sets evaluated, and gene set enrichment analysis (GSEA), which evaluates for enrichment of specified gene sets across the entire ranked gene list. Although alternatives to GSEA have been described in an attempt to improve sensitivity or performance for specific applications (Luo et al., 2009), GSEA has remained among the most widely used and accepted tools in current bioinformatics for genome-wide evaluation of biological function, since its description in 2005, with over 7000 references to date (Subramanian et al., 2005). Both Spearman's correlation analysis and GSEA identified the same animal models as providing the strongest correlation with human neurodegenerative disease, including Tg4510, Dicer KO mice and late-stage SOD1-93A mice. However, “best” is a relative term, as the human transcriptional response to neurodegeneration was more closely related to that of even “healthy” human aging than any of the available mouse models of AD, PD, ALS or HD. In evaluating potential alternate mouse models for the neurodegenerative process, both GSEA and Spearman's rank correlation analysis demonstrated prion diseaseþ LPS, and prion disease alone, to provide the best and second best correlations, respectively, with the human transcriptomes for AD, ALS, PD and HD. Indeed, only these models were able to achieve correlation scores matching those of human aging samples for each neurodegenerative disease (Table 4A–D and Supplemental Tables S5 and S6). In addition to parallel analysis of human and mouse transcriptional responses via GSEA, we built upon the ranked gene list of principles of GSEA to perform a direct comparison of the mouse and human transcriptional responses using a rank difference strategy. This enabled us to identify significant differences between data sets with highly overlapping transcriptomes. Rather than restricting our functional analysis to well established gene ontology terms or predefined MSigDB collections, such as C2, we utilized gene sets derived from analysis of individual CNS cell types as well as modules from WGCNA analysis of healthy and diseased human brains. This analysis enabled us to detect and characterize differences between human and mouse transcriptional responses that would otherwise not have been possible. For example, using even the MSigDB C2 collection, comprising over 4000 gene sets, only 6 gene sets showed significant differences in the rank difference list comparing mice and human brain responses to aging at FDRo0.05. Conversely, our analysis of manually curated gene sets based upon WGCNA modules from Alzheimer's and healthy brain tissue as well as cell type-specific signatures (Supplemental Table S7) revealed 12 of 33 gene sets differentially expressed between aged human and aged mouse

13

brains, offering novel opportunities to functionally dissect the difference between mouse and human CNS responses to disease. The contribution of ailing stress responses to human neurodegeneration and cognitive impairment is increasingly recognized (Lu et al., 2014). Our data illustrate the challenge of recapitulating the multiple molecular “failures” of human neurodegeneration in mouse models. In fact, many mouse models, actually showed augmentation, rather than failure of stress responses. Most mouse models of neurodegeneration are induced by a single mutation, leaving open the potential for adaptive mitigating responses. In contrast, human neurodegeneration, typically occurs in response to multiple exposures and genetic polymorphisms, which appear to collectively inhibit appropriate stress responses. Of note, compared to the “best” mouse models of neurodegeneration, human disease signatures revealed relatively greater convergent impacts of bioenergetic failure, impaired protein catabolism, impaired RNA processing, and impaired DNA repair, all of which constitute components of an appropriate stress response. Interestingly, a pattern was seen even in comparison of mouse and human aging. Given that stronger enrichment for human aging signatures was seen in datasets representing the oldest mice (Table 3), the difference observed between human and mouse aging (Table 8A and Supplemental Table S4), may suggest the mice collectively analyzed were not sufficiently “old” compared to the human data. However, it should be noted that even those diseases that are induced by a single mutation, such as Huntington's disease, also elicited a more robust neuroprotective response in mice than humans (Table 4D). This raises the possibility that mice may be inherently more resistant to neurodegenerative processes than humans. 4.2. Robust models of Alzheimer's disease AD animal models that most closely correlated with human AD in this study included Tg4510 mice, as confirmed in 2 independent data sets, and the Dicer KO mice. Despite a difference in enrichment signal between the 2 Tg4510 mice, which was stronger in E-GEOD-53480 than in E-GEOD-56772 mice, the pattern of enrichment for numerous gene sets in these data were extremely similar, and their transcriptomes clustered together in hierarchical clustering analysis, confirming the robust nature of our data analysis strategy. Tg4510 mice express a repressible human tau variant and progressively develop neurofibrillary tangles, memory impairment and neuronal loss (Santacruz et al., 2005). Interestingly, these mice do not develop amyloid plaques, the clearance of which has been the focus of much recent unsuccessful drug development (Doody et al., 2014; Salloway et al., 2014). Of note, repression of the tau variant in this model reverses memory dysfunction, although neurofibrillary tangles continue to accumulate (Santacruz et al., 2005), further cautioning against interpretation of pathological hallmarks as etiological contributors to neurodegenerative disease (Krstic and Knuesel, 2013a, 2013b). Dicer is a type III RNase enzyme responsible for processing and functional maturation of microRNA molecules that are essential for neuronal integrity (Hebert et al., 2010). Dicer KO mice demonstrate tau hyperphosphorylation, neuroinflammation, impaired neurogenesis and neurodegeneration(Hebert et al., 2010; Nowakowski et al., 2013). Although these mice were never intended to provide an etiologically accurate model of human AD, they aptly illustrate the potential functional importance of altered RNA processing in the cascade of neurodegeneration (Li et al., 2014). 4.3. Microglial activation in animal models of neurodegeneration Among mouse models of AD, PD, ALS and HD, the ALS model SOD1-G93A was one of the 3 identified neurodegenerative disease models that most accurately replicated the neurodegenerative


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132

14

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66


disease transcriptome. Neuroinflammation and microglial activation are prominent features across all neurodegenerative diseases, and are particularly strongly featured in ALS (Li et al., 2014). As such, strategies have been developed to block microglial activation in order to attenuate the neurodegenerative process. The translational relevance of this approach, however, is unclear. For example, minocycline, which blocks microglial activation, actually worsened outcomes in a large randomized human clinical trial (Gordon et al., 2007). Interestingly, subsequent attempts to reproduce the therapeutic effects of minocycline in SOD1-G3A ALS mice using rigorously optimized standards for preclinical testing failed to elicit a therapeutic effect (Scott et al., 2008), highlighting yet another hurdle of reproducibility in the translational process. In further contrast to preclinical studies supporting the inactivation of microglia, attempts to decrease microglial activation in ME7 prion mice via microglial Cx3cr1 actually accelerated neurodegeneration (Grizenkova et al., 2014). Similar results were also reported in Niemann–Pick mice, wherein Ccl3 deletion failed to delay neurodegeneration and worsened clinical outcomes (Lopez et al., 2012). Continued efforts are ongoing to refine strategies to “reprogram” microglia, so as to attenuate their presumably harmful actions whilst promoting their evidently beneficial activities. Gene sets related to glial and microglial identity and activation were strongly enriched in both human neurodegenerative disease— particularly so in ALS—as well as in transcriptionally robust animal models, such as SOD1-93A, Tg4510 and Dicer KO, along with identified robust “alternate models,” including prion and Niemann– Pick. At first glance, this may raise hopes that successful microglial repolarization in animal models should be detected in the resulting transcriptome with strategic insights obtained from the correlation between mice and human transcriptional data. However, similar upregulation of these same gene sets was actually also seen in purified motor neurons from ALS mice (Table 4B). This could be due to diseased neurons up-regulating similar stress response programs that drive gliosis and microglial activation, or due to epigenetic alterations and chromatin remodeling leading to the partial loss of their unique neuronal transcriptional identity. Either way, attempts to de-convolute whole tissue transcriptome data to infer cell-typespecific gene expression profiles may prove particularly challenging in tissue from neurodegenerative diseases (Shen-Orr et al., 2010). As such, we suggest that informed development of therapies to modulate microglial or other cell-type-specific activities will require available transcriptome data from prospectively glial and neuronal subtypes from both humans and mice. Indeed, until these data are available, the extent to which the glial activation profiles of animal models reflect human disease will remain enigmatic. 4.4. Toward better models and more successful therapies One of the most striking findings from our analysis was the frequent up-regulation of putative neuroprotective stress response genes in animal models—genes that are uniformly downregulated in human disease. Even among the “best” animal models, the degree of downregulation of such genes failed to match that of human disease. Based upon several complementary line of evidence, including GSEA analysis of aging VTA versus SN-enriched genes, functional analysis of human-downregulated gene sets upregulated in mice, and rank-difference analysis of mouse versus human transcriptome responses, a picture emerges of even the best mouse models failing to encompass all of the convergent alterations of human neurodegeneration. Likely in part large part due to their etiology as typically single mutation models, they not only maintain a more functional stress response, but fail to exhibit the extent of bioenergetic failure, impaired DNA damage responses, and dysfunction of protein catabolism that synergistically characterize human neurodegeneration. Moreover, differences in the mouse

neuroinflammatory response are observed, with reproducible trends toward more exuberant TGFB1, STAT3, and NFKB signaling in human disease than mouse models. To date, virtually all therapies that have shown promise in preclinical trials have failed in clinical trials. Our data suggest this may be due in part to animal models providing a poor representation of human diseases. It is well known that non-fatal stressors, such as hypoxia induce upregulation of neuroprotective transcriptional programs that enhance resilience to subsequent stressors. Given that genes downregulated in human neurodegeneration were upregulated in many mouse models, as well as in aging VTA—a structure uniquely resilient to stress—it is conceivable that these stress response mechanisms may act synergistically with administered preclinical pharmacotherapies in mice to impart a therapeutic effect, but fail in humans wherein neuroprotective pathways are down-regulated, rather than upregulated. We hypothesize that a “multiple hit” model of neurodegeneration that combines, for example, a chronic tauopathy with measures to induce mitochondrial and proteasomal dysfunction, interfere with RNA processing, induce DNA damage and oxidative stress perhaps via low dose irradiation, and/or exacerbate neuroinflammation via chronic low dose LPS exposure, may collectively yield a more human-like neurodegenerative state complete with a dysfunctional stress responses. Conversely, the fact that mouse models of neurodegeneration generally do upregulate neuroprotective responses, and actually can be successfully treated via an array of previously tested pharmacotherapies raises the question of whether efforts to revitalize a healthy stress response could provide a more successful and generalizable approach to neurodegeneration than attempts to block specific disease-associated pathways. For example, recent data suggest that treatment of aged mice with plasma from young mice reverses several parameters of aging-induced cognitive decline (Villeda et al., 2014). Given the close transcriptional overlap between aging and neurodegeneration-induced cognitive decline, it will be of interest to determine if augmentation of stress responses are in part responsible for such young plasma-induced cognitive “rejuvenation”. The challenge of overcoming neurodegenerative disease requires valiant efforts from all possible angles. The availability of iPS cells offers unprecedented potential to model human diseases in vitro; more representative animal models of human brain aging and neurodegenerative disease almost certainly exist than mice (Languille et al., 2012). Nevertheless, murine studies have been and will likely continue to be an important contributor to neurodegeneration research. As such, we feel a strong obligation to critically evaluate mouse models in light of available human data to optimize therapeutic insights obtained from preclinical studies. To our knowledge, this work represents the largest comparative mouse versus human transcriptome analysis of neurodegenerative diseases to date. Nevertheless, it captures only a small subset of animal models currently available. For example the Jackson Laboratory currently offers over 100 different mouse models for AD alone. We hope that our analysis will not only provide a data resource for translational studies, but will inspire increased efforts to prospectively obtain region and cell-type-specific data from post-mortem tissue of humans and mice alike.

5. Conclusions We here provide an analytical platform based upon GSEA and Spearman correlation to reproducibly evaluate transcriptional differences between mouse models and human disease. We find that most available models of neurodegeneration fail to recapitulate the salient features of the corresponding human neurodegenerative diseases and that aberrant animal models frequently upregulate


67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Q4 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 Q2 57 58 59 60 61 62 63 64 65 66

stress response genes that are normally downregulated in human neurodegenerative diseases. We identify Tg4510, Dicer KO and endstage SOD1-93A as the 3 most transcriptionally robust models of neurodegeneration, showing significant up- and down-regulation of gene sets that are similarly regulated in human disease. We identify Niemann–Pick mice as a similarly robust model, but conclude that prion disease, particularly prion diseaseþLPS, best replicates the transcriptome of human neurodegeneration. We provide evidence that even the best available mouse models fall short as models of neurodegeneration, being only as similar to human neurodegeneration as human aging, and we suggest that the array of deficiencies in these models, including insufficient defects in mitochondrial function, protein catabolism, RNA processing and DNA repair, could be improved by use of “multiple hit” rather than single mutation models of neurodegeneration. This may be particularly important in mice, which appear inherently more resistant to neurodegenerative disease processes than humans. With continued advances in bioinformatics tools we are optimistic that increasingly relevant and meaningful data will be derived from both human and preclinical transcriptome studies to help develop novel therapies for these devastating diseases.

Conflicts of Interest None.

Acknowledgments We thank Cindy H. Samos for Editorial assistance. Funding support was provided by the California Institute of Regenerative Medicine (TCB).

Appendix A. Supporting information Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.ejphar.2015.03.021. References Alam, M.S., Getz, M., Safeukui, I., Yi, S., Tamez, P., Shin, J., Velazquez, P., Haldar, K., 2012. Genomic expression analyses reveal lysosomal, innate immunity proteins, as disease correlates in murine models of a lysosomal storage disorder. PloS One 7, e48273. Barger, J.L., Kayo, T., Vann, J.M., Arias, E.B., Wang, J., Hacker, T.A., Wang, Y., Raederstorff, D., Morrow, J.D., Leeuwenburgh, C., Allison, D.B., Saupe, K.W., Cartee, G.D., Weindruch, R., Prolla, T.A., 2008. A low dose of dietary resveratrol partially mimics caloric restriction and retards aging parameters in mice. PloS One 3, e2264. Berchtold, N.C., Cribbs, D.H., Coleman, P.D., Rogers, J., Head, E., Kim, R., Beach, T., Miller, C., Troncoso, J., Trojanowski, J.Q., Zielke, H.R., Cotman, C.W., 2008. Gene expression changes in the course of normal brain aging are sexually dimorphic. Proc. Natl. Acad. Sci. USA 105, 15605–15610. Blalock, E.M., Geddes, J.W., Chen, K.C., Porter, N.M., Markesbery, W.R., Landfield, P.W., 2004. Incipient Alzheimer's disease: microarray correlation analyses reveal major transcriptional and tumor suppressor responses. Proc. Natl. Acad. Sci. USA 101, 2173–2178. Burns, T.C., Verfaillie, C.M., 2015. From mice to mind: strategies and progress in translating neuroregeneration. Eur. J. Pharmacol. (in press). Cavanaugh, S.E., Pippin, J.J., Barnard, N.D., 2014. Animal models of Alzheimer disease: historical pitfalls and a path forward. Altex 31, 279–302. Chen, E.Y., Tan, C.M., Kou, Y., Duan, Q., Wang, Z., Meirelles, G.V., Clark, N.R., Ma’ayan, A., 2013. Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinf. 14, 128. Colantuoni, C., Lipska, B.K., Ye, T., Hyde, T.M., Tao, R., Leek, J.T., Colantuoni, E.A., Elkahloun, A.G., Herman, M.M., Weinberger, D.R., Kleinman, J.E., 2011. Temporal dynamics and genetic control of transcription in the human prefrontal cortex. Nature 478, 519–523. Croft, D., O’Kelly, G., Wu, G., Haw, R., Gillespie, M., Matthews, L., Caudy, M., Garapati, P., Gopinath, G., Jassal, B., Jupe, S., Kalatskaya, I., Mahajan, S., May, B., Ndegwa, N., Schmidt, E., Shamovsky, V., Yung, C., Birney, E., Hermjakob, H., D’Eustachio,

15

67 P., Stein, L., 2011. Reactome: a database of reactions, pathways and biological processes. Nucl. Acids Res. 39, D691–D697. 68 De Oliveira, G.P., Alves, C.J., Chadi, G., 2013. Early gene expression changes in spinal 69 cord from SOD1(G93A) Amyotrophic Lateral Sclerosis animal model. Front. Cell. 70 Neurosci. 7, 216. Delcuve, G.P., Khan, D.H., Davie, J.R., 2012. Roles of histone deacetylases in 71 epigenetic regulation: emerging paradigms from studies with inhibitors. Clin. 72 Epigenet. 4, 5. 73 Doody, R.S., Thomas, R.G., Farlow, M., Iwatsubo, T., Vellas, B., Joffe, S., Kieburtz, K., 74 Raman, R., Sun, X., Aisen, P.S., Siemers, E., Liu-Seifert, H., Mohs, R., Alzheimer's Disease Cooperative Study Steering, C., Solanezumab Study, G., 2014. Phase 75 3 trials of solanezumab for mild-to-moderate Alzheimer's disease. N. Engl. J. 76 Med. 370, 311–321. 77 Fossale, E., Seong, I.S., Coser, K.R., Shioda, T., Kohane, I.S., Wheeler, V.C., Gusella, J.F., MacDonald, M.E., Lee, J.M., 2011. Differential effects of the Huntington's disease 78 CAG mutation in striatum and cerebellum are quantitative not qualitative. 79 Hum. Mol. Genet. 20, 4258–4267. 80 Gao, L., Hidalgo-Figueroa, M., Escudero, L.M., Diaz-Martin, J., Lopez-Barneo, J., 81 Pascual, A., 2013. Age-mediated transcriptomic changes in adult mouse substantia nigra. PloS One 8, e62456. 82 Gispert, S., Brehm, N., Weil, J., Seidel, K., Rub, U., Kern, B., Walter, M., Roeper, J., 83 Auburger, G., 2015. Potentiation of neurotoxicity in double-mutant mice with 84 Pink1 ablation and A53T-SNCA overexpression. Hum. Mol. Genet. 24, 85 1061–1076. Gladstone, D.J., Black, S.E., Hakim, A.M., Heart, Stroke Foundation of Ontario Centre 86 of Excellence in Stroke, R., 2002. Toward wisdom from failure: lessons from 87 neuroprotective stroke trials and new therapeutic directions. Stroke 33, 88 2123–2136. Godbout, J.P., Chen, J., Abraham, J., Richwine, A.F., Berg, B.M., Kelley, K.W., Johnson, 89 R.W., 2005. Exaggerated neuroinflammation and sickness behavior in aged 90 mice following activation of the peripheral innate immune system. FASEB J. 19, 91 1329–1331. 92 Gordon, P.H., Moore, D.H., Miller, R.G., Florence, J.M., Verheijde, J.L., Doorish, C., Hilton, J.F., Spitalny, G.M., MacArthur, R.B., Mitsumoto, H., Neville, H.E., Boylan, 93 K., Mozaffar, T., Belsh, J.M., Ravits, J., Bedlack, R.S., Graves, M.C., McCluskey, L.F., 94 Barohn, R.J., Tandan, R., Western, A.L.S.S.G., 2007. Efficacy of minocycline in 95 patients with amyotrophic lateral sclerosis: a phase III randomised trial. Lancet Neurol. 6, 1045–1053. 96 Grizenkova, J., Akhtar, S., Brandner, S., Collinge, J., Lloyd, S.E., 2014. Microglial 97 Cx3cr1 knockout reduces prion disease incubation time in mice. BMC Neurosci. 98 15, 44. 99 Hebert, S.S., Papadopoulou, A.S., Smith, P., Galas, M.C., Planel, E., Silahtaroglu, A.N., Sergeant, N., Buee, L., De Strooper, B., 2010. Genetic ablation of Dicer in adult 100 forebrain neurons results in abnormal tau hyperphosphorylation and neuro101 degeneration. Hum. Mol. Genet. 19, 3959–3969. 102 Hodges, A., Hughes, G., Brooks, S., Elliston, L., Holmans, P., Dunnett, S.B., Jones, L., 103 2008. Brain gene expression correlates with changes in behavior in the R6/1 mouse model of Huntington's disease. Genes Brain Behav. 7, 288–299. 104 Kohman, R.A., Rodriguez-Zas, S.L., Southey, B.R., Kelley, K.W., Dantzer, R., Rhodes, 105 J.S., 2011. Voluntary wheel running reverses age-induced changes in hippo106 campal gene expression. PloS One 6, e22654. Krstic, D., Knuesel, I., 2013a. The airbag problem-a potential culprit for bench-to107 bedside translational efforts: relevance for Alzheimer's disease. Acta Neuro108 pathol. Commun. 1, 62. 109 Krstic, D., Knuesel, I., 2013b. Deciphering the mechanism underlying late-onset 110 Alzheimer disease. Nat. Rev. Neurol. 9, 25–34. Kuhn, A., Goldstein, D.R., Hodges, A., Strand, A.D., Sengstag, T., Kooperberg, C., 111 Becanovic, K., Pouladi, M.A., Sathasivam, K., Cha, J.H., Hannan, A.J., Hayden, M.R., 112 Leavitt, B.R., Dunnett, S.B., Ferrante, R.J., Albin, R., Shelbourne, P., Delorenzi, M., 113 Augood, S.J., Faull, R.L., Olson, J.M., Bates, G.P., Jones, L., Luthi-Carter, R., 2007. Mutant huntingtin's effects on striatal gene expression in mice recapitulate 114 changes observed in human Huntington's disease brain and do not differ with 115 mutant huntingtin length or wild-type huntingtin dosage. Hum. Mol. Genet. 16, 116 1845–1861. 117 Kurosawa, M., Matsumoto, G., Kino, Y., Okuno, M., Kurosawa-Yamada, M., Washizu, C., Taniguchi, H., Nakaso, K., Yanagawa, T., Warabi, E., Shimogori, T., Sakurai, T., 118 Hattori, N., Nukina, N., 2015. Depletion of p62 reduces nuclear inclusions and 119 paradoxically ameliorates disease phenotypes in Huntington's model mice. 120 Hum. Mol. Genet. 24, 1092–1105. 121 Kurz, A., Wohr, M., Walter, M., Bonin, M., Auburger, G., Gispert, S., Schwarting, R.K., 2010. Alpha-synuclein deficiency affects brain Foxp1 expression and ultrasonic 122 vocalization. Neuroscience 166, 785–795. 123 Labbadia, J., Cunliffe, H., Weiss, A., Katsyuba, E., Sathasivam, K., Seredenina, T., 124 Woodman, B., Moussaoui, S., Frentzel, S., Luthi-Carter, R., Paganetti, P., Bates, G.P., 2011. Altered chromatin architecture underlies progressive impairment of 125 the heat shock response in mouse models of Huntington disease. J. Clin. Invest. 126 121, 3306–3319. 127 Languille, S., Blanc, S., Blin, O., Canale, C.I., Dal-Pan, A., Devau, G., Dhenain, M., 128 Dorieux, O., Epelbaum, J., Gomez, D., Hardy, I., Henry, P.Y., Irving, E.A., Marchal, J., Mestre-Frances, N., Perret, M., Picq, J.L., Pifferi, F., Rahman, A., Schenker, E., 129 Terrien, J., Thery, M., Verdier, J.M., Aujard, F., 2012. The grey mouse lemur: a 130 non-human primate model for ageing studies. Ageing Res. Rev. 11, 150–162. 131 Li, M.D., Burns, T.C., Kumar, S., Morgan, A.A., Sloan, S.A., Palmer, T.D., 2015. AgingQ3132 like changes in the transcriptome of irradiated microglia. Glia (in press).


16

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48


Li, M.D., Burns, T.C., Morgan, A.A., Khatri, P., 2014. Integrated multi-cohort transcriptional meta-analysis of neurodegenerative diseases. Acta Neuropathol. Commun. 2, 93. Lopez, M.E., Klein, A.D., Hong, J., Dimbil, U.J., Scott, M.P., 2012. Neuronal and epithelial cell rescue resolves chronic systemic inflammation in the lipid storage disorder Niemann–Pick C. Hum. Mol. Genet. 21, 2946–2960. Lu, T., Aron, L., Zullo, J., Pan, Y., Kim, H., Chen, Y., Yang, T.H., Kim, H.M., Drake, D., Liu, X.S., Bennett, D.A., Colaiacovo, M.P., Yankner, B.A., 2014. REST and stress resistance in ageing and Alzheimer's disease. Nature 507, 448–454. Lu, T., Pan, Y., Kao, S.Y., Li, C., Kohane, I., Chan, J., Yankner, B.A., 2004. Gene regulation and DNA damage in the ageing human brain. Nature 429, 883–891. Lukas, T.J., Luo, W.W., Mao, H., Cole, N., Siddique, T., 2006. Informatics-assisted protein profiling in a transgenic mouse model of amyotrophic lateral sclerosis. Mol. Cell. Proteomics 5, 1233–1244. Lunnon, K., Teeling, J.L., Tutt, A.L., Cragg, M.S., Glennie, M.J., Perry, V.H., 2011. Systemic inflammation modulates Fc receptor expression on microglia during chronic neurodegeneration. J. Immunol. 186, 7215–7224. Luo, W., Friedman, M.S., Shedden, K., Hankenson, K.D., Woolf, P.J., 2009. GAGE: generally applicable gene set enrichment for pathway analysisBMC Bioinf. 10161 Mielcarek, M., Landles, C., Weiss, A., Bradaia, A., Seredenina, T., Inuabasi, L., Osborne, G.F., Wadel, K., Touller, C., Butler, R., Robertson, J., Franklin, S.A., Smith, D.L., Park, L., Marks, P.A., Wanker, E.E., Olson, E.N., Luthi-Carter, R., van der Putten, H., Beaumont, V., Bates, G.P., 2013. HDAC4 reduction: a novel therapeutic strategy to target cytoplasmic huntingtin and ameliorate neurodegeneration. PLoS Biol. 11, e1001717. Moisoi, N., Klupsch, K., Fedele, V., East, P., Sharma, S., Renton, A., Plun-Favreau, H., Edwards, R.E., Teismann, P., Esposti, M.D., Morrison, A.D., Wood, N.W., Downward, J., Martins, L.M., 2009. Mitochondrial dysfunction triggered by loss of HtrA2 results in the activation of a brain-specific transcriptional stress response. Cell Death Differ. 16, 449–464. Noh, H., Park, C., Park, S., Lee, Y.S., Cho, S.Y., Seo, H., 2014. Prediction of miRNAmRNA associations in Alzheimer's disease mice using network topology. BMC Genomics 15, 644. Nowakowski, T.J., Mysiak, K.S., O’Leary, T., Fotaki, V., Pratt, T., Price, D.J., 2013. Loss of functional Dicer in mouse radial glia cell-autonomously prolongs cortical neurogenesis. Dev. Biol. 382, 530–537. O’Collins, V.E., Macleod, M.R., Donnan, G.A., Horky, L.L., van der Worp, B.H., Howells, D.W., 2006. 1,026 experimental treatments in acute stroke. Ann. Neurol. 59, 467–477. Ogata, H., Goto, S., Sato, K., Fujibuchi, W., Bono, H., Kanehisa, M., 1999. KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucl. Acids Res. 27, 29–34. Oldham, M.C., Konopka, G., Iwamoto, K., Langfelder, P., Kato, T., Horvath, S., Geschwind, D.H., 2008. Functional organization of the transcriptome in human brain. Nat. Neurosci. 11, 1271–1282. Panza, F., Logroscino, G., Imbimbo, B.P., Solfrizzi, V., 2014. Is there still any hope for amyloid-based immunotherapy for Alzheimer's disease? Curr. Opin. Psychiatry 27, 128–137. Pereson, S., Wils, H., Kleinberger, G., McGowan, E., Vandewoestyne, M., Van Broeck, B., Joris, G., Cuijt, I., Deforce, D., Hutton, M., Van Broeckhoven, C., Kumar-Singh, S., 2009. Progranulin expression correlates with dense-core amyloid plaque burden in Alzheimer disease mouse models. J. Pathol. 219, 173–181. Perrin, F.E., Boisset, G., Docquier, M., Schaad, O., Descombes, P., Kato, A.C., 2005. No widespread induction of cell death genes occurs in pure motoneurons in an amyotrophic lateral sclerosis mouse model. Hum. Mol. Genet. 14, 3309–3320. Phani, S., Gonye, G., Iacovitti, L., 2010. VTA neurons show a potentially protective transcriptional response to MPTP. Brain Res. 1343, 1–13. Polito, V.A., Li, H., Martini-Stoica, H., Wang, B., Yang, L., Xu, Y., Swartzlander, D.B., Palmieri, M., Di Ronza, A., Lee, V.M., Sardiello, M., Ballabio, A., Zheng, H., 2014. Selective clearance of aberrant tau proteins and rescue of neurotoxicity by transcription factor EB. EMBO Mol. Med. 6, 1142–1160.

Salloway, S., Sperling, R., Fox, N.C., Blennow, K., Klunk, W., Raskind, M., Sabbagh, M., Honig, L.S., Porsteinsson, A.P., Ferris, S., Reichert, M., Ketter, N., Nejadnik, B., Guenzler, V., Miloslavsky, M., Wang, D., Lu, Y., Lull, J., Tudor, I.C., Liu, E., Grundman, M., Yuen, E., Black, R., Brashear, H.R., Bapineuzumab, Clinical Trial, I., Clinical Trial, I., 2014. Two phase 3 trials of bapineuzumab in mild-tomoderate Alzheimer's disease. N. Engl. J. Med. 370, 322–333. Santacruz, K., Lewis, J., Spires, T., Paulson, J., Kotilinek, L., Ingelsson, M., Guimaraes, A., DeTure, M., Ramsden, M., McGowan, E., Forster, C., Yue, M., Orne, J., Janus, C., Mariash, A., Kuskowski, M., Hyman, B., Hutton, M., Ashe, K.H., 2005. Tau suppression in a neurodegenerative mouse model improves memory function. Science 309, 476–481. Scott, S., Kranz, J.E., Cole, J., Lincecum, J.M., Thompson, K., Kelly, N., Bostrom, A., Theodoss, J., Al-Nakhala, B.M., Vieira, F.G., Ramasubbu, J., Heywood, J.A., 2008. Design, power, and interpretation of studies in the standard murine model of ALS. Amyotroph. Lateral Scler. 9, 4–15. Seok, J., Warren, H.S., Cuenca, A.G., Mindrinos, M.N., Baker, H.V., Xu, W., Richards, D.R., McDonald-Smith, G.P., Gao, H., Hennessy, L., Finnerty, C.C., Lopez, C.M., Honari, S., Moore, E.E., Minei, J.P., Cuschieri, J., Bankey, P.E., Johnson, J.L., Sperry, J., Nathens, A.B., Billiar, T.R., West, M.A., Jeschke, M.G., Klein, M.B., Gamelli, R.L., Gibran, N.S., Brownstein, B.H., Miller-Graziano, C., Calvano, S.E., Mason, P.H., Cobb, J.P., Rahme, L.G., Lowry, S.F., Maier, R.V., Moldawer, L.L., Herndon, D.N., Davis, R.W., Xiao, W., Tompkins, R.G., Inflammation, Host Response to Injury, Large Scale Collaborative Research Program, 2013. Genomic responses in mouse models poorly mimic human inflammatory diseases. Proc. Natl. Acad. Sci. USA 110, 3507–3512. Shen-Orr, S.S., Tibshirani, R., Khatri, P., Bodian, D.L., Staedtler, F., Perry, N.M., Hastie, T., Sarwal, M.M., Davis, M.M., Butte, A.J., 2010. Cell type-specific gene expression differences in complex tissues. Nat. Methods 7, 287–289. Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette, M.A., Paulovich, A., Pomeroy, S.L., Golub, T.R., Lander, E.S., Mesirov, J.P., 2005. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. USA 102, 15545–15550. Takao, K., Miyakawa, T., 2015. Genomic responses in mouse models greatly mimic human inflammatory diseases. Proc. Natl. Acad. Sci. USA 112, 1167–1172. Usenovic, M., Krainc, D., 2012. Lysosomal dysfunction in neurodegeneration: the role of ATP13A2/PARK9. Autophagy 8, 987–988. Valor, L.M., Guiretti, D., Lopez-Atalaya, J.P., Barco, A., 2013. Genomic landscape of transcriptional and epigenetic dysregulation in early onset polyglutamine disease. J. Neurosci. 33, 10471–10482. Villeda, S.A., Plambeck, K.E., Middeldorp, J., Castellano, J.M., Mosher, K.I., Luo, J., Smith, L.K., Bieri, G., Lin, K., Berdnik, D., Wabl, R., Udeochu, J., Wheatley, E.G., Zou, B., Simmons, D.A., Xie, X.S., Longo, F.M., Wyss-Coray, T., 2014. Young blood reverses age-related impairments in cognitive function and synaptic plasticity in mice. Nat. Med. 20, 659–663. Yu, M., Eckart, M.R., Morgan, A.A., Mukai, K., Butte, A.J., Tsai, M., Galli, S.J., 2011. Identification of an IFN-gamma/mast cell axis in a mouse model of chronic asthma. J. Clin. Invest. 121, 3133–3143. Zamanian, J.L., Xu, L., Foo, L.C., Nouri, N., Zhou, L., Giffard, R.G., Barres, B.A., 2012. Genomic analysis of reactive astrogliosis. J. Neurosci. 32, 6391–6410. Zhang, B., Gaiteri, C., Bodea, L.G., Wang, Z., McElwee, J., Podtelezhnikov, A.A., Zhang, C., Xie, T., Tran, L., Dobrin, R., Fluder, E., Clurman, B., Melquist, S., Narayanan, M., Suver, C., Shah, H., Mahajan, M., Gillis, T., Mysore, J., MacDonald, M.E., Lamb, J.R., Bennett, D.A., Molony, C., Stone, D.J., Gudnason, V., Myers, A.J., Schadt, E.E., Neumann, H., Zhu, J., Emilsson, V., 2013. Integrated systems approach identifies genetic nodes and networks in late-onset Alzheimer's disease. Cell 153, 707–720. Zhang, Y., Chen, K., Sloan, S.A., Bennett, M.L., Scholze, A.R., O’Keeffe, S., Phatnani, H.P., Guarnieri, P., Caneda, C., Ruderisch, N., Deng, S., Liddelow, S.A., Zhang, C., Daneman, R., Maniatis, T., Barres, B.A., Wu, J.Q., 2014. An RNA-sequencing transcriptome and splicing database of glia, neurons, and vascular cells of the cerebral cortex. J. Neurosci. 34, 11929–11947.


49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95

Mouse models of human diseases.

Preclinical mouse models of osteosarcoma.

Correction for Takao and Miyakawa, Genomic responses in mouse models greatly mimic human inflammatory diseases.

Mouse models for human infectious diseases.

Preclinical mouse cancer models: a maze of opportunities and challenges.

Evaluation of traditional medicines for neurodegenerative diseases using Drosophila models.

Olfactory assays for mouse models of neurodegenerative disease.

Genetic screens in Caenorhabditis elegans models for neurodegenerative diseases.

Mouse models of human cancer.

Preclinical non-human models to combat dementia.

Rodent models for human diseases.

Generation of a neuro-specific microarray reveals novel differentially expressed noncoding RNAs in mouse models for neurodegenerative diseases.

Mouse models of the nuclear envelopathies and related diseases. Preface.

Genetically engineered humanized mouse models for preclinical antibody studies.

Predicting Drug Response in Human Prostate Cancer from Preclinical Analysis of In Vivo Mouse Models.

Genetically engineered mouse models for drug development and preclinical trials.

Human models of human diseases; breast cancer and the lymphomas.

Significance of Mouse Models in Dissecting the Mechanism of Human Eosinophilic Gastrointestinal Diseases (EGID).

Pathology of the Aging Brain in Domestic and Laboratory Animals, and Animal Models of Human Neurodegenerative Diseases.

Mouse Genetic Models of Human Brain Disorders.

Humanized mouse models of human cytomegalovirus infection.

Towards authentic transgenic mouse models of heritable PrP prion diseases.

Brain Trauma: From Preclinical Models to Clinical Applications.

Genetically engineered pig models for human diseases.