Neurogenetics DOI 10.1007/s10048-014-0409-x
ORIGINAL ARTICLE
‘Neuroinflammation’ differs categorically from inflammation: transcriptomes of Alzheimer's disease, Parkinson's disease, schizophrenia and inflammatory diseases compared Michaela D. Filiou & Ahmed Shamsul Arefin & Pablo Moscato & Manuel B. Graeber
Received: 30 May 2014 / Accepted: 2 June 2014 # Springer-Verlag Berlin Heidelberg 2014
Abstract ‘Neuroinflammation’ has become a widely applied term in the basic and clinical neurosciences but there is no generally accepted neuropathological tissue correlate. Inflammation, which is characterized by the presence of perivascular infiltrates of cells of the adaptive immune system, is indeed seen in the central nervous system (CNS) under certain conditions. Authors who refer to microglial activation as neuroinflammation confuse this issue because autoimmune neuroinflammation serves as a synonym for multiple sclerosis, the prototypical inflammatory disease of the CNS. We have asked the question whether a data-driven, unbiased in silico approach may help to clarify the nomenclatorial confusion. Specifically, we have examined whether unsupervised analysis of microarray data obtained from human cerebral cortex of Alzheimer's, Parkinson's and schizophrenia patients would reveal a degree of relatedness between these diseases and recognized inflammatory conditions including multiple sclerosis. Our results using two different data analysis methods provide strong evidence against this hypothesis demonstrating that very different sets of genes are involved. Consequently, the designations Electronic supplementary material The online version of this article (doi:10.1007/s10048-014-0409-x) contains supplementary material, which is available to authorized users. M. D. Filiou Max Planck Institute of Psychiatry, Kraepelinstraße 2, 80804 Munich, Germany A. S. Arefin : P. Moscato Centre for Bioinformatics, Biomarker Discovery and Information-based Medicine, Hunter Medical Research Institute, University of Newcastle, Kookaburra Circuit, New Lambton Heights NSW 2305, Australia M. B. Graeber (*) Brain and Mind Research Institute, Faculty of Medicine and Faculty of Health Sciences, University of Sydney, 94 Mallett Street, Camperdown, Sydney NSW 2050, Australia e-mail:
[email protected]
inflammation and neuroinflammation are not interchangeable. They represent different categories not only at the histophenotypic but also at the transcriptomic level. Therefore, non-autoimmune neuroinflammation remains a term in need of definition. Keywords Bioinformatics . Inflammation . Microarrays . Microglia . Neurodegeneration . Neuroinflammation Abbreviations AD Alzheimer’s disease CNS Central nervous system DM Dermatomyositis EAE Experimental autoimmune encephalomyelitis IBD Inflammatory bowel disease JDM Juvenile dermatomyositis MHC Major histocompatibility complex MS Multiple sclerosis PD Parkinson’s disease SCZ Schizophrenia UC Ulcerative colitis
Introduction The term ‘neuroinflammation’ has rapidly gained popularity in recent years with approximately 1,500 articles currently in PubMed that have either the words neuroinflammation or neuroinflammatory in their title. A citation analysis reveals a steep increase in the number of publications mentioning neuroinflammation and their citations ESM1. However, there is a growing unease about the discrepancy between what this rapidly proliferating body of literature [1] implicitly seems to suggest, namely that an understanding of ‘neuroinflammation’ does exist, and the findings of careful central nervous system
Neurogenetics
(CNS) tissue analyses, which consistently demonstrate the absence of credible inflammation pathology from brain tissue in Alzheimer’s disease (AD), Parkinson’s disease (PD) and schizophrenia (SCZ). Multiple sclerosis (MS) is a CNS disorder where inflammation represents a morphologically leading, pathogenetically crucial and nosologically recognized feature. The histological phenotype of this disease is fully in agreement with what the definition of inflammation holds for peripheral organs [2]. Specifically, the presence of perivascular infiltrates of cells of the adaptive immune system is conspicuous whereas such cells are characteristically absent from brain tissue affected by AD, PD and SCZ. Autoimmune neuroinflammation is used as a synonym for MS [3] and its animal model, experimental autoimmune encephalomyelitis (EAE) [4]. Other examples of true inflammation in the CNS include brain abscess [5], various infections, stroke and trauma. However, the designation ‘neuroinflammation’ as referring to microglial (and astrocytic) activation, which effectively describes mild gliosis in neuropathological diagnostic terms, is now increasingly being applied not only to AD [6] but to a rapidly widening spectrum of etiologically highly diverse conditions that include obesity [7], SCZ [8], autism [9], major depression [10], chronic functional bowel syndrome [11], sepsis [12], anxiety [13], bipolar disorder [14], air pollution [15], pain [16], delirium [17], sleep loss [18] and epilepsy [19]. Even a clinical journal with the term ‘neuroinflammation’ in its title has been launched [20]. One possible reading of this development could be that novel technologies allow the detection of disease features with greater sensitivity and thus force a change of concepts. However, alternatively, an unfounded but fashionable idea may become promoted, gain popularity and create a trend simply because there are not enough hands to perform the necessary reality checks in a timely fashion. For instance, imaging-tissue correlations of microglial activation lag years behind. In addition, there are only few brain researchers with an opportunity to see and compare diseased brain tissue, notably inflammatory conditions, and from human brains in particular. As an explanation, neuropathology as a fully developed and independent medical specialty only exists in a small number of countries. Even economic aspects may play an unfortunate role because biomedical research is increasingly seen and conducted as a business with the generation of commercially exploitable intellectual property being a major goal [21]. It is worth noting in this context that several already approved drugs are usable in the huge market that is created if inflammation is seen as relevant to both neurodegeneration and brain aging. Yet, warning examples of past scientific error exist in the neurosciences like the widespread reporting of apoptosis in neurodegenerative diseases about a decade ago and the significant failed research effort of the 1990s aiming to show that
astrocytes are the most important antigen-presenting cells of the CNS. Moreover, inflammation itself, which clearly is a household name, is also an abused term [22]. Taken together, wherever activated microglial cells are currently sighted, chances are high that their presence will be equated with the presence of ‘neuroinflammation’. We have therefore asked the question whether a data-driven, unbiased in silico approach may help to clarify the existing nomenclatorial confusion. Specifically, using two different data analysis methods, unsupervised hierarchical clustering and Craig–Moscato scoring [23], we have tested the hypothesis that the popular ‘neuroinflammatory' conditions, AD, PD and SCZ are related to recognized inflammatory diseases based on their transcriptomic expression profile.
Materials and methods Datasets Ten publicly available microarray datasets (Table 1) representing classical inflammatory conditions (inflammatory bowel disease [IBD], juvenile dermatomyositis [JDM], MS, ulcerative colitis [UC]) and neurodegenerative/ neurodevelopmental diseases (AD, PD, SCZ), respectively, were downloaded from GEO (http://www.ncbi.nlm.nih.gov/ gds/), a database that stores curated gene expression data in the Gene Expression Omnibus repository. Genes represented on the Affymetrix U133A chip were used in this study, as corresponding probe values could be extracted from all datasets. The names of the individual .CEL files used are provided as ESM2.
Selection of inflammation-associated genes Pathway Studio (http://www.elsevier.com/online-tools/ pathway-studio/about, originally by Ariadne Genomics), provides a database of known biological associations. We have previously used the desktop version of this program [24–31]. Our search for inflammation-related genes in the Pathway Studio Database yielded 2,848 entries corresponding to 4,905 probe sets on the U133A gene chip and generated our inflammation gene list. For each dataset, the corresponding. CEL files were imported into Pathway Studio Version 9 (desktop version) as a separate experiment and the respective differential gene expression values (two-class unpaired t-test, with or without multiple testing correction, Benjamini– Hochberg/FDR) were exported to Microsoft Excel. The list of inflammation-related probe sets (U133A) is provided as ESM3 and the inflammation-related differential gene expression values across all datasets as ESM4.
Neurogenetics Table 1 Cortical microarray datasets used for unsupervised hierarchical clustering and CM_2 score analyses Dataset No
Dataset name
Disease studied
No of disease and control samples included
References
1 2 3 4 5 6 7 8 9 10a
PD_Cortex_All PD AD_Blalock AD_Liang MS Schiz IBD DM UC PD_Cortex_Pure
PD PD AD AD MS SCZ IBD JDM UC PD
3 PD, 5 CT 14 PD, 15 CT 22 AD, 9 CT 11 AD, 23 CT 1 MS (active plaque), 2 CT 30 SCZ, 29 CT 15 IBD, 8 CT 19 JDM, 4 CT 8 UC, 18 CT 3 PD, 3 CT
[28, 98] [98, 99] [100] [101, 102] [103] [104] [105, 106] [107] [108] [28]
a
Two control cases (CT) were excluded from this dataset due to the presence of AD pathology (see Moran et al. [28])
Unsupervised hierarchical clustering Unsupervised hierarchical clustering of the ‘inflammation component’ of the ten datasets was carried out using open source software (http://bonsai.hgc.jp/~mdehoon/software/ cluster/software.htm). Data were analysed using centroid linkage by means of Cluster 3.0 and visualized using TreeView. Both programs provide a computational and graphical environment for analysing data from DNA microarray experiments (http://bonsai.hgc.jp/~mdehoon/ software/cluster/manual/Introduction.html#Introduction). Michael Eisen and Alok Saldanha, respectively, originally wrote the programs (http://bonsai.hgc.jp/~mdehoon/ software/cluster/software.htm). Computation of CM_2 scores CM (Craig–Moscato)_1 score computation is a new method for the identification of individualizing markers [23]. Like the CM_1 score, the CM_2 score is based on the ratio of two
CM 2ðp; X ; Y Þ ¼
values computed over sets of samples, but the CM_2 score takes into account information on the range of variation of values in both sets of samples. Similar to the Student’s t-test, the CM_2 score is basically the ratio between two values with the numerator being the difference between the observed means between the two groups of interest (X and Y) and the denominator being the function of the range of values of one of the groups rather than combined standard deviation values in the sets X and Y as in Student’s t-test. For instance, let S be a set of samples of interest. For illustration purposes, we are considering the partition of them in two subsets: X (the class of interest, i.e., disease samples) and Y (the class of reference, which in this example would be healthy controls) in such a way that X ∪ Y=S (all samples in S are either control or disease samples) and X ∩ Y=ϕ (no sample has received two different labels). For the sake of simplicity, let {X(p)} be the set of values that probe p has on disease samples and, analogously, let {Y(p)} be the set of values that probe p has on control samples. Then the CM_2 scores for each probe set p are computed as follows,
AVGfX ðpÞg−AVGfY ðpÞg MINð1 þ MAXfX ðpÞg−MINfX ðpÞg; 1 þ MAXfY ðpÞg−MIN fY ðpÞgÞ
where MIN() is the minimum function of its arguments, so MIN(a,b) returns a if a