Article Text
Abstract
We are still very limited in management strategies for dementia, and establishing effective disease modifying therapies based on amyloid or tau remains elusive. Neuroinflammation has been increasingly implicated as a pathological mechanism in dementia and demonstration that it is a key event accelerating cognitive or functional decline would inform novel therapeutic approaches, and may aid diagnosis. Much research has therefore been done to develop technology capable of imaging neuroinflammation in vivo. The authors performed a systematic search of the literature and found 28 studies that used in vivo neuroimaging of one or more markers of neuroinflammation on human patients with dementia. The majority of the studies used positron emission tomography (PET) imaging of the TSPO microglial marker and found increased neuroinflammation in at least one neuroanatomical region in dementia patients, most usually Alzheimer's disease, relative to controls, but the published evidence to date does not indicate whether the regional distribution of neuroinflammation differs between dementia types or even whether it is reproducible within a single dementia type between individuals. It is less clear that neuroinflammation is increased relative to controls in mild cognitive impairment than it is for dementia, and therefore it is unclear whether neuroinflammation is part of the pathogenesis in early stages of dementia. Despite its great potential, this review demonstrates that imaging of neuroinflammation has not thus far clearly established brain inflammation as an early pathological event. Further studies are required, including those of different dementia subtypes at early stages, and newer, more sensitive, PET imaging probes need to be developed.
- DEMENTIA
- ALZHEIMER'S DISEASE
- NEUROIMMUNOLOGY
- LEWY BODY
- PET, LIGAND STUDIES
Statistics from Altmetric.com
Introduction
Dementia is a chronic condition in which progressive cognitive impairment leads to functional disability. Alzheimer's disease (AD) is the most frequent cause of dementia, but other common types include vascular dementia, ‘mixed’ dementia, frontotemporal lobar dementia (FTLD), dementia with Lewy bodies (DLB) and Parkinson's disease dementia (PDD).1 Our understanding of the aetiology and pathogenesis of dementia is constantly increasing,2 but we are still very limited in our ability to accurately differentiate between dementia subtypes at early stages of the disease. This in turn limits our ability to give accurate diagnostic and prognostic information,3 particularly in the situation of overlapping phenotypes,4 and to develop treatments that prevent the progression of dementia when it is mild.
As well as tau and amyloid pathology, neuroinflammation is a disease process that has been increasingly implicated as a pathological mechanism in dementia.5 Unlike neurological conditions such as multiple sclerosis and autoimmune encephalitis in which there is an established pathogenic role of the adaptive immune system, neuroinflammation in dementia is thought to be mediated predominantly by aberrant activation of the brain's innate immune system, namely microglia.5 Microglia are central nervous system (CNS) resident macrophages derived from haematopoietic stem cells whose processes rapidly converge on damaged cortex in vitro.6 Ligands to the peripheral benzodiazepine receptor (PBR, also known as the ‘18 kDa Translocator Protein’ or ‘TSPO’) found on activated microglia amplify apoptosis of rat neurones in vitro7 and delivery of microglia inhibitory factor significantly reduces microglial activation and the volume of damage caused by fibrillary Abeta in aged primate cortex.8 Part of the neurodegeneration in dementia might be mediated through microglial release of pro-inflammatory cytokines such as interleukin-1B.9
However, neuroinflammatory mechanisms are complex and incompletely understood, and neuroinflammation might also be neuroprotective in dementia under certain conditions and stages of the disease. It has been suggested that microgliosis in AD might initially be neuroprotective (through the degradation of amyloid plaques) but that impaired clearance of amyloid due to age or genetic predisposition leads to amyloid accumulation and an exaggerated microglial response that is ultimately neurodegenerative.10 Astrocytes are the most numerous and diverse glial cells in the CNS and they phagocytose and degrade amyloid plaques in cultured mouse models of AD,11 suggesting that, as with microglial activation, reactive astrogliosis in dementia might initially be a protective response to the primary neuropathological insult of amyloid plaque formation.12 Furthermore, in vivo imaging of TSPO in rats after ethanol-induced neuronal insults suggests that activated astrocytes might limit the expansion of microglial-induced scarring.13
The above studies are invariably in-vitro, postmortem or animal studies and predominantly in models of AD rather than other forms of dementia. The ability to measure neuroinflammation in humans in vivo might therefore provide novel insights into the neuropathology of numerous types of dementia, the development of novel preventative therapies as well as enabling early diagnosis and prognosis of this extremely important disease.
Attempts have been made to measure serum and cerebrospinal fluid (CSF) biomarkers of neuroinflammation, but these have proven unable to provide sufficiently detailed information regarding the extent and distribution of neuroinflammation in vivo. Non-invasive imaging of neuroinflammation has therefore been heralded as a potential method of providing such information. Structural neuroimaging studies using for example, CT, MRI have made limited progress to date, but functional neuroimaging technologies using for example, positron emission tomography (PET)14 enable the imaging of specific molecules associated with inflammation and thus, potentially, detailed information regarding the neuroanatomical distribution of specific biomarkers of inflammation.
Much research has therefore been done to develop technology capable of imaging neuroinflammation in vivo.15 However, efforts to develop PET ligands capable of quantifying regional neuroinflammation are hampered by our incomplete understanding of its pathophysiology. As mentioned above, the TSPO is thought to be a marker of activated microglia. It is a phylogenetically conserved receptor present on the outer mitochondrial membrane that has been implicated in cholesterol transport, cell replication and apoptosis; immunohistochemical studies have demonstrated minimal expression of TSPO in ependymal cells of healthy brain and significantly upregulated TSPO expression in microglia, macrophages and astrocytes of diseased brain.16 Most of the plethora of PET ligands targeting putative inflammatory targets that have hitherto been used in animals or humans (see online supplementary file) have labelled the TSPO on activated microglia.16 Table 1 is a summary of the techniques that have been used to image neuroinflammation in humans to date; whilst most target the TSPO, several ligands are available that target alternative putative indices of neuroinflammation including monoamine oxidase B (MAO-B) in activated astrocytes, peripheral macrophages that are thought to have infiltrated the CNS and even metabolite levels such as arachidonic acid (AA) or N-acetylaspartate that were thought to be non-specific markers of neuroinflammation. Increasingly, many studies have also investigated inflammatory neuroimaging in human dementia in vivo.17 However, no comprehensive review has been recently published on this important topic.
The authors therefore performed a systematic search of the literature for all studies that imaged neuroinflammation in human patients with dementia in order to investigate whether neuroinflammation imaging might potentially be of use in the clinical management of dementia.
Methods
To investigate the techniques used in neuroinflammation imaging, the literature was initially searched for studies that imaged neuroinflammation in vivo in any neurological or psychiatric disorder in either humans or animal models. Thirty-three distinct imaging methodologies utilising PET, single-photon emission computed tomography (SPECT) or MRI in either animals or humans were found and are listed in the online supplementary table; table 1 contains the sixteen imaging methodologies that have been used in humans to date.
To investigate the imaging of neuroinflammation in human patients with dementia, on 7/9/2014 the databases AMED, EMBASE, HMIC, MEDLINE, PsycINFO, BNI, CINAHL, HEALTH BUSINESS ELITE were searched for articles with the following terms:
((dementia OR alzheimer* OR mild cognitive impairment OR Parkinson* OR front* OR vascular) AND (PET OR SPECT OR imaging OR neuroimaging) AND (inflammation OR PK11195 OR DAB OR PBR OR TSPO OR DPA*)).af
There were 15 720 results that reduced to 2128 after excluding duplicates. References from this electronic search, as well as reference lists of reviews and important papers in the field, were screened on the basis of relevance of title and abstract to the review. This yielded 100 potentially included references. Eight of these references were only published in abstract/poster form—the authors of all of these references were contacted twice but only three authors replied explaining a full length article had not been published, so all eight of these references were excluded because the full length article could not be obtained or was not yet published. After further screening on the basis of full text, 28 of these 92 references were included in the review and are included in table 2. In order to be included, studies had to use in vivo neuroimaging of one or more markers of neuroinflammation on human patients with dementia, which included mild cognitive impairment (MCI) as well as AD, vascular dementia, mixed dementia, DLB, FTLD or PDD.
Results
After screening on the basis of full text, 28 studies (see table 2) were considered to meet the inclusion criteria and used in vivo neuroimaging of one or more markers of neuroinflammation on human patients with dementia, which was taken to include MCI as well as AD, vascular dementia, mixed dementia, DLB, FTLD or PDD.
Twenty-seven of these studies used PET and only one study18 used SPECT as the imaging modality to measure neuroinflammation.
Seven ligands were used; 19 studies used PK11195,18–36 two studies used DAA1106,37–38 one study used vincopetine,39 two studies used DED,40–41 two studies used PBR28,42–43 one study used FEDAA110644 and one study used AA.45 Five of these seven ligands (PK11195, DAA1106, vincopetine, PBR28, FEDAA1106) bind the TSPO on activated microglia, one (DED) binds to MAO-B in activated astrocytes and one (AA) images a metabolic by-product of microglial-induced neuroinflammation.45 No study was found that used two ligands together to measure neuroinflammation.
The mean number of control participants was 13.1 (range 1–24) and the mean number of participants in a dementia group was 8.9 (range 1–22).
All 28 studies had a cross-sectional design to at least part of the trial; 25 of the 28 studies were purely cross-sectional whilst one study had both cross-sectional and case–control components20 and two studies had cross-sectional and cohort components.36 ,38
All 28 studies included healthy control participants while 14 studies had comparator groups with AD,18–19 ,23 ,25–29 ,32 ,37 ,39–40 ,44–45 11 studies had AD and MCI comparator groups,20 ,22 ,24 ,30 ,31 ,33 ,36 ,38 ,41 ,42 ,43 one study had FTLD as the comparator,21 one study had PDD34 and one had DLB.35 Six of the 28 included studies investigated neuroimaging methodologies and did not compare inflammatory ligand binding between dementia and control patients;22 ,23 ,25–27 ,43 their results are thus not pertinent to the question of whether inflammation neuroimaging might potentially be of use in the clinical treatment of dementia. The present review therefore found 22 studies which compared inflammatory ligand binding (and thus neuroinflammation) between control and dementia patients. A minority (6) of these 22 studies did not find a statistically significant difference in inflammation between control and dementia patients in at least one region of the brain;19 ,24 ,31 ,33 ,39 ,44 each of these six ‘negative’ cross-sectional studies compared control patients with AD+/− MCI patients, therefore no ‘negative’ cross-sectional studies were found that investigated other dementia subtypes. The majority (16) of the 22 included studies which compared neuroinflammation between control and dementia patients found a statistically significant difference in inflammation between control and dementia patients in at least one brain region.18 ,20 ,21 ,28 ,29 ,30 ,32 ,34–38 ,40 ,41 ,42 ,45 Unexpectedly, one of these 16 cross-sectional studies reported decreased inflammation in the right hippocampus in AD relative to controls,41 but the remainder found increased neuroinflammation in dementia patients.
Nine included studies compared MCI with healthy age-matched controls.20 ,24 ,30 ,31 ,33 ,36 ,38 ,41 ,42 Three of these studies24 ,31 ,33 found no statistically significant difference in neuroinflammation in AD or MCI relative to age-matched controls, two of these studies found increased neuroinflammation in AD (but not MCI) relative to controls,36 ,42 two of these studies found increased neuroinflammation in AD and MCI relative to controls,20 ,38 one study found increased neuroinflammation in MCI relative to controls but did not give data for AD relative to MCI or controls30 and one study found increased neuroinflammation in MCI relative to both AD and controls.41
One study20 used discriminant function analysis to show, retrospectively, that increased inflammation in the left inferior temporal lobe discriminated patients with AD from age-matched controls with a sensitivity of 75% and with no false categorisations of controls.
One study20 found that neuroanatomical regions with high inflammation demonstrated the highest rate of atrophy over 1–2 years, while Yasuno et al38 found that four of the five MCI patients who were followed up had whole measured region neuroinflammation more than 0.5 SDs greater than the control mean and all four developed dementia within 5 years. However, Schuitemaker et al36 did not find significant differences in neuroinflammation between MCI patients who progressed to dementia and MCI patients who did not.
Several of the included studies investigated whether regional neuroinflammation correlated with dementia severity, as measured by either cognition18 ,28 ,30 ,32 ,34 ,36–38 ,42 or age of dementia onset.30 ,42 Three of these studies found no statistically significant association between Mini-Mental State Examination (MMSE) and ligand binding in any neuroanatomical region in AD.36–38 Five studies found statistically significant correlations between standardised cognitive questionnaire scores and regional neuroinflammation in AD.18 ,28 ,30 ,32 ,42 The neuroanatomical regions demonstrating an association between inflammation and cognition varied between studies but included frontal cortex,18 ,28 ,32 parietal cortex,18 ,28 ,42 temporal cortex,30 cingulate cortex,28 ,30 ,32 precuneus32 and hippocampus.32 Furthermore, one study found a statistically significant association between cognition and neuroinflammation in the frontal, parietal, temporal and occipital cortex in PDD.34 No studies were found by this review investigating an association between cognition and neuroinflammation in any other dementia type.
Only two studies assessed whether regional neuroinflammation correlated with age of dementia onset in AD. Okello et al30 found no correlation between duration of symptoms and neuroinflammation in patients with AD who were amyloid positive, while Kreisl et al42 found that patients with early onset (<65) AD had increased global neuroinflammation relative to patients with late-onset (>65) AD. Kreisl et al42 also found that neuroinflammation in the parietal cortex and striatum correlated with lower age of dementia onset in patients with AD.
Seven studies were found that used multiple PET ligands to image not just neuroinflammation but also amyloid load (eg, Pittsburgh Compound B PiB binding). Four of these seven studies found no statistically significant correlation between inflammatory ligand binding and PiB binding in AD.28 ,30 ,31 ,41 Three of these seven studies32 ,40 ,42 found inflammatory ligand binding to positively correlate with PiB binding (and thus, presumably, amyloid load) in at least one neuroanatomical region (the parietal cortex,42 temporal cortex,42 occipital cortex,40 cingulate cortex,32 hippocampus42 and precuneus42) in AD.
Discussion
The present review contributes to the existing literature on neuroinflammation imaging, and thus is of interest not just for dementia but also for the diverse and increasing number of disorders that have a putative inflammatory component; this includes depression,46 delirium47 and schizophrenia.48
The majority of the studies to date of in vivo imaging of neuroinflammation in dementia have used PET imaging of the TSPO microglial marker. TSPO is an attractive imaging target because if it were found to be correlated with the development of dementia in vivo it would be relatively simpler to attempt to modify this single molecular target than if multiple separate markers of neuroinflammation were each shown to be partly involved. However, the almost exclusive focus on TSPO is a weakness within the current imaging literature since microglial activation is unlikely to represent the totality of the neuroinflammatory response in dementia. Indeed, astrocytes have been implicated in the neuroinflammation of dementia5 but only two40 ,41 of the 28 studies included in the present review used radioligands targeting a putative marker of astrocyte activation, namely MAO-B. While the majority of PET ligands found to have been used in the imaging of neuroinflammation in any neuropsychiatric disorder in humans also target the TSPO (see table 1), ligands are available from animal studies that target alternative putative indices of neuroinflammation for example, COX-1, MPO, macrophage infiltration and even metabolite levels such as AA or N-acetylaspartate (see online supplementary file). Future studies should therefore investigate the utility of such alternative ligands, either in isolation or in combination with TSPO-targeting ligands, in dementia patients.
An additional cause for concern by the current focus on TSPO imaging in the literature is the finding that the rs6971 TSPO single nucleotide polymorphism alters binding affinity of second generation radioligands for TSPO between healthy individuals independently of the degree of inflammation. Indeed, in vivo uptake in healthy individuals of the PBR28 radioligand for TSPO was 40% greater in homozygotes than heterozygotes for the high affinity TSPO protein.49 However, only one43 of the six37–39 ,42–44 studies using second generation TSPO radioligands that were included in the present review measured participant TSPO binding affinity genotype. Furthermore, a possible role for the rs6971 TSPO single nucleotide polymorphism as a modifier of neuroinflammation in dementia has not been extensively investigated, despite in vitro evidence that high affinity rs6971 TSPO homozygotes have significantly elevated levels of the neurosteroid precursor, pregnenolone, in peripheral cells compared to low affinity homozygotes or heterozygotes.50 It is therefore imperative that future imaging studies control for the relative proportions of low and high affinity TSPO participants.43
While a minority of the included studies did not find a statistically significant difference in inflammation between control and dementia patients in at least one region of the brain,19 ,24 ,31 ,33 ,39 ,44 the studies included in this review were not adequately powered to conclude that there is no significant difference in neuroinflammation between dementia and controls. Furthermore, the majority (15) of included studies found increased neuroinflammation in at least one neuroanatomical region in dementia patients relative to controls. AD was the most commonly studied dementia type, with increased inflammation in virtually all neuroanatomical locations, both cortical (frontal/parietal/temporal/occipital/cingulated/hippocampal/entorhinal/fusiform gyri/parahippocampal gyri) and subcortical (amygdala, pallidum, thalamus, striatum). There were significant differences between trials in the anatomical structures (and their laterality) demonstrated to have increased inflammation; comparison of results between studies was not helped by the frequent use of different neuroanatomical ‘regions of interest’ between studies. Thus, whilst the majority of studies investigating neuroinflammation imaging in AD demonstrated increased neuroinflammation relative to controls, it is not possible to state with certainty the neuroanatomical location where such increased neuroinflammation occurs. This is in contrast with for example, 18-fluoro-deoxyglucose PET imaging studies demonstrating reproducible reductions in glucose metabolism in frontal, parietal, temporal and posterior cingulate cortices of patients with AD51 and suggests either that existing imaging studies do not accurately portray the distribution of neuroinflammation in vivo or that neuroinflammation is globally increased in AD. Furthermore, if neuroinflammation is globally increased in AD this might suggest that neuroinflammation does not contribute to the early stages of its disease pathogenesis, since early AD typically involves preferential atrophy of the temporal cortex and hippocampus.4 Only three studies were found that investigated non-Alzheimer's types of dementia (FTLD,21 PDD34 and DLB35), but all three, similarly to the findings in AD, found increased neuroinflammation in a variety of cortical and subcortical structures. Therefore a plethora of neuroanatomical structures have been implicated as having increased neuroinflammation in all dementias studied, but the published evidence to date does not indicate whether the regional distribution of neuroinflammation differs between dementia types or even whether it is reproducible within a single dementia type between individuals. It also raises the tantalising possibility that neuroinflammation might be a common pathological mechanism in all types of dementia.
Nine included studies compared MCI patients with healthy controls.20 ,24 ,30 ,31 ,33 ,36 ,38 ,41 ,42 Taken together, the results of these nine studies are equivocal; five of these studies found no statistically significant difference in neuroinflammation in MCI relative to control patients24 ,31 ,33 ,36 ,42 while four of these studies found increased neuroinflammation in MCI relative to controls in frontal cortices,38 ,30 ,41 parietal cortices,38 ,41 temporal cortices,20 ,38 anterior cingulate cortices,38 fusiform gyri,20 left parahippocampus,20 striatum38 and cerebellum.38 Thus, it is less clear that neuroinflammation is increased relative to controls in MCI than it is for dementia, and therefore it is unclear whether neuroinflammation is part of the pathogenesis in early stages of dementia. However, the regional distribution of any neuroinflammation that might occur appears to be similar to that in dementia, that is, in a variety of cortical and subcortical structures.
Multitracer PET studies included in this review found only an inconsistent and weak correlation between neuroinflammation and other, better established markers of AD load (such as PiB). This suggests that neuroinflammation imaging might highlight novel aspects of dementia not revealed by existing neuroimaging ligands and therefore be more useful in the clinical management of dementia than existing neuroimaging methods have proven to be. Indeed, several of the included studies attempted to investigate neuroinflammation imaging for potential clinical management purposes in dementia. One study20 found, retrospectively, that increased neuroinflammation in the left inferior temporal lobe distinguished patients with AD from controls with no false categorisations of controls and a sensitivity of 75%; further studies are now needed that prospectively trial the use of neuroinflammation imaging as a diagnostic tool in early stages of dementia, including MCI, when a clinical diagnosis is difficult to make. Furthermore, three of the included studies investigated whether neuroinflammation was prognostic in AD;20 ,36 ,38 Cagnin et al20 found that neuroanatomical regions with high inflammation demonstrated the highest rate of atrophy over 1–2 years, whilst Yasuno et al38 found that four out of the five MCI patients who were followed up had whole measured region neuroinflammation more than 0.5 SDs greater than the control mean and all four developed dementia within 5 years. However, Schuitemaker et al36 did not find significant differences in neuroinflammation between MCI patients who progressed to dementia and MCI patients who did not; further research is therefore needed before deciding whether neuroinflammation imaging might potentially be used as a prognostic tool in the clinical management of dementia. Similarly, several of the included studies attempted to correlate neuroinflammation with dementia severity, as measured by either standardised cognition questionnaire scores or age of onset. The results were mixed and not convincingly reproducible; three studies found no association between cognition and regional neuroinflammation in AD,36–38 five studies found significant correlations between cognition and regional neuroinflammation in AD,18 ,28 ,30 ,32 ,42 one study found a statistically significant association between cognition and neuroinflammation in PDD,34 one study found no correlation between age of onset and neuroinflammation in AD30 while one study found that neuroinflammation in the parietal cortex and striatum correlated with earlier disease onset in AD.42 Therefore the existing literature does not unequivocally demonstrate that neuroinflammation is an objective correlate of dementia severity in vivo.
Despite its great potential, the lack of reproducibility of positive findings in this review therefore demonstrates that neuroinflammation imaging has not thus far been convincingly used for diagnosis, prognosis or severity profiling in any dementia subtype. The observed heterogeneity of results between studies of the same type of dementia (AD) is likely due to multifactorial causes that include greater variability in the pathogenesis of neuroinflammation than is currently understood; since different inflammatory mechanisms might be responsible for different phases of neuroinflammation in dementia, it is not entirely unsurprising that a single imaging modality detecting one aspect of neuroinflammation (such as microglial activation in the case of the PK11195 ligand) might be unable to detect similar differences between patients with AD at different stages of the disease. Furthermore, existing studies have generally been small and inadequately powered, and whilst PK11195 is a robust marker of neuroinflammation its small signal-to-noise ratio might make it unable to detect subtler patterns of regional neuroinflammation that might exist in dementia. In order to further investigate the role of neuroinflammation imaging in the clinical management of dementia, adequately powered trials are therefore needed which take into account the stage of dementia, patient TSPO genotype and use more sensitive TSPO imaging ligands, ideally in conjunction with alternative markers of neuroinflammation.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Data supplement 1 - Online supplement
Footnotes
Funding The authors are supported by the NIHR Biomedical Research Unit in Dementia and the Biomedical Research Centre awarded to Cambridge University Hospitals NHS Foundation Trust and the University of Cambridge, and the NIHR Biomedical Research Unit in Dementia and the Biomedical Research Centre awarded to Newcastle upon Tyne Hospitals NHS Foundation Trust and the Newcastle University.
Competing interests JO has acted as a consultant for GE Healthcare and Avid/Lilly.
Provenance and peer review Not commissioned; externally peer reviewed.