A unified model of dementias and age-related neurodegeneration

…but ignorance more frequently begets confidence than does knowledge: it is those who know little, and not those who know much, who so positively assert that this or that problem will never be solved by science. Those working with Alzheimer's and other dementias have been frustrated by the implacability of these diseases. Regardless of limited symptomatic treatment,1 there are no proven disease-modifying interventions. Despite huge and growing costs of care,2 a pipeline of candidate drugs,3 >400 registered trials,4 tens of thousands of patients,5 billions of dollars in both US federal6 and pharmaceutical company investment,7, 8 more than a century of clinical expertise, and thousands of professional careers, dozens of pharmaceutical and biotechnology firms have foundered and failed9 in attempts to prevent, slow, or alter the course of the dementias. This article presents a novel model to explain the relationships between age-related neurodegenerative disorders (eg, dementias) and the underlying molecular mechanisms of the aging process. The hypothesis is prompted by the fact that accepted conceptual models have failed to yield effective interventions for Alzheimer's or other dementias.10 This article is a specific response to the Alzheimer's & Dementia editorial of November 2015,11 which called for a systemic re-evaluation of our current models and their ability to answer fundamental questions regarding complex brain disorders and their relationship to clinical dementia, as well as the failure to yield effective clinical interventions. The article is divided into three parts. The first part explores current models of age-related neurogenerative diseases. The second part proposes a specific model and details both its working and its implications. The third part applies the model to answering the 10 key questions proposed by the Alzheimer's & Dementia editorial. The intent is to provide a conceptual model that accords with known data and proposes a novel point of clinical intervention. The model is intended to provoke discussion and provide a point of departure, rather than to offer a complete and final model for age-related neurodegenerative disease. The value of any such model rests on the outcome of clinical trials, but the model offers potentially innovative pathways to such trials. Although there is no general explanation for the failure to identify an effective intervention, there is growing consensus that a major factor is the lack of a comprehensive model for the dementias.12 Over the past century,13 several models have attained varying degrees of acceptance. Such models generally assume (or imply) a predominant single cause of Alzheimer's disease (AD; Figure 1). Such models suggest that, although several factors may contribute to the pathology, most Alzheimer's pathology results from a single predominant cause, such as amyloid plaque. In Figure 1, A (the predominant cause) is the leading causal factor, whereas B, C, D, and E represent contributing (but less significant or even incidental) factors. Until recently, the foremost among such explanations has been amyloid β (Aβ) theory,14-17 in various iterations. Other candidates have been tau tangles,18-20 mitochondrial dysfunction,21, 22 or other etiologies. Despite academic argument, and in line with Alois Alzheimer's original warning about confusing histological findings with causation,13 these models have targeted biomarkers (whether classic biomarkers such as amyloid and tau protein or less common biomarkers such as inflammation,23, 24 or mitochondrial dysfunction), but a comprehensive, systems model explaining causation often remains unclear. Some models assume or imply a single causal agent which, in turn, drives other common pathological findings (and biomarkers) of AD, although this view (which assumes a predominant interacting cause) is seldom explicit. One leading model used by many pharmaceutical firms, for example, has been that AD is caused by Aβ deposition and other markers of AD are secondary, implying that an intervention targeting the primary cause (Aβ) would prove effective in dealing with secondary factors (tau protein changes, and so on). Equally, pharmaceutical interventions targeting other putatively primary causes, such tau proteins, mitochondrial dysfunction, and microglial activation, imply that their primary target lies “upstream” and is causal not only for dementia, but for other, secondary pathological findings and biomarkers (Figure 2). Thus, for example, primary abnormalities in tau protein physiology might result in secondary Aβ plaque formation and other secondary biomarkers. Such models have failed to result in effective interventional human trials and, despite the suspicion that the underlying pathology of AD might well share mechanisms with other dementias, none of these models accounts for potentially shared mechanisms in all dementias, such as Parkinson disease, frontotemporal dementia (FTD), vascular dementias, mixed dementias, Lewy body dementia, primary progressive aphasia (PPA), LATE (limbic-predominant age-related TDP-43 encephalopathy),26 LOAD (late-onset AD), and so on (hereafter referred to collectively as dementias). Current models also fail to account for age-related cognitive decline in animals. Lack of an overarching model for both age-related human dementias and cognitive decline in animals likely plays a role in our consistent clinical failures in human trials. Lacking a comprehensive model, we fail to develop effective interventions. In AD, there has been a growing, if informal, consensus27 that a more fundamental upstream mechanism underlies the pathognomonic findings. The mechanism is often attributed to glial cell dysfunction,28-33 although there is no agreement on the glial cell changes that result in downstream findings.34-38 This type of model (Figure 3) can be represented by a single fundamental upstream mechanism causing multiple downstream findings, including amyloid plaque, tau tangles, mitochondrial dysfunction,39 lipid processing,40 TDP-43 proteinopathy,41 immune function, and synaptic loss.42 A more general model would encompass all dementias, offering an underlying and shared upstream mechanism expressed variably in specific dementias. The putative upstream, fundamental mechanism is shared, with clinical expression depending on its coupling with individual genetic and epigenetic variables, which differ between patients (Figure 3). Optimally, it would encompass age-related human dementias and age-related cognitive decline in animals. The bane of human trials based on animal studies is that “everything works in mice, nothing works in humans.” This caveat suggests that although age-related cognitive decline is common in both animals and humans, the expression of such decline—and hence the underlying mechanisms of such decline—might preclude effective animal models.43-47 Just as age-related human dementias may share a fundamental upstream mechanism, whose expression (depending on its coupling with patient-specific genetic and epigenetic differences) varies between patients, so too the age-related cognitive declines in animals, including non-human primates,48-50 may share a fundamental upstream mechanism, the expression of which (depending on its coupling with species-specific genetic and epigenetic differences) varies between species (Figure 3). Although the fundamental upstream mechanism may be shared—for example, between mice and humans—the genetic and epigenetic differences between species will not only result in different biomarker findings (eg, the degree of Aβ plaque or the preeminence of tau tangles) but will also result in therapeutic failures when we decline to acknowledge these markedly different intermediate pathways. For example, if glial cell dysfunction is a shared upstream mechanism causing cognitive decline in mice and dementia in humans, we cannot expect (a priori) that the two species would share identical findings in terms of the neuronal effects (eg, amyloid metabolism, and tau protein metabolism) or the neuroanatomical locations of the most common lesions (eg, medial temporal lobe, neocortex, and subthalamic nuclei). To the contrary, if we attend solely to the intermediate expression of an underlying mechanism, we should expect poor extrapolation from mouse models to human trials. A model that can clarify, define, and allow confirmation or disproof could explain both the lack of extrapolation from animal models to human trials51 and the failure of human trials generally. A fundamental upstream mechanism that is variably expressed in multiple biomarkers and in an array of neuropathological findings (in coupling with the varying genetic and epigenetic landscape) implying that any intervention targeting exclusively downstream biomarkers (eg, Aβ plaques, tau tangles, and mitochondrial dysfunction) will fail precisely because each such biomarker is merely one among several downstream findings with none of these downstream findings being causal per se. Treating a single biomarker (eg, amyloid) will not necessarily have a significant beneficial impact on another downstream finding (eg, tau tangles). We might target multiple downstream biomarkers simultaneously, but the optimal target lies upstream, at a more fundamental level of the cascade of pathology (Figure 4). Symptomatic interventions, such as neurotransmitter drugs, should have no significant impact on disease progression, as neuronal dysfunction is attributable to intermediate processes (eg, amyloid and tau changes), which are themselves attributable to a fundamental upstream mechanism. Current clinical targets (eg, monoclonal antibody approaches) aim at intermediate targets and likewise should have no significant impact on disease expression, because neuronal dysfunction is the combined result of multiple intermediate processes (including plaque or soluble52 Aβ, tau tangles, soluble or hyperphosphorylated tau,53 and mitochondrial dysfunction) rather than a single intermediate process, and since these intermediate processes are driven by a more fundamental upstream mechanism. We fail to affect AD because we aim at the wrong targets. To cure, prevent, or even slow age-related CNS disease—in humans or animals—requires that we identify the optimal target. Too often, as one editor lamented,54 our focus has been on neuropathology rather than the function of the system. Geneticists see alleles; pathologists see pathology. Rather than a systems approach, we often adopt a component approach, employing a narrow view that ultimately results in clinical failure and lost investment. A systems approach10, 55 suggests an optimal upstream target, with the downstream findings (often used as interventional targets) being merely symptomatic results of the upstream process. There is a growing interest in glial cell dysfunction as playing a central role in such a systems approach. Glial cells demonstrate prominent changes in age-related cognitive failure, for example, memory loss, in both humans and animals,56 including glial cell activation,57, 58 inadequate amyloid turnover,59-62 tau proteins,63, 64 methylation,65 homeostatic changes,66 and so on. Although the cascade of causation remains undefined, the indictment of glia as key players has become common in the literature. Clinical failure suggests that we “reassess current ideas on biological underpinnings … [and] … reinforces the necessity for a reexamination of the neurobiological premises for therapy development.”67 A comprehensive model must account for the gamut of age-related CNS dysfunction in both humans and animals, and encompass current clinical, pathology, and other data, including neuronal and glial changes, as well as the physiological and epigenetic findings. Current models not only fail to account for the gamut of age-related human CNS disease and parallel age-related animal changes but, more critically, fail to intervene effectively in human disease. The need for an effective model has prompted delineation of the required characteristics of such a model, accounting for known risk factors, particularly biological age and genetic predisposition but also head trauma, gum disease, sleep disturbances, behavioral factors, education, cognitive reserve, and so on. The influence of comorbid conditions such as tobacco use, hypertension, diabetes, and cardiovascular disease (CVD) should also be incorporated.68, 69 One example includes 10 questions that a comprehensive model should address (Figure 5). The first point raised by Khachaturian et al.70 is that the model must account for the role of the aging process itself. The single most reliable risk factor is not genetic or environmental, but biological age.71, 72 Age is the best predictive independent variable for (appropriately termed) “age-related” dementias.73 Although the central importance of the aging process is often glossed over or even ignored, it is central to not only the demography, but also the conceptual basis of any model of the dementias. As Leonard Hayflick aptly observed: “The cause of aging is ignored by the same people who argue that aging is the greatest risk factor for their favorite disease.”74 A systems model must therefore: (1) account for the role of aging; (2) account for known human and animal data, including demographic, genetic, clinical, anatomic, and laboratory data; and (3) offer an effective point of intervention. One example of a systems model posits a central role for cell senescence with downstream effects on cell function (Figure 6), resulting in secondary alterations in cell and tissue function, which then result in tertiary clinical findings. This model was outlined initially to account for age-related human pathology as well as the basic laboratory findings associated with cell senescence75 and aging phenomenon but has grown to encompass recent laboratory and clinical findings as well.76 Cell senescence was first described >50 years ago,77, 78 and its implications for age-related disease were outlined in both the lay79 and medical literature >20 years ago80, 81 and repeatedly implicated since.82 Cell senescence was first shown to correlate with,83-85 then to be the causal result of, telomere shortening and amenable to resetting in vitro.86 Subsequently, resetting of telomere length was shown to reset function in aging human tissue both ex vivo87 and in vitro,88, 89 further supporting cell senescence as a viable point of clinical intervention.90 A textbook describes the role of cell senescence in detail, tissue-by-tissue, including the role of cell senescence in age-related CNS disease, as well as potential techniques for intervention.91 A subsequent article reviewed,92 independent animal studies supported,93, 94 and a recent book95 outlined the potential for cell senescence as an effective point of intervention, and other authors made the same suggestion for etiology and intervention.96 Cell senescence correlates not with absolute but with relative changes in telomere length. Key changes in cell function occur because changes in telomere lengths result in changes in gene expression.97, 98 This phenomenon, TPE (telomere position effect) is insufficiently understood, but can result in both local (subtelomeric) and distant changes in gene expression,99 including SAGE (senescence-associated gene expression) and SASP (senescence-associated secretory phenotype). The effects of cell senescence on dividing cells (eg, glial cells) can accrue in non-dividing cells (eg, neurons), the normal function of which depends critically on dividing cells. Although these issues are addressed in detail in subsequent text, a few points should be made here. As glial cells transition to senescence, secondary dysfunction in local neuron populations occurs. Age-related CNS dysfunction is associated with glial cell activation, inflammation, and deceleration in glial turnover of local protein pools, such as Aβ, which affect neuronal function. Discussion will focus on glial cells (often microglia), but references to glia should be construed to include not only microglia, but also astroglia, oligodendrocytes, ependymal cells, and others. More direct effects of cell senescence, due to neuronal cell division, are probably minimal in comparison, although the issue remains controversial. Central to the cell senescence model is that it is not changes in telomere length per se, but rather the subsequent changes in gene expression (epigenetic changes) that play the functional role in cell senescence and in subsequent clinical disease. This is in line with the current emphasis on regulatory rather than coding variables, that is, EQTLs (expression quantitative trait loci), particularly in the context of age-related dementias.100, 101 A natural implication of the central role of epigenetic shifts is that both the cell's underlying genetics and its prior epigenetic state will strongly affect the functional outcome of these epigenetic shifts. Because of different genes, different species will express different cellular outcomes despite equivalent changes in telomere lengths. Consequently, even with equivalent levels of cell senescence, species will differ in the type and degree of cell and tissue dysfunction and will have different clinical presentations. Given the parallel process of cell senescence in the CNS, two different species may both show cognitive decline as they age, but (given those genetic differences) still vary markedly in species-specific biochemical, histological, and clinical behavioral findings. Cell senescence may be a shared mechanism, but the outcomes may vary. Pari passu, individuals within a species (with different alleles and with different epigenetic baselines) demonstrate differing cellular, pathological, and clinical outcomes. Although the underlying process—cell senescence—is shared among different species and among different individuals within the same species, the outcomes will vary between and within those species. Mice, for example, may share both the fundamental mechanisms of cell senescence and the behavioral outcome (age-related cognitive decline), yet the intermediate mechanisms—Aβ plaques, tau tangles, and so on—vary from those seen in humans. Although the variance in intermediate mechanisms accounts for the frustration in taking translational research from one species (eg, Mus musculus) to another (eg, Homo sapiens), the shared fundamental mechanism (eg, cell senescence and epigenetic shifts) suggests a viable point of intervention. Likewise, although human patients may share the same fundamental mechanisms, the individual outcome—AD, Parkinson disease, frontotemporal dementia, and so on—will depend on genetic and epigenetic differences between patients. Shifts in epigenetic expression102 resulting from the senescence process are gradual and subtle but still cause progressive cell dysfunction. The rate of turnover is surprisingly critical103 and can be defined and quantified numerically by formula.104 The process itself, however, is simple. Molecular pools (for example, Aβ) are not static, but in dynamic equilibrium as the molecules turn over, and turnover slows as cell senescence progresses. Even if the rate of molecular damage (oxidation, denaturation, and so on) remains constant, the slower rate of molecular turnover results in a gradual increase in the accrual of damage (as a percentage of the molecular pool). The percentage of damaged or dysfunctional molecules increases with cellular aging not as a result of chronological age per se, but as a result of cell senescence, which correlates with, but is distinct from, chronological age. Biological or cellular age (rather than chronologic age) drives the age-correlated increase in molecular damage, the loss of cell function, and the clinical progression of disease. Cell senescence results in age-related damage, not the other way around. This effect—slower rates of molecular turnover as cell senescence increases—renders the cell increasingly dysfunctional. The slowed turnover affects protein pools (such as Aβ), mitochondrial function (ATP and reactive oxygen species [ROS] production), lipids (eg, mitochondrial and nuclear membranes), DNA repair (increasing DNA damage especially with shorter telomeres105), and so on. In the case of Aβ, we see slower binding,59 uptake (eg, endocytosis106), ingestion, and degradation61 of Aβ in older glial cells and that these effects can be normalized by implantation of younger microglia.60 The slower molecular turnover in glial cell senescence not only results in amyloid molecules with longer “lifetimes” but increases the percentage of denatured molecules as senescence progresses. The outcome is an increasingly inefficient and dysfunctional cell. The dynamic (rather than static) nature of protein pools explains the failure of previous interventions that focused on removal of denatured molecules (eg, Aβ plaque) rather than on increasing the recycling rate of the molecular pool. Targeting amyloid plaques or tau tangles as static targets will prove ineffective, because we need to target the dynamic process that creates and dynamically maintains such plaques and tangles. Cell senescence has a species-specific basal rate roughly correlating with chronological age (the lifetime cumulative number of cell divisions), but even in a single genetically homogeneous species, the senescence of equivalent cell populations may vary. A plethora of upstream, independent variables have a significant impact on the cell division rates. Although the basal rate of cell senescence correlates with chronological age within a species, these upstream variables can accelerate the basal rate of cell division and thus cell senescence. These effects have been documented in many organs and tissues, and in the CNS. For example, damage to and loss of glial cells, secondary to infection, trauma, or radiation, results in glial cell replacement via cell division, accelerating the basal rate of cell senescence. Many independent variables can affect (almost always increase) the rate of cell senescence including genes,40, 107-109 chemotherapy,110, 111 toxins,112-114 trauma,115-117 hypertension,118, 119 stroke,120 hyperglycemia,121, 122 microbiome,123, 124 stress,125, 126 hormones,127-129 infection,130, 131 senolytic therapy,132, 133 and so on (Figure 7). These variables may be subdivided into subcategories, with supportive data, such as the host of possible infectious etiologies due to bacterial,134 chlamydial,135 fungal,136 viral,137-140 or prion-related141, 142 causes. Despite the number of upstream, independent variables, the mechanism of telomere shortening and changes in epigenetic expression is shared and provides a common diagnostic, prognostic, and therapeutic focal point. In AD, for example, data show telomere shortening in older glial cells143 and correlations between glial (or other) telomere lengths and AD status144-150 (as well as Parkinson disease151, 152) and lifespan,153 as well as epigenetic changes in neurons.154, 155 Not surprisingly,156 studies looking at inappropriate cells (such as peripheral leukocytes)157, 158 often find inclusive or misleading results. One animal study had contradictory results, as shortened telomeres are associated with early onset behavioral problems in normal mice, but if transgenic mice have been altered to produce Alzheimer's precursor protein (APP), then shortened telomeres appear to blunt the effects of the APP gene normally seen in the transgenic mice, although the relevance of the transgenic background in this context can be argued.159 Regarding glia, the clustering of microglia around Aβ-containing senile plaques has long been noted,160 and the argument made that aged microglia are unable to clear amyloid,161 with deleterious consequences to both glia and neurons. Synaptic loss, a hallmark of declining memory function with aging, may be linked to impairment of neuronal and/or glial function. Neuronal integrity and function, in turn, are highly dependent on fully functional glia. In the normal CNS, microglia engage in continuous monitoring of neuronal well-being. To ensure proper neuronal functioning, complex molecular and cellular interactions occur between neurons and glia.162 Because glia are capable of producing both neuroprotective and neurotoxic molecules, depending on neuronal signals,163 any impairment in glial function due to cellular senescence may impair neuronal activity and cognitive function in the aging brain.164-166 Over time, glia senesce, becoming less able, or unable, to maintain neuronal health. When sufficient glia senesce, neurons they once supported may become dysfunctional and ultimately die due to diminished support and maintenance. Neuronal cell death and synaptic loss results in the memory loss and other clinical findings in the dementias. Although not a dementia, a similar process may be relevant in other neurodegenerative diseases, such as amyotrophic lateral sclerosis (ALS).167 Within the adult CNS, progressive senescence in both glia and vascular endothelial cells—which routinely divide in the adult organism—can be expected to result in dysfunction and secondary loss of neurons. Neurons themselves divide less frequently in adult humans, although (despite earlier reports168) neurogenesis occurs in the adult hippocampus169 and arguably other areas, with age-related changes in the rate of neuronal cell division. Furthermore, such neurogenesis drops sharply in patients with AD170, 171 or (in mice) with short telomeres172 or increased age.105 Moreover, telomerase expression may be protective, particularly against hyperphosphorylated tau production in both humans and mice.173 More importantly, telomerase interventions spur neural stem cell production and differentiation, improve brain volume, and reverse age-related behavioral changes.93, 94 Absent such intervention, telomerase expression in human neural stem cells falls with recurrent cell division.174 Any model of age-related dementias should demonstrate predictive validity, that is, should predict and account for the outcomes of previous or proposed clinical interventions, for example, the numerous monoclonal antibody trials targeting Aβ. In predicting such results, the cell senescence model takes multiple factors into account. These include the scope of possible downstream outcomes of cell senescence, such as the effects on Aβ, tau proteins, mitochondrial dysfunction, and inflammation. This model predicts that any intervention (eg, monoclonal antibodies [MAB] or beta-site APP-cleaving protein [BACE]175, 176) aimed solely at amyloid faces two obstacles: (1) amyloid is only one among many cascades of downstream pathology resulting from cell senescence and (2) amyloid turnover is a dynamic process, and although MAB interventions may transiently improve extant plaques, they will not affect the recurrence and turnover of such plaques. The cell senescence model therefore predicts that any intervention aimed at a single portion of the cascade of pathology (ie, Aβ plaques) will show a small and transient delay in the disease course, but not affect the subsequent vector of the disease. Specifically, the cell senescence model predicts that monoclonal antibody interventions aimed at Aβ plaques will result in a very small temporal displacement in the disease process with subsequent resumption of the underlying vector of disease progression (Figure 8), with no change in mortality rate. This is precisely what clinical trials show. The cell senescence model is a comprehensive systems model, consistent with known clinical and animal data, demonstrating predictive validity, and offering a novel point of intervention for human clinical trials that may be uniquely effective and without precedent. Ideally, a comprehensive model of dementias should be capable of addressing the 10 questions raised by Khachaturian et al. In this section, we consider each question from the perspective of the cell senescence model of dementia. That cell aging drives the pathology underlying the dementias is central to the cell senescence model and the role of aging per se is therefore both natural and fundamental to the cell senescence model. Clinical observation of age-related pathology and the model are both consistent with viewing chronological and biological age as correlated but distinct. Age is the major risk factor for dementias, but there is substantial individual age variance in the onset and progression of dementias. This is equally true of cell senescence, in which age itself is the major predictor, but there is substantial individual variance in the onset and progression of cell senescence. The cell senescence model proposes that in any given tissue, the lifetime events—including diabetes, infection, poor nutrition, inflammation, trauma, toxins, and so on—provide such variance in the trajectory of cell senescence within a tissue and a correlated variance in the trajectory of age-related disease, including dementias. The model therefore suggests that chronological age correlates with and to an extent drives the trajectory of dementias as a result of a basal rate of cell loss and division,177, 178 but also accounts for individual variance through relevant genetic differences40, 179-181 such as APP and presenilins, as well as lifetime events (with epigenetic consequences182) that may accelerate the underlying basal rate of advancing cell senescence. In short, cells age with advancing years, but any number of factors may accelerate the underlying aging process. Both the underlying aging process and factors that accelerate cell senescence determine the onset and progression of dementia in each patient. There are three operative variables: (1) genetic background, (2) epigenetic background, and (3) epigenetic changes secondary to cell senescence per se. “Average/usual agin