Use of theragnostic markers to select drugs for phase II/III trials for Alzheimer disease

In a slowly progressive disorder like Alzheimer disease, evaluation of the clinical effect of novel drug candidates requires large numbers of patients and extended treatment periods. Current cell- and animal-based disease models of Alzheimer disease are poor at predicting a positive treatment response in patients. To help bridge the gap between disease models and large and costly clinical trials with high failure rates, biomarkers for the intended biochemical drug effect may be of value. Such biomarkers may be called 'theragnostic'. Here, we review the literature addressing the prospective value of these biomarkers.


Background
Th ree decades of multidisciplinary research have resulted in detailed knowledge of the molecular pathogenesis of Alzheimer disease (AD) [1]. We know that the symptoms of AD are caused by synaptic dysfunction and neuronal death in the areas of the brain that are involved in memory consolidation and other cognitive functions [1]. Th is neurodegeneration is fi rmly asso ciated with aggre gation of the 40-to 42-amino acid amyloid beta (Aβ) peptide into senile plaques, phosphorylation and aggre ga tion of tau proteins that form neurofi brillary tangles, and microglial activation that may be a protective response or contribute to the neuronal dysfunction and damage [2]. Th e relative impor tance of these processes to the clinical presentation of the disease remains uncertain.
Clinical trials of novel anti-AD drugs face at least two major challenges. First, the new types of drug candidates that attack basic disease processes are likely to be most eff ective in early stages of the disease, before neuronal degeneration has become too widespread and severe [3]. However, clinical methods that recognize early AD are lacking. Second, the drug candidates may slow down the degenerative process without having any immediate and easily recognizable symptomatic eff ect [4]. Th is makes evaluation of the drug eff ect diffi cult. Th eragnostic biomarkers (that is, biomarkers that detect and monitor biochemical eff ects of the drug) may help solve some of these problems. Here, we review three pathological processes that are thought to be involved in the complex surge of AD -namely the amyloid cascade, abnormal tau phosphorylation, and microglial activation with neuroinfl ammation -and the currently available biomarkers thought to refl ect them ( Figure 1).

Core biomarkers of Alzheimer disease
It is well established that cerebrospinal fl uid (CSF) levels of total tau (T-tau), phospho-tau (P-tau), and the 42-amino acid fragment of Aβ (Aβ42) refl ect core elements of the AD process [3]. T-tau is a marker of cortical axonal degeneration and disease activity [5][6][7]. P-tau refl ects neurofi brillary pathology [8,9]. Aβ42 is a marker of plaque pathology [9][10][11][12]. Together, these biomarkers identify AD and predict AD in mild cognitive impairment (MCI) with a sensitivity and specifi city of 75% to 95% [3]. Th e predictive power is, however, sub optimal in general populations as compared with MCI cohorts because of the lower prevalence of incipient AD in this group [13]. Plasma biomarkers refl ective of patho physiological changes in the AD brain are highly warran ted, the subject of intense research, but unfor tunately still lacking [3].

Amyloid
Experimental data, as well as longitudinal studies in humans, suggest that certain forms of Aβ may act as initiators in the disease process with potent toxic eff ects at the synaptic level [2]. Based on this knowledge, novel treat ments aimed at inhibiting Aβ toxicity have been developed and are being tested in patients [14]. Th ese include secretase inhibitors and modulators that aff ect the production of Aβ from amyloid precursor protein (APP), immunotherapy aimed at increasing the clearance

Abstract
In a slowly progressive disorder like Alzheimer disease, evaluation of the clinical eff ect of novel drug candidates requires large numbers of patients and extended treatment periods. Current cell-and animalbased disease models of Alzheimer disease are poor at predicting a positive treatment response in patients. To help bridge the gap between disease models and large and costly clinical trials with high failure rates, biomarkers for the intended biochemical drug eff ect may be of value. Such biomarkers may be called 'theragnostic' . Here, we review the literature addressing the prospective value of these biomarkers.
of Aβ from the brain, and Aβ aggregation inhibitors that should prevent pathological build-up of the peptide in the brain [14].

Tau
Among the typical brain lesions in AD are neurofi brillary tangles that consist of abnormally phosphorylated forms of the microtubule-stabilizing protein tau [15]. Tau expres sion is high in non-myelinated cortical axons, especially in the regions of the brain (such as the limbic cortex, including the hippocampus) which are involved in memory consolidation [16]. Hyperphosphorylation of tau causes the protein to detach from the microtubules and destabilizes the axons [17]. Th is process promotes axonal and synaptic plasticity in the developing brain [17] but may be pathological in the adult brain and specifi cally related to a group of disorders referred to as tauopathies; this group includes AD and some forms of frontotemporal dementia [15]. Inhibiting tau phos phory lation or aggregation has been considered a promising strategy to slow down the neurodegeneration in AD. Drug candidates intervening in tau-related disease pro cesses (for example, inhibitors of the tau kinase GSK3β and tau aggregation inhibitors) exist but are still in an early phase of development [14].

Microglial activation
Microglia are the resident immune cells of the central nervous system (CNS) [18] and are macrophages of myeloid lineage and invade the CNS during embryogenesis. Th ese innate immune cells perform the majority of the immunological surveillance in the CNS. However, in certain conditions such as multiple sclerosis or neuroborreliosis, infi ltration of T cells but also B cells into the CNS occurs. Microglia are usually in a resting state but at any time may become activated in response to infection or injury [18]. Th e key question of microglia in AD is whether the infl ammation mediated via micro glia is benefi cial or not. Th e capability of microglia to release reactive oxygen species, nitric oxide, interleukin-1-beta (IL-1β), and tumor necrosis factor-alpha (TNFα) is benefi cial in response to invading pathogens. However, these compounds are also neurotoxic and collateral damage to neurons is frequent during infections. Th e same may occur in AD because plaques function as immunological triggers for the activation and recruitment of microglia, which may result . Gamma-secretase inhibitors should reduce Aβ1-40 and Aβ1-42 and increase Aβ1-14, Aβ1-15, and Aβ1-16. Both Aβ immunotherapy and anti-aggregation agents might be monitored by CSF levels of Aβ1-40 and Aβ1-42. Therapy-induced Aβ degradation might be monitored by CSF levels of diff erent Aβ peptides, depending on the proteolytic pathway used for degradation. Aβ effl ux from the brain to the blood might be monitored by measurement of Aβ in CSF and plasma. Infl ammatory markers in plasma and CSF as well as CSF levels of CCL2 and chitotriosidase activity are putative markers of microglial activity and may change in response to treatments that infl uence microglial activity. Treatment with tau hyperphosphorylation inhibitors might be monitored with CSF phospho-tau (P-tau) levels. Downstream eff ects on axonal degeneration from disease-modifying treatments could be monitored by using the axonal damage markers CSF total tau (T-tau) and neurofi lament light protein (NFL).
in neuron loss [19]. On the other hand, microglia have been shown to clear deposits of Aβ through the Toll-like receptor 4 (TLR4), and AD mice with a defective TLR4 have increased deposits of Aβ [20].

Other drug targets
Besides the three targets mentioned above, several other approaches are aimed at improving neural transmission and memory consolidation in AD. Th ese include nerve growth factor gene therapy, stimulation of nicotinergic acetylcholine receptors by varenicline, protein kinase C activation by bryostatin 1, and many more [21]. Th eragnostic biomarkers for each of these drugs may be diff erent from those reviewed below and are specifi cally related to the mode of action of the drug.

General issues
Th eragnostic markers have accelerated the development of treatments in some types of cancer, HIV infection, atherosclerosis, and multiple sclerosis, and cancerspecifi c fusion transcripts or mutations, viral load, plasma levels of low-density lipoprotein cholesterol, and brain MRI (magnetic resonance imaging) white matter lesion burden, respectively, have been used to ascertain that the drug candidate is benefi cial [22]. Th ese examples indicate that theragnostic markers may be useful in evaluating novel therapeutics also in AD. Furthermore, such studies may help to bridge the gap between animal studies that are poor at predicting treatment success in humans and large clinical trials [1]. Sometimes, these types of bio markers are referred to as surrogate markers of patho genic processes. However, the term surrogate marker often indicates a marker that is (i) a validated substitute for a clinically meaningful endpoint and (ii) expected to predict the eff ect of therapy [23,24]. Th is defi nition goes beyond a mere correlation between a laboratory measure ment and a clinical outcome or a pathogenic process since a fully validated surrogate marker also requires proof that intervention on the surrogate marker predicts the eff ect on the clinical outcome [25]. If applied in full by regulatory authorities, very few biomarkers in medicine live up to these requirements, which may obstruct implementation of surrogate biomarkers in large-scale clinical trials. However, this circumstance does not hinder the use of non-validated surrogate markers when deciding upon the most promising drug candidates in early stages of drug development. Rather, this approach is advocated by the US Food and Drug Administration [26].

Are they useful?
To date, only preliminary reports suggest that CSF biomarkers may be useful in detecting and monitoring biochemical eff ects of novel drugs against AD. With regard to biomarkers for amyloid pathology, the many factors that infl uence steady-state levels of Aβ in CSF (production, aggregation, enzymatic clearance, and bidirectional transport across the blood-brain barrier) make it diffi cult to predict what diff erent amyloidtargeting treatment paradigms might do to CSF Aβ concen trations. In fact, any treatment-induced change to an amyloid-related biomarker which is informative with respect to clinical outcome would be a major step forward. So far, data from animal studies show that γ-secretase inhibitor treatment results in a reduction in cortical, CSF, and plasma levels of Aβ [27,28]. Similarly, treatment of monkeys with a BACE1 inhibitor reduced the CSF levels of Aβ42, Aβ40, and β-sAPP [29]. Other promising biomarkers that are closely linked to the amyloidogenic process in AD are CSF BACE1 (the major β-secretase) concentration and activity, CSF levels of α-and β-cleaved soluble APP, and Aβ oligomers [30][31][32]. Th ese biomarkers appear to provide information of limited diagnostic usefulness but may turn out to be important for identifying treatment eff ects of drugs that are meant to inhibit βsecretase or break up amyloid aggregates.
In patients with AD, it is uncertain how CSF Aβ42 may respond to treatment with effi cacious anti-Aβ drugs. A phase IIa study of the Aβ clearance-enhancing compound PBT2 showed a signifi cant dose-dependent reduction in CSF Aβ42 levels during treatment [33]. Data from a clinical study on the amyloid-targeting drug phenserine also showed changes in CSF Aβ levels in response to treatment [34]. However, in the interrupted phase IIa AN1792 trial of active immunization against Aβ, no signifi cant treatment eff ect on CSF Aβ42 was found [35]. A clinical study on γsecretase inhibitor treatment also failed to detect any eff ect on CSF Aβ42 levels [36]. Nevertheless, when the eff ect of this drug on Aβ produc tion rate by the use of a stable isotope-labeling kinetic tech nique was evaluated, a clear inhibitory eff ect of γ-secretase inhibition on Aβ production was identifi ed [37]. Recent data show that shorter Aβ peptides in CSF -namely Aβ1-14, Aβ1-15, and Aβ1-16represent a novel APP-processing pathway [38] that is upregulated in a dose-dependent manner in response to γsecretase inhibition [39].
Given longitudinal studies of conditions involving acute neuronal injury [40] and data from the interrupted phase IIa AN1792 trial [35], T-tau should decrease toward normal levels if a treatment is successful in inhibit ing the neurodegenerative process in AD. Th e same may be expected for P-tau, as suggested by two recent pilot studies on memantine [41,42].
Currently, there are no established CSF biomarkers for microglial activation which could be used as theragnostic markers in trials aimed at inhibiting, boosting, or modulating microglial activity in AD. Chemokine (C-C motif ) ligand 2 (CCL2) (also called monocyte chemoattractant protein-1, or MCP-1) and chitotriosidase are fi rmly associated with macrophage activation in the periphery [43,44] and may be considered promising markers of microglial activation in the CNS, but studies in relation to AD are scarce [45]. However, several biomarkers for general infl ammation exist. Pilot studies showed increased CSF levels of transforming growth factor-beta (TGFβ) in AD as compared with controls [46,47]; this result was recently confi rmed in a meta-analysis of cytokines in AD [48]. Other classical markers such as IL-1β, IL-6, and TNFα were not altered in the CSF of patients with AD. Th e plasma levels of several cytokines such as IL-1β, IL-6, IL-12, IL-18, TNFα, and TGFβ -but not IL-4, IL-8, IL-10, interferon-γ, or C-reactive proteinwere increased in AD. Together, these data argue for an infl ammatory component in AD. However, the results of anti-infl ammatory therapy in AD have been contradictory [49]. As explained above, the link between infl am m ation and other core disease processes in AD remains elusive.

Concluding remarks
Th eoretical reasoning suggests that theragnostic biomarkers could play a major role in drug development against AD, but, admittedly, the body of literature supporting this view is limited at present. We know quite a lot about central pathogenic features of the disease, and several biomarkers that monitor these features exist. A number of phase 0-I clinical trials indicating small but statistically signifi cant eff ects on theragnostic biomarkers, mostly in relation to axonal integrity and amyloid pathology, have been published. Interpreting these biomarker results is, however, complicated by the fact that none of the studies was designed to detect clinical eff ects. Th is circumstance precludes analyses of whether the patients with biomarker changes imposed by the treatment were those with the clearest clinical benefi t.
Th e recent interruption of the phase III trials (IDENTITY [Interrupting Alzheimer's Dementia by Evalu ating Treatment of Amyloid Pathology] and IDENTITY-2) of the γ-secretase inhibitor semagacestat (LY450139) (Eli Lilly and Company, Indianapolis, IN, USA) may be considered a blow to the fi eld of theragnostic biomarkers. Despite compelling evidence in cell and animal models, as well as plasma Aβ data [36] and Aβ turnover rates [37] in humans, suggesting that the compound reduces Aβ production, cognition declined faster in the treatment arms compared with placebo. In our view, these data should spur us to continue developing more biomarkers for APP-and Aβprocessing for other desired drug eff ects such as improvement of neural transmission as well as for undesired eff ects (for example, inhibition of Notch signaling). For another recently failed trial (tarenfl urbil, which is supposed to act as a γ-secretase modulator), there were plenty of bio marker data suggesting that the drug did not hit its target in the human brain [50]. Th ese data could have curbed the enthusiasm to move to phase III and thus saved a lot of money.
Several other clinical trials on disease-modifying drug candidates which include biomarkers as readouts are currently ongoing. Th ese trials will provide more evidence on whether biomarkers will be useful as tools to select the most promising drug candidates for phase II/ III trials for AD.

Competing interests
HZ has served on an advisory board for GlaxoSmithKline (Uxbridge, Middlesex, UK). KB has served on an advisory board for Innogenetics (Gent, Belgium). The other authors declare that they have no competing interests.