Neuroimaging outcomes in clinical trials in Alzheimer's disease

Abstract

The first disease modifying drugs targeting beta amyloid that were tested in phase II and III clinical trials have been disappointing. We believe that failures descended from a leaky drug development pipeline where insufficient attention has been devoted to valid animal models and valid imaging markers of disease progression. In the future, valid animal models will need to take into greater consideration the natural and molecular history of AD, where both beta amyloid and tau play a key role. Valid imaging markers of disease progression will need to be identified in humans and translated into animal versions. Future testing of putative disease modifying drugs in valid animal models with valid imaging markers of disease progression will allow to maximize the predictability of their effect in phase II and III clinical trials.

The detection of disease modification in Alzheimer’s disease

Disease modification is the modern holy grail of drug treatment for Alzheimer’s disease but, so far, no drug has convincingly shown disease modifying effects. Disease modification implies an effect on the very core of disease neuropathology such that cognitive deterioration can be slowed or halted altogether and adverse outcomes such as disability and institutionalization may be delayed ideally indefinitely.

Showing disease-modifying effects of a drug poses specific challenges. While it might be contended that correction of the etiological agent may lead to halting of the degenerating process, system boosting by neuronal plasticity, and immediate cognitive improvement followed by stabilization, it may also be that decades of ongoing neurodegeneration (1) will have led to impoverished plasticity to an extent that short term improvement may not take place and stabilization would be the only beneficial effect. The latter worst-case scenario implies that the design of trials for disease modifying drugs should be radically different from that of symptomatic drugs, where the trajectories of treated and untreated diverge as early as 4 weeks after trial inception but remain parallel thereafter. In the case of disease modifying drugs, trajectories are supposed to diverge slowly since trial inception and the gap between treated and untreated widens with time. Here, for a given effect size to be demonstrated, group size is proportional not only to drug efficacy (as is the case of symptomatic drug trials), but also to the length of follow-up.

This is the reason why phase II trials of the disease modifying drugs currently in phase III have involved hundreds of patients followed for as long as 18 months – nevertheless almost invariably failing to detect clear-cut positive clinical effects. The usefulness of disease markers sensitive to disease progression lies in the huge reduction of group size and length of follow-up, which could greatly hasten the drug discovery process (3).

Atrophy rate is a marker sensitive to disease progression

Progression of atrophy on MR-based structural imaging is the most extensively validated surrogate outcome for trials of disease modifying drugs in AD. The Guideline on Medicinal Products for the Treatment of Alzheimer’s Disease and Other Dementias issued by the Committee for Medicinal Products for Human Use (CHMP) of the EMEA (European Medicines Agency) (4) states that a disease modifying effect can be claimed if “delay in the natural course of progression of the disease based on clinical signs and symptoms […] can be established […] supported by a convincing package of biological and/or neuroimaging data, e.g. showing delay in the progression of brain atrophy”.

However, the phase III AN1792 vaccine trial, employing atrophy rates as surrogate outcome measure, has returned unexpected results (5). Treated patients with a positive response to the Abeta antigene and delayed deterioration of a set of cognitive tests (6) showed greater atrophy rates than those who did not respond to treatment. A number of ex post explanations have been provided, among which the physical space occupied by the amyloid being cleared by the immunologic cascade (5), but the fact remains that a phase III trial in humans has given results exactly opposite to expectations. Seeing as the steps of the pipeline leading to the development of disease modifying drugs has not changed greatly since the unsuccessful AN1792 trial, this observation raises worries about the predictability of the results of the clinical trials of disease modifying drugs currently in or about to enter phase III.

The ideal pipeline of drug development

The ideal pipeline of drug development involves the following steps: (i) identification of a valid marker of disease progression (VMDP) of disease progression in humans; (ii) development of animal models where the valid marker can be translated and has similar natural history (valid animal model, VAM); (iii) test of the efficacy of potential drugs in VAMs with the VMDP as surrogate outcome; and (iv) phase I/II/III studies in humans. This logical chain has been followed very loosely in the development of AN1792 and the other Alzheimer’s disease-modifying drugs due to the use in the preclinical phase of markers of disease progression of doubtful validity, if any, and the lack of consistent animal and human versions of VMDP.

The pipeline followed by current putative Alzheimer’s disease-modifying drugs

Identify a valid marker of disease progression (VMDP) in humans

Fox and colleagues (2) have provided evidence that yearly brain tissue loss is closely related to yearly loss of points on the Mini Mental State (figure 1). Seeing as AD is defined as a condition featuring progressive cognitive deterioration, this represents a strong argument that progressive atrophy may be a VMDP.

Figure 1.

Evidence of validity of rate of whole brain atrophy as a marker to track disease progression. Rates of atrophy are strongly related to rates of cognitive change (2)

Rates of hippocampal atrophy have been shown to correlate with cognitive deterioration and cognitive stability in patients with mild cognitive impairment and healthy elders (7). However, observing the pathway of neurodegeneration in the AD brain (Braak stages at the neuropathological level, Delacourte stages at the biochemical level) (8, 9) it is clear that the hippocampus may not the earliest atrophic structure. Atrophy develops hierarchically, starting in the entorhinal cortex, then extending to the hippocampus, and later involving then temporal pole and inferior temporal cortex. Moreover, atrophy may be a late disease marker in that when atrophy is observed in the neocortex, tau pathology is likely well developed and may be irreversible (1, 10). A significant step forward would be represented by the development of a panel of markers (imaging and non imaging) that change consistently over time and are sensitive to disease progression in humans. Testing putative disease modifying drugs with such panel of markers, rather than a single marker, in animal models and back to human would greatly enhance the predictability of the effect of the drug in phase II and III clinical trials. Interestingly, it is worth underlining that the sensitivity to change of CSF markers such as abeta42, total tau, and phospho-tau (11) is doubtful.

Develop animal models where the VMDP can be translated and has similar natural history (valid animal model - VAM)

Although measuring whole brain and hippocampal atrophy in animal models is technically feasible, few studies have addressed this issue, possibly because currently available animal models have little if no atrophy.

Brain volume changes with ageing have been studied is the APP transgenic mouse (12). Here, at 1 year of age no brain volume differences could be detected versus wild type mice in any of the studied regions.

Figure 2.

Whole brain volume at 12 months in APP transgenic mice (red arrow) expressed as proportion of brain volume of wild type mice (12)

A more sophisticated design was implemented in the study by Lau and colleagues (13) who have measured whole brain and hippocampal volumes in double APP/PS1 transgenic mice at birth and at 4, 6, and 8 months of age (figure 3). The animals have been shown to have brain development similar to wild type mice, but brain volumes were lower at all ages. This indicates a developmental rather than ageing-associated phenomena. The weakness of this study is that 8 months’ observation may not be enough to detect ageing-associated neurodegeneration in animals that can live over 24 months.

Figure 3.

Brain volume changes over 8 months in double APP/PS1 transgenic (13)

One of the few regions where an interaction was detected between genotype and ageing (i.e. the statistical effect that would be expected if the transgenic animals had age-associated neurodegeneration) was the entorhinal cortex. Here, increasing volume was detected in wild type mice with time (interpretable as brain maturation) but no volume change with ageing in transgenic mice. However, this is not what one expects to find if the latter is a model of neurodegeneration, where initial brain volume increase should be followed by brain volume decrease. Thus, even this effect can be parsimoniously be interpreted as being due to neurodevelopmental rather than neurodegenerative factors. Alternatively, it might be hypothesized that 8 months’ observation may not be enough to detect ageing-associated neurodegeneration in animals that can live over 24 months.

Also studies of CSF markers in animal models are not numerous. Kawarabayashi and colleagues (14) have studied concentrations of Abeta42 in the brain and CSF of Tg2576 mice over time. They have found that increased brain deposition of Abeta42 can be detected from week 9 and concentrations continue to rise until at least week 23. Interestingly, they have found a symmetric and chronologically consistent decrease of Abeta42 in the CSF. The problem with CSF markers is that we have much fewer information on the natural history of biomarkers in humans with AD than we have on the natural history of structural changes, and therefore the validity of CSF markers for tracking disease progression is more doubtful.

These observations raise serious concerns that the above represent VAMs to test Alzheimer’s disease modifying drugs. In addition and in the same direction, the nature of atrophy in humans is clearly related to tau pathology. Indeed, the hippocampus and antorhinal cortex are the first regions to be affected by tauopathy (8, 9) and the first regions with significant atrophy in the course of AD in humans (15). At variance, in transgenic mice medial temporal atrophy is related to amyloid burden (huge excess of APP transgene and Abeta deposits) since no tauopathy here is developed (16). The ultimate goal for a VAM is its relevance to AD physiopathology not only at the phenotype level (plaques and tangles), but also at the mechanistic level, with an APP/Tau synergy, as observed in the human brain (17, 18).

The effect of potential drugs on the VMDP should be tested in VAMs

Of the disease modifying drugs currently in phase III or about to enter phase III, for only a couple data on their ability to modify a putative marker of AD in an animal model is available. The peripheral administration of the antibody m266, currently undergoing human testing, leads to reduce brain Abeta burden in the brain and increased Abeta42 in the CSF of the PDAPP mouse. The mechanism is believed to lie in altered clearance of Abeta from the central nervous system to the plasma (19). Notably, the gamma secretase inhibitor LY450139 also has been studied in APP transgenic mice, but it has been shown to cause further decrease of CSF Ab42 (20), consistently with its putative effect of decreased production of Abeta in the central nervous system. Unfortunately, we have no information on whether the above animal models develop brain atrophy and the effects of the above drugs on progression of atrophy.

It should be noted that two drugs supposed to have disease modifying effects have been shown to lead to opposite effects on a putative marker of disease progression such as CSF Abeta42. This suggests that the effect of disease modifying drugs on VMDPs, whether structural or biochemical, may not be a mere reversion of the marker to normal or flattening out of the slope of VMDP changes over time, but may yield a time course specific to the mechanism of action of a given drug.

Phase I/II/III studies

It can be inferred from the above sections that of all the disease modifying drugs currently undergoing human testing, even those with the most extensive animal data – at least those publicly available – have only loosely followed the logical chain that should allow to predict human results based on animal testing.

Indeed, the initial faulty step is the availability of a VAM for Alzheimer’s disease, as most available models do not show atrophy and some do not even show cognitive-behavioural age-associated deterioration. Most researchers tend to agree that currently available models are mechanistic factories of enhanced amyloid production. It is therefore not surprising that phase II and III trials give unexpected results on imaging and biological markers.

Conclusion

It is time to develop diagnostic and therapeutic strategies that take into greater consideration the natural and molecular history of AD, rather than relying on pure brain amyloidosis. While a valid marker of Alzheimer’s disease progression is available in humans, still a number of issues need to be addressed before disease modifying drugs can be confidently translated from preclinical to clinical studies. Specifically: (i) VAMs need to be developed where the natural history disease markers mirrors that in humans, (ii) animal versions of the human VMDP need to be developed and translated to the VAMs, and (iii) candidate drugs need to be tested in the VAMs with the animal version of the VMDP. This pipeline should maximize the predictability of the effect of candidate drugs in phase II and III clinical trials.

Financial disclosure: None of the authors had any financial interest or support for this paper.