An often overlooked problem is that certain neurotransmitter systems are differently wired in human vs. rodent settings. In some instances, even certain drug receptor subtypes only exist in human brain and not in mouse brain. A typical example is the human 5HT1D-R, site of action for many marketed antimigraine drugs. However, in the mouse, the 5-HT1D receptor is downregulated at birth and no functional expression is seen in vivo (4). For this type of pharmacology, mice are clearly not the right animal.

Other dramatic differences can be found in actual receptor localization with Table 2 giving a number of examples.

The 5-HT6 receptor is of particular interest, because changes have been reported in the brain of AD patients (9) and a number of 5-HT6 antagonists are currently in development for cognitive deficits. The observation of a different mouse 5-HT6 receptor distribution limits dramatically the testing of 5-HT6 drugs in transgene Alzheimer mouse models.

Advances in genomic technology have identified key functional genotypes with clinical effects in human subjects with the APOE isoform as an obvious example in the Alzheimer field. The presence of the APOE4 gene confers a higher risk for younger age of onset, although the natural course as measured by the ADAS-Cog does not seem to be influenced by the presence of the APOE4 allele (10). The APOE gene in mouse does not have these different genotypes (see (11) for a discussion of the consequences in mouse models). In addition, there are important differences in the mouse vs. the human promoter (12), which would in theory necessitate the inclusion of the promoter in all transgene constructs. Despite a lot of effort in transgene animal models, the biological role of APOE4 in the human disease has not been elucidated completely.

Other examples with relevance to cognitive performance include the Val158Met SNP in the Catechol-O-methyl-transferase (COMT) gene, the protein product of which is involved in degradation of catecholamines. Subjects with the Met/Met allele have a less stable COMT gene and lower protein expression, which leads to higher free levels of dopamine and noradrenaline in the prefrontal cortex. A number of clinical studies have shown that subjects with the Met/Met allele perform consistently better on a variety of cognitive tasks (for a review, see (13)). The functional effect suggests that this genotype affects the clinical outcome of dopaminergic or adrenergic drugs, as shown in the effect of olanzapine on an N-back working memory test on schizophrenic patients (14).

Another example is the recent clinical observations that clinical antidepressant response of certain drugs is associated with particular genotypes of the ABCB1 drug transporter, when the drug is a substrate of this transporter (15). This suggests a more complex degree of interaction at the blood–brain barrier than can be achieved in rodent models.

Because rodents often do not have similar functional genotype, it is close to impossible to assess its effect on the functional effect of a candidate drug. Failure to recognize this issue might lead to an imbalance in the differently powered arms in clinical trials.

Drug metabolism in rodents is usually much faster than in humans. Many years of expertise in pharmacokinetics has allowed to often successfully extrapolating the PK profile observed in rodent models to human situations. However, predicting the human pharmacodynamics (i.e. the pharmacological activity) has been much more difficult.

This has been illustrated in a systematic study of different antipsychotics in rats vs. humans (16). In this case, the actual brain D2R occupancy can be readily measured using PET imaging of ¹¹C-raclopride. Using in vivo radiography in the rat model, in order to achieve similar exposure as seen in patients, repeated drug applications or minipump solutions needed a dose at least five times higher than the optimal single-dose concentration. This is because metabolism of these drugs in rats is four to six times faster than in humans. As a consequence, many preclinical drug studies often use concentrations that are by no means relevant for the clinical situation.

The availability of an imaging tracer or a functional CNS readout, like pharmacoEEG, which links rodent plasma concentration to a functional brain readout in order to determine the active dose-range allows for a correct extrapolation to the human situation; however, such an ideal situation is seldom achieved. Furthermore, in many cases, the translational link with the clinical situation is not appreciated.

Another problem in the predictability of animal models is the possible difference in blood–brain barrier (BBB) permeability of certain drugs. While the make-up of the rodent BBB is very similar to the human, a species-specific drug–PgP interaction sometimes makes drugs behave differently. Again, having an imaging tracer or a functional CNS readout available in the clinical setting allows one to rapidly determine how readily the compound gets into the brain.

Over the last years, numerous transgene animal models have been developed where the genetic make-up of the animal is modified to introduce certain wild-type or mutated human genes, involved in the familial forms of AD. Cross-breeding of different lines in theory leads to the expression of different neuropathology forms in the same animal. Examples include the triple transgene 3xTg mice with PS1, APP, and tau mutations (17), currently the model, which captures the broadest AD neuropathology. While this mouse model is an important research tool that has clarified aspects of the interaction between Aβ deposits and tangle formation, it is by no means a complete model of the human Alzheimer neuropathology. For instance, in the 3xTg mouse, an age-dependent decrease in α7 nicotinic acetylcholine receptor (nAChR) density has been reported (18), but no α4β2 nAChR loss. This is different from the clinical AD pathology, in which consistent loss of α4β2 nAChRs has been seen, even in vivo (19) and sometimes, but not always loss of α7 nAChRs. Some of these pathologies interact with each other; for instance, there is mounting evidence for a direct interaction between nAChRs and brain microglia that modulate the degree of brain inflammation (20). With many of the documented pathologies interacting with each other, it is clear that the limited ability to model these interactions appropriately in animals has important consequences for the predictability of clinical efficacy.

Animal models try to capture the long-life dynamics of the pathology in a few months; this compression necessarily leads to a limited simulation of the full spectrum of the human pathology. For instance, many mice models using hAPP overexpression already report “cognitive” deficits well before plaques are present (21), suggesting a disconnection between functional deficits and amyloid aggregation. Such data have been explained by the fact that smaller nonfibrillar Aβ oligomer aggregates are toxic to neuronal synapses and have been interpreted as a pre-Alzheimer or an MCI (i.e., mild cognitive impairment) stage.

In contrast, in the human situation recent imaging studies using the amyloid agent PIB-1 have revealed that in normal volunteers and MCI, the degree of retention of the tracer (which is related to the amyloid plaque density) is well correlated with the degree of cognitive episodic memory impairment, as measured by the Rey Complex Figure test and the California Verbal learning test, both at long delays (22). There was no correlation in the case of Alzheimer patients, presumably because of a ceiling effect.

A large majority of AD animal models are based on genetic modification using human genes, which predispose to early familial type of the disease. The dynamics of this familial type are very likely different from the sporadic case, for there is between a 20 and 40 year difference in age of onset. Of note is that there are other “more sporadic” animal models of AD, such as a rabbit model fed on a diet of 2% cholesterol with traces of copper (23), which might contribute more to understanding the interplay between neuropathology, cognitive deficit, and therapeutic response for the sporadic case.

Both in real-life clinical situations and in well-controlled clinical trial studies, patients are allowed to take additional medication. For instance, many AD patients are treated with acetylcholinesterase (AChE)-inhibitors or NMDA antagonists defining a placebo active comparator arm. However, some of these medications do have indirect but modest effects on some of the intracellular pathways involved in amyloid pathology.

As an example, increasing cholinergic tone by reducing ACh hydrolysis activates m1 muscarinic receptors, which have been shown in preclinical models to modulate both tau and amyloid related pathways (for a recent review see (24)). Galantamine has been shown to have pleiotropic activities in vitro and in vivo, due to its allosteric modulation of the nAChR (25). Similarly, meta­botropic glutamate receptors, which are indirectly modulated by nAChR-dependent changes in presynaptic glutamate release or by antagonism at the NMDA-R, can modulate APP expression (26).

In many clinical trials, antidepressant medication is allowed and recent well-reproduced data suggest a direct effect of antidepressants on hippocampal neurogenesis (27). Also, atypical antipsychotics, which are sometimes used to address behavior problems, have been shown to be involved in neurogenesis, stimulating proliferation of neuronal stem cells and microglia (for a review, see (28)). This could have important effects on various effects of AD neuropathology. These data show that the existing medications, which are often used in combination with the investigative drug, can influence various aspects of AD pathology. Yet, in many preclinical animal models, candidate drugs are often selected based on experiments in which no real polypharmacy is simulated. Although such combination studies would delay the discovery project, assessment of possible pharmacodynamic interactions can be very important. In addition, multitarget-directed ligands, which combine symptomatic improvement (for instance using AChE inhibition) with a disease-modifying action in one molecule, will have a much larger chance of successful clinical development. However, this necessitates modifying the current mantra of “one drug, one target,” currently adhered to in many medicinal chemistry departments in pharmaceutical companies.