Simulating the Effects of Common Comedications and Genotypes on Alzheimer’s Cognitive Trajectory Using a Quantitative Systems Pharmacology Approach

Abstract

Background:

Many Alzheimer’s disease patients in clinical practice are on polypharmacy for treatment of comorbidities.

Objective:

While pharmacokinetic interactions between drugs have been relatively well established with corresponding treatment guidelines, many medications and common genotype variants also affect central brain circuits involved in cognitive trajectory, leading to complex pharmacodynamic interactions and a large variability in clinical trials.

Methods:

We applied a mechanism-based and ADAS-Cog calibrated Quantitative Systems Pharmacology biophysical model of neuronal circuits relevant for cognition in Alzheimer’s disease, to standard-of-care cholinergic therapy with COMTVal158Met, 5-HTTLPR rs25531, and APOE genotypes and with benzodiazepines, antidepressants, and antipsychotics, all together 9,585 combinations.

Results:

The model predicts a variability of up to 14 points on ADAS-Cog at baseline (COMTVV 5-HTTLPRss APOE 4/4 combination is worst) and a four-fold range for the rate of progression. The progression rate is inversely proportional to baseline ADAS-Cog. Antidepressants, benzodiazepines, first-generation more than second generation, and most antipsychotics with the exception of aripiprazole worsen the outcome when added to standard-of-care in mild cases. Low dose second-generation benzodiazepines revert the negative effects of risperidone and olanzapine, but only in mild stages. Non APOE4 carriers with a COMTMM and 5HTTLPRLL are predicted to have the best cognitive performance at baseline but deteriorate somewhat faster over time. However, this effect is significantly modulated by comedications.

Conclusion:

Once these simulations are validated, the platform can in principle provide optimal treatment guidance in clinical practice at an individual patient level, identify negative pharmacodynamic interactions with novel targets and address protocol amendments in clinical trials.

Keywords

ADAS-Cog clinical variability pharmacodynamic interactions polypharmacy protocol amendments

INTRODUCTION

In clinical practice, Alzheimer’s disease (AD) patients are treated with standard-of-care, if available, but additional comedications are added often to address other comorbidities with, for instance, an estimated 40% and 20% of AD patients taking antidepressants and antipsychotics, respectively, for behavioral disturbances [1]. Current guidelines are in place for Pharmacokinetic Drug-Drug interactions (PK DDI) whereby one drug affects the metabolism of the other drugs or where a specific genotype of the metabolizing enzymes determines individual drug dose. However, the sometimes negative pharmacodynamic Drug-Drug interactions (PD DDI) are not very well studied in detail, most likely leading to a less than optimal treatment paradigm. As an example, the use of drugs with direct or an indirect anticholinergic activity leads to a higher risk of AD [2, 3] and the number of medications in elderly patients is correlated with higher incidence of unplanned hospitalization [4], while polypharmacy is associated with lower cognitive performance [5]. Furthermore, polypharmacy might mimic glucose metabolism changes of dementia in elderly subjects [6] and neurocognitive effects of comedications have been suggested as a possible reason for the failure of clinical trials [7].

Common genotype variants such as the COMTVal158Met rs4680 [8], 5-HTTLPR rs25531 s/L [9], and APOE that can affect cognitive state and their effects on dopamine and serotonin dynamics (important for cognition) and on synapse density [10 –12] have been documented. In contrast to PK DDI interactions (i.e., the impact of one drug on the metabolism and plasma exposure of a second drug), the interaction of CNS active drugs acting on the same neuronal circuits in the human brain (i.e., pharmacodynamic interactions) have not been studied to the same degree. The mere number of possible combinations, especially when taking dose into account together with the impact of common genotype variants and disease state as reflected in amyloid and tau status, results in more possible combinations than there are subjects available, suggesting that each patient is unique (N = 1 situation).

To address these issues, we applied the novel technology of Quantitative Systems Pharmacology (QSP) basically a biophysically realistic computer model of neuronal circuits relevant for cognition [13, 14]. By explicitly modeling the neurophysiological impact of the genotypes from clinical imaging observations and introduction of the pharmacology of the comedications, a QSP model, in principle can estimate the actual impact of these interactions at the individual patient level. The platform has demonstrated its predictive validity in a prospective prediction of an unexpected clinical outcome for a novel pro-cognitive target in AD [15]. With regard to PD DDI interactions in clinical psychiatry practice, the platform has been able to derive a predictive classifier for motor side-effects in subjects taking two antipsychotics and without training set that was 35% better than the state-of-the art calculations [16]. Here we address the interactions between acetylcholine inhibitors (AChE-I; donepezil and galantamine at two doses), SSRI antidepressants (4 doses), first- and second-generation benzodiazepines (each 4 doses), and six antipsychotics (aripiprazole, haloperidol, risperidone, olanzapine, paliperidone, and quetiapine) at the high dose usually prescribed in schizophrenia patients and a lower dose (1/6 of the higher dose).

It has to be noted that this report simulates all possible combinations of standard-of-care with various comedications and genotypes irrespective of their distribution in clinical practice. The platform can be combined with any specific distribution in a specific patient population (such as determined by inclusion/exclusion criteria of a clinical trial protocol) to simulate the specific PD DDI.

METHODS

Defining target exposure in quantitative systems pharmacology model

The receptor competition model [17, 18] calculates the degree of activation of various postsynaptic receptors (dopamine, serotonin, norepinephrine, and cholinergic neurotransmitters) with various drugs (see the Supplementary Material for a list of the different parameters). The affinity of the parent molecule and its major metabolite for both pre- and postsynaptic receptors, derived from the Psychoactive Drug Screening Program [19] is used to calculate the competition with endogenous neurotransmitters. The presynaptic autoreceptor neurophysiology properties are calculated from preclinical data using fast-cyclic voltammetry constrained with clinical imaging data [15]. The functional intrasynaptic concentration of antipsychotics, antidepressants, and AChE-I is determined from calculating the concentration that corresponds to the clinically observed displacement of a radio-active D₂R specific PET tracer, such as raclopride [20], ¹¹C-DASB [21], or ¹¹C-PMP [22], respectively.

Calibrated model for ADAS-COG readout

The calibrated QSP model for cognition in AD has been extensively described before [14]. Basically, the model consists of a biophysically realistic network of 80 prefrontal cortex pyramidal glutamatergic and 40 GABAergic interneurons, with the effects of dopaminergic, serotonergic, noradrenergic and cholinergic modulation (see also the Supplementary Material) and is based on the stability of a memory trace within a working memory paradigm. The model takes into account a progressive neuropathology using a time-dependent elimination of synapses and neurons in addition to a cholinergic deficit. Furthermore, this QSP model has been calibrated using 28 different drug-dose-duration interventions with acetylcholinesterase inhibitors and 5-HT₆ antagonists [14].

Implementation of AChE-I and comedications

Target engagement of donepezil, an AChE-I with a K_i of 20 nM [23], is derived from imaging studies with ¹¹C-PMP [24], corresponding to brain AChE-inhibition levels of 35% at 10 mg [25, 26]. This leads to synaptic ACh half-lives of 6.9 and 7.7 ms for donepezil at 5 and 10 mg and to half-lives of 5.9, 6.8, and 7.7 ms for galantamine at 8, 16, and 24 mg which are applied to the receptor competition model. The subsequent changes in ACh half-life affects activation levels of muscarinic and nicotinic receptors, leading to corresponding modifications in glutamate and GABA (see Supplementary Table 1 for biological references).

In addition, galantamine has a small allosteric potentiating effect on nAChR [23], which we implemented as a 5, 10, or 15% (respectively for 8, 16, and 24 mg) relative increase in both α₇ nAChR and α₄β₂ nAChR activation levels.

First-generation benzodiazepines (lorazepam) are implemented as non-selective agonists at the GABA-A α1 and GABA-A α2 receptors. The α1 receptor is localized both on inhibitory-excitatory and on inhibitory-inhibitory GABAergic synapses, while α1 is predominantly located at the inhibitory-excitatory synapses. Second-generation benzodiazepines such as midazolam more preferentially affect the α2R over the α1R. Four doses are simulated: 2.5, 5, 7.5, and 10% increase for the inhibitory-excitatory GABAergic synapses with a corresponding 1.75, 3.5, 5.25, and 7.5% (for the first-generation benzodiazepines) and 1.25, 2.5, 3.75, and 5% (for the second-generation benzodiazepines) increase on inhibitory-inhibitory GABAergic synapses. These effects were selected based on clinical calibration of cognitive effects of benzodiazepines in schizophrenia [27].

Anti-depressants are implemented by increasing the half-life of serotonin with four dose-levels, i.e., 6.25, 12.5, 18.75, and 25% increase with corresponding effects on all the 5-HTR implemented in the model (5-HT_1A, 5-HT_1B, 5-HT_2A, 5HT₃, 5HT₄, and 5HT₆), calculated from the receptor competition model (see above).

In terms of antipsychotic comedication, we simulated risperidone, paliperidone, olanzapine, quetiapine, aripiprazole, and Haldol both at half the dose and a low dose, corresponding to 1/6 of the dose used in schizophrenia. The affinity parameters for each individual drug and neurotransmitter for all the human receptors were derived from the standardized PDSP database (https://pdsp.unc.edu/) [19]. Importantly, the active moiety of antipsychotics, taking into account the pharmacology of metabolites was used. We took great care in determining the functional intrasynaptic concentration of the various antipsychotics using published ¹¹C-raclopride displacements observed with specific antipsychotic dose combinations using the receptor competition model (see Supplementary Material).

Implementation of genotypes

We study all possible combinations of the following genotypes: COMTVal158Met, 5-HTTLPR rs25531, and APOE (all together 27 cases). The genotypes are COMTMM, COMTMV, and COMTVV (also abbreviated to MM, MV, and VV, respectively) for COMTVal158Met; 5-HTTLPRLL, 5-HTTLPRLs, and 5-HTTLPRss (also abbreviated to LL, Ls, and ss, respectively) for 5-HTTLPR rs25531; and APOE 4/4, APOE 4/X, and APOE X/X where X = 2,3.

The same receptor competition model can be used to determine the pharmacodynamic effect of genotypes (see Supplementary Material). To reproduce experimental findings that the COMTVal158Met genotype affects the displacement of the D₁R PET radiotracer NNC-112 in healthy unmedicated volunteers [28], the synaptic half-life of dopamine in the VV case was adjusted to 100 ms, 130 ms in the MV, and 160 ms in the MM case. Similarly, the displacement of the 5-HT₄ PET tracer [¹¹C]SB207145 is dependent upon the 5-HTTLPR s/l isoform [29], resulting in a half-life of 55 ms for the LL case, 75 ms for the Ls case, and 100 ms for the ss case.

We implemented the APOE genotype using different excitatory-excitatory synapse densities with APOE 4/4 a 20% lower and APOE X/X had a 20% higher synapse density compared to APOE 4/X genotype [10 –12].

We assume a Hardy-Weinberg distribution for all genotypes [30] except for APOE where the allele frequency of the E4 allele increases from 0.16 in controls to 0.40 in AD [31].

Pharmacodynamic interaction between genotypes and different comedications and their combination

In order to simulate the cognitive trajectory of different genotypes (see Methods, above) in the presence of different comedication combinations (see Methods, above), we proceeded as follows:

Calculate the cognitive performance using the ADAS-Cog model at 0, 12, 26, 52, and 78 weeks following placebo, galantamine (2 doses) or donepezil (2 doses) for each of the 27 different genotypes.

Calculate a normalization factor for each of these cases by dividing the simulated cognitive performance by the outcome for a COMTMV, 5HTTLPRLs, and APOE 4/X (triple heterozygous) virtual subject.

Use this normalization factor to derive the cognitive performance of the various combinations, obtained in the Methods (that have been simulated for a COMTMV, 5HTTLPRLs, and APOE 4/X virtual subject) for each of the 27 different genotypes.

To validate this normalization process we used the QSP model to explicitly simulate all combinations of comedications (Methods, above) but now for both a COMTMM-5HTTLPRLL-APOE 2/2 and a COMTVV-5HTTLPRss-APOE 4/4 genotype and compared this outcome with the predicted outcomes of step 3 for the COMTMM-5HTTLPRLL-APOE 2/2 and COMTVV-5HTTLPRss-APOE 4/4 genotypes.

Virtual patient simulation

For each of the individual combinations we calculated the anticipated absolute ADAS-Cog values at five different time points (0, 12, 26, 52, and 78 weeks) in a hypothetical clinical trial of mild-to-moderate AD patients. The global rate of cognitive deterioration was then calculated as the average of the changes in ADAS-Cog divided by duration, i.e., the slopes between baseline and each of the four time points.

RESULTS

We simulated the effect of unique combinations in ‘virtual patients’ taking a combination of no AChE-I or 2 doses of galantamine (16 and 24 mg) or 2 doses of donepezil (5 and 10 mg) as standard-of-care, 4 doses of antidepressants (serotonin transporter antagonists), 4 doses of first-generation benzodiazepines or 4 doses of second-generation benzodiazepines or 2 doses (half or 1/6 of normal dose used in schizophrenia) for each of the antipsychotics risperidone, paliperidone, olanzapine, haldol, quetiapine, and aripiprazole. Each virtual patient was treatment naïve or had one of the four standard-of-care treatments and we simulated a large number of possible combinations of 1, 2, and 3 additional medications from the three classes mentioned above. We further investigate the effect of these combinations against a background of different APOE, COMTVal158M et, and 5-HTTLPR rs23351 genotypes.

Effect of single comedications added to standard-of-care on cognitive trajectory

In this section, we focus on the effects of adding a compound of each of the 3 classes to virtual patients that were on the standard-of-care.

As expected, adding antidepressants or benzodiazepines dose-dependently worsens cognitive readout (up to 1.5 points) compared to the standard-of-care, measured both as changes from baseline as well as absolute values at time 0. This worsening is about 1 point at 12 weeks and 3 points at 26 weeks for the highest dose of antidepressants and around 1.5 points at 26 weeks for the highest dose of benzodiazepines). However, at later stages of the disease, the simulations suggest that the negative benzodiazepines effect declines in a dose-dependent way. Also, second-generation benzodiazepines seem to have a slightly smaller effect on cognitive worsening (0.5 point less on the ADAS-Cog) at earlier stages of the disease, and in later stages of the disease slightly improve cognition, especially at higher doses.

Adding antipsychotics to standard-of-care worsen the baseline ADAS-Cog substantially in the cases of risperidone, paliperidone and olanzapine (up to 1 point on ADAS-Cog) while Haldol, aripiprazole, and quetiapine have no effect or a small improvement. The antipsychotics-induced changes in ADAS-Cog are not dependent upon the trial duration.

Combining different classes of medications

Here we study the effect of augmentation therapy when adding a drug from the three classes (antidepressants, benzodiazepines, and antipsychotics) to any combination of the two other classes in the presence of standard-of-care and focus on the difference of this individual add-on strategy. This allows us to identify the conditions where augmentation therapy with one class of medications might improve or impede cognitive readout. We illustrate this with a few examples, but there are obviously many more combinations and outcomes.

The effect of adding antidepressants to patients already on benzodiazepines is shown in Fig. 1A. The simulations suggest that it is important to start at a low dose of antidepressant to reduce the negative impact on cognitive readout. Interestingly, while at an early stage of the disease, low benzodiazepine concentration is preferable, in later stages of the disease, high doses of benzodiazepines tend to have a smaller negative effect, irrespective of the dose of the antidepressant.

Fig.1

Simulated effects of adding antidepressants, benzodiazepines, or antipsychotics to existing treatment with AChE-I in combination with any of the other medications, illustrating the unique pharmacodynamic interactions between the different classes of medications here for a subject with COMTMV, 5-HTTLPRL/s, and APOE 4/X. A) Antidepressants most markedly worsen cognition in early stages of the disease and their effect is exaggerated with a high dose of benzodiazepines but the opposite is true at later stages. B) First-generation benzodiazepines when added to other combinations in general worsen cognition mostly in early stages (weeks 12 and 26) and much less in later stages (weeks 52 and 78) of the disease. High dose second-generation benzodiazepines tend to slightly stabilize or even improve cognition at earlier time points in specific combinations with other medications (see text), but not at later stages, while the opposite is true for low dose second-generation benzodiazepines. C) When added to antipsychotics with 10 mg of donepezil, benzodiazepines in general dose-dependently tend to worsen cognitive readout, except for risperidone and olanzapine at lower doses of benzodiazepines. D) At early stages of the disease in combination with AChE-I only, or also with antidepressants or benzodiazepines, risperidone, paliperidone, and olanzapine worsen cognition, while Haldol and quetiapine have little effect, and aripiprazole provides a modest improvement.

When benzodiazepines are added to any combination of other drug classes (Fig. 1B), on average high doses of first-generation benzo reduced cognitive performance significantly, especially at earlier stages (between 1.0 and 1.5 points) with a much smaller effect at later stages (worsening only about 0.5 points). Interestingly, on average second-generation benzodiazepines have a better profile, especially at lower doses and at earlier time points, and also at higher doses later in the disease. For instance, a low dose second generation benzodiazepine stabilizes response between 0 and 12 weeks, in the presence of donepezil and paliperidone, but has a substantial worsening of more than 2 points in the presence of olanzapine and donepezil.

Figure 1C shows the complex interaction between benzodiazepines and antipsychotics in virtual patients on 10 mg donepezil. While there is a general tendency for dose-dependent cognitive worsening, low doses of benzodiazepines are beneficial when added to risperidone and olanzapine in combination with AChE-I.

When adding antipsychotics to existing treatment at earlier stages of the disease, the simulations suggest a substantial worsening for all antipsychotics except aripiprazole and to a lesser degree Haldol, especially when added to a combination of AChE-I and antidepressants (Fig. 1D). Again, the effects are more pronounced at earlier stages of the disease (results from later stages not shown). Interestingly as shown above, when adding to a combination of AChE-I and benzodiazepines, the negative effects are greatly reduced.

Effect of genotypes on cognitive trajectory

We subsequently simulate the pharmacodynamic interactions between the 27 different genotype combinations with placebo, antidepressants (4 doses), first and second generation benzodiazepines (4 doses each), galantamine and donepezil (2 doses each), and antipsychotics (6 drugs at 2 doses each), and a large number of combinations between these drugs. (all together 9585 different outcomes for the cognitive trajectory over 78 weeks). As mentioned in the Methods, these are calculated by applying a correction factor to the outcomes of all comedication combinations (see Results) derived from the genotype outcomes.

We first compared these “normalized” outcomes with real simulations for the two extreme COMTMM-5HTTLL-APOE 2/2 and COMTVV-5HTTss-APOE 4/4 genotypes. Differences between the two outcomes was about 2% or 0.45 points, suggesting that this correction is a reasonable approximation, given the fact that the dynamic range for the different ADAS-Cog outcomes is in the range of 10–12 points, about 20–25 times larger.

Figure 2 suggests that the rate of cognitive worsening can vary widely between 0.034 points ADAS-Cog/week (a COMTMV-5HTTss-APOE 4/4 subject on 16 mg galantamine and 1 mg risperidone) and 0.22 points ADAS-Cog/week (a COMTMM-5HTTLL-APOE 2/2 subject on donepezil 5 mg and highest dose of second-generation benzodiazepine), almost a six-fold difference. In general, the 5-HTTLPRLL genotype tends to have a statistical genotype-dependent faster progression rate, except for all COMTVV genotypes irrespective of the APOE genotype. Note the large overlap between the groups.

Fig.2

Box-and whisker plot for baseline cognitive status(left) and rates of progression (right) over 78 weeks for the 27 different combinations of APOE, COMTVal158Met, and 5-HTTLPR rs23351 genotypes. For each genotype combination, comedications (AChE-I, antidepressants, first- and second-generation benzodiazepines, and six different antipsychotics) and combinations hereof are simulated (355 cases). The data suggest that the 5-HTTLPR rs23351 genotype is most responsible for changes in the progression rate with the 5-HTTLPRLL genotype having the highest progression rate, except for subjects with a COMTVV genotype and irrespective of the APOE genotype. With regard to baseline cognition, the data suggest that the 5-HTTLPR rs23351 genotype is driving most of the differences with the 5-HTTLPRss genotype having the worst baseline cognitions.

Similarly, Fig. 2 suggests that cognitive performance at baseline significantly worsens genotype-dependently with the 5-HTTLPR rs23351 genotype with the 5-HTTLPRss genotype being the worst (as defined by a higher ADAS-Cog). Here the effect is present for all combinations of APOE and COMTVal158Met genotypes.

Slow progressors are usually associated with worse baseline performance (Fig. 3), although it is clear that for the same baseline cognitive performance the rate of progression can vary significantly depending upon the comedications and genotypes (a two-fold difference for baseline ADAS-Cog values between 20 and 28). This is likely because subjects with reduced cognitive performance at baseline (i.e., have the largest ADAS-Cog values) end up with smaller dynamic range for deterioration or that the negative effect of these genotypes could overlap with the underlying pathology.

Fig.3

Relation between baseline ADAS-Cog value and natural rate of progression for 78 weeks for all possible combinations of APOE, COMTVal158Met, and 5-HTTLPR rs23351 with different comedications such as AChE-I, benzodiazepines, antidepressants, and antipsychotics and combinations thereof (all together 9,585 cases). Higher ADAS-Cog baseline is associated with lower rates of cognitive deterioration, although there is substantial overlap. There is a 14-point range of baseline ADAS-Cog and almost a six-fold range for the rate of progression. The data also suggest a large variability of progression rates for the same baseline cognition, suggesting a strong pharmacodynamic interaction with comedication and genotypes.

DISCUSSION

This simulation study using an ADAS-Cog validated computer model of relevant neuronal circuits in AD cognition, reveals interesting and unexpected pharmacodynamic interactions between various comedications used in clinical practice. From a large meta-analysis of ADNI data [32], it is increasingly becoming evident that cognition in placebo patients can vary widely due to a multitude of factors, notably amyloid and tau status and various comorbidities in addition to the comedications and genotypes studied in this report. In this report we focus on common genotype variants and comedications, but amyloid, tau status and other comorbidities play an important role as well. Focusing on possible negative pharmacodynamic interactions by comedications is of major interest as, unlike with many of the other factors, these can be mitigated in clinical practice.

Basically, the simulations recapitulate the observations that adding antidepressants, benzodiazepines and antipsychotics to standard-of-care in general worsens cognitive performance, as recently documented in a trial with citalopram to address behavioral deficits [33]. However, there are interesting differences which might have impact on prescribing in clinical practice. For instance, if antipsychotics need to be prescribed for behavioral problems, this study suggests that aripiprazole would have minimal impact on cognitive functioning. The differential effect of antipsychotics on cognitive readout in AD is likely due to the individual pharmacological profile where many receptors affected by antipsychotics also drive firing in neuronal networks associated with cognitive readouts. In particular, the substantial 5-HT1A agonist effect and to a lesser extent, the relatively weaker antagonistic effect at the D2R of aripiprazole might explain its beneficial effect as documented in schizophrenia patients [34].

Furthermore, if benzodiazepines need to be prescribed, the study suggests that second-generation low doses might be a good start. Special care has to be taken when prescribing these medications for intermediate stages of the disease where the negative impact on cognition is the greatest. Interestingly, benzodiazepines reduce the cognitive deficit introduced by risperidone, paliperidone, and olanzapine. As each antipsychotic has its individual pharmacological profile covering a number of dopaminergic, serotonergic, adrenergic, and cholinergic neurotransmitter receptors, a number of non-linear interactions will drive the clinical outcome. For instance, drugs with a substantial D1R antagonism such as risperidone and paliperidone acting on pyramidal cells, reduce excitatory tone and affect cognitive readout negatively. This probably can lead to reduced pyramidal drive of GABA interneurons which can be countered by benzodiazepine mediated allosteric GABA modulation, restoring the original excitation-inhibition balance.

As the disease progresses and more pyramidal cells get affected, the excitatory tone in the network can becomes smaller than the inhibitory tone. Benzodiazepines increase GABA tone not only on pyramidal cells (further reducing excitatory tone), but also on the fast firing interneurons, providing a compensatory reduction in inhibitory firing and GABA tone.

The observation that low dose benzodiazepines might be slightly beneficial in very early stages of the disease is in line with recent studies suggesting a dysfunctional hyperactivity in AD patients [35] and in a number of transgene mouse models [36]. While benzodiazepines might not be the first choice of treatment, they tend to dampen the excitatory hyperactivity through an increase in GABA tone and can possibly revert the dysfunctional hyperactivity. In line with this hypothesis, a case report documents the appearance of Myoclonus Status epilepticus in an AD patient after abrupt withdrawal of alprazolam [37]. In general, epidemiological studies increasingly identify negative PD DDI (for a review, see [38]), especially with antipsychotics but systematic large-scale studies are still lacking.

Even if we explicitly implemented the allosteric potentiating effect of galantamine on nicotinic receptors using quantitative preclinical data [39], the simulations suggest a very small and clinically undetectable beneficial effect compared to the pure AChE-I donepezil. However, it is possible that different cognitive tasks in different populations can unmask galantamine’s allosteric effect, as documented in the synergistic interaction with nicotine on reaction time in a visuospatial attention task in healthy volunteers [40].

Studies on the impact of the COMTVal158Met genotype on cognition in AD research have yielded mixed results (for an overview, see [41]) with some studies suggesting a positive effect of the COMTMM genotype and others not showing any effect at all. In our simulations on average COMTMM subjects tend have a better baseline performance but there is a strong negative interaction with benzodiazepines, antidepressants and some antipsychotics, so that there is a substantial overlap with other COMT genotypes. Accounting for these comedications can likely unmask the differential performance of homozygote COMTMet158Met subjects.

The observation that patients with better cognitive performance at baseline have a greater rate of progression fits with the concept of cognitive reserve [42], in that they tend to stave off the cognitive decline associated with dementia, but have a faster deterioration once these compensatory mechanism are overwhelmed. In this simulation, the 5-HTTLPR gene polymorphism does not protect against the ongoing neuropathology, thus providing no neural reserve and protection. We speculate that the concept of cognitive reserve might be partially driven by certain neurotransmitter properties that also could protect against cognitive deficit in other diseases, such as schizophrenia. Another explanation could be related to the “regression to the mean”, as the dynamic range for changes becomes smaller when the baseline is higher.

In this study, we define disease state (early versus late) as a certain period of time (either between 0 and 12 weeks or between 52 and 78 weeks) after a state of mild-to-moderate AD as typically encountered in clinical trials, because the platform has been calibrated using well-controlled Phase III AD clinical trials. Whether this can be generalized to clinical practice situations is currently unknown. We acknowledge that the rate of pathology progression might be different for each patient, due to different additional impacts of amyloid and tau pathology and that the model in its current form does not capture this additional pathology-related variability.

It has to be emphasized that we assume all simulations of comedication combination for the same time point assume an identical pathology. In clinical practice, however, standard-of care AChE-I are often prescribed for rapidly deteriorating patients [43]; therefore those patients are likely on a different pathological trajectory as compared to subjects without AChE-I [44].

A better understanding of negative pharmacodynamic interactions could possibly lead to improved clinical trial design for novel drugs in AD. Study of PK DDI interactions are part of the standard clinical development plans, but neglecting the impact of pharmacodynamic interactions can have similar if not greater impact on trial outcomes [45]. A recent review paper [46] documents some observed anecdotical PD DDI interactions between standard-of-care in AD and other medications, but highlights the need for a more systematic study of the different interactions. It has to be noted that the observed effects of comedications and genotypes on ADAS-Cog are in the range of reported differences for AChE-I in clinical trials (2–3 points) and higher than the aducanumab outcomes (1.4 point difference in the successful EMERGE but only 0.6 point in ENGAGE trial).

In principle, this platform can also be used to generate a historical patient cohort based on their genotype distribution, actual use of medications, and pathology markers as a placebo treatment group. In addition, the platform could simulate the pharmacodynamic interactions with an active molecule targeting a novel pathway if that target is quantitatively implemented. Taking these interactions into account would allow to improve power calculations for any novel target.

In clinical practice where polypharmacy is quite common, a better identification of these PD DDI interactions can lead to rational treatment guidance, improving the clinical improvement for patients while at the same time reducing side effects and costs. The same QSP approach when applied to schizophrenia patients in clinical practice treated with a combination of two antipsychotics as a prospective classifier, was validated with actual electronic health records and has identified a number of medications with high tendency for motor side-effects [16].

In contrast to this mechanism-based model, other data-driven platforms have been proposed. Based on longitudinal cognitive trajectories from the ADNI and CAMD (Coalition Against Major Diseases), a robust predictive model was developed [47] highlighting the large variability of placebo responses. Interestingly the authors observed that cognitive status at baseline was a major determinant in the functional progression. It is of interest to note that this led to a Clinical Trial Simulation Tool that was subsequently approved by regulatory agencies.

Machine learning methods based on historical datasets of clinical progression in unmedicated AD patients are an alternative approach to predict the cognitive outcome of “virtual patients” in clinical trials. These techniques traditionally can substantially improve the diagnosis of AD subtypes from imaging and biochemical biomarkers, and predictive classifiers have been developed for progression of AD using the concept of digital twins [48]. Another interesting approach using a generalized metric Learning Vector Quantization approach suggests that introduction of biomarker data significantly improves the prediction of individual cognitive trajectories in mild cognitive impairment patients [49] with relatively few CNS active comedications.

The generalizability of these predictors is strongly dependent upon the training set and individual virtual patients are often generated in a stochastic way from a distribution with the same properties as the patients in the database. Also, the sheer number of combinations far outnumbers the number of possible subjects in any database. For instance, just assuming 5 different standard-of-care medications, 2 doses of first- and second generation benzodiazepines, 4 doses of antidepressants, and 6 antipsychotics at two doses each, together with the combination of the three genotypes mentioned in this report combined with 3 different brain atrophy, 4 amyloid, and 3 tau loads, this rapidly gets to over a million unique combinations. Moreover, such a data-driven prediction cannot account for ad-hoc changes in medication or clinical trial protocol amendments.

Obviously, validation at the single patient level is a key prerequisite for any model. Unfortunately, there are not many databases that cover all characteristics (medications, genotypes and disease states) covered in this paper. The C-path Alzheimer database (https://c-path.org/programs/cpad/) is more of an enriched clinical trial population and provides data on comedications and APOE genotype, but not on disease state or any of the two other genotypes. The natural history ADNI database (http://adni.loni.usc.edu/) provides information on co-medications and disease state and APOE, but not on the two other genotypes studied in this paper.

There are major limitation of this study. First, only neuropathology progression in terms of synapse and neuronal cell loss has been simulated but no amyloid or tau pathology mechanism have been incorporated. While it is certainly possible to include aspects of amyloid-β effects on glutamatergic or nicotinic neurotransmission [13] or the effect of tau pathology on voltage-gated ion channels [50], the pharmacodynamic interaction with comedications and genotypes has not yet been simulated. However, in the absence of any disease-modifying therapeutic interventions, amyloid or tau changes are relatively slow in comparison to the time span simulated here. Longitudinal studies of amyloid-β load suggest already a substantial load in many prodromal mild cognitive impairment patients and a gradual near-saturation at later stages in the disease [51], suggesting that amyloid pathology does not change much in the simulated patient population. Tau pathology arises somewhat later than amyloid-β accumulation and its progression is more complex [52], likely affecting different brain regions at different times. One extension of this work has addressed the impact of the three genotypes and standard-of-care on the dose-response of amyloid modulating agents [53] and future applications of this work will focus on the impact of comedications and genotypes on tau modulating agents.

Secondly, we limited our analyses to the classes of serotonin transporter blockers, first- and second generation benzodiazepines and a select number of antipsychotics. In clinical practice additional classes of medications are used such as anti-epileptics, statins, metabolism affecting drugs and nutriceuticals. With the exception of anti-epileptics, the pharmacological interaction of these medications with neuronal circuits relevant for cognition is much less established. In addition, we limited the analysis to 3 common genotype variants for which quantitative clinical imaging data were available; it is conceivable that other common genotypes can play an important role in affecting cognitive trajectory.

Furthermore, this approach does not take into account factors such as age, education status, and gender which can be important in cognitive outcome such as the impact of APOE [54]. However, in principle, the effect of age and gender can be introduced in the platform using documented changes in neurotransmitters, for example the impact of age and sex on 5-HT levels as reported by 5-HT₄ binding by ¹¹C-SB207145 [55]. Education status is likely associated with neuroplasticity capacity with impact on synapse functioning [56].

In summary, this approach of mechanism-based modelling of virtual patients with identical characteristics as the real patients is complementary to other data-driven approaches but is more flexible with regard to changes in medications and protocol amendments in clinical trials and allows one to better understand the underlying biology of the individual patient variability.

DISCLOSURE STATEMENT

Authors’ disclosures available online (https://www.j-alz.com/manuscript-disclosures/20-0688r1).

Footnotes

The supplementary material is available in the electronic version of this article: .

References

Clague

, Mercer

, McLean

, Reynish

, Guthrie

(2016) Comorbidity and polypharmacy in people with dementia: Insights from a large, population-based cross-sectional analysis of primary care data, Age Ageing 46, 33–39.

Gray

, Anderson

, Dublin

, Hanlon

, Hubbard

, Walker

, Yu

, Crane

, Larson

(2015) Cumulative use of strong anticholinergics and incident dementia: A prospective cohort study, JAMA Intern Med 175, 401–407.

Risacher

, McDonald

, Tallman

, West

, Farlow

, Unverzagt

, Gao

, Boustani

, Crane

, Petersen

, Jack

Jr , Jagust

, Aisen

, Weiner

, Saykin

(2016) Association between anticholinergic medication use and cognition, brain metabolism, and brain atrophy in cognitively normal older adults, JAMA Neurol 73, 721–732.

Wimmer

, Bell

, Fastbom

, Wiese

, Johnell

(2016) Medication regimen complexity and number of medications as factors associated with unplanned hospitalizations in older people: A population-based cohort study, J Gerontol A Biol Sci Med Sci 71, 831–837.

Lau

, Mercaldo

, Shega

, Rademaker

, Weintraub

(2011) Functional decline associated with polypharmacy and potentially inappropriate medications in community-dwelling older adults with dementia, Am J Alzheimers Dis Other Demen 26, 606–615.

Riepe

, Walther

, Vonend

, Beer

(2015) Drug-induced cerebral glucose metabolism resembling Alzheimer’s Disease: A case study, BMC Psychiatry 15, 157.

Amanatkar

, Papagiannopoulos

, Grossberg

(2017) Analysis of recent failures of disease modifying therapies in Alzheimer’s disease suggesting a new methodology for future studies, Expert Rev Neurother 17, 7–16.

Lachman

, Morrow

, Shprintzen

, Veit

, Parsia

, Faedda

, Goldberg

, Kucherlapati

, Papolos

(1996) Association of codon 108/158 catechol-O-methyltransferase gene polymorphism with the psychiatric manifestations of velo-cardio-facial syndrome, Am J Med Genet 67, 468–472.

Heils

, Teufel

, Petri

, Seemann

, Bengel

, Balling

, Riederer

, Lesch

(1995) Functional promoter and polyadenylation site mapping of the human serotonin (5-HT) transporter gene, J Neural Transm Gen Sect 102, 247–254.

10.

Kim

, Yoon

, Basak

(2014) Apolipoprotein E in synaptic plasticity and Alzheimer’s disease: Potential cellular and molecular mechanisms, Mol Cells 37, 767–776.

11.

Chung

, Verghese

, Chakraborty

, Joung

, Hyman

, Ulrich

, Holtzman

, Barres

(2016) Novel allele-dependent role for APOE in controlling the rate of synapse pruning by astrocytes, Proc Natl Acad Sci U S A 113, 10186–10191.

12.

Dumanis

, Tesoriero

, Babus

, Nguyen

, Trotter

, Ladu

, Weeber

, Turner

, Xu

, Rebeck

, Hoe

(2009) ApoE4 Decreases spine density and dendritic complexity in cortical neurons in vivo, J Neurosci 29, 15317–15322.

13.

Geerts

, Spiros

, Roberts

(2018) Impact of amyloid-beta changes on cognitive outcomes in Alzheimer’s disease: Analysis of clinical trials using a quantitative systems pharmacology model, Alzheimers Res Ther 10, 14.

14.

Roberts

, Spiros

, Geerts

(2012) Simulations of symptomatic treatments for Alzheimer’s disease: Computational analysis of pathology and mechanisms of drug action, Alzheimers Res Ther 4, 50.

15.

Nicholas

, Sridhar Duvvuri

, Leurent

, Raunig

, Rapp

, Iredale

, Rowinski

, Carr

, Patrick

, Spiros

, Geerts

(2013) Systems pharmacology modeling in neuroscience: Prediction and outcome of PF-04995274, a 5-HT4 partial agonist, in a clinical scopolamine impairment trial, Adv Alzheimer Dis 2, 83–98.

16.

Kadra

, Spiros

, Shetty

, Iqbal

, Hayes

, Stewart

, Geerts

(2018) Predicting parkinsonism side-effects of antipsychotic polypharmacy prescribed in secondary mental healthcare, J Psychopharmacol 32, 1191–1196.

17.

Spiros

, Carr

, Geerts

(2010) Not all partial dopamine D(2) receptor agonists are the same in treating schizophrenia. Exploring the effects of bifeprunox and aripiprazole using a computer model of a primate striatal dopaminergic synapse, Neuropsychiatr Dis Treat 6, 589–603.

18.

Spiros

, Geerts

(2012) A quantitative way to estimate clinical off-target effects for human membrane brain targets in CNS research and development, J Exp Pharmacol 4, 53–61.

19.

Besnard

, Ruda

, Setola

, Abecassis

, Rodriguiz

, Huang

, Norval

, Sassano

, Shin

, Webster

, Simeons

, Stojanovski

, Prat

, Seidah

, Constam

, Bickerton

, Read

, Wetsel

, Gilbert

, Roth

, Hopkins

(2012) Automated design of ligands to polypharmacological profiles, Nature 492, 215–220.

20.

Spiros

, Roberts

, Geerts

(2012) A quantitative systems pharmacology computer model for schizophrenia efficacy and extrapyramidal side effects, Drug Dev Res 73, 196–213.

21.

Smith

, Kahn

, Sacher

, Rusjan

, van Eimeren

, Flint

, Wilson

(2011) Serotonin transporter occupancy and the functional neuroanatomic effects of citalopram in geriatric depression, Am J Geriatr Psychiatry 19, 1016–1025.

22.

Kadir

, Darreh-Shori

, Almkvist

, Wall

, Grut

, Strandberg

, Ringheim

, Eriksson

, Blomquist

, Langstrom

, Nordberg

(2008) PET imaging of the in vivo brain acetylcholinesterase activity and nicotine binding in galantamine-treated patients with AD, Neurobiol Aging 29, 1204–1217.

23.

Woodruff-Pak

, Lander

, Geerts

(2002) Nicotinic cholinergic modulation: Galantamine as a prototype, CNS Drug Rev 8, 405–426.

24.

Geerts

, Roberts

, Spiros

(2015) Assessing the synergy between cholinomimetics and memantine as augmentation therapy in cognitive impairment in schizophrenia. A virtual human patient trial using quantitative systems pharmacology, Front Pharmacol 6, 198.

25.

Darreh-Shori

, Kadir

, Almkvist

, Grut

, Wall

, Blomquist

, Eriksson

, Langstrom

, Nordberg

(2008) Inhibition of acetylcholinesterase in CSF versus brain assessed by 11C-PMP PET in AD patients treated with galantamine, Neurobiol Aging 29, 168–184.

26.

Shinotoh

, Aotsuka

, Fukushi

, Nagatsuka

, Tanaka

, Ota

, Tanada

, Irie

(2001) Effect of donepezil on brain acetylcholinesterase activity in patients with AD measured by PET, Neurology 56, 408–410.

27.

Geerts

, Roberts

, Spiros

(2013) A quantitative system pharmacology computer model for cognitive deficits in schizophrenia, CPT Pharmacometrics Syst Pharmacol 2, e36.

28.

Slifstein

, Kolachana

, Simpson

, Tabares

, Cheng

, Duvall

, Frankle

, Weinberger

, Laruelle

, Abi-Dargham

(2008) COMT genotype predicts cortical-limbic D1 receptor availability measured with [11C]NNC112 and PET, Mol Psychiatry 13, 821–827.

29.

Fisher

, Holst

, Mc Mahon

, Haahr

, Madsen

, Gillings

, Baare

, Jensen

, Knudsen

(2012) 5-HTTLPR status predictive of neocortical 5-HT4 binding assessed with [(11)C]SB207145 PET in humans, Neuroimage 62, 130–136.

30.

Lesch

, Bengel

, Heils

, Sabol

, Greenberg

, Petri

, Benjamin

, Muller

, Hamer

, Murphy

(1996) Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region, Science 274, 1527–1531.

31.

Roses

, Saunders

, Alberts

, Strittmatter

, Schmechel

, Gorder

, Pericak-Vance

(1995) Apolipoprotein E E4 allele and risk of dementia, JAMA 273, 374–375; author reply 375-376.

32.

Veitch

, Weiner

, Aisen

, Beckett

, Cairns

, Green

, Harvey

, Jack

Jr , Jagust

, Morris

, Petersen

, Saykin

, Shaw

, Toga

, Trojanowski

; Alzheimer’s Disease Neuroimaging Initiative (2019) Understanding disease progression and improving Alzheimer’s disease clinical trials: Recent highlights from the Alzheimer’s Disease Neuroimaging Initiative, Alzheimers Dement 15, 106–152.

33.

Porsteinsson

, Drye

, Pollock

, Devanand

, Frangakis

, Ismail

, Marano

, Meinert

, Mintzer

, Munro

, Pelton

, Rabins

, Rosenberg

, Schneider

, Shade

, Weintraub

, Yesavage

, Lyketsos

(2014) Effect of citalopram on agitation in Alzheimer disease: The CitAD randomized clinical trial, JAMA 311, 682–691.

34.

Sumiyoshi

, Higuchi

, Uehara

(2013) Neural basis for the ability of atypical antipsychotic drugs to improve cognition in schizophrenia, Front Behav Neurosci 7, 140.

35.

Bakker

, Krauss

, Albert

, Speck

, Jones

, Stark

, Yassa

, Bassett

, Shelton

, Gallagher

(2012) Reduction of hippocampal hyperactivity improves cognition in amnestic mild cognitive impairment, Neuron 74, 467–474.

36.

Harris

, Devidze

, Verret

, Ho

, Halabisky

, Thwin

, Kim

, Hamto

, Lo

, Yu

, Palop

, Masliah

, Mucke

(2010) Transsynaptic progression of amyloid-beta-induced neuronal dysfunction within the entorhinal-hippocampal network, Neuron 68, 428–441.

37.

Vicente Ferreira Naves

, Calderaro

, Caboclo

(2018) De Novo Myoclonic status epilepticus after benzodiazepine withdrawal, Clij Neuropharmacol 41, 142–144.

38.

Pasqualetti

, Tognini

, Calsolaro

, Polini

, Monzani

(2015) Potential drug-drug interactions in Alzheimer patients with behavioral symptoms, Clin Interv Aging 10, 1457–1466.

39.

Samochocki

, Hoffle

, Fehrenbacher

, Jostock

, Ludwig

, Christner

, Radina

, Zerlin

, Ullmer

, Pereira

, Lubbert

, Albuquerque

, Maelicke

(2003) Galantamine is an allosterically potentiating ligand of neuronal nicotinic but not of muscarinic acetylcholine receptors, J Pharmacol Exp Ther 305, 1024–1036.

40.

Hahn

, Shrieves

, Olmstead

, Yuille

, Chiappelli

, Pereira

EFR

, Albuquerque

, Fawcett

(2019) Evidence for positive allosteric modulation of cognitive-enhancing effects of nicotine in healthy human subjects, Psychopharmacology (Berl) 237, 219–230.

41.

Perkovic

, Strac

, Tudor

, Konjevod

, Erjavec

, Pivac

(2018) Catechol-O-methyltransferase, cognition and Alzheimer’s disease, Curr Alzheimer Res 15, 408–419.

42.

Stern

(2002) What is cognitive reserve? Theory and research application of the reserve concept, J Int Neuropsychol Soc 8, 448–460.

43.

Samtani

, Xu

, Russu

, Adedokun

, Lu

, Ito

, Corrigan

, Raje

, Brashear

, Styren

, Hu

(2015) Alzheimer’s disease assessment scale-cognitive 11-item progression model in mild-to-moderate Alzheimer’s disease trials of bapineuzumab, Alzheimers Dement (N Y) 1, 157–169.

44.

Huisa

, Thomas

, Jin

, Oltersdorf

, Taylor

, Feldman

(2019) Memantine and acetylcholinesterase inhibitor use in Alzheimer’s disease clinical trials: Potential for confounding by indication, J Alzheimers Dis 67, 707–713.

45.

Geerts

, Gieschke

, Peck

(2018) Use of quantitative clinical pharmacology to improve early clinical development success in neurodegenerative diseases, Expert Rev Clin Pharmacol 11, 789–795.

46.

Caraci

, Sultana

, Drago

, Spina

(2017) Clinically relevant drug interactions with anti-Alzheimer’s drugs, CNS Neurol Disord Drug Targets 16, 501–513.

47.

Ito

, Corrigan

, Romero

, Anziano

, Neville

, Stephenson

, Lalonde

(2013) Understanding placebo responses in Alzheimer’s disease clinical trials from the literature meta-data and CAMD database, J Alzheimers Dis 37, 173–183.

48.

Fisher

, Smith

, Walsh

(2019) Machine learning for comprehensive forecasting of Alzheimer’s disease progression, Sci Rep 9, 13622.

49.

Giorgio

, Landau

, Jagust

, Tino

, Kourtzi

(2020) Modelling prognostic trajectories of cognitive decline due to Alzheimer’s disease, Neuroimage Clin 26, 102199.

50.

Hill

, Karikari

, Moffat

, Richardson

MJE

, Wall

(2019) Introduction of tau oligomers into cortical neurons alters action potential dynamics and disrupts synaptic transmission and plasticity, eNeuro 6, ENEURO.016619.2019.

51.

Jack

Jr. , Therneau

, Wiste

, Weigand

, Knopman

, Lowe

, Mielke

, Vemuri

, Roberts

, Machulda

, Senjem

, Gunter

, Rocca

, Petersen

(2016) Transition rates between amyloid and neurodegeneration biomarker states and to dementia: A population-based, longitudinal cohort study, Lancet Neurol 15, 56–64.

52.

Chiotis

, Saint-Aubert

, Rodriguez-Vieitez

, Leuzy

, Almkvist

, Savitcheva

, Jonasson

, Lubberink

, Wall

, Antoni

, Nordberg

(2018) Longitudinal changes of tau PET imaging in relation to hypometabolism in prodromal and Alzheimer’s disease dementia, Mol Psychiatry 23, 1666–1673.

53.

Geerts

, Spiros

(2020) Learning from amyloid trials in Alzheimer’s disease. A virtual patient analysis using a quantitative systems pharmacology approach, Alzheimers Dement 16, 862–872.

54.

Williams

, An

, Armstrong

, Shafer

, Helphrey

, Kitner-Triolo

, Ferrucci

, Resnick

(2019) Apolipoprotein E epsilon4 allele effects on longitudinal cognitive trajectories are sex and age dependent, Alzheimers Dement 15, 1558–1567.

55.

Madsen

, Haahr

, Marner

, Keller

, Baare

, Svarer

, Hasselbalch

, Knudsen

(2011) Age and sex effects on 5-HT(4) receptors in the human brain: A [(11)C]SB207145ET study. J Cereb Blood Flow Metab 31, 1475–1481.

56.

Porto

, Fox

, Tusch

, Sorond

, Mohammed

, Daffner

(2015) In vivo evidence for neuroplasticity in older adults, Brain Res Bull 114, 56–61.