Animal models of Alzheimer's disease: modeling targets, not disease

Animal models of Alzheimer's disease (AD) pathogenesis range from Caenorhabditis elegans to aged non-human primates, but by far the most widely used are rodent models. Most animal models used for drug discovery over-express proteins with familial AD mutations (Table 1). While these models develop certain characteristics of AD-like pathology, they do not recapitulate the entirety of the human disease. Furthermore, it is unclear to what extent the pathogenic pathways in rodents mirror those in human AD. Other challenges in translation include mouse/human species differences (for example, differences in cerebrovascular anatomy, neuronal network complexity, connectivity and disease susceptibility, white/gray matter ratios, cellular redox conditions, and dynamics of drug/target interactions [1]). Nonetheless, rodent models offer a means for testing pharmacodynamic properties of candidate molecules on drug targets that may be involved in AD pathogenesis.

Table 1 Animal models for use in Alzheimer's disease preclinical studies
Full size table

This target-driven approach in animal models has already translated to therapeutic studies in humans. In the amyloid-beta (Aβ) immunotherapy trial of bapineuzumab, for example, the immunotherapy cleared plaques in both mice and humans [2, 3]. Gamma-secretase inhibitors developed at Eli Lilly and Company (Indianapolis, IN, USA) and Bristol-Myers Squibb Company (Princeton, NJ, USA) (semagacestat and BMS-708163, respectively) showed good target-focused preclinical animal data, reducing Aβ levels in mice and in the spinal fluid of human patients in a phase 2 study [4, 5]. Demonstration of positive effects on cognitive outcomes from treatment with bapineuzumab of patients with AD is in the final stages of clinical testing. The phase 3 clinical trial of semagacestat was terminated prematurely because of lack of efficacy as well as serious side effects [6], whereas clinical testing of BMS-708163 is in progress. Thus, while these examples provide reassurance that well-executed preclinical studies can translate to human patients with regard to pathological targets, they also highlight our limited understanding between causative pathways and clinical decline of cognitive function in AD and our inability to accurately model all aspects of the disease in animals.

Therefore, animal models appear more useful as models of specific disease targets and pathways than of the complete human disease. To optimize their use in that manner, our advisory panel recommended choosing models for preclinical studies that exhibit significant and well-characterized pathology relevant to the disease process of interest (that is, amyloid plaques, tau pathology, neuronal loss, oxidative stress/inflammatory changes, and so on). In addition, models that do not rely solely on mutated human genes to induce pathology are currently underused and can be quite informative. These include aged rodents, pharmacologically and surgically induced models, and other non-transgenic models (Table 1). Since there is no one model for AD, hypothesis testing in multiple models is preferable in order to provide better preclinical validation. In the following sections, we present the panel's recommendations and guidelines for the design, execution, and interpretation of preclinical studies. The objective of this panel was to improve the predictive value of animal models for clinical benefit.

Know your model

Many transgenic lines show high variability in the extent and time course of expression of disease phenotypes. Table 2 illustrates common factors affecting phenotype variability, including environmental factors, age, sex, genetic background, litter, transgene copy number, and health status. Not all of these variables can be avoided, but measures can be taken so that phenotype changes due to such factors can be properly noted and potentially corrected [7–9].

Table 2 Major factors affecting phenotypic variability in mice
Full size table

Important points to keep in mind

  • Maintain good communication among laboratory members to track deviations from expected phenotypes. Keep careful records to track whether a change in phenotype occurs.

  • Identify issues with breeding, such as longer litter intervals, smaller litter sizes, and fewer pregnancies. Identifying such problems early will help keep production on track.

  • Screen gene copy numbers and transgene expression level regularly. Document and report.

  • Freeze embryos early during characterization of the transgenic line in case phenotypic drift necessitates rederivation of colony.

  • Consider your genetic background: Mice may be healthier and more viable on a hybrid background, but genetic drift must be controlled to avoid confounding variables. Keep in mind that certain inbred strains are more prone to characteristics like blindness, hearing loss, and aggression.

  • If working with an outside breeder or contract research organization, ask to see historical data on the colony. These data should include rearing conditions such as light cycle, housing type, diet, and health status as well as breeding schemes to assess genetic management of the strain background(s) in the colony.

Improving rigor in study design

Many animal studies are flawed by methodological weaknesses that compromise study validity and reproducibility. In fact, it was reported that the majority of published effects in amyotrophic lateral sclerosis mice, for example, were likely measurements of noise in the sample population as opposed to actual drug effects [10]. By paying careful attention to study design before starting experiments, investigators can save time and money as well as minimize the probability of false-positive or false-negative results. Table 3 outlines key study design considerations. In addition, performance on behavioral assays can be highly sensitive to protocol design. For the Morris water maze, for example, variables that can affect performance include water tank size, number and kinds of visual cues, training protocol, how long animals are acclimated to the test room before testing, and strain differences (that can be differentially affected by genetic alterations or the aging process or both). It is important to consider and control for these variables in experimental design and use multiple overlapping tests to substantiate behavioral changes.

Table 3 Key considerations for preclinical animal studies
Full size table

Develop and employ translatable biomarkers for animal preclinical studies

Biomarkers have been instrumental in revolutionizing the way we think about human AD and have allowed us to improve clinical trial design and assess target engagement and response to treatments. Animal preclinical studies can also benefit immensely from the use of biomarkers to assess target engagement of investigative treatments, monitor biological responses to treatment in real-time, characterize the translatability of AD models, and determine the translatability of a novel therapeutic if the same biomarker can be used in a human clinical trial. Although more validation is needed, biomarker methods under development in rodents include imaging - magnetic resonance imaging (MRI), magnetic resonance spectroscopy, functional MRI, arterial spin labeling MRI, flouro-2-deoxy-D-glucose-positron emission tomography (FDG-PET), PET amyloid imaging, PET tau imaging, single-photon emission computed tomography/computed tomography, and others - and biochemical assays on biological fluids such as plasma and cerebrospinal fluid [11–13]. It is important to be aware of the limitations of these biomarkers in rodents, however. For example, functional imaging in mice can be affected by the requirement for either anesthesia or restraint stress. Drawing cerebral spinal fluid from mice is difficult but doable, although it is important to avoid blood contamination [14]. Rat models are becoming more popular and may have advantages in these types of biomarker studies. In any case, whenever possible, biomarker measurements should be incorporated into the study design.

Timing of treatment

Treatment timing should depend on whether the therapeutic goal is disease prevention, therapeutic intervention (that is, slowing/reversal of established pathology), or symptomatic relief. Tissue should be collected from a proper cohort of animals at the time when treatment is initiated to determine whether the treatment reduced pre-existing pathology in the brain or simply slowed its age-associated accumulation. The degree to which disease stages in mouse models correlate with those in humans is currently unclear. Amyloid mice which do not show tangles or neuronal loss may be representative of presymptomatic or early-stage AD, although this idea is not universally accepted [15]. Where a longitudinal assessment is possible (that is, using peripheral biomarkers, imaging, and certain behavioral responses), taking repeated measures of the same animal can be especially informative and add statistical power. Treatment should be timed on the basis of the optimal stage of pathology development in the animal, which will allow acceptable signal-to-background ratio and dynamic range for experimental treatments. Optimally, demonstration of assay validation should be a prerequisite to embarking on therapeutic studies. Because pathology can vary widely with animal age, control and treatment groups should be age-matched to the greatest extent possible (that is, within days of one another). Pathology and biochemical readouts can also vary widely among animals within a genetically engineered line. The variability in pathology with age and in outcome measures must be assessed in order to power the animal studies properly.

Pharmacokinetics/Pharmacodynamics, ADME-Toxicology

Studies should include pharmacokinetics (PK) and pharmacodynamics (PD) assessments to determine whether the compound exposure is sufficient and whether it is interacting with the target of interest. Depending on whether a study is exploratory or therapeutic (see 'Exploratory versus therapeutic studies'), the degree to which absorption, distribution, metabolism, excretion, and toxicity (ADMET) are profiled should be considered as part of the prospective study design. In therapeutic studies, it is critical (a) to demonstrate that the test compound has the capacity to reach its target with sufficient concentration and stability to be relevant to prior in vitro studies and (b) to guide the dosing concentrations and frequency to optimize the chance of achieving therapeutic effects. More information about these types of studies can be found in the Alzheimer's Drug Discovery Foundation/Institute for the Study of Aging/Alzheimer's Research Forum Drug Development Tutorial [16]. It is important to note that genetically engineered models may not always be the most cost-effective and translatable models for measuring PK/PD. Wild-type mice are often preferable for use in these studies, but correspondence with genetic background strains in the transgenic studies should be considered.

Statistical analysis plan and methods

Statistical methods should be chosen before a study is begun, with the anticipated direction of change (one-sided or two-sided) in mind. Statistical considerations should be clearly stated in the Methods section of all data reporting. Assessment of endpoint variability in a large sample size is necessary and should be considered in the choice of statistical tests, as the type of variability (normal distribution versus skewed) dictates a parametric versus non-parametric statistical analysis of the data. Guidance or consultation of a statistician should be enlisted in the design of the study once the endpoint variability has been characterized.

Proper quantification

Both the area and magnitude of pathology should be quantified and reported. Adequate tissue sampling is critical for accurate estimation of pathological burden. For imaging, typically at least six or seven fields per section and six or seven sections per mouse (sampled across multiple affected brain regions) should be measured. The use of unbiased stereology and the optical fractionator method is critical to determining an accurate and statistically reliable neuronal count in brain sections [17]. Staining and field sampling methods should always be stated in the Methods section, and sampling should be guided by statistical considerations of the variability in the endpoint being interrogated. Analysis and quantification of pathology should be conducted by an individual who is blind to the treatment condition.

Sample size

Animal studies are frequently underpowered. This was reported to be the single most important factor in influencing spurious research results with animal models [10]. Minimum sample size depends on the expected magnitude of the biological effect, the inherent variability of the target being measured (for example, cerebral spinal fluid Aβ is much more variable than hippocampal Aβ), variability in behavioral measures or other outcomes, and other factors such as variations in survival within the particular cohort of animals. It is critical to be aware of the natural variability within and among animals in outcome measures in non-treated animals in order to determine the number of animals required for proper statistical powering of therapeutic effects. The sample size needed to achieve significant differences given the variability of disease outcomes in most AD mouse models has been estimated to be on the order of 20 to 30 per group, rarely achieved in most published mouse studies.

Exclusion criteria

Animals whose physiological condition appears to be compromised by factors unrelated to the normal progression of the disease should be excluded from the study. A statistical analysis plan should be developed to address dropouts and death. Exclusion criteria should be established prior to the study and not on a post hoc basis. Records should be kept of which animals were excluded and why, and such information should be reported explicitly in the Methods section of data reporting.

Balancing and randomization

Sex-matching and age-matching are critical in study design as both of these factors significantly affect pathological expression. For example, Aβ plaque loads can increase exponentially during the first stages of plaque deposition, and spurious drug effects may be seen in animals analyzed at this stage unless control and treatment groups are age-matched to within days of one another. Mice should be separated into groups by sex, age, and litter and then randomly assigned to either control or treatment groups. In addition, wild-type or young controls or both should be included in study design as a reference point.

Blinding

Individuals conducting the experiments and those analyzing the results should be blinded to treatment. In the event that a test compound has a readily obvious phenotypic impact on the treated animals, these potentially unblinded observations should be noted by the animal handler but kept segregated to the degree possible from the analyst until the experiment is unblinded. If this is not possible, a full re-design of the experiment may be required. For example, a compound that results in reduced feeding activity (and the phenotypic observation of reduced rate of weight gain) may have an impact on Aβ levels for reasons unrelated to its therapeutic target.

Reporting

Investigators should report full details of target assay methods and detailed information on the animal model used, including genetic background, copy number, exclusion criteria, and statistical analyses. For behavioral assays, training as well as testing phases should be reported. When possible, scatter-plots should be shown rather than, or in addition to, bar graphs.

Publication bias fueled by a decreased ability or desire to publish negative results represents a huge problem for the field [18]. To increase efficiency, decrease redundant efforts, and learn from others' experiences, it is crucial that negative results be reported. Forums for discussing the quality of negative results, and results that differ from laboratory to laboratory, would aid in the interpretation of negative studies.

Exploratory versus therapeutic studies

Many investigators, particularly in academic settings, lack the infrastructure and budget to perform the extensive preclinical studies incorporating all of the design, methodological, and statistical considerations recommended here. In addition, comprehensive analyses are not always warranted when the compound or target is being assessed in early stages. As a result, we propose to distinguish between exploratory and therapeutic studies (Table 4).

Table 4 Exploratory versus therapeutic preclinical studies
Full size table

Exploratory studies

Exploratory studies should demonstrate that a particular molecular target is involved in a disease process. While exploratory studies do not require the extensive lead optimization, PK/PD, and toxicity analyses undertaken in therapeutic studies, they nonetheless should provide sufficient data to inform the decision of whether to proceed to a therapeutic animal study. Exploratory studies should contain a tolerability/toxicity assay to verify that selected doses are not causing an adverse effect. Multiple doses below the toxicity/tolerability range should be incorporated into an exploratory study, as doses approaching tolerability limits can frequently impact phenotypic outcomes unrelated to the therapeutic target being investigated. Furthermore, terminal blood and brain tissue samples should be collected for possible PK verification later, as the half-life of the test compound may or may not have been consistent with the timing of the putative therapeutic readout.

Therapeutic studies

Therapeutic studies should be compound-focused and include a full PK/PD and ADMET profile to ensure appropriate dosing and timing of outcomes with respect to exposure of the compound. Toxicity considerations are particularly critical in this context to minimize potential off-target phenotypic impacts on outcome measures. The design, conduct, analysis, and reporting of a therapeutic animal study should be analogous in rigor to those required for human clinical trials.

Future directions

Below, we list some overall recommendations and challenges that we hope will significantly advance the field by making animal studies more consistent and predictive of future clinical outcomes.

Improve access, characterization, and standardization of existing Alzheimer's disease mouse and rat models

The field should identify a few models in which key disease phenotypes are well replicated and characterize these models fully with regard to major targets and how they are affected by major biological and experimental variables (for example, age, gender, and housing conditions). Government funding of preclinical animal cores could improve availability and standardization of models. In addition, intellectual property creates obstacles to model access. This is a major impediment that needs to be addressed by the scientific and business communities.

Develop more animal models to non-traditional targets and make more use of available non-transgenic models

New animal models that better recapitulate the full complement of human AD pathology and novel non-amyloid targets are needed (Box 1). Aged rodent and non-transgenic models should also be better used as described above.

Standardize commonly used protocols

To be better able to compare and pool research results, it is important to improve quality control measures across laboratories. Efforts to standardize biomarker protocols have been widely successful [19]. Standardizing protocols for common assays, such as Aβ/tau extraction, behavioral assays, and measures of neuroprotection and neurodegeneration, in preclinical studies could rapidly advance the interpretation of the testing of novel treatments.

Develop new and higher throughput methods for measuring disease-related outcomes

Research efforts and funding should be targeted toward better characterizing and developing new methods for targets and outcome measures that are higher throughput for drug discovery. For example, for oxidative stress and inflammation, more emphasis should be placed on the development of pharmacological and metabolic imaging, gene expression, and proteomic screens that provide a broader view of redox and inflammatory changes and allow incorporation of both pro- and anti-oxidant or pro-and anti-inflammatory species. For behavioral tests, the field would benefit from new translatable behavioral tests that are sensitive to early changes and progression in cognitive function across lifespan and those that lend themselves to repeated testing for optimal within-subject experimental design. Standardization of behavioral methodologies should be attempted to the highest possible degree in order to improve the ability for comparisons of results from multiple laboratories.

Focus on novel targets and outcome measures

More emphasis should be placed on non-amyloid disease processes and pathways (Box 1), including those related to neuroprotection, synaptic plasticity, oxidative stress, inflammation, vascular targets, mitochondria, and energy use. Assays are available to measure these alternative outcomes.

Standardize review of animal studies in grant applications and scientific publications

Our advisory panel recommended developing a study design checklist of items for consideration by funding agencies and scientific journals (Table 3). A rating system on study design quality of published papers and a centralized website that lists study results with a forum for comments and feedback would help to improve the selection of compounds for promotion into clinical trials. Furthermore, it would increase the incentive among academic researchers to improve study design.

Establish a public data repository for animal studies

A public data repository for both positive data and negative data from animal studies would help to improve research efficiency and disseminate negative data. Given that it is often difficult to distinguish a true negative result from a poorly designed study, the critical challenge of such a resource would be quality control. The considerations listed above will be important in providing this sort of distinction and will enable analyses of studies with various strengths and weaknesses in design. Such a repository could help to identify translatable biomarkers in animal models that can also be used in human clinical trials to better predict outcomes.

Box 1. Thinking beyond amyloid: diversifying across disease processes

While many reasons, including clinical study design, may have contributed to the high-profile clinical trial failures with anti-amyloid treatments to date, the AD field would benefit from diversifying its research portfolio to include non-amyloid disease pathways. Understudied disease processes include neuronal function, vascular changes, oxidative stress and inflammation, mitochondria function, and lipid metabolism. Here, we list methods that can be used to assess some of these processes and that should be more fully exploited.

Neuronal function

As Alzheimer's disease (AD) progresses, loss of synapses, shrinkage of dendrites, and neuronal death occur. It is these pathological changes that are most closely associated with the cognitive decline seen in the human disease. They can be modeled to some degree in a number of different animal models and are amenable to experimental monitoring as outlined below.

  • Measure synaptic density by immunohistochemistry for the synaptic markers, such as synaptophysin. Non-homogeneity of synaptic markers in tissue surrounding plaques can present challenges in analysis. Also, PSD-95, AMPA-R, immediate early genes, and others are better markers for synaptic function than synaptophysin.

  • Assess dendritic branching, neuronal structure volume, and total neuron numbers with careful stereology [22].

  • Use T2-weighted magnetic resonance imaging (MRI) to indirectly assess neurodegeneration in vivo by measuring structure volumes, and use 1H magnetic resonance spectroscopy (1H MRS) to quantify N-acetyl-aspartate levels [23].

  • Use electrophysiology to measure long-term potentiation in hippocampal slices.

  • Employ behavioral studies to assess neuronal function. The Morris water maze, a spatial memory test that can be very sensitive to hippocampal function, is most commonly employed, but alternatives are available that may detect more subtle changes or may be more readily translatable - including attentional set shifting, delayed non-match-to-sample, recognition memory (novel object recognition), discrimination and reversal learning, contextual fear conditioning, and olfaction-based assays - or both [24–26]. Multiple behavioral tests may be needed to fully capture potential therapeutic effects. The use of multiple tests also helps control for factors, such as motivation and overall health, that may influence performance.

Vascular targets

Vascular pathology in human AD has received little attention, despite increasing evidence that vascular and neuronal dysfunction are closely intertwined and mutually exacerbating in the human disease. For example, cerebral amyloid angiopathy is seen in over 75% of patients with AD and can lead to vessel rupture, microbleeds, and hemorrhagic stroke [27, 28]. Other vascular changes include reduced cerebral blood flow, degeneration of vascular endothelium, basement membrane and smooth muscle, and pathological changes in the neurovascular unit associated with astrocytes, pericytes, and microglia [29]. Attention to vascular targets is further warranted by evidence that amyloid-beta (Aβ) immunotherapy exacerbates cerebral amyloid angiopathy and microhemorrhages in both mouse AD models and human AD [30]. Experimental methods for monitoring vascular pathology include the following:

  • Detect microhemorrhages with Prussian blue and double-stain for vascular Aβ. T2-weighted MRI can also be used.

  • Assess blood vessel area and patency (quantify using stains, such as tomato lectin, that bind to endothelial cells, together with imaging analysis). Counterstain for components of healthy blood vessels (for example, smooth muscle actin).

  • Quantify retinal hemodynamics, which may be translatable as a biomarker in humans [31].

  • Measure blood flow directly by arterial spin labeling and dynamic susceptibility contrast MRI and indirectly by FDG-PET or SPECT imaging. Blood volume can be sensitively measured by monocrystalline iron oxide nanoparticle-enhanced MRI.

  • Assess the structural integrity of the neurovascular unit by glial fibrillary acetic protein staining and counting total numbers of astrocytic end-feet and the number in contact with blood vessels.

  • Assay neurovascular unit function by immunocytochemistry, quantitative polymerase chain reaction, or Western blot measurement of aquaporin 4 or potassium channels (Kir4.1, BK calcium-dependent potassium channel) that are enriched in astrocytic end-feet. It is important to do immunohistochemistry in addition to biochemical measurements as channel distribution can be altered without changes in total levels.

Oxidative stress/Inflammation

Oxidative stress and inflammation are known to be associated with AD and are relevant targets for drug development, in particular for sporadic AD. However, detecting reliable changes in AD models can be quite difficult.

High oxidative stress is not seen in the most commonly used amyloid precursor protein transgenic (Tg2576) but can be seen in more recent models in which redox pathways have been genetically manipulated [32–34]. These models also show more aspects of AD pathology. Different animal models vary in their upregulation of specific inflammatory profiles; tau models, in particular, show high levels of inflammation in association with neurodegeneration. The time points in which these pathways are assessed is critical since oxidative and inflammatory processes that are toxic at one time point may be protective at others; their levels and effects may also vary among brain regions.

Because reactive species are labile and cannot be measured directly, oxidative stress must be measured via surrogate markers:

  • Chronic oxidative stress can be detected by monitoring lipid peroxidation or oxidized proteins (by carbonyl assay or by high-performance liquid chromatography) or by quantifying changes in genes and proteins known to increase or decrease with cellular redox balance [35].

  • Commonly used markers also include oxidized DNA and anti-oxidant proteins (such as glutathione, which is known to be directly affected by Aβ).

  • Free iron, copper, and zinc can be measured. Free iron is a linear indicator of disease progression in many neurodegeneration models and is an indicator of free radical generation around Aβ.

  • Many of these assays have low sensitivity and so likely will detect only robust drug effects. The more sensitive assays require mass spectrometry or high-performance liquid chromatography or 31P and/or 1H MRS.

  • Emerging technologies include two-photon imaging with novel probes for reactive oxygen species, which could be used to separately monitor changes occurring in the parenchyma and vasculature.

Inflammation can be assessed by the following:

  • Measuring invasion of peripheral cells or microglial activation or both. The number and morphology of microglia in tissue sections can be quantified.

  • In vivo imaging of microglial activation markers (for example, PK11195 positron emission tomography imaging, autoradiography, or tissue sample scintillation counting). Not all of these markers are well characterized in terms of when during the inflammatory process microglia express them.

  • Testing for blood-brain barrier leakage, which can be detected by steady-state gadolinium-enhanced T1-weighted MRI or classical intravenous bovine serum albumin and/or Evans blue brain parenchyma quantification. Related cerebral edema can be quantified by diffusion or absolute T2-weighted MRI or both.

FDG-PET, fluro-2-deoxy-D positron emission tomography; SPECT, single-photon emission computed tomography.

Conclusions

Our advisory panel produced recommendations in regard to the measurement, analysis, and reporting of relevant targets in AD animal models. These recommendations stressed the need for quality control measures in breeding and colony maintenance to manage phenotypic variability and outlined key issues related to preclinical animal study design. Distinguishing between exploratory and therapeutic animal studies will aid in defining the scope of the study and the interpretation of results and hopefully will bring some of the rigor of industry preclinical testing to the academic space.

Whereas Aβ likely plays a key role in the development of AD, amyloid deposition alone does not represent the entirety of the disease process or the totality of targets worth investigating for therapeutic intervention. To propel innovation, we should broaden our focus to additional disease-relevant pathways and processes (such as tauopathy, neuroprotection, synaptic plasticity, oxidative stress, inflammation, vascular changes, and mitochondrial dysfunction). Furthermore, non-transgenic models of disease are often underused. Aging is the greatest risk factor for AD; hence, aged rodent models that show cognitive impairment may provide a useful choice for testing investigational therapies targeting mechanisms of neuroprotection, learning, and memory. In addition, models that demonstrate clear neurodegeneration and cell death (such as tau transgenic models and pharmacologically induced models) may be better for testing neuroprotective therapies in general. There is no standard model or set protocol for testing investigational treatments in AD, and it is critical to tailor the choice of model, experimental plan, and outcome measures specifically for the therapy's target or proposed mechanism of action. Animal models of disease will never be able to predict all possible outcomes in humans.

While these recommendations are specifically geared toward AD, they echo many of the sentiments raised in other recent consensus efforts for related diseases such as stroke, vascular cognitive impairment, and amyotrophic lateral sclerosis [10, 20, 21]. These efforts, as a whole, high-light common challenges in animal model selection, study design, interpretation, and reporting which go beyond individual disease states. While each therapeutic area will undoubtedly have its own unique issues, there is much to be learned from shared barriers in translational research. Hopefully, these collective efforts will raise the bar for preclinical studies and will aid in designing animal studies to optimize their interpretability, improve their predictive value, and drive innovation, ultimately improving our efficiency in bringing effective treatments to patients.