Cognitive impairment, one of the key non-motor comorbid symptoms of PD, is not only a risk factor for early mortality but also has a greater impact on quality of life than motor symptoms (Oosterveld et al., 2015; Saredakis et al., 2019). Cognitive impairment in PD can be divided based on severity into mild cognitive impairment (PD-MCI) and dementia (PDD). PD-MCI, an intermediate cognitive stage on the continuum between normal cognitive function and dementia, can be identified in 15%–40% of PD patients at the time of diagnosis (Aarsland, 2016; Monastero et al., 2018). The prevalence of PD-MCI and PDD in the PD population is 40% and 30%, respectively (Aarsland and Kurz, 2010; Baiano et al., 2020). PDD occurs in 75%–90% of patients who have had PD for 10 years or more (Gratwicke et al., 2015). Understanding the pathophysiology of PD-MCI and PDD is critical to the development of disease-modifying treatments. Although only 5%–15% of PD cases are attributed to clear genetic factors, several studies have shown that genetic factors such as α-synuclein (α-Syn) variants play an important role in cognitive impairment (Ball et al., 2019; Chen et al., 2020). The 85%–95% remaining idiopathic cases of PD can be linked to environmental factors, such as heavy metals, pesticides, and illicit drugs, whose important role cannot be overlooked when studying cognitive impairment (Ball et al., 2019; Cariccio et al., 2019; Agnihotri and Aruoma, 2020). Systemic medical factors are also associated with PD cognitive impairment, including lower uric acid levels and the presence of an orthostatic heart rate or orthostatic hypotension (McDonald et al., 2016; Veselý et al., 2019; Tanaka et al., 2020). Gene–environment–medical history interactions may play a vital role in the pathogenesis of PD-MCI and PDD. Because the development of cognitive impairment in patients with PD is a multi-factor interaction, the heterogeneous pathophysiologic mechanisms underlying PD-MCI and PDD remain unclear. Recent studies have shown that inhibitors, amino acids, and actively targeted gold nanoparticle composites could improve the cognitive flexibility of experimental models. There are very few therapeutic options for PD patients with cognitive impairment due to the limited efficacy and significant side effects of therapeutic methods (Hsu et al., 2018; Byers et al., 2020; Liu L. et al., 2020; Wang et al., 2021). In order to identify risk factors, mechanisms, and treatment strategies for PD-MCI and PDD, research using appropriate experimental models is invaluable.

Experimental models, an important part of basic science research that emulates human physiology or pathology, are crucial to improving our understanding of PD-MCI and PDD pathophysiology and to the preclinical testing of novel therapeutics. Prior use of multitudinous PD experimental models has assessed the different symptoms and stages of PD and includes classical neurotoxin-induced, transgenic, and combined models (Kin et al., 2019). Some experimental PD models replicate PD cognitive impairment, which is very useful for studies on PD-MCI and PDD. These models include genetic and toxin-based exposure to a variety of animals (Prediger et al., 2010; Ba et al., 2019; Singh et al., 2019). However, each model exhibited varying degrees of motor and cognitive dysfunction symptoms or only produced mild symptoms of cognitive dysfunction alone because of the tangle modeling methods that were used. These experimental models, which were assessed with uneven behavioral tests such as the NOR, MWM, Y-maze test, T-maze test, and Barnes maze test demonstrated short-term or long-term spatial memory deficits, episodic-like memory deficits and recognition memory deficits (Huang et al., 2015; Singh et al., 2019; Ding et al., 2020). Regrettably, an ideal and representative experimental model that replicates all of the phenotypic and pathologic features of PD-MCI and PDD is still unavailable. However, each existing model has its own benefits and drawbacks that are important to consider when selecting an appropriate model based on the specific objectives of the study (Grandi et al., 2018). It is therefore vital to have a comprehensive understanding of the neuropathology of each model, permitting an accurate explanation of research results and a reasonable estimate of how well the model will correlate with human PD-MCI and PDD. In this review, we analyze trends in PD cognitive impairment modeling and discuss the features and limitations of current models based on the clinical features of PD-MCI and PDD.

To clarify trends in experimental modeling of PD cognitive impairment, we systematically searched PubMed using the following search terms: (“Parkinson’s disease [Title/Abstract]” AND “cognitive function [Title/Abstract]”) OR (“Parkinson’s disease [Title/Abstract]” AND “cognitive impairment [Title/Abstract]”) OR (“Parkinson’s disease mild cognitive impairment” AND “model”) OR (“Parkinson’s disease dementia” AND “model”). The publication range was from the start of the database to 30 May 2021. Two investigators (YF and JH) reviewed each topic and abstract independently to select references that met the inclusion criteria. Inconsistencies encountered by the reviewers would be resolved after consultation with a third reviewer (DC). The selection process is shown in Figure 1. We identified 5432 articles, of which 738 were duplicates. A total of 227 articles met our inclusion criteria and were included in the analysis.

Information on the experimental models used in each work was extracted. Models were divided into three categories based on model generation: neurotoxin-induced experimental model, transgenic experimental model, and combined experimental model. Models using any kind of neurotoxic substance were defined as a neurotoxin-induced experimental model. The transgenic experimental model group included models that involved gene manipulation. Models that included two or more methods were assigned to the combined experimental model group. Figure 2 demonstrates an increased focus on cognitive impairment in PD over time. Although neurotoxin-induced experimental models were the most prevalent during every time period, transgenic experimental models and combined experimental models have received increased recent attention.

Only 5%–15% of PD cases are attributed to clear genetic factors; however, transgenic experimental models play a significant role in the study of idiopathic PD-MCI and PDD. In addition to common familial PD-linked genes, which include α-Syn, parkin, PTEN-induced putative (PINK1), protein deglycase (DJ-1), and LRRK2, 11 other transgenic models were used in a total of 38 articles to replicate PD-MCI or PDD (Table 5). The genes that these models interfered with included Pitx3, dopamine transporter (DAT), mitochondrial transcription factor A (TFAM), β-synuclein (β-Syn), A2A receptors, Atp13a2, B4galnt, Ifnb, midkine (Mdk), Ndufs4, tau, and Washc4 (Prediger et al., 2011; Lei et al., 2012; Sekiyama et al., 2012; Li et al., 2013; Moon et al., 2013; Schultheis et al., 2013; Ejlerskov et al., 2015; Choi et al., 2017; Hwang et al., 2019; Wu et al., 2020; Courtland et al., 2021). Rodent, NHP, and drosophila animal models based on α-Syn are currently available. Rodents and drosophila have been used in LRRK2 transgenic models. However, other rodent transgenic models have also been established.

α-Synuclein, a small neuronal protein present at presynaptic terminals, regulates membrane and vesicular dynamics (Kin et al., 2019). Transgenic models of α-Syn, such as those with mutations (A53T and A30P) or gene knockout, have developed cognitive deficits similar to PD-MCI or PDD (Freichel et al., 2007; Kokhan et al., 2012; Singh et al., 2019). As tau oligomers are elevated by α-Syn in the hippocampus, pons, and cerebellum, A53T mice showed not only freezing behavior and loss of motor coordination but also long- and short-term spatial learning and memory deficits at 12 months (but not 6 months) of age (Singh et al., 2019). Several mechanisms are related to the motor and cognitive deficits seen in α-Syn transgenic models, such as reduced dendritic spine density of the hippocampal and caudate putamen medium spiny neurons, lower synaptophysin and synapsin I levels in the frontal cortex and hippocampus, and impaired microglial response (Table 5). Moreover, α-Syn transgenic NHP models of PD develop more replicable and stable phenotypes and pathologic changes compared with those of toxin-treated NHP models of PD (Niu et al., 2015). Interestingly, α-Syn knockout mice also developed long-term learning and spatial memory deficits, which suggests that appropriate α-Syn is necessary for cognitive function of the hippocampus and limbic systems (Kokhan et al., 2012). Therefore, human α-Syn A53T and A30P appear to play a key role in the mechanisms of PD-MCI and PDD.

R1441C/G and G2019S are the two most common LRRK2 mutations that are related to the pathologic mechanisms of the autosomal dominant inheritance of familial PD. LRRK2 mutation models may be most useful for studying early compensation in the nigrostriatal DA system because they demonstrate loss of motor coordination and spatial recognition memory deficits consistent with late-stage DA transmission deficits in the dorsal striatum and low expression of PSD-95 in the hippocampus (Sloan et al., 2016; Adeosun et al., 2017).

Gene knockout (KO) models, including parkin, PINK1, DJ-1, Pitx3, Mdk, and Ndufs4, are useful models of the earliest abnormalities in the nigrostriatal DA system during PD pathogenesis. MitoPark (DAT^+/cre-Tfam^loxP/loxP) mice and TGR (NSEhA2A) rats may play an important role in studies that involve the DA transporter. Transgenic models that more closely resemble human PD physiology and progression are valuable in the development of preventative treatments. However, it is not clear if results that are obtained from transgenic models can be applied directly to humans because most of the splicing events between humans and rodents are conserved (Kin et al., 2019). On the other hand, some types of transgenic models can serve as novel tools for generating different phenotypes of cognitive deficits and unstable motor symptoms of PD, although the repeatability and stability of these phenotypes require further verification. Thus, we should objectively analyze the results of transgenic models in future work, such as by observing the relationship between cognitive impairment and changes in different regions of the brain during transgenic animal development and aging using advanced imaging.

In order to imitate the complex pathogen and neuropathology of PD-MCI and PDD, combined experimental models have become increasingly popular. The number of articles that utilized models has gradually increased. A total of 24 kinds of combined experimental models were reported in 30 articles and can be divided into three categories: dual/multiple toxin-induced models, models combining toxins and transgenes, and models combining toxins and surgery (Tables 6–8).

There are currently six kinds of dual/multiple toxin-induced models of PD with cognitive deficits (Table 6). 6-OHDA- and LPS-combined lesions may provide a new model for verifying associations between memory process vulnerabilities and the stress reactivity of the temporal brain in PD (Hritcu et al., 2011). The hypocretin/orexin (HO) system plays a central role in memory processes. The model induced by 6-OHDA and saporin anti-orexin-B showed deficits in long-term and spatial memory, but not short-term and working memory, which suggests that the HO system is functionally involved in memory processes and consolidation in PD (Oliveira et al., 2020). A PD-MCI model induced by paraquat and maneb administered twice a week for a total of 12 times was characterized by deficits in recognition memory, spatial learning, and memory by damaging synapses in the hippocampus and DA neurons in the SN (Ding et al., 2020). It is a relatively mature combined model that has been widely used in pre-clinical studies of potential therapies for PD-MCI (Wang et al., 2021). On this basis, N-(2-chloroethyl)-N-ethyl-2-bromobenzylamine (DSP-4), which damages locus coeruleus/norepinephrine neurons and depletes norepinephrine, combined with paraquat and maneb, could exacerbate spatial learning and memory via ferroptosis and microglia-mediated neuroinflammation (Hou et al., 2019). Moreover, the model induced by DSP-4 and 6-OHDA developed a similar phenotype (Rampersaud et al., 2012). An LPS and recombinant monomeric human α-Syn co-induced model that has the pathologic features of LB-formation and neuroinflammation could potentiate the effects of each vector and may be a high potential model for studying PD-MCI and PDD (La Vitola et al., 2021).

There are currently 12 kinds of models that combine toxins and transgenes (Table 7). Because both genetic makeup and environmental factors have a crucial role in the early onset of PD, these combined rodent models are useful tools for assessing the effects of environmental factors on genetically susceptible individuals. The combination of one kind of toxin, such as 6-OHDA, MPTP, reserpine, α-Syn oligomers, human α-Syn preformed fibrils, and LPS, and one kind of transgenic mouse, including APP_swe/PS1△E9 mice, CHT^HET mice, mice expressing the human E4 isoform, hTau mice, TKO mice, Nestin-GFP mice, hD1 mice, Prnp^0/0 mice, TLR4^–/– mice, 5xFAD Tg mice, and A53T mice resulted in memory, attention, and motor defects (Zurkovsky et al., 2013; Melief et al., 2015; Klein et al., 2016; Bruns et al., 2018; La Vitola et al., 2018, 2019, 2021; Gratuze et al., 2019; Bassil et al., 2020; Torres et al., 2020). These models, which are more closely related to the pathogenesis of PD-MCI and PDD, may be more suitable for assessing the efficacy of a therapeutic agent. However, as generating models that combine a toxin and transgene is difficult, cumbersome, and cost-consuming, significant additional research and discussions are required.

Six kinds of models combined a toxin and surgery to understand the impact of the relationship between risk factors and cognitive deficits in PD, yielding a total of 10 articles (Table 8). For instance, the model combining a toxin (6-OHDA or MPTP) and a bilateral ovariectomy replicated PD-MCI and PDD phenotypes, indicating that parkinsonism aggravates during the premenstrual period in the setting of the lowest levels of estrogen and progesterone (Arbabi et al., 2016; Yadav et al., 2017). Maternal separation, a method that imitates a state of stress, worsens cognitive function in a PD model (Dallé et al., 2017). A model that combined MPTP and narrowing of the bilateral common carotid arteries explored the relationship between orthostatic hypotension and cognitive impairment in PD, reporting that hypotension aggravates cognitive impairment in PD by exacerbating hippocampal damage and white matter lesions. Finally, a model combining rotenone and sleep deprivation reported that the combination of pesticide exposure and insomnia is a key risk factor for cognitive deficits in PD (Gao et al., 2019; Kmita et al., 2019; Lv et al., 2019).

Various experimental models for PD-MCI or PDD have been generated using multiple methods and diverse animal species. Unfortunately, there is no perfect experimental model for studying PD-MCI or PDD, and five pivotal problems that are summarized in the proceeding paragraphs require further consideration and resolution.

One principal challenge is that the difference between PD-MCI and PDD has not been standardized in an experimental model, although there are analogous phenotypic, behavioral, biochemical, and anatomical impairments that represent both PD-MCI and PDD. A clear distinction between PD-MCI and PDD will help transform the results of experimental studies into clinical applications more easily. Based on the diagnostic criteria of PD-MCI and PDD, the severity of cognitive impairment is a key distinction between these pathologies (Dubois et al., 2007; Litvan et al., 2012). A PD model could be used as a PD-MCI model if any one of the following cognitive impairments appears: attention, executive function, memory, or visuospatial function as assessed using different kinds of behavioral tests. When the model has two or more of the above cognitive deficits, it can be considered a model of PDD. Importantly, researchers need to understand the various testing methods that are used to assess PD symptoms, such as dyskinesia and cognitive impairment, from different perspectives. In assessing dyskinesia, the foot-printing test, the rotarod test, the pole test, and the balance beam are commonly used and authoritative methods in rodents. Among these, the balance beam, a simple and direct balance and motor coordination test, is superior to the rotarod test for testing fine movement defects. For NHP, we can assess dyskinesia based on the Unified Parkinson’s Disease Rating Scale. Moreover, we can evaluate the motor function of the drosophila or zebrafish by analyzing their movements paths. For single-sided injections, such as 6-OHDA, specific stereotyped movements after injection of apomorphine is an important test for measuring dyskinesias (Smith et al., 2017; Real et al., 2019). As the DA content in the striatum of the injured side decreases, the number and sensitivity of DA D2 receptors increase significantly. Apomorphine, a DA receptor agonist, can make the excitatory effect of the injured side dominate, thereby inducing the spontaneous rotational movement of models. In addition, VDR, CPT, MDR, SDR, ASST, MWM, NOR, PAT, Barnes maze test, T-maze test, and Y-maze test are the authoritative and representative methods for measuring cognitive impairment. Among these, MWM, a common test to investigate spatial learning and memory in rodents, can be used to assess lesions in distinct brain regions, such as the hippocampus, striatum, cerebral cortex, basal forebrain, and cerebellum (D’Hooge and De Deyn, 2001). In addition to detecting spatial learning and memory functions, the Y maze can detect the ability of animals to recognize spatial position and orientation, the ability to avoid conditioned reflexes, spatial working memory, and fragmented memory. The Barnes maze test is used to evaluate spatial memory and non-spatial memory of experimental animals and can distinguish reference memory from working memory. NOR, a fine and sensitive behavioral method, uses rodents’ instincts to approach and explore novel objects to detect recognition and memory. Thus, we should select the corresponding method based on the different experimental requirements, conditions, and objects. This also requires the evaluation of motor and cognitive function in novel models using different types of behavioral experiments. Forming a set of plans that integrate symptoms, biochemistry, neuroimaging, and other objective indicators to judge and identify the success of a proposed novel model is therefore needed.

An additional challenge is that most novel models simply develop a cognitive deficit that is measurable by a single or analogous behavioral test in the setting of changes in pathology. No model has been integrated to evaluate phenotypes, such as motor deficits, cognitive impairment, and biomarkers in biofluids, and neuroimaging. It is therefore difficult to determine whether a model represents PD-MCI or PDD. The lack of a comprehensive evaluation tool is a principal obstacle to the mutual transformation of research results from basic experiments to clinical trials, especially with respect to clinical examination and radiology. For example, the CSF of rodents, drosophila, and zebrafish is difficult to obtain. MRI and PET instruments for animals are rare and expensive. However, these problems will be solved in the future with the development of technology. Thirdly, molding rate and mortality, which are significant indicators of the repeatability and stability of a novel model, are rarely mentioned. One example is a work that identified a 60%–80% molding rate in 6-OHDA-induced models (Marshall et al., 2019). Some MPTP-induced models also appeared to have normal cognitive function (Gao et al., 2019). However, the combination of MPTP and pesticides resulted in a high mortality rate in rodents. Mortality would be expected to increase in combined experimental models, and the molding rate and mortality of such models should be calculated in future studies.

Fourth, NHP, the closest species to humans, developed cognitive deficits in MPTP-treated and α-Syn transgenic models. However, the mechanism behind the cognitive impairment that results from these models is unclear (Schneider and Kovelowski, 1990; Vezoli et al., 2011; Niu et al., 2015). Moreover, handling and breeding NHP processes are difficult, time-consuming, and expensive. Thus, how to effectively use NHP for PD research remains a significant issue. We could observe the dynamic transition process from PD-MCI to PDD in NHPs, such as symptoms, biomarkers in biofluids, and neuroimaging, and provide new objective evidence for distinguishing PD-MCI from PDD. Furthermore, in addition to the striatum, SN, and hippocampus, researchers need to pay attention to lesions of the PFC, amygdala, anterior commissure, locus coeruleus, and putamen in the NHP model. Importantly, we need to use a neural network perspective to study the pathological mechanisms of PD-MCI and PDD in the NHP model, given their complex pathogenesis (Gratwicke et al., 2015). Finally, cellular models are necessary for mechanism research but have not been used in studies on PD-MCI or PDD. Abnormal aggregation of α-Syn and human α-Syn oligomers were positively correlated with the severity of cognitive deficits in PD (Finkelstein et al., 2016; Liu L. et al., 2020). Inducible α-Syn overexpression and SH-SY5Y-A53T cellular models are good candidate models of PD-MCI or PDD (Zhang et al., 2020; Zhu et al., 2021). However, it is currently difficult to distinguish PD-MCI from PDD at the cellular level.

While each of the models discussed in the present work has their own disadvantages, there have been a tremendous number of advances in PD cognitive impairment models. Almost every model can verify a hypothesis about PD-MCI or PDD. Based on previous research, we identified a set of plans that integrate symptom, biochemistry, neuroimaging, and other objective indicators to judge and identify that the novel model plays a key role in new strategies for developing representative models. Moreover, we may develop representative models based on the different clinical features of PD-MCI and PDD. Simultaneously, inducible α-Syn overexpression and SH-SY5Y-A53T cellular models and good candidate models of PD-MCI or PDD are important directions for the development of novel models. With the development of improved technology and in-depth research methods, novel models will become increasingly in line with the complex neurodegenerative symptoms and mechanisms of PD-MCI and PDD.

YF and MZ designed the research. YF and JH analyzed the data. YF and LZ wrote and reviewed the manuscript. ZH, XH, and YJ reviewed the manuscript. XH, CW, and PW edited figures and tables. DC and MZ supervised the research and reviewed the manuscript. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.