**1. Introduction**

After decades of research, there is still a huge unmet medical need for novel interventions to treat dementia-like disorders. In 2019 more than 50 million people were a ffected by dementia and the

number could increase to about 152 million by 2050 [1]. Alzheimer's disease (AD) is the most common type of dementia, accounting for 2/3 of all cases [2]. More than 400 failures in drug development during the last decades [3] have led to the consideration of alternative intervention options, e.g., repurposing, combinatory approaches and preventive treatments [4].

The complex pathology of the disease is characterized by several hallmarks, such as prominent extracellular amyloid plaques [5,6]. According to the amyloid cascade hypothesis, an alteration of amyloid-β (Aβ) metabolism is the central pillar of AD pathology and crucially influences and initiates other hallmarks [7]. In AD, initial pathologic processes progress decades before the first cognitive symptoms appear in patients, a stage entitled preclinical Alzheimer's [8]. Disruptions in amyloid metabolism, as one of the first chronological hallmarks, potentially represent a relevant target for preventive interventions in AD.

In order to further elucidate disease mechanisms and identify novel treatment options, the group of Takaomi Saido at the RIKEN Center for Brain Science has developed a new generation of AD mouse models. These knock-in (KI) mice provide advantages compared to transgenic models, which are based on massive amyloid-β protein precursor (AβPP) overexpression with the result of artificial phenotypes due to overproduction of other AβPP fragments aside from Aβ. In the *AppNL-G-F* model, the murine AβPP sequence is humanized and three mutations are introduced. Swedish (NL), Arctic (G) and Beyreuther/Iberian mutations (F) increase the total amount of Aβ and the Aβ42/Aβ40 ratio, show pro-inflammatory e ffects and finally result in a three times faster memory impairment [9].

Elevated levels of the endogenous amino acid homocysteine (HCys), called hyperhomocysteinemia, have been described as another hallmark of AD [10]. HCys is increased significantly in AD patients, whereas levels of di fferent B-vitamins are reduced compared to controls [11,12]. A remaining question is whether hyperhomocysteinemia is merely a marker or whether it contributes causally to AD pathology, thereby providing options for therapeutic intervention. Some authors describe the role of plasma HCys as an independent risk factor for memory deficits and AD [13,14]. Consequently, B-vitamin supplementation as a HCys-modifying intervention was proposed previously [15]. According to Smith et al., B-vitamins lowered HCys levels and subsequently slowed the rate of brain atrophy and cognitive decline in patients [16,17]. However, a causal link between hyperhomocysteinemia and Alzheimer's disease, called the "homocysteine hypothesis", has been a source of controversy for years. Kennedy teased out the equivocal results of numerous studies in detail [18]. Several studies neither support an association of HCys with AD nor an improvement of cognitive performance by B-vitamin treatment [19,20]. Meta-analyses were conducted to assess this topic, challenging the homocysteine hypothesis and amelioration of cognitive functions by the use of folate and other B-vitamins [21,22]. In the context of an international consensus statement, researchers assessed the homocysteine hypothesis as being plausible and considered hyperhomocysteinemia a modifiable risk factor for dementia. Furthermore, they recommended considering polyunsaturated fatty acids (PUFAs) in addition to B-vitamins for future trials [23]. PUFAs such as docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA) are also suggested to be linked to AD pathology and HCys metabolism [24,25], i.e., elevated HCys impairs the formation of PUFAs and leads to a lower availability of PUFAs in the brain. B-vitamin treatment might only be successful when PUFA plasma concentrations are in the upper normal range [26]. A more recent systematic review points out that the evidence for nutrient supplementation remains limited and indicates that more research is needed to assess preventive measures in dementia [27].

Transsulfuration and re-methylation are major metabolic pathways for HCys (Figure 1), being dependent on an adequate supply of B-vitamins, particularly B6, B12 and folate [28]. As illustrated, the relevant B-vitamins play key roles in intrinsically decreasing HCys levels and therefore correlate negatively with HCys. Vitamin B12 and folate are crucial in providing methyl groups in the context of the re-methylation cycle, whereas the transsulfuration pathway depends on vitamin B6 as an essential enzymatic cofactor. Disturbed HCys metabolism (Figure 1) is likely to be linked to AD pathology by direct and indirect neurotoxic pathways [24]. Neurotoxicity is caused by excitotoxicity via

N-methyl-D-aspartate receptor (NMDA) activation and by increased levels of reactive oxygen species promoting oxidative stress. Furthermore, excess HCys and subsequently a lack of methionine and S-Adenosyl-L-methionine (SAM), as well as elevated S-Adenosyl-L-homocysteine (SAH), are associated with a reduced methylation capacity and the inhibition of methylation reactions, which is suggested to exacerbate amyloid and tau pathologies in AD. Moreover, HCys results in an activated immune system, damages cerebral vessels and disrupts the blood-brain-barrier [24,29]. Both homocysteine and its oxidative metabolite homocysteic acid (HCA) are considered neurotoxic [30,31], but HCA is suggested to be the more potent species [32–34] and might contribute to dementia through oxidative stress and excitotoxicity by NMDA activation. Both mechanisms have been considered relevant for AD pathology [5,24].

**Figure 1.** Homocysteine (HCys) and homocysteic acid (HCA): metabolic role and link to Alzheimer's disease; involved enzymes (black boxes) linked to relevant B-vitamins (green boxes) functioning as coenzymes or methyl donor (C1); SAM = S-Adenosyl-L-methionine; SAH = S-Adenosyl-L-homocysteine.

The present exploratory animal study concentrates on the role of hyperhomocysteinemia, driven by vitamin B deficiency, in the context of AD. Therefore, we used the novel and not ye<sup>t</sup> fully characterized *AppNL-G-F* knock-in mouse as a model of the disease. The *AppNL-G-F* mouse is expected to display a mildly impaired phenotype, simulating the very early preclinical period of AD pathology and thus should provide the possibility of assessing preventive interventions adequately. A versatile behavioral test battery should firstly assess potential deterioration of cognitive performance by hyperhomocysteinemia. Secondly, behavioral testing should clarify whether special diets enhance cognition and potentially could serve as preventive measures for AD. Here, we compared B-vitamins and PUFAs with a more complex micronutrient mixture similar to Fortasyn® Connect [35]. HCys and HCA levels were

measured in urine and serum using a validated LC-MS/MS method (liquid chromatography-tandem mass spectrometry) and the quantity of Aβ plaques in the brains was assessed.

#### **2. Materials and Methods**

A detailed description of all experimental procedures including the single behavioral testing systems, analytical methodologies and quality parameters of the current study can be found in Appendix A.

#### *2.1. Animals and Experimental Diets*

All experimental procedures were carried out in compliance with the '3R' and in accordance with the Principles of Laboratory Animal Care (National Institutes of Health publication no. 86-23, revised 1985), the DIRECTIVE 2010/63/EU and the regulations of GV-SOLAS and were approved by the local Ethics Committee for Animal Research in Darmstadt, Germany (approval number: F152/1011; approval date: 31.07.2017). In the current study, 16 C57BL/6J wild type mice (WT) and 96 homozygous *AppNL-G-F* knock-in (KI) mice, consisting equally of males and females, were included.

AIN93M chow served as a basis for the experimental diets and was modified, defining the different groups of *AppNL-G-F* mice (Table 1). The exact composition of the diets is summarized in Table A1. Each mouse received four grammes of diet per day, except for the period of food restriction for males during the touchscreen PAL-task. Water was available ad libitum, except for the period of temporally conditioned water access for females during the IntelliCage experiment.


**Table 1.** Details of the experimental groups.

#### *2.2. Behavioral Testing*

The testing battery we conducted consisted of diverse behavioral tests investigating different domains of cognition in the animals (Figure 2). At the age of 15 weeks, resp. 10 weeks on diet, the mice were first tested in the open field, followed by the elevated zero maze, Barnes maze and social interaction test. Finally, males were tested in a touchscreen task and females in the IntelliCage system.

Outcomes of every behavioral experiment were assessed automatically by camera or transponder detection. All experiments were performed between 8 a.m. and 3 p.m. during the light phase. After each trial, testing systems were cleaned with 70% ethanol to remove odors in the devices and to achieve comparable conditions for each animal.

#### *2.3. Sample Collection*

As illustrated in Figure 2, serum and 24-h urine of the mice were sampled after 8 and 30 weeks on experimental diets, resp. 13 and 35 weeks of age. The biological matrices were stored at −80 ◦C for subsequent analysis of HCys and HCA. At the end of the study, we euthanized all animals at the age of 35 weeks in order to harvest the brains. Brains were removed and post-fixed in 4% paraformaldehyde, followed by a stepwise dehydration, and embedding in paraffin. Ten μm thick sections were cut and mounted on glass slides for subsequent immunohistochemical analysis.

**Figure 2.** Time line of the study course; open field test (OF), elevated zero maze (EZM), Barnes maze (BM), social interaction test (SI), touchscreen paired associates learning (PAL) inclusive training phase, IntelliCage place learning task (PL) and reverse learning task (RL) inclusive habituation period; see Appendix A for detailed explanations of the single tests.

#### *2.4. Biochemical and Immunohistochemical Analyses*

The determination of HCA was performed as previously described in detail [36] using a combination of protein precipitation and solid phase extraction for sample preparation followed by an LC–MS/MS analysis applying a combination of a HILIC separation and tandem mass spectrometry. HCys was analyzed using protein precipitation in combination with reversed phase chromatography and tandem mass spectrometry.

Brain sections were immunohistochemically stained for amyloid-β peptides (Aβ) using an ABC/DAB protocol that is described in detail in Appendix A. After digitization of the sections, we analyzed the resulting images for the area of Aβ plaques in several regions of interest (ROI; Table A2), using ImageJ software.

#### *2.5. Statistical Analyses*

All experiments were statistically analyzed using IBM SPSS Statistics 25 (Ehningen, Germany). For each test, we conducted an outlier analysis in order to exclude extreme outliers (more than three times the interquartile range). Shapiro Wilk tests revealed whether Gaussian distribution could be assumed or not. Because of several data sets, which did not show a normal distribution, testing of statistically significant di fferences was computed by non-parametric Mann-Whitney-U-tests (comparison 1: C57BL/6J (group 1) versus *AppNL-G-F* control (group 2); comparison 2: *AppNL-G-F* control (group 2) versus *AppNL-G-F* on special diets (groups 3–7)). A *p* value lower than 0.05 was considered statistically significant. Results were expressed as median ± interquartile range (IQR). Where applicable, medians were further compared to hypothetical medians using the non-parametric one-sample Wilcoxon signed rank test.

Graphical presentation was performed using GraphPad Prism 7 software (San Diego, CA, USA).
