**3. Results**

#### *3.1. Homocysteine and Homocysteic Acid*

LC-MS/MS analysis was performed in order to measure HCys and its oxidative metabolite HCA in serum and urine samples. Vitamin B deficiency resulted in an elevation of both HCys and HCA serum levels in males and females after 8 weeks on experimental diet (HCys male *p* < 0.001, female *p* = 0.001; HCA (pooled) *p* < 0.001) (Figure 3A,C). A consistent statistically significant di fference between C57BL/6J wild type (WT) and *AppNL-G-F* knock-in (KI) mice was not observed. Dietary interventions resulted in decreased serum levels of HCys (PUFA-ENR male *p* = 0.001, female *p* = 0.005; B+PUFA-ENR male *p* < 0.001, female 0.026; FC male & female *p* < 0.001). Serum samples had to be pooled for an adequate analysis of HCA because of low sample volumes obtained by vena facialis puncture (Figure 3C). Because of the resulting decreased number of observations, data are not depicted separately for males and females in this case. After 30 weeks on the diet, vitamin B deficient males remained significantly hyperhomocysteinemic (*HCys p* = 0.001; *HCA p* = 0.001), although to a lower extent, compared to 8 weeks on the diet, whereas females returned to baseline level due to the maintenance chow they received during the IntelliCage tasks. Analysis of 24-h urine samples delivered data that were largely comparable to the results from the serum samples. After 8 weeks on the diets (Figure 3E,G), both urinary HCys and HCA were significantly elevated because of the vitamin B deficient chow (HCys male & female *p* < 0.001; HCA male *p* = 0.001, female *p* = 0.035), whereas a genotype e ffect was not detectable. Experimental diets resulted in decreased amounts of HCys (PUFA-ENR female *p* = 0.014) and HCA (B-ENR female *p* = 0.001; PUFA-ENR female *p* = 0.022; FC female *p* = 0.040) in the urine compared to KI control mice. After 30 weeks on diets (Figure 3F,H), males deficient in vitamin B6, B12 and folate displayed elevated urinary amounts of HCys (*p* = 0.001) and HCA (*p* = 0.003), but to a lower extent compared to that after 8 weeks on the diets. Vitamin B deficient females showed equal quantities to the control groups due to the maintenance chow they had received during the IntelliCage tasks.

#### *3.2. Open Field*

This behavioral test aimed to evaluate locomotion, anxiety, and habituation behavior of the mice during a 30-min session in the open field boxes. The total distance moved revealed no statistically significant di fferences (Figure 4A). Consequently, locomotion activity was not influenced by genotype or dietary intervention. The time the animals spent in the inner zone of the box, an indicator of anxiety, was not a ffected by genotype or diet (Figure 4B). As a third parameter, the amount of intrasession habituation was expressed by a habituation ratio (Equation (1)):

$$\text{ratio instances} \\ \text{withination} = (\text{5 min(final)}) / ((\text{5 min(final)} + \text{5 min(initial)})) \tag{1}$$

A ratio lower than 0.5 indicates habituation; a ratio of 0.5 means no change in activity, i.e., that no habituation occurred as in the case of groups 2–6 in males and groups 3 and 5–6 in females. Females fed with a vitamin B deficient chow displayed the least tendency to habituate; however, e ffects of experimental diets did not reach statistical significance in comparison to the KI control group. Female *AppNL-G-F* control mice displayed a significantly lower level of habituation compared to the C57BL/6J WT control (*p* = 0.009), indicating an impact of the genotype (Figure 4C).

#### *3.3. Elevated Zero Maze*

We tested anxiety behavior of each mouse for a session duration of 5 min. C57BL/6J WT and *AppNL-G-F* KI control mice moved equal distances in the maze; only male *AppNL-G-F* mice fed with a vitamin B and PUFA enriched diet moved less than *AppNL-G-F* controls (*p* = 0.003) and thus displayed lower locomotion activity (Figure 5A). The time spent in the open corridors of the maze was an index for open space-induced anxiety in mice (Figure 5B). No genotype e ffect was observed between C57BL/6J and *AppNL-G-F* mice, whereas di fferent dietary interventions showed a reduction of cumulative time in open corridors. Particularly, male mice fed with the combination of PUFA and vitamin B enriched chow as well as with FC-like, spent significantly less time in the open corridors (*B*+*PUFA-ENR p* < 0.001; *FC p* = 0.021) and thus displayed increased anxiety. In females, a reduced time in the open corridors was observed in the vitamin B deficient group (*p* = 0.040) compared to KI control mice.

**Figure 3.** Homocysteine (HCys) and homocysteic acid (HCA) urine and serum levels in 13 and 35 weeks old C57BL/6J and *AppNL-G-F* mice; 8 resp. 30 weeks on experimental diet; all samples analyzed by LC-MS/MS; data presented as median ± IQR; outliers beyond threefold IQR removed; *p* < 0.05 (Mann-Whitney-U-test) considered statistically significant (\*). (**A**) HCys serum levels; 8 weeks on diet. (**B**) HCys serum levels; 30 weeks on diet. (**C**) HCA serum levels; 8 weeks on diet (males and females pooled); samples pooled for analytical method due to low volumes obtained by vena facialis puncture. (**D**) HCA serum levels; 30 weeks on diet. (**E**) HCys urine levels; 8 weeks on diet. (**F**) HCys urine levels; 30 weeks on diet. (**G**) HCA urine levels; 8 weeks on diet. (**H**) HCA urine levels; 30 weeks on diet.

**Figure 4.** Open field test (30 min) in 15 weeks old C57BL/6J and *AppNL-G-F* mice; 10 weeks on experimental diet; data presented as median ± IQR; outliers beyond threefold IQR removed; *p* < 0.05 (Mann-Whitney-U-test) considered statistically significant (\*). (**A**) Total distance moved. (**B**) Percentage of time spent in inner zone. (**C**) Intrasession habituation expressed as ratio between the distance moved in the final time block divided by the sum of the final and the initial time block. (#) Ratio different from 0.5 (one-sample Wilcoxon signed rank test).

**Figure 5.** Elevated zero maze test (5 min) in 16 weeks old C57BL/6J and *AppNL-G-F* mice; 11 weeks on experimental diet; data presented as median ± IQR; outliers beyond threefold IQR removed; *p* < 0.05 (Mann-Whitney-U-test) considered statistically significant (\*). (**A**) Total distance moved. (**B**) Percentage of time spent in open corridors.

#### *3.4. Barnes Maze*

To investigate spatial memory and learning, the Barnes maze test was implemented in this study. In the first part of the test, the acquisition phase, the mice had to learn and remember the location of the escape box at the target hole. Figure 6A shows the latencies the mice needed to reach the target hole on subsequent days of training in the acquisition phase. The graph indicates a learning curve in every group. Tests on statistical significance were carried out for day 4 and revealed no differences at this stage of the test. In the probe trial on day 5 (Figure 6B), the reference memory of the previously learned target hole was tested. At this time, female *AppNL-G-F* controls needed significantly longer to reach the target hole compared to the C57BL/6J WT control animals (*p* = 0.016). Vitamin B deficiency and corresponding hyperhomocysteinemia did not result in a worse performance at any stage of the Barnes maze test.

**Figure 6.** Barnes maze test in 17–18 weeks old C57BL/6J and *AppNL-G-F* mice; 12–13 weeks on experimental diet; data presented as median ± IQR; outliers beyond threefold IQR removed; *p* < 0.05 (Mann-Whitney-U-test) considered statistically significant (\*). (**A**) Latency to target hole; training days 1–4; 180 s per trial; acquisition phase; test on statistical significance carried out for day 4. (**B**) Latency to target hole; day 5; 90 s per trial; probe trial.

#### *3.5. Social Interaction Test*

Testing social behavior proceeded in two subsequent phases. At first, we assessed sociability, describing the curiosity of the animals towards the stimulus mouse in the testing system (Equation (2)) (Figure 7A).

$$\text{A ratio socialibility} = \text{(time social edge)} \text{((time social edge + time empty cage))} \tag{2}$$

No statistically significant difference was observed between C57BL/6J WT and *AppNL-G-F* control animals. Experimental diets also had no impact on the social ability of the mice. Medians were statistically unequal to 0.5 except for group 2, 4 and 6 (males) and group 2 and 3 (females). A ratio of 0.5 means that contact times with the conspecific stimulus mouse and the empty cage were equal.

In the second phase of the test, we assessed the social recognition performance of the animals (Equation (3)) (Figure 7B).

ratio social recognition = (time novel animal)/((time novel animal + time familiar animal)) (3)

As for sociability, neither genotype nor experimental diets had an influence on social recognition in the different experimental groups. In neither phase of the test did hyperhomocysteinemia aggravate the cognitive performance of the mice. Except for group 1 (males) and group 5 and 6 (females), medians of the other groups did not differ significantly from 0.5.

**Figure 7.** Social interaction test (25 min) in 19–20 weeks old C57BL/6J and *AppNL-G-F* mice; 14–15 weeks on experimental diet; data presented as median ± IQR; outliers beyond threefold IQR removed; *p* < 0.05 (Mann-Whitney-U-test) considered statistically significant (\*); (#) ratio different from 0.5 (one-sample Wilcoxon signed rank test). (**A**) Sociability expressed as the ratio between the contact time with the stimulus mice and the sum of the contact times with the stimulus mice and the empty cage. (**B**) Social recognition expressed as the ratio between the contact time with novel stimulus mice and the familiar ones.

#### *3.6. Paired Associates Learning (PAL) Task*

The touchscreen PAL was used to assess potential cognitive impairment of the male mice (about five to eight months of age). Both the session duration and the number of trials completed per session, as well as the percentage of correct trials per session were analyzed (Figure 8). The resulting learning curves revealed no statistically significant difference in these parameters between C57BL/6J WT mice and *AppNL-G-F* KI mice in the final phase of the test (block 6). Hyperhomocysteinemic *AppNL-G-F* mice did not perform worse than *AppNL-G-F* control mice. Other experimental diets also had no benefit on the cognitive abilities of *AppNL-G-F* mice at this age. The C57BL/6J WT group showed a smaller variability in the touchscreen chambers in comparison to the *AppNL-G-F* KI groups. This effect was particularly observed in the parameter trials completed (Figure 8B). Vitamin B deficient animals showed a tendency to perform better at the beginning of the test (trials completed, block 1) and thus did not display a learning curve like that of *AppNL-G-F* control mice. However, no effects reached statistical significance in block 6. Animals fed with a vitamin B and PUFA combination diet did not reach the maximum number of trials per session. Therefore, the session duration scarcely also decreased over time in this group. The proportion of correct and incorrect trials was not affected.

**Figure 8.** Touchscreen paired associates learning test (PAL) in 21–34 (incl. training phase) weeks old C57BL/6J and *AppNL-G-F* mice; 16–29 weeks on experimental diet; only males; data summarized in blocks of 6 sessions and presented as median ± IQR; outliers beyond threefold IQR removed; test on statistical significance carried out for block 6. (**A**) Session duration: time (s) needed to complete 36 trials per session; maximum 3600 s. (**B**) Amount of trials completed per session; maximum 36 trials. (**C**) Percentage of correct trials per session.

#### *3.7. Place Learning (PL) and Reversal Learning (RL) Task*

Learning and memory performance of the females at the age of about six to eight months was finally tested using two tasks in the IntelliCage system. We detected the visits of the mice to the drinking corners and analyzed the percentage of correct visits during the drinking sessions in the place learning (PL) and the reversal learning (RL) tasks. Three points in time along the course of the tasks are illustrated in Figure 9. Statistical analysis of the late phase of this course in both (Figure 9A) PL and (Figure 9B) RL (session 31; resp. 23) revealed no significant differences between *AppNL-G-F* and age-matched C57BL/6J mice. In comparison to the *AppNL-G-F* KI control group, none of the groups fed with experimental diets showed improved or impaired memory abilities.

**Figure 9.** IntelliCage place learning (PL) and reversal learning (RL) task in 27–34 (incl. habituation period) weeks old C57BL/6J and *AppNL-G-F* mice; 22–29 weeks on experimental diet; only females; data are shown for three points in time along the course of the task and presented as median ± IQR; outliers beyond threefold IQR removed; test on statistical significance carried out for the final point in time. (**A**) Percentage of correct visits during drinking sessions in the PL. (**B**) Percentage of correct visits during drinking sessions in the RL.

#### *3.8. Immunohistochemical Analysis*

Brain sections of all animals were immunohistochemically stained and analyzed in order to semi-quantify the amount of amyloid plaques. For this purpose, we assessed the area (percentage) occupied by plaques in images of several regions of interest (ROI). The positions of the different cortical and hippocampal ROI (Table A2) are marked in Figure 10.

Figure 10 illustrates examples of brain sections of a C57BL/6J WT mouse and an *AppNL-G-F* KI mouse. Aβ plaques, indicated by characteristic brown staining, occurred abundantly and diffusely in the brain sections of the KI animals (Figure 10B), whereas WT mice did not show any signs of Aβ deposition at all (Figure 10A). The differences in the Aβ burden between the C57BL/6J and *AppNL-G-F* genotype, as well as a potential impact of the experimental diets, were further analyzed using ImageJ software. Semi-quantification of the Aβ burden confirmed a significant difference between WT and KI control groups (Figure 11) in all ROI (*p* < 0.001; *p* = 0.002; *p* < 0.001; *p* < 0.001; *p* < 0.001; *p* < 0.001; *p* < 0.001; *p* < 0.001).

There was no statistically significant difference in the plaque area between the diet groups and the *AppNL-G-F* control group in the single ROI and in total. However, the immunohistochemical results indicate prominent plaque formation in all *AppNL-G-F* groups at about 8 months of age.

**Figure 10.** Immunohistochemically stained sections of mouse brains; regions of interest (ROI) in cortical and hippocampal areas are marked in whole brain images (100× magnification) and depicted separately for further semi-quantification of amyloid plaques; 35 weeks of age, 30 weeks on experimental diet. (**A**) Exemplary section of a C57BL/6J wild type animal. (**B**) Exemplary section of an *AppNL-G-F* knock-in animal.

**Figure 11.** Semi-quantitative analysis of amyloid-β (Aβ) in immunohistochemically stained brain sections; results are shown for single regions of interest (ROI) and in total; 35 weeks of age, 30 weeks on experimental diet (males and females pooled); data presented as median ± IQR; outliers beyond threefold IQR removed; *p* < 0.05 (Mann-Whitney-U-test) considered statistically significant (\*).

## **4. Discussion**

The current preclinical study investigated the impact of an induced hyperhomocysteinemia in the *AppNL-G-F* knock-in mouse model for AD, as well as potentially preventive benefits of di fferent micro-nutritional interventions. In order to characterize the phenotypes of the mice, we conducted a versatile behavioral test battery, accompanied by an analysis of HCys-/HCA levels and of the Aβ plaque burden. However, despite successful induction of prominent cerebral plaque deposition and hyperhomocysteinemia, merely subtle impairments were observed in the *AppNL-G-F* mice.

C57BL/6J mice, a frequently studied mouse strain and background strain of the *AppNL-G-F* knock-in (KI) model, served as an age-matched wild type (WT) control group in this study. Hence, results in these mice indicated a reference behavior and enabled subsequent assessment of the *AppNL-G-F* genotype in the KI mice. In the open field, we focused on the intrasession habituation of the mice, which is one form of learning. Intrasession habituation describes a decreasing level of exploration of a new environment over time in a single session which can typically be detected in C57BL/6J mice [37]. This is in accordance with our finding that the habituation ratio in C57BL/6J was significantly lower than 0.5 and therefore indicated intrasession habituation. As expected, C57BL/6J mice demonstrated spatial learning and memory ability on consecutive days of training in the Barnes maze [38]. In a test for sociability and social recognition [39], C57BL/6J mice preferred to spent time with a conspecific (ratio sociability > 0.5) [40]. However, they did not prefer the novel conspecific in the second part of the test (ratio social recognition = 0.5). In the touchscreen PAL [41], male WT animals completed the maximum number of all 36 trials per session, accompanied by decreasing session duration. The increase in the percentage of correct trials is in accordance with observations in a similar study [42]. In the IntelliCage setup [43,44], learning curves indicated a constant learning e ffect in the female WT animals.

For several reasons, we decided to use an AβPP-based KI mouse model for AD in this study. Firstly, the novel KI models provide the advantage of not overexpressing AβPP in comparison to the more established transgenic models. Consequently, artificial phenotypes due to an overproduction of AβPP fragments besides the Aβ peptide should be avoided [9]. Secondly, an increased anabolism of Aβ levels is primarily a hallmark of hereditary- or early-onset AD [3]. Hyperhomocysteinemia, which is especially prominent in older people [45], is supposed to be a risk factor for AD [24]. Therefore, elevated HCys and HCA have been regarded as a hallmark of sporadic- or late-onset AD. The late-onset form a ffects the vast majority of AD patients [3]. By combining both the increased Aβ anabolism as a feature of hereditary AD and the detrimental e ffects of excess HCys as a feature of sporadic AD, we attempted to simulate cognitive decline more comprehensively. Thirdly, in order to investigate preventive treatments, it is mandatory to use a model displaying subtle phenotypes corresponding to a very early stage of the disease. According to a review by Zahs and Ashe, AβPP-based mouse models simulate the early phase of AD and thus are adequate for preventive interventions [46]. In the current study, a very subtle phenotype, i.e., very mild cognitive deficits, was observed. For each analysis, we compared C57BL/6J WT animals with *AppNL-G-F* KI control animals. Both groups received the same control diet. KI mice displayed an impaired habituation behavior in the open field. Male mice of the two control groups habituated equally to the new environment, whereas females di ffered significantly. Data from the probe trial in the Barnes maze confirmed this finding: *AppNL-G-F* KI mice needed longer to locate the former target hole than WT mice. As in the open field, this e ffect reached statistical significance only in females. Previous clinical studies sugges<sup>t</sup> that a reduced cognitive reserve in women might explain the female vulnerability to develop a more severe phenotype of AD, a disorder affecting more women than men [47,48]. Other behavioral tests did not reveal di fferences caused by the KI genotype. However, data from the PAL test indicated an increased variance of results (higher IQR) in the *AppNL-G-F* versus WT mice. WT animals showed a clearer performance curve with regard to the session duration and the number of trials completed along the course of the test, meaning that WT mice did not need as long as the KI mice to fulfil the 36 trials in a 1-h session. This enhanced e fficiency might be the result of a higher motivation of the WT animals. Nevertheless, e ffects at the final stage of the test (block 6) did not indicate a significant impact of the genotype.

Other groups reported similar findings in *AppNL-G-F* mice, indicating a very subtle phenotype. Two recent publications summarized these findings in tabular overviews, considering also sex and age of the mice of the included studies [49,50]. Latif-Hernandez and colleagues showed that the behavior of *AppNL-G-F* mice was largely una ffected at the age of 3–10 months [51]. Similarities with our study can also be found in a publication by Whyte et al., who observed no di fferences between C57BL/6J and *AppNL-G-F* mice in di fferent cognitive tests at the age of 6 months [52]. Sakakibara and colleagues tested *AppNL-G-F* mice at a higher age (15–18 months) and reported an intact learning ability but also recommended *AppNL-G-F* as an AD model for preventive studies [53]. One year later, Jacob et al. observed neither consequences on cognitive performance in a touchscreen task nor age-dependent changes in a phase-amplitude coupling analysis, which was used as a measure of neurophysiological functioning, in 4.5 month old *AppNL-G-F* mice. In accordance with our findings in the *AppNL-G-F* model, these mice displayed a higher variability than WT control mice [42].

The question remains whether the KI mice were too young to display clear impairments. Further investigations are required to test the combination of the *AppNL-G-F* genotype with our experimental diets in older mice. However, other groups detected significant cognitive deficits in the *AppNL-G-F* model [9,49,50,54]. As summarized elsewhere [49], the majority of studies in the field investigated only male animals. Hence, a 1:1 comparison of these studies with our results comprising both sexes is di fficult. Furthermore, a review of the topic described a relatively high level of variability in AβPP KI models between di fferent laboratories [55]. Staining results of *AppNL-G-F* brain sections showed prominent plaque deposition throughout the brain, as previously reported in similar studies [49,52], and thus indicate amyloid pathology as a central hallmark of early AD.

In order to investigate potentially detrimental e ffects of elevated HCys and HCA levels, one group of *AppNL-G-F* mice received a special diet deficient in vitamin B6, B12 and folate. The resulting hyperhomocysteinemic state was confirmed in serum and urine prior to the start of behavioral tests. Our behavioral testing data obtained in the social interaction test, PAL and in the IntelliCages revealed no deficits in hyperhomocysteinemic mice and therefore do not support previous findings (e.g., [56]). The open field test and Barnes maze indicated subtle deficits in habituation behavior and spatial learning and memory, but these e ffects did not reach statistical significance. Only the elevated zero maze revealed an increased anxiety in hyperhomocysteinemic females. This observation might be of translational relevance, because anxious behavior is also one aspect of the AD phenotype [57]. Various preclinical studies in the field indicate a significant impact of hyperhomocysteinemia on plaque burden [58,59]. Other groups reported no such e ffects, which is in accordance with our immunohistochemical results in the *AppNL-G-F* model [60,61]. In conclusion, despite severely elevated levels of HCys and HCA over a longer period of their life span, *AppNL-G-F* mice showed neither a modified plaque burden nor significant cognitive deficits due to hyperhomocysteinemia. A majority of preclinical data published in the field indicate behavioral deficits in animal models caused by increased HCys (e.g., [56,59,62]). However, we assume that the evidence might be biased to some extent. On the one hand, behavioral data obtained in transgenic models based on massive AβPP overexpression might be somewhat artificial because of an overproduction of other AβPP fragments aside from Aβ [9]. It should also be considered that negative results are often not published, although equally important as positive results. The publication bias, meaning the reduced publishing of negative or null results, is not restricted to the field of AD research, but is rather a general problem [63].

Hyperhomocysteinemia is referred to as a hallmark of AD [10], but its impact on the disease is still under discussion. From a translational point of view, this experimental group simulates the portion of elderly people who are deficient in B-vitamins [64]. Preclinical evidence [65] and clinical evidence [45] confirm an age-related elevation of HCys levels. An impaired vitamin status is one reason amongs<sup>t</sup> others for hyperhomocysteinemia in the elderly [66]. In the present study, the lack of vitamin B6, B12 and folate in combination with 1% sulfathiazole sodium to inhibit bacterial folate synthesis in the gu<sup>t</sup> [58], led to a "severe" hyperhomocysteinemic state, according to a classification used in other publications [67]. Consequently, our vitamin B deficient mice displayed high HCys serum

concentrations (45,760 ng/mL ≈ 339 μmol/L) in comparison to our *AppNL-G-F* KI control (1054 ng/mL ≈ 8 μmol/L) and in comparison to elevated HCys levels in similar studies (e.g., [56,62,68]). Fuso and colleagues also reached high plasma total HCys (>400 μmol/L ≈ 54,000 ng/mL) in their study with TgCRND8 mice and explained the relatively high levels by not fasting the mice before sacrifice and by inhibiting both the re-methylation and the transsulfuration pathway [58].

Vitamin B deficient chow resulted in ~50 fold higher serum and urinary HCys and ~10–20 fold higher serum and urinary HCA compared to animals fed with control diet for 8 weeks. About 0.1% of HCys molecules were oxidized to HCA in serum (42.9 ng/mL ≈ 0.23 μmol/L) and excreted in urine (1184 ng) in 24 h. Only free HCys can be oxidized to HCA, which is suggested to be the main neurotoxic species [32–34]. In the current study, we did not measure the free form but the levels of total HCys by adding a reduction step (TCEP-solution) in the analytical method. In vivo, most HCys molecules are protein-bound or dimerized; only about 1% are available in the free thiol form [12]. Hasegawa et al. reported cognitive impairment in transgenic 3xTg-AD mice, triggered by elevated HCA in the brain [69].

We also investigated other experimental diets besides the vitamin B deficient chow discussed above. Group 4 received a vitamin B enriched diet containing a particular high content of folate, B6 and B12 compared to both the control diet and the FC-like diet. The goal of this diet was to investigate whether an additional increase, specifically of B-vitamins, in comparison to the FC-like diet could provide further benefits in the outcome of the study. Therefore, the difference in B-vitamin contents should simulate a potentially different effectiveness between FC (Souvenaid®) and existing higher dosed vitamin B preparations as human treatment options. In accordance with a recent international consensus statement [23], PUFAs (DHA + EPA) have been suggested to be beneficial for cognitive functioning in general and might be additionally linked to AD pathology [25,70]. Because single nutrient intervention studies often failed to show beneficial effects on cognitive function, it has been suggested that it might be important to investigate combinatory approaches [35]. For this purpose, we combined the high content vitamin B enrichment with the supplementation of PUFAs (group 6). Finally, group 7 received the FC-like diet, a complex mixture of ingredients (Table A1), which we implemented due to positive previous findings (e.g., [35,71]).

Supplementation of B-vitamins and PUFAs, as well as combinatory approaches and the FC-like mixture, were capable of lowering HCys and HCA below the levels of the *AppNL-G-F* control mice fed with a standard rodent chow. However, by taking both sampling points (8 and 30 weeks on diet) as well as behavioral testing data into consideration, results appear inconsistent. In the open field, anxiety-related behavior did not differ between groups fed with B-vitamins, PUFAs or a mixture and *AppNL-G-F* control animals. However, the elevated zero maze revealed increased anxiety in males fed with the combination diets. Especially the mice supplemented with both B-vitamins and PUFAs were more anxious and stayed in the closed corridors of the zero maze, but it has to be emphasized that these mice also displayed a reduced locomotion activity during the test. In the Barnes maze, experimental chow did not affect latencies to target at day 4 of training. Other researchers too did not observe benefits of PUFA-supplementation in cognitive tasks [72]. We confirmed the lack of dietary effects on cognitive performance in the social interaction test, the IntelliCage and PAL. Although not significant in the final block of 6 sessions, the session duration and trials completed indicate a worse learning curve for group 6 (B+PUFA-ENR) in the PAL test. This might be due to a lack of motivation in these mice receiving a high number of vitamins and PUFAs, which possibly lowered their affinity to the milk reward in the PAL task. One reason could be that the food restriction was not strict enough for this group. The FC-like diet did not prove beneficial in any test in comparison to the control chow. This is in accordance with some clinical studies, which do not support the benefit of the FC diet and thus indicate equivocal evidence [73,74]. In conclusion, the beneficial tendencies we observed did not mostly reach statistical significance in behavioral tests and biochemical-/immunohistochemical analyses and consequently do not sugges<sup>t</sup> a clear beneficial effect of B-vitamins or PUFAs in this mouse model at the investigated age and diet duration. It is important to question here whether it is

possible to observe amelioration through dietary intervention when merely a subtle behavioral deficit is induced in the KI mouse model.

Overall, this mouse model, simulating amyloid pathology without AβPP overexpression, merely displays a very mild phenotype despite massive cerebral Aβ deposition at the age of 35 weeks. The amyloid hypothesis has been questioned frequently because of the disappointing track record in clinical trials of drugs that target Aβ despite decades of extensive research in the field [7,75]. In addition, in some cases, substantial plaque deposition does not even cause dementia-like symptoms [76]. However, the window for potentially preventive measures is limited to an early stage of AD, where cerebral amyloidosis remains the central hallmark of the pathology [3]. Despite all criticism of the amyloid hypothesis, beneficial effects were recently observed using the human anti-Aβ monoclonal antibody aducanumab [77], confirming a causal role of Aβ in AD pathogenesis.
