Alzheimer’s disease (AD) is the most common form of dementia (Kawas and Corrada, 2006), resulting in a progressive decline of cognitive functions (Blennow et al., 2006). While the cellular and molecular mechanisms that underlie AD are still not well understood, some neuropathological hallmarks were already identified. These include neurofibrillary tangles consisting of aggregated hyperphosphorylated tau protein and neuritic plaques characterized by an overload of mainly insoluble Aβ peptides (e.g., Hardy and Higgins, 1992). Mutations in the amyloid precursor protein (APP) and the presenilin genes (PS1 and PS2), which are both linked to the familiar form of AD (e.g., Goate et al., 1991), have been described to lead to a massive increase in cytotoxic Aβ₄₀ and Aβ₄₂ oligomers (e.g., Walker et al., 2005; Theuns et al., 2006).

One of the main objectives of Alzheimer research is to identify novel therapeutic approaches alleviating cognitive deficits. In this context, focusing on neurotrophin signaling appears to be a promising approach. Brain-derived neurotrophic factor (BDNF) is known to play a critical role in the survival, differentiation and synaptic plasticity of mammalian neurons (see e.g., Cowansage et al., 2010; Edelmann et al., 2014). The cellular actions of BDNF are mediated mainly through TrkB-receptors (summarized e.g., in Park and Poo, 2013). In the adult mammalian brain BDNF is most prominently expressed in the entorhinal cortex and in the hippocampus (Phillips et al., 1990; Ferrer et al., 1999; Quartu et al., 1999), brain areas that are also affected by neuronal loss in AD (Peng et al., 2005). Furthermore, AD patients have been shown to exhibit decreased BDNF levels in these regions (Phillips et al., 1991; Murray et al., 1994; Amoureux et al., 1997), which is mirrored by reduced blood serum BDNF levels in AD patients (Laske et al., 2006). In vitro experiments demonstrated that BDNF exerts several neuroprotective effects by reducing the cytotoxic effects of Aβ₄₂ (Arancibia et al., 2008) and by stimulating the non-amyloidogenic pathway, resulting in a reduction of toxic Aβ species (Scheuner et al., 1996; Fu et al., 2002; Nishitomi et al., 2006; Thornton et al., 2006; Rohe et al., 2009). In rodent and primate models of AD it has been shown that acute application of BDNF protein can partially rescue Aβ₄₂-induced neurodegenerative changes (Nagahara et al., 2009). Furthermore, studies in AD mouse models revealed that activation of the TrkB-receptor via application of the recently discovered TrkB-receptor agonist 7,8-Dihydroxyflavone rescued learning and memory deficits and reduced Aβ levels in the brain (Devi and Ohno, 2012; Castello et al., 2014; Zhang et al., 2014).

Taken together, several lines of evidence suggest that direct application of BDNF or activation of TrkB-receptors exert a protective effect against AD pathology and that the expression of BDNF seems to be down-regulated in AD patients. However, the mechanisms explaining how changes in BDNF protein levels might contribute to the occurrence and severity of AD pathology are still unknown. To further investigate how altered BDNF levels might contribute to the progression of AD pathology, we created a novel mouse model that combines AD pathology with chronically reduced BDNF levels. We crossed an APP/PS1 mouse model (Radde et al., 2006) with a heterozygous BDNF knock out (BDNF^+/−) mouse (Korte et al., 1995), in which BDNF protein levels are reduced to roughly 50% of the BDNF levels observed in wildtype animals (see e.g., Psotta et al., 2013). We hypothesized that, if endogenous BDNF has a protective impact against AD pathology, AD-like symptoms should occur in more severity and/or occur earlier in live history in this novel mouse model compared to the classical APP/PS1-mouse line. Therefore, we tested adult (5 months) and aged (12 months) animals for their cognitive abilities in the novel object recognition (NOR) task, the two-way active avoidance paradigm and the Morris water maze (MWM). Also, open field (OF) and elevated plus maze (EPM) tests were conducted, to test for possible differences in explorative activity and anxiety levels. For a more detailed analyis of the neuropathological characteristics in the investigated mouse lines, amyloid plaque load in the hippocampus and prefrontal cortex were analyzed by quantitative immunhistochemistry.

Since there is evidence for an association between altered BDNF signaling and the occurrence and/or intensity of AD pathology, the aim of the present study was to combine these two factors in a novel animal model to study the impact of chronically reduced endogenous BDNF levels on the development of AD pathology in vivo. By using this novel mouse model we could show, that the reduction of endogenous BDNF within the APP/PS1-mouse model causes an accelerated onset of learning impairments in a two-way active avoidance task but not in spatial learning in the MWM. Also short term memory for object recognition remained unaffected in APP/PS1-BDNF^+/−-mice. Quantitative analysis of plaque pathology in the two AD mouse lines revealed no differences between the APP/PS1- and APP/PS1-BDNF^+/−-mice and also no age-dependent increase between 5 and 12 months of age. However, for both AD mouse lines we found that amyloid plaque density was significantly higher in the mPFC compared to the hippocampus.

Our working hypothesis claimed that BDNF deficiency should accelerate the onset of cognitive impairment and amplify the cognitive deficits displayed by the APP/PS1-mice. As predicted, APP/PS1-BDNF^+/−-mice showed an earlier onset of impairments in active avoidance learning compared to APP/PS1- and BDNF^+/− -mice. At five months of age, avoidance learning was only impaired in APP/PS1-BDNF^+/−-mice and not yet detectable in APP/PS1- or BDNF^+/− -mice, whereas at twelve months of age APP/PS1- and APP/PS1-BDNF^+/−-mice displayed significant learning deficits compared to wildtype animals. Previous studies testing APP/PS1-mice reported similar age-dependent deficits in avoidance learning in 9 months old mice and in eye blink conditioning at the age of twelve months, which were not yet detectable in three month old animals (Gruart et al., 2008; Lok et al., 2013). Interestingly, no study has tested BDNF^+/−-mice in the TWAA paradigm so far, even though this task is known to induce BDNF protein expression in several brain areas (Ulloor and Datta, 2005). Even though 12 months old BDNF^+/−-mice did not perform significantly worse than their wildtype littermates, our data still show a strong trend to an impaired learning performance in these animals at this age. This observation of a probable age-dependent learning decline would be in line with other studies reporting age-dependent learning deficits for BDNF^+/−-mice in different learning paradigms. For example, a deficit in fear extinction learning was observed in 7 but not 2 months old BDNF^+/−-mice (Psotta et al., 2013) and BDNF^+/−-mice exhibit a deficit in amygdala-dependent fear learning, when the animals are older than 3 months of age (Endres and Lessmann, 2012). Together these findings suggest that different learning paradigms, depending on distinct brain areas, all suffer from BDNF deficiency only at later ages.

Importantly, the observed age-dependent decline in the BDNF^+/−-mice seems to appear later and to a weaker extent than in the APP/PS1- or APP/PS1-BDNF^+/−-mice, suggesting that the observed accelerated onset of impaired TWAA learning in the young triple transgenic animals is rather due to an accelerated AD-like pathology in combination with chronically reduced BDNF protein levels than simply due to the chronic BDNF reduction alone.

Importantly, for the investigated basal behavioral parameters we didn’t observe any changes of explorative activity and no obvious motoric deficits in the APP/PS1-BDNF^+/−-mice. In contrast, our EPM data indicate a low anxiety phenotype in the triple mutants. However, it has to be pointed out that this finding was specific for the EPM, no signs of enhanced anxiety-like behavior could be observed in the OF. Taken together, it appears unlikely that the observed learning deficits are due to alterations in basal explorative activity of motor deficts but we cannot completely rule out that the reduced anxiety levels may contribute to the observed learning impairments in the TWAA paradigm. However, this appears unlikely since other studies have shown that low anxiety rats as defined by an enhanced number of entries into the open arms of an EPM performed better in a two-way active avoidance task than high anxiety rats, that showed reduced entries into the open arms of an EPM (Ho et al., 2002).

In contrast to the findings in our new triple mutant mouse model, we found reduced locomotor activity in the 12 months old APP/PS1-mice in the OF as well as in the EPM tests. According to these results one could argue that the BDNF deletion in APP/PS1 mice restored to normal levels of locomotor or exploratory behavior in the APP/PS1-BDNF^+/−-mice.

However, in the NOR task we didn’t observe any signs of reduced motor activity and explorative behavior in the APP/PS1-mice compared to the other tested animals. Also, exploratory and motor behavior appeared to be normal during MWM, since swimming speed wasn’t reduced in APP/PS1-mice and the latency to find the platform was comparable to the wildtype animals throughout the whole training trials, especially during the first trials that reflect more general locomotor than learning abilities (Figure 3). Thus, our findings, which are supported by previous observations (Radde et al., 2006; Lo et al., 2013a), suggest that there is no general deficit in motor and explorative behavior in APP/PS1-mice.

Also in the MWM paradigm, we observed age-dependent learning deficits. However, in contrast to TWAA learning these learning deficits occurred to a different extent in all tested transgenic animals. While training performance (acquisition) was fully intact in 5 months old animals, the APP/PS1- as well as the APP/PS1-BDNF^+/−-mice showed significantly reduced training performances at the age of 12 months, thereby the APP/PS1-BDNF^+/−-mice displayed the worst performance since they showed the highest latencies during the training phase at that age.

For the spatial memory tested in the probe trial in 5 months old animals a reduced learning performance was observed in APP/PS1-mice but not in the other genotypes. However, this finding in the APP/PS1-mice does not reflect a completely impaired memory, since these animals still exhibited a strong tendency towards spending more time in the previous target quadrant (p = 0.08). At 12 months of age all tested mutant mouse lines, the BDNF^+/−-mice, the APP/PS1-mice as well as the triple mutant APP/PS1-BDNF^+/− showed an impaired performance in the probe trial when compared to the wildtype animals. But again, there was no complete memory blockade since all animals still showed a trend towards spending more time in the previous target quadrant.

In general, these findings are in line with other studies demonstrating age-dependent spatial learning deficits in APP/PS1-mice occurring around 8/9 months (Radde et al., 2006; Lo et al., 2013a,b), as well as in BDNF^+/−-mice, occurring around 10–12 months (Linnarsson et al., 1997; Rantamäki et al., 2013). Interestingly, another study using a similar mouse model reported an aggravated AD pathology in the MWM by chronic BDNF reduction (Rantamäki et al., 2013). However, a direct comparison between the two mouse models is difficult since we used a different AD mouse model harboring different APP and PS1 mutations under control of a different promoter, resulting probably in different expression patterns of Aβ and subsequent AD-like pathology.

The quantification of amyloid plaque density in the two AD mouse models revealed no differences between the APP/PS1-mouse and our newly created APP/PS1-BDNF^+/−-mouse and also no age-dependent differences (5 vs. 12 months). This finding indicates that the amyloid plaque pathology in our new mouse model was not directly affected by the reduced BDNF level, a result that is in line with findings described in the study of Rantamäki et al. (2013).

However, our findings revealed for both analyzed AD mouse models that the amyloid plaque load was higher in the mPFC compared to the hippocampus. These results are comparable to findings of previous studies showing that the classical APP/PS1-mice exhibits a higher (approx. three fold) plaque pathology in the cortex compared to the hippocampus (Radde et al., 2006; Scholtzova et al., 2008; Gengler et al., 2010). In addition it was shown that the onset of plaque pathology occurs later in the hippocampus than in the cortex (Radde et al., 2006). These regional differences in amyloid pathology might explain our finding of an aggravated learning deficit in the TWAA task but not in spatial learning, since these different learning tasks rely on the recruitment of different neuronal networks. Whereas the spatial MWM learning relies mainly on the basal ganglia and hippocampal function (Morris et al., 1982; Cho et al., 1999; Devan and White, 1999; Barry and Commins, 2011), the more complex TWAA learning requires besides several limbic structures (Cain and LeDoux, 2008; Choi et al., 2010) also network activity of higher cortical areas (Martinez et al., 2013; Moscarello and LeDoux, 2013).

Finally, major dysfunctions for short-term memory abilities were neither observed in our novel triple mutant, nor in the two respective single transgenic mouse lines (i.e., BDNF^+/− and APP/PS1), regardless of the age of the animals. This finding is in line with other observations indicating that BDNF signaling is required for long-term but not short-term object memory (Seoane et al., 2011; Callaghan and Kelly, 2012). However, this speculation has to be assessed in future experimental approaches applying specific long-term and short-term memory tasks.

In conclusion, we could demonstrate that according to our hypothesis the chronic reduction of BDNF levels in our novel AD-mouse model indeed accelerated the onset of learning deficits in two way active avoidance learning. However, we could identify this acceleration only for TWAA learning but not for spatial learning in the MWM and for short-term memory in an object recognition task in the analyzed age groups. Overall, the creation of the triple transgenic mouse model was a further step to study the interrelation of BDNF and Aβ pathology in vivo. As the effects of chronic BDNF reduction vary between learning tasks requiring different complex brain networks this novel mouse model provides an interesting tool to further understand how differently regulated Aβ and BDNF expression relate to cognitive declines in AD pathology. Furthermore, this mouse model might be a promising model system to test BDNF-related treatment strategies against AD pathology. Thus, the APP/PS1-BDNF^+/−-mouse model created in this study provides new opportunities to investigate the underlying molecular and cellular mechanisms of AD in more detail.

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.