To suppress neurogenesis in adult mice we established a new pharmacological approach based on four cycles of treatment with the DNA-alkylating agent temozolomide (TMZ; Figure 1). TMZ shows an excellent blood brain barrier passage and has a very advantageous side effect profile [18]–[20]. In a dosage finding experiment we found a clear dose-dependency after monocyclic treatment. Using 25 mg TMZ per kg body weight for three single daily injections was found to reduce proliferation (number of BrdU-positive cells) in the dentate gyrus by more than 80% (277±23 vs. 1589±134 cells in vehicle controls, t-test: t(9) = 3.14, p<0.05). Because doubling that dose did not result in a significant higher reduction of BrdU-positive cells TMZ was applied at a dose of 25 mg/kg for all following experiments. After four treatment cycles precursor cell proliferation was decreased by >90% (1347±178 BrdU⁻positive cells in controls vs. 84±93 for TMZ treated mice, resulting in a substantial reduction of recruitable newborn granule cells (t-test: t(9) = 2.94, p<0.001).

In contrast, TMZ-induced myelosuppression had fully recovered after 4 weeks, when the animals were tested in the water maze (Figure 2A, B). Increase of body weight, locomotor skills and motor activity were all normal (data not shown and Figure 2C). No differences in exploratory behavior were found between groups (Figure 2D). There was neither morphological activation of microglia nor an increase in microglia numbers (Figure 2E, F). The 4-week interval thus ensured that at the time of behavioral testing only a minimum of potentially recruitable neurons would be present [21] but all potentially relevant transient side effects had ceased.

To assess hippocampal function at the systems level we investigated LTP in the DG and CA1. In the DG, LTP is physiologically inhibited by GABAergic interneurons. Still lacking that inhibition the immature adult-born neurons, in contrast to older granule cells, show a reduced threshold for LTP induction [22], [23]. Consequently, unless the GABAergic inhibition of granule cells is blocked, stimulating the medial perforant path with high-frequency bursts elicits a weak LTP in the DG that is based on the newly generated neurons [2], [24]. In TMZ-treated mice this characteristic weak LTP was not detectable (Figure 3A, C, D, F(1,198) = 252.8, p<0.05, ANOVA). After application of GABA receptor antagonist bicuculline, however, strong LTP could be elicited in both groups, indicating physiological synaptic plasticity in mature neurons (Figure 3B). LTP in CA1 was unaffected by TMZ treatment (Figure 3E). This implied that after TMZ treatment, only LTP in the DG was influenced and only in a manner consistent with the reduced level of ongoing neurogenesis.

We trained mice on a reference memory version of the MWM task with 6 trials per day for 5 days and a new platform position (reversal) on the beginning of day 4 [13]. Although TMZ-treated mice performed significantly worse than controls for the entire experiment (latency, F(1,590) = 34.95, p<0.001, repeated measures ANOVA), both TMZ-treated mice and controls (CTR) successfully learned to navigate to the hidden platform during the first three days (Figure 4A). At the end of this period there were no significant differences between the groups in path length and latency time to reach the hidden platform. Although TMZ-treated animals and controls generally succeeded in learning the MWM, significant differences appeared on day two of the acquisition phase regarding both latency and path length (Figure 4A, t(9) = 3.42 p<0.001 for latency and t(9) = 2.77, p<0.01 for path length, Student's t-test).

Swim speed of both groups did not differ throughout the experiment (data not shown). In the 1st trial of day 4 (1st probe trial), when the platform had been moved to the opposite quadrant, mice from both groups spent significantly more time in the previous goal quadrant than in the other quadrants, indicative of successful spatial learning (t(9), p<0.05 for comparison of the NE quadrant with each other quadrant and for each group, Student's t-test, Figure 4D, E).

After reversal, further significant differences were found. As apparent in the heat maps visualizing the probability for the mice to be present at a given location, TMZ-treated mice took significantly longer to develop a clear preference for the first goal and persevered swimming to that platform position even after two days of training following reversal indicating a lack of cognitive flexibility to cope with the altered challenge (Figure 4B, C). A second probe trial at the end of the reversal phase showed that, whereas controls had successfully learned to navigate reliably and precise to the new platform position, TMZ-treated mice largely failed to adapt to the changed situation and to develop an appropriate new spatial preference (Figure 4F, G).

To assess the qualitative aspects of learning the water maze task we analyzed the respective search strategies displayed by the mice to locate the hidden platform (Figure 5). Parameter-based classification of search patterns turned out to be reliable for both groups indicating that treated animals used identical basic principles to implement the full range of strategies (Figure 6A).

Before and after reversal we found a clear progression from thigmotaxis to direct swimming for both groups (Figure 6B; thigmotaxis F(4,72) = 6.37, p<0.001, random search F(4,72) = 11.17, p<0.001, scanning F(4,72) = 6.14, p<0.001, directed search F(4,68) = 3.95, p<0.001, focal search F(4,72) = 8.63, p<0.001, direct swimming F(4,72) = 7.8, p<0.001 ANOVA, inner-subject test). However, treated animals and controls differed significantly in using the respective strategies to master the task. The inter-subject test revealed highly significant differences for focal search (F(1,18) = 14.82, p<0.001) and direct swimming (F(1,18) = 23.31, p<0.001) as well as significant differences for scanning (F(1,18) = 5.78, p<0.05) and perseverance (F(1,18) = 9.87, p<0.05). A highly significant interaction effect day×group was found for direct swimming (F(4,72) = 6.67, p<0.001) and perseverance (F(4,72) = 8.34, p<0.001). For scanning (F(4,72) = 3.17, p<0.05) and directed search (F(4,72) = 2.66, p<0.05) the interaction was significant. Group specific differences firstly appeared only on the level of scanning, when the search becomes increasingly systematic. Importantly, the group differences turned out to be especially large when hippocampus-mediated spatial precision became essential. Over time, trial-wise contribution of strategies differed significantly between treated mice and controls for scanning (p<0.05, Mann-Whitney on trials 3, 11, 12 and 13), focal search (p<0.05 on trial 14), directed search (p<0.05 on trials 10, 17 and 18), direct swimming (p<0.001 on trials 11, 16 and 17) and perseverance (p<0.05 on trials 19, 20, 22 and 23). All ANOVAs were done using the Bonferroni correction.

Because the distribution of differences in strategy use appeared to reflect an underlying temporal pattern we looked for systematical changes in the respective contributions to group performance in treated animals compared to controls (Figure 6C). In the first trial treated mice and controls differ only to a small degree with respect to random and scanning strategies. Importantly, when exposing the animals to the pool for the very first time no differences were found for thigmotaxis behavior (Figure 6C left). In general, the use of directed and place specific search patterns turned out to be at least delayed in mice with suppressed adult neurogenesis. Furthermore, to find the hidden platform TMZ mice consistently relied on scanning and directed search at the expense of focal search and direct swimming, the latter representing the truly allocentric search patterns with the highest spatial precision. On day 3, in the late phase of the first acquisition period, both focal search and direct swimming were found only sparsely among treated animals. Surprisingly, on that day the absolute mean difference in latency between groups was found to be minimal, whereas the sum of relative differences in strategy contribution was high for TMZ mice.

Our finding that treated mice reached comparable performance levels at the end of the first acquisition phase raised the important question whether progression to the spatially most precise, allocentric strategies is mandatory for learning the reference memory version of the water maze. Alternatively, the task might be mastered by simply improving the proficiency of using generally less effective strategies. Therefore, we used convolution analysis to test the hypothesis that effectively switching to the most effective search patterns is the learning strategy generally underlying acquisition of the MWM task (Figure 6D–F, see methods section for details).

By comparing the values measured with predictions from two models holding constant either the frequency or efficiency with which each strategy was applied, linear regression analysis revealed an effective progression towards direct swimming to be the general learning strategy in both groups (CTR: r = 0.94 compared to r = 0.68; TMZ: r = 0.83 compared to r = 0.54; Figure 6 E, F). This finding indicates that treated mice, too, solved the task by trying to switch to the more advanced spatially precise learning strategies.

For all experiments we used 6–8 weeks old female C57BL6 mice. To label adult generated cells animals received one single intraperitoneal injection of bromodeoxyuridine (BrdU, 50 µg/kg body weight, Sigma). Tissue preparation was done as described [37]. The mice were kept at the animal facility of the Max Delbrück Center for Molecular Medicine (MDC) Berlin-Buch, Germany. All local and federal regulations of animal welfare in research were followed. The experiment was approved by the responsible authority, Landesamt für Technische Sicherheit und Gesundheit (LaGetSi) Berlin.

Temozolomide (TMZ, Temodal®, SP Europe, Belgium) is an alkylating agent intended for the treatment of recurrent malignant glioma. In a dosage finding experiment we found a dose of 25 mg/kg body weight to be effective in suppressing adult neurogenesis by more than 80% after monocyclic (3 days) of treatment (Figure 1A).

Consequently, to suppress adult neurogenesis, mice from the treatment group (TMZ) received injections of TMZ at 25 mg/kg (i.p., 2,5 mg/ml in 0.9% NaCl), whereas the control group (CTR) received sham injections of saline only. This regimen was given on the first three days of a week for 4 weeks to resemble paradigms used for glioma treatment in humans (Figure 1B). Behavioral testing was performed 4 weeks after the final TMZ injection.

For adult-born granule cells to become recruitable the sequence of proliferation, differentiation and maturation requires approximately up to 28 days. Thus, suppressing adult neurogenesis for at least 4 weeks combined with a reconstitution period of 4 more weeks ensured that most of the cells borne immediately before onset of TMZ treatment would have been already used or eliminated by apoptosis. Using intercalating convalescence times, we minimized the risk of confounding side effects during behavioral testing.

It has recently been shown that 6–8 weeks old, adult generated granule cells become selectively recruited during acquisition of a spatial learning task [21]. Therefore, our primary aim was to minimize the number of potentially recruitable adult-born neurons exactly at that time, when the new granule cells should be of particularly high relevance for task acquisition. Because we suppressed proliferation for at least 4 weeks and behavioral testing began an additional 4 weeks later, at the time-point of the behavioral analysis the subpopulation of 5–7 week old adult generated neurons is primarily affected by TMZ. An immediate recovery of the stem cell niche seemed unlikely as it was found that after wiping out proliferation in the SGZ by irradiation reconstitution occurs only slow and on a prolonged timescale [38]. Consequently, at the time, when the mice were learning the water maze task only very few (or no) new granule cells were available to be functionally integrated into existing circuits.

Histology procedures were performed as described previously (2). Briefly, animals were perfused with 4% paraformaldehyde, the brains removed from the skulls, cryoprotected and sectioned at 40 µm on a dry ice-cooled sliding microtome. For immunostaining we used rat anti-BrdU (1∶500, Biozol), goat anti-Iba1 (1∶250, SantaCruz) as primary and biotinylated-anti-rat (1∶500, Dianova) as secondary antibodies.

For visualization of BrdU incorporation, DNA was denatured in 2N HCL for 30 minutes at 37°C. Free floating sections were then rinsed in 0.1 M borate buffer, pH 8.5, and thoroughly washed in tris-buffered saline (TBS), pH 7.4. To block endogenous peroxidase reactions, sections were pretreated with 0.6% H₂O₂. Sections were incubated with primary antibody against BrdU in TBS supplemented with 0.1% TritonX-100 and 3% donkey serum (TBS-plus) at 4°C overnight. After rinsing the sections with TBS and a blocking step with TBS-plus, incubation with the biotinylated secondary antibody (1∶500, Dianova) in TBS-plus followed. ABC reagent (Vectastain Elite, Vector Laboratories) was applied for 1 hour at a concentration of 9 µl/ml per reagent. Diaminobenzidine (DAB, Sigma) was used as a chromogen at 0.25 mg/ml in TBS with 0.01% H₂O₂ and 0.04% nickelchloride followed by rinsing with tap water and TBS. The sections were mounted on gelatine-coated glass slides and coverslipped with Neomount.

The number of BrdU-positive cells was analyzed following the standard routine of our laboratory. The method is a simplified version of the optical fractionator principle. In a complete series of 40 µm sections, 240 µm apart, BrdU-positive cells in the SGZ and the granule cell layer were counted exhaustively, but cells in the uppermost focal plane (at 40× magnification) were disregarded to avoid oversampling at the cutting surfaces. All counts were done with the experimenter ignorant of the treatment group of the specimen. A total of 10 animals were used.

Blood cell counts were obtained using a Beckman Coulter Ac-T Diff Hematology Analyzer with a software version for veterinary applications. After anesthetizing the animals with diethylether, blood samples were taken from the retro-orbital sinus. Blood samples were taken 1 day before, 1 day after and 4 weeks after treatment and collected using EDTA coated MiniCollect® tubes (Greiner bio-one, Kremsmünster, Austria). A total of five animals were used.

350 µm-thick horizontal slices containing the entorhinal cortex, the subiculum, and the hippocampus were prepared from 3 to 4 months-old female mice. The slices were transferred to an interface recording chamber continuously perfused with an aerated (95% O₂, 5% CO₂), prewarmed (32°C) artificial cerebrospinal fluid (ACSF) containing (in mM) NaCl 129, Na2PO4 1.25, NaHCO3 26, KCl 3, CaCl2 1.6, MgSO4 1.8, glucose 10 at a pH of 7.4. After 2 h of equilibration the medial perforant path was stimulated and the evoked potentials in the dentate gyrus were recorded in the molecular layer above the upper blade by using a glass capillary microelectrode filled with artificial cerebrospinal fluid (tip resistance 2–3 MΩ). Microelectrodes were prepared from borosilicate glass tubes. After 10 min of stable baseline response to test stimulation (once every 30 s), the ability to elicit LTP was assessed. To induce LTP, four tetani of high-frequency stimulation were applied at 100 Hz for 1 s with 10 s intertrain intervals. Responses were recorded every 30 s for 30 min after LTP induction. In a parallel set of experiments, the ACSF contained 5 µM bicuculline (BCM, Sigma, Deisenhofen, Germany) to block GABA_A receptor-mediated inhibitory activity.

To record field EPSPs (fEPSPs) in the CA1 region of the hippocampus, afferent fibers of the Schaffer collateral pathway were stimulated and recordings were made in the CA1 pyramidal cell layer. For LTP experiments, a 10-min baseline was recorded by stimulating every 30 s. LTP was induced by using a 100 Hz stimulation (four trains, 1 s in duration), after which responses were elicited once every 30 s at the same stimulation intensity for 30 min. Signals were filtered at 3 kHz and sampled at 10 kHz using a TIDA interface card (HEKA, Lambrecht, Pfalz, Germany).

All data were analyzed offline using TIDA software. Amplitudes of evoked field potentials were measured from an average of 4 peaks. Data were expressed as means±SEM and statistical comparison was done by repeated-measures ANOVA on the last 10 min (Aabel Software, Gigawiz). Significance level was set to p<0.05.

The rotating rod apparatus (Columbus Instruments) was used to measure the ability of the animals to improve their locomotor skills with training [39]. 1 day before, 1 day after and 4 weeks after TMZ/sham treatment exploratory activity was evaluated using an open field test [40].

For the rotarod task mice were placed on the rod (3 cm in diameter) for three trials per day for three consecutive days. Each trial lasted a maximum of 5 min, while the rotating rod underwent a linear acceleration from 4 to 40 rpm. Animals were scored for their latency to fall for each trial. Between the trials the animals rested a minimum of 10 min to avoid motoric fatigue. All animals were subjected to the rotarod task 1 day before, 1 day after and 4 weeks after TMZ/vehicle treatment.

For the open field test each animal was placed in the center of a white plastic chamber (60×60×20 cm) under standard room-lighting conditions. Overall activity and the time spent by each animal in the center and/or the wall zone was measured and analyzed using a digital camera on the ceiling operated via Ethovision (Noldus, Netherlands). The time spent in the center was used as an index of general anxiety levels.

Two weeks after end of treatment, mice were trained in the reference memory version of the Morris water maze task [41] to locate a hidden escape platform in a circular pool (1.89 m diameter). Water was made opaque with non-toxic white paint and kept at a temperature of 19–20°C. Each mouse was given 6 trials a day for 5 consecutive days with an inter-trial interval (ITI) of 30 mins. Platform position was changed after day 3 (Figure 4). Mice were released from one of four possible starting points and allowed to search up to 120 s for the platform. During each day the starting position remained constant. Irrespective of trial performance mice were guided to the platform and allowed to remain there for at least 15 s. Swim paths were recorded using Ethovision (Noldus) and further analyzed using Matlab (The Mathworks, USA).

General spatial learning was analyzed using classical parameters like latency to reach the platform, swim path length and relative time spent in the four quadrants (Figure 4).

To characterize the development of a spatial preference for the goal platform the MWM pool was divided into 10×10 cm wide sectors allowing the presence probability of an animal in each sector to be represented as heat map-like occupancy plots.

To assess the qualitative aspects of learning the MWM task we analyzed the search strategies used by the animals to locate the hidden platform. Originally, the existence of different behavioral strategies in the context of spatial learning was demonstrated by Wolfer and Lipp [14], [16], [17], [42]. Here the concept of different search strategies was modified as shown in Figure 5. Briefly, swim path data from Ethovision (Noldus, NL) were used to derive the time-tagged xy-coordinates for classifying the respective predominant search strategies by an algorithm implemented in Matlab. Classification criteria served an only descriptive function and were not meant to prove the involvement of particular cognitive processes or brain structures.

In the course of learning the MWM, the animals showed a sequential use of different search patterns ranging from initially almost undirected to spatially precise, highly efficient and allocentric strategies. Immediately after being introduced to the MWM pool the first time some animals showed a behavior known as “wall hugging” or “thigmotaxis”. Thigmotaxis is considered to be a first and highly emotional response to a new and stressful situation. Usually thigmotaxis is rapidly overcome and replaced by the “random search” strategy covering the entire pool surface. As the animals gathered more knowledge of the MWM arena their search behavior became increasingly restricted to the central pool area where the availability of distant visual cues is maximal. Because in this stage the mice scan the environment for landmarks this strategy is called “scanning”. For both random search and scanning any spatial or even directional preference is absent. “Chaining” is proposed to be a successor of scanning where the animals show a clear preference for the goal annulus having learned the correct distance of the goal platform to the wall. Once the animals develop a directional preference the use of distant visual landmarks is evident and thus their behavior albeit still being egocentric has become hippocampus-dependent. This point is marked by the “directed search” strategy where the search becomes directional restricted in a triangular fashion pointing from the starting position towards the actual goal. Although path lengths swum may be initially not significantly shorter than for the scanning or chaining strategy directed search is usually the beginning of a subsequent refinement in the efficiency to reach the goal platform from each possible start position. This phase has to be considered as transition from egocentric to allocentric navigation. Therefore the two most efficient behavioral strategies to navigate to the hidden platform are “focal search” where the animals spend most of the time in the nearest neighborhood of the goal and “direct swimming” where they navigate directly to the goal regardless of the actual starting position used. The development of a precise and place specific preference depends on the integration of the relevant cues into an allocentric cognitive representation. Principally “perseverance” can be found among all groups irrespective of treatment and represents a still present preference for the previous goal position until the animal's behavior is changed as a sign of functional plasticity.

The classification process was implemented as a script in Matlab and relied on a set of numerical parameters (Figure 5) and applied those criteria to each trial of the MWM experiment therefore providing an unbiased objective method to identify certain qualitative properties of the recorded swim paths. All criteria represent the common abstract properties of the respective strategy and correspond directly to their special functional demands. To avoid misclassifications due to shared features among multiple strategy definitions the classification process excludes the respective patterns serially. Thus search patterns all of which are enclosed by other patterns as special cases have to be sorted first followed by the less precisely defined ones. The algorithm used in our study was tested on 4320 example trials showing an overall reliability of more than 90% (trials not recognized at all were interpreted as errors).

Generally, mice can use different higher-level strategies to learn the MWM task. As an alternative to an effective progression towards the principally most precise search strategies animals could also just improve their proficiency in applying less precise strategies. Which of those higher-level strategies is preferentially used makes a fundamental difference for interpreting the true hippocampal learning. Especially under the highly reductionistic conditions found for the MWM task just practicing non-spatial strategies such as scanning or chaining can result in comparable latency times without any contribution of hippocampus-specific behaviors. Therefore we evaluated the respective contribution of both general strategies using convolution analysis [14].

Learning performance in a spatial task is usually measured using parameters like latency or path length. Common to those parameters is their dependence on more basic variables that cannot be easily assessed directly. Accordingly both latency and path length can be interpreted as values of convoluted functions. Given the concept of distinct qualitative search strategies, learning performance in the MWM depends on the respective frequency with which a respective strategy is applied (function f(x)) and the quality that is found in applying a certain search pattern as measured by the respective path length for that pattern (function e(y)). Thus for the animals' task performance we get p(t) = (f * e)(t) as the convoluted function of f(x) and e(y).

To estimate the relative contribution of each respective function to the learning performance p(t) we compared the measured path length for each search strategy (p(t)) with a prediction derived from a model where either the frequency (f(x) dx) or quality (e(y) dy) is held constant for the whole experiment. The respective means of frequency and path length over all days were used as constants to calculate those predictions (Figure 6D).