Adult-generated hippocampal neurons allow the flexible use of spatially precise learning strategies

Abstract

Despite enormous progress in the past few years the specific contribution of newly born granule cells to the function of the adult hippocampus is still not clear. We hypothesized that in order to solve this question particular attention has to be paid to the specific design, the analysis, and the interpretation of the learning test to be used. We thus designed a behavioral experiment along hypotheses derived from a computational model predicting that new neurons might be particularly relevant for learning conditions, in which novel aspects arise in familiar situations, thus putting high demands on the qualitative aspects of (re-)learning.In the reference memory version of the water maze task suppression of adult neurogenesis with temozolomide (TMZ) caused a highly specific learning deficit. Mice were tested in the hidden platform version of the Morris water maze (6 trials per day for 5 days with a reversal of the platform location on day 4). Testing was done at 4 weeks after the end of four cycles of treatment to minimize the number of potentially recruitable new neurons at the time of testing. The reduction of neurogenesis did not alter longterm potentiation in CA3 and the dentate gyrus but abolished the part of dentate gyrus LTP that is attributed to the new neurons. TMZ did not have any overt side effects at the time of testing, and both treated mice and controls learned to find the hidden platform. Qualitative analysis of search strategies, however, revealed that treated mice did not advance to spatially precise search strategies, in particular when learning a changed goal position (reversal). New neurons in the dentate gyrus thus seem to be necessary for adding flexibility to some hippocampus-dependent qualitative parameters of learning.Our finding that a lack of adult-generated granule cells specifically results in the animal's inability to precisely locate a hidden goal is also in accordance with a specialized role of the dentate gyrus in generating a metric rather than just a configurational map of the environment. The discovery of highly specific behavioral deficits as consequence of a suppression of adult hippocampal neurogenesis thus allows to link cellular hippocampal plasticity to well-defined hypotheses from theoretical models.

Figure 1. Effective suppression of adult hippocampal neurogenesis in C57BL/6 mice using temozolomide (TMZ).

(A) Suppression of proliferation in the dentate gyrus was dosage dependent. After monocyclic treatment (one daily injection on three consecutive days) using 25 mg/kg TMZ total numbers of proliferating cells were reduced by more than 80%. Because doubling the dose did not result in significant lower numbers of BrdU⁺ cells, we used 25 mg/kg as the standard dose for all other experiments in this study. Proliferation was detected by incorporation of BrdU following a single injection using 50 mg/kg i.p. BrdU four days after end of TMZ treatment. (B) One treatment cycle consisted of single daily injections on three consecutive days followed by a resting period of four days. After four cycles of TMZ treatment proliferation was reduced by more than 90% (t(9) = 2.94, p<0.001). A single injection of BrdU was given four days after end of TMZ treatment. (C and D) BrdU⁺ cells in the dentate gyrus of mice after multicyclic treatment using either saline or TMZ (25 mg/kg), respectively.

Figure 2. At the time point of behavioral testing no potentially confounding side effects were found.

(A and B) As expected, hematology only revealed a significant leucopenia immediately after four cycles of TMZ (t-test: t(4) = 3.59, p<0.05). The number of leukocytes fully recovered after four weeks of resting. No differences were found for erythrocytes, hemoglobin and platelets. (C and D) Locomotor abilities and exploratory behavior appeared to be unimpaired after TMZ treatment. (E and F) Expression of Iba-1 was used to identify activated microglia in the dentate gyrus of C57BL/6 mice. No elevated numbers of Iba-1⁺ cells were found for any of the TMZ doses used in the dosage finding experiment indicating the absence of inflammatory processes. The right micrograph represents microglia in mice treated with 50 mg/kg TMZ.

Figure 3. LTP in C57BL/6 mice with suppressed adult neurogenesis.

(A) In TMZ mice treatment prevented LTP in the dentate gyrus of hippocampal slices perfused with ACSF, but LTP was induced in control mice (ANOVA F(1,198) = 252.8, p<0.05). Previous studies showed that new, adult generated granule cells facilitate the weak LTP observed, while LTP induction in mature neurons is blocked by GABAergic inhibition under physiologic conditions (16, 17). (B) Using GABA_A antagonist biculline a strong LTP was observed in both treated mice and controls. (C) Significant differences between TMZ mice and controls appeared only for ACSF-LTP. (D) A cumulative histogram for both ACSF (right) and bicuculline (left) perfused hippocampal slices. Dashed lines represent TMZ mice, solid lines controls. (E) LTP in hippocampal region CA1 was not affected by TMZ.

Figure 4. Spatial learning in C57BL/6 mice with suppressed adult neurogenesis.

(A) Both groups successfully learned to navigate to the hidden goal, but TMZ treated mice needed longer to locate the hidden platform. While the TMZ group reached comparable latency times at the end of the first acquisition phase, mice with suppressed adult neurogenesis performed worse both transiently before and consistently after reversal. (B) For controls, the occupancy plot shows the rapid development of a place-specific preference for the platform position both before and after reversal. (C) In TMZ treated mice learning a correct place-specific response for the first goal position was significantly delayed but still effective. After platform reversal mice with suppressed adult neurogenesis showed a profoundly perseverating preference for the previous platform position. (D–G) Probe trials indicated successful spatial learning of the first goal for both groups. After reversal treated animals failed to develop a preference for the new goal quadrant.

Figure 5. Algorithm-based classification of search strategies.

(A) Determination of variables used in the classification process. The pool was divided into distinct zones to calculate the amount of time spent in the respective areas (left). Average distance of all datapoints of a given swim path to its centroid and to the present or previous goal position was used during the classification process (middle). Search patterns based on a directional preference for the goal position were identified using a triangular shaped corridor expanding from the starting point with its bisecting line towards the platform (right). (B) Each strategy was identified by one or two parameters representing their respective abstract key properties. (C) Because some search patterns were defined less specific than others exclusion of strategies had to be achieved in a particular order. Search patterns not recognized were classified by hand.

Figure 6. Qualitative analysis of spatial learning in mice with suppressed adult hippocampal neurogenesis.

(A) Examples of search strategies recognized by the classification algorithm used. (B) Both groups showed a clear progression towards increasingly hippocampus-dependent strategies. The top row represents the basic experimental protocol including the hidden platform position and the four starting positions used (arrows). Contribution of respective strategies to group performance in learning the MWM task was analyzed by classifying all trials using a parameter-based algorithm (Figure 5). Strategies are color coded. See text for statistical comparisons. (C) Compared to controls in treated animals respective strategies contributed differentially to group performance. Differences are shown as percent of contribution relative to controls. The color code used is the same as in (B). The black line shows the mean difference between groups in latency to reach the hidden platform. (D) Principle of convolution analysis according to Brody et al. . (E and F) Comparison of measured mean path lengths with predictions derived from convolution analysis using a constant-frequency model and a constant-quality model. An efficient progression towards directed or place-specific navigation strategies was found to be the general underlying learning scheme in both groups.