Distinct Properties of Long-Term Potentiation in the Dentate Gyrus along the Dorsoventral Axis: Influence of Age and Inhibition

The hippocampus is important for spatial navigation, episodic memory and affective behaviour. Increasing evidence suggests that these multiple functions are accomplished by different segments along the dorsal-ventral (septal-temporal) axis. Long-term potentiation (LTP), the best-investigated cellular correlate of learning and memory, has distinct properties along this axis in the CA1 region, but so far, little is known about longitudinal differences in dentate gyrus (DG). Therefore, here we examined potential dorsoventral differences in DG-LTP using in vitro multi-electrode array recordings. In young mice, we found higher basal synaptic transmission in the dorsal DG, while the LTP magnitude markedly increased towards the ventral pole. Strikingly, these differences were greatly reduced in slices from middle-aged mice. Short-term plasticity, evaluated by paired-pulse ratios, was similar across groups. Recordings in the presence and absence of GABAA-receptor blocker picrotoxin suggested a higher inhibitory tone in the ventral DG of young mice, confirmed by an increased frequency of miniature inhibitory postsynaptic currents. Our findings support the view that the hippocampus contains discrete functional domains along its dorsoventral axis and demonstrate that these are subject to age-dependent changes. Since these characteristics are presumably conserved in the human hippocampus, our findings have important clinical implications for hippocampus- and age-related disorders.

Dorsoventral differences in electrophysiological parameters are age-dependent. In a second series of experiments, we employed the recently developed 'multi-electrode-optrode array' (MEOA) from imec (Heverlee, Belgium) 19 , which features a densely-spaced, combined electrical and optical LED array and improved signal-to-noise ratio. For this study, we only utilized the electrical array. We again investigated input/output relationships and LTP from dorsal and ventral hippocampal slices, prepared as described before, but now from young (2-3 months) and middle-aged (9-12 months) mice (Fig. 2). In these experiments, we did not add any PTX to maintain a more physiological excitation/inhibition balance. We also included paired-pulse stimulation to assess a short-term, presynaptic form of plasticity.
In slices prepared from young animals, we could replicate our previous finding of higher basal synaptic transmission in the dorsal DG (F 1,19 = 14.46, p = 0.0012; Fig. 2a). Paired-pulse ratios indicated no significant differences in short-term plasticity (Fig. 2b). Subsequent LTP experiments (Fig. 2c) showed the expected smaller LTP magnitude in the absence of PTX as compared to the experiments with 10 μM PTX before. Importantly, however, the dorsoventral difference in potentiation was maintained (F 1,14 = 11.88, p = 0.0039). Linear regression for the induction phase (Supplementary Table S2) revealed significant differences for both the intercept (F 1,124 = 5.07, p = 0.0261) and rising slope (F 1,124 = 10.07, p = 0.0019), which is already apparent in Fig. 2c. These results are indicative of marked differences in potentiation induction kinetics between dorsal and ventral DG. Nonlinear regression analysis for the LTP decay phase confirmed the principal differences between both segments by yielding the best fit of ventral LTP with a one-phase exponential decay function (F 1,294 = 71.97, p < 0.0001), as before, but dorsal LTP decay with a straight line (linear regression) (Supplementary Table S2).
When the same electrophysiological parameters were examined in middle-aged animals, we could no longer obtain any differences in basal transmission (Fig. 2d). Noteworthy, this seems to be due to an age-dependent decrease of basal synaptic effectiveness in the dorsal DG, as evidenced by a statistical comparison of dorsal input/ output curves from young versus middle-aged mice (F 1,18 = 10.05, p = 0.0053). While paired-pulse values showed the same pattern as in young mice (Fig. 2e), the dorsoventral difference in LTP magnitude was clearly reduced, and only significant during induction (0-20 min: F 1,12 = 9.20, p = 0.0104); no longer in the maintenance phase (20-180 min: F 1,12 = 0.8261, p = 0.3813; Fig. 2f). Linear regression analysis of the induction phase indicated that the intercepts were significantly different (F 1,108 = 6.14, p = 0.0148), similarly as in young mice, but the slopes were the same (F 1,108 = 3.65, p = 0.0588) (Supplementary Table S3). For both dorsal and ventral DG, the LTP decay phase was now best fitted by linear regression, with significantly different slopes (F 1,458 = 7.98, p = 0.0049) and intercepts (F 1,458 = 22.34, p < 0.0001) between the segments (Supplementary Table S3).
Scientific RepoRts | 7: 5157 | DOI:10.1038/s41598-017-05358-1 GABA A -receptor antagonism differentially affects dorsal and ventral dentate gyrus. Next, we wanted to investigate in more detail how disinhibition by the GABA A -receptor antagonist picrotoxin (PTX) contributes to the observed dorsal-ventral differences in young mice. Hereto, we compared the levels of potentiation obtained at 20 min (i.e. initial potentiation, 5 min after the last HFS) and at 180 min for both conditions, with and without PTX in the recording solution (as derived from Figs 1c and 2c, respectively; Fig. 3a). We found that the addition of PTX leads to higher potentiation levels at 20 min in both regions (condition effect: F 1,27 = 34.85, The grey lines superimposed on the data points denote the best-fitted functions for the induction of potentiation and subsequent decay. See Supplementary Table S1 for all equation parameters. Two-way RM-ANOVA was used for statistical analysis (** indicates p < 0.01 and *** indicates p < 0.001).
p < 0.0001; two-way ANOVA), but also that it has a markedly stronger effect on the ventral DG (interaction effect: F 1,27 = 14.09, p = 0.0008). At 180 min, the ventral -but not dorsal -potentiation level is still greatly affected by PTX application during induction (condition effect: F 1,27 = 30.76, p < 0.0001; interaction effect: F 1,27 = 31.38, p < 0.0001). For simplicity, the graph in Fig. 3a only shows the region effects (left to right: t 5.07 = 4.14, p = 0.0088; t 11.02 = 3.71, p = 0.0034; t 7.93 = 6.41, p = 0.0002; t 13.17 = 1.33, p = 0.21; unpaired t tests with Welch's correction). Furthermore, to directly compare potentiation levels between the two experimental conditions, we calculated mean ventral/dorsal ratios (see Methods for details). Statistical comparison revealed that this ratio was significantly affected by PTX application, both at 20 min (t 74.10 = 7.88, p < 0.0001) and at 180 min (t 63.61 = 10.61, p < 0.0001). These results were confirmed by an independent statistical approach using a bootstrapped resampling method (see Methods and Supplementary Fig. S1 for details). Altogether, these findings suggest a stronger sensitivity to PTX of ventral DG-LTP, likely caused by a higher GABA A receptor-mediated inhibitory tone in this region compared to the dorsal DG.
Higher frequency, but similar amplitude of miniature inhibitory postsynaptic currents in ventral compared to dorsal granule cells. To further investigate the apparently higher inhibitory tone in the ventral DG, we performed measurements of miniature inhibitory postsynaptic currents (mIPSCs) in dorsal and ventral granule cells of young mice (Fig. 4a). We found that ventral granule cells have a significantly higher mIPSC frequency (shorter inter-event intervals), with a median inter-event interval of 409 ms compared to 524 ms in dorsal cells (Kolmogorov-Smirnov (KS) test, p < 0.0001, 17 neurons per region from 6 mice; Fig. 4b). In

Discussion
In this study, we report an age-dependent dorsoventral gradient in basal synaptic transmission and LTP in the mouse DG. In young (2-3 months old) mice, we obtained markedly stronger LTP in the ventral as compared to the dorsal hippocampus. Strikingly, this difference was found to be almost faded away in middle-aged (9-12 months old) animals, apparently due to a decline in strength of ventral LTP to a similar magnitude as dorsal LTP. The larger ventral DG-LTP is diametrically opposed to studies in the CA1 region, where the dorsal segment has been reported to have a higher ability to undergo LTP [10][11][12][13][14] . In accordance with a dorsoventral gradient rather than a clear-cut segregation, our LTP data of the intermediate segment fall right in between those of dorsal and ventral slices, supporting the presence of an intermediate region that expresses "partly overlapping characteristics with its neighbour regions", as proposed by Fanselow and Dong (2010) 1 . In vivo findings of Kenney and Manahan-Vaughan (2013) indicate the existence of a similar dorsoventral LTP gradient in the DG of rats, as they reported a lower LTP induction threshold and stronger potentiation in the intermediate compared to the dorsal part 16 .
In addition to the dorsoventral difference in LTP magnitude, we found that the LTP induction and decay kinetics vary along the dorsoventral hippocampal axis, as assessed by linear and non-linear regression analyses. Although potentiation in both regions converged into similar levels of stable potentiation from about two hours onwards, the marked differences in dynamics point to a clear dorsoventral segregation of the mechanisms that sustain LTP.
In agreement with the latter, molecular, anatomical and functional studies have reported strong differences between dorsal and ventral hippocampus (as reviewed in refs 1 and 2). For example, gene expression studies revealed dorsoventral variation for all subfields of the hippocampus [20][21][22] . The expression of several proteins related to synaptic transmission and signal transduction was found to differ along the dorsoventral axis 21 , providing one putative substrate for the observed LTP differences. Furthermore, the subunits of AMPA and NMDA glutamate receptors, important postsynaptic components in synaptic plasticity, are differentially distributed along the dorsoventral axis in the rat hippocampus 23 . Several other channels that are important for intrinsic excitability are also differentially expressed or activated along the dorsoventral axis 14,[24][25][26][27] . This results in decisive differences in neuronal parameters which control the excitability and firing properties, and hence the basal input/output properties and LTP magnitude obtained in response to particular induction protocols 13,25,28 . Although almost all of these investigations have been performed in CA1 pyramidal cells, it is reasonable to assume that similar factors underlie the different LTP properties between dorsal and ventral DG. The opposite dorsoventral gradients of LTP properties in the CA1 region versus DG are most likely due to a different excitation/inhibition balance.
In line with this, the DG is known to maintain a higher inhibitory tone than other hippocampal subfields 29,30 . Noteworthy, we obtained a higher LTP magnitude in the ventral DG despite the fact that most types of GABAergic interneurons were reported to increase in number along the dorsoventral axis 31,32 . This underlines that the magnitude of potentiation does not depend solely on the sheer number of interneurons but rather on the overall excitatory/inhibitory balance. While detailed investigations of dorsovental differences regarding inhibitory processes in the DG are missing, studies in the CA1 region suggest marked differences in GABA A -receptor mediated recurrent inhibition 33 , which are best explained by dorsoventral differential expression of particular subunits, leading to contrasting properties of the synaptic pentameric GABA A -receptor protein complexes 34 . On the other hand, distinct proportions of certain subunits (such as α4, β3 and δ) in extrasynaptic GABA A -receptors could result in dorsoventral differences in tonic inhibition 34,35 , but our measurements did not provide evidence for this.
We did find a higher frequency of mIPSCs in ventral granule cells, which points to a higher presynaptic GABA release in this segment 36 , and is in agreement with the stronger potentiating effect of PTX application on ventral DG-LTP. Noteworthy, the low concentration of PTX (10 μM) used in our experiments did not affect baseline responses and caused only partial inhibition.
In the present study, we did not detect any differences in paired-pulse responses between dorsal and ventral DG. However, substantial differences in paired-pulse inhibition and facilitation have been reported for rats in vitro 37 and in vivo 16 .
Another factor that may contribute to the dorsoventral LTP gradient in DG and its opposite direction in CA1 is adult neurogenesis, since the DG is one of the rare neurogenic niches in the adult brain. A pool of radial glia-like stem cells in the subgranular zone regularly undergoes asymmetric divisions and gives rise to new neurons, even in aged animals. It is hypothesized that these adult-born cells play an important role in hippocampal functioning and synaptic plasticity, both during maturation and after integration into pre-existing networks (as reviewed in ref. 38). The maturation itself takes several weeks, during which the young neurons display an enhanced ability to undergo LTP 39,40 . Interestingly, the induction of LTP without the use of GABA-receptor antagonists (and thus under more physiological conditions) was reported to depend on young, adult-born neurons, in contrast to LTP induced in the presence of PTX 41,42 . As such, the reduced ventral DG-LTP that we reported here in middle-aged animals, recorded in the absence of PTX, may directly reflect decreased neurogenesis in this segment. Indeed, the dynamics of neurogenic processes are reported to differ between the dorsal and ventral hippocampus. The maturation of new granule cells in the ventral DG was shown to be slower 43,44 , giving them a wider time window to express enhanced plasticity. Furthermore, although the net neurogenic capacity is similar between both regions, adult neurogenesis happens more continuously and less disciplined in the dorsal hippocampus 45 . Dorsal adult-born cells were shown to be crucial for contextual discrimination, whereas those in the ventral DG are required for anxiolytic effects 46,47 . Interestingly, manipulations of adult neurogenesis, e.g. by antidepressants, environmental enrichment or physical exercise, may selectively target either the dorsal or ventral segment 43,47,48 . All of these aspects seem to reflect a different functional role for adult neurogenesis in the dorsal and ventral hippocampal segments, which may contribute to the observed differences in synaptic plasticity.
Since an age-dependency of hippocampal dorsoventral segmentation has not been reported before, the underlying mechanisms for the marked gradient in LTP seen in young but not in middle-aged mice are difficult to delineate. Many studies agree that electrophysiological properties of neurons such as resting membrane potential, membrane time constant and input resistance remain the same during normal ageing, but some report an increase in Ca 2+ conductance and after-hyperpolarizing potential in aged neurons that may cause disturbances in Ca 2+ homeostasis and reduced neuronal excitability (as reviewed in refs 49 and 50). With regard to the effects of ageing on synaptic plasticity, published data support the view that the induction of LTP in CA1, CA3 and DG is unchanged in response to strong high-frequency standard protocols, but that LTP maintenance tends to be impaired. When weak induction protocols are used, aged animals often show a deficit in LTP induction 50 . In previous studies, we found impaired LTP maintenance in the CA1 region of 20 months old mice in vitro 51 , but no LTP deficits in the dorsal DG of middle-aged 12-16 months old rats in vivo 52 , in line with the findings reported here. Recent work of our laboratory indicated that ageing goes along with certain changes in the lipid composition of neuronal membranes that lead to an attenuation or impairment of LTP and other types of synaptic plasticity 51,53 , and are associated with cognitive deficits, but whether these age-dependent changes in membrane composition differ between dorsal and ventral hippocampal segments is unknown. Hence, which mechanisms underlie the age-dependent diminution of the dorsoventral LTP gradient remains elusive.
The human hippocampus can also be segmented into an anterior (ventral) and posterior (dorsal) part, and longitudinal differences in anatomy and function are very likely conserved between rodents and humans 54,55 . Many neuroimaging studies have reported structural and functional connectivity differences along the longitudinal human hippocampal axis (as reviewed in refs 2 and 54). Considering that DG-LTP could also be induced in human hippocampal tissue, where it was shown to have nearly identical properties as in rodents 56,57 , it is tempting to speculate that a similar dorsoventral gradient in DG-LTP is present in humans.
Ageing also has a profound impact on the dorsoventral functional specialization in humans and it appears to predominantly target the DG over other hippocampal subfields 58,59 . Moreover, the anterior (ventral) segment seems to undergo the largest age-related volume loss 60,61 . In line with this, functional connectivity was shown to undergo a shift from anterior to posterior hippocampal dominance during ageing 62 .
A human functional anterior-posterior gradient also has major clinical implications, given the broad variety of hippocampus-related neurological disorders that may preferentially target either segment. For example, cognitive decline, dementia and Alzheimer's disease are all associated with a profoundly higher atrophy in the anterior segment 61,[63][64][65] . This is not surprising given that ageing also preferentially affects this segment, as discussed above, and these conditions are typically age-dependent. Other diseases that can be linked to a specific part of the hippocampal longitudinal axis include depression, medial temporal lobe epilepsy, schizophrenia and ischemia (as reviewed in refs 2, 59 and 66). The results of our study may directly reflect a higher vulnerability of the ventral DG to age-and disease-related disturbances, possibly including or being (partially) caused by reduced adult neurogenesis. Altogether, the dorsal-ventral (posterior-anterior) distinction is an important factor to take into account when studying underlying mechanisms, diagnostics and potential treatments.
Finally, all above-mentioned disorders are thought to be based on synaptic dysfunctions that are becoming overt as deficits in synaptic plasticity 67,68 . It will be interesting to examine in future research whether the dorsoventral differences in synaptic plasticity and their age-and inhibition-dependency observed in this study, are maintained in animal models for human diseases and in human patients.

Methods
Animals. C57BL/6J mice were bred and raised in our own animal facility. All animals were group-housed with ad libitum access to food and water, and kept at constant temperature and humidity in a normal 12 h light/ dark cycle (lights on from 8 am-8 pm). In MEA experiments, young (2-3 months old) and middle-aged (9-12 months old) mice of both genders were used and data were pooled, since preliminary experiments indicated no gender-influences on the examined parameters. In patch-clamp experiments, only young (2-3 months old) male mice were used. The housing conditions and procedures to prepare acute brain slices were approved by the KU Leuven Ethical Committee (P203/2012) and in accordance with European Directive 2010/63/EU. Acute hippocampal slice preparation. Mice were rapidly killed by cervical dislocation followed by decapitation. The brain was immediately removed and submerged in ice-cold artificial cerebrospinal fluid (ACSF), saturated with carbogen gas (95% O 2 , 5% CO 2 ). ACSF consisted of (all in mM): 124.0 NaCl, 4.9 KCl, 1.3 MgSO 4 , 2.5 CaCl 2 , 1.2 KH 2 PO 4 , 25.6 NaHCO 3 and 10.0 glucose (pH 7.4). After isolating the hippocampus, transverse slices (300 µm thick) were prepared using a custom-made tissue chopper, alternatingly starting from the dorsal or ventral pole. Both left and right hippocampi were used, with a similar left/right ratio in all experimental groups. Slices were placed into different sectors of an incubation chamber based on their origin relative to the dorsoventral axis, constantly supplied with carbogen and allowed to recover at room temperature for at least 90 min.
Multi-electrode array recordings. After incubation, up to two slices were placed simultaneously on a multi-electrode array (MEA) chip and perfused with ACSF (same composition as above) at 2.5 ml/min and a constant temperature of 32 °C. Slice positioning was guided by an inverted light microscope (Leica DM IL LED, Leica, Germany) or upright stereo light microscope (Olympus SZ61, Olympus, Japan) and pictures were taken for each experiment using a 3.0 MP microscope camera (Moticam 3, Motic, China). A harp slice grid (ALA Scientific Instruments, USA) was put on top of the slice(s) to assure immobilization and optimal contact with electrodes. Two types of MEAs, both containing 60 electrodes, were used: the standard planar 60MEA200/30iR-Ti type from Multi Channel Systems (Germany) and the recently developed multi-electrode-optrode array (MEOA) from imec (Belgium) 19 . Both types of MEAs contain 60 electrodes in a 8 × 8 layout, and are compatible with the MEA system by Multi Channel Systems (Germany), consisting of a MEA1060-BC amplifier, SG4002 stimulus generator, TC02 temperature controller and software (MEA_Select, MC_Stimulus, MC_Rack). Data streams were sampled at 10 kHz. For each slice, a single electrode was selected for stimulation. Biphasic, constant voltage pulses Scientific RepoRts | 7: 5157 | DOI:10.1038/s41598-017-05358-1 (100 µs pulse-width) were applied to evoke field excitatory postsynaptic potentials (fEPSPs) from the medial perforant path in the suprapyramidal blade of the DG. Correct placement in this pathway was confirmed by the observation of paired-pulse depression responses at an interpulse interval of 40 ms. Hereafter, an input/output curve with stimulation intensities ranging from 0.5 to 4.0 V (in steps of 0.5 V), each applied twice with 30-120 s interval, was established. The intensity evoking 40% of the maximal fEPSP slope was used for further stimulation. Where mentioned, an extended series of paired-pulse stimulation, at interpulse intervals of 10,20,50,100,200 and 500 ms, was recorded. After obtaining a stable baseline of at least 30 min, LTP was induced by four trains of high-frequency stimulation at 100 Hz, 1 s in duration and 200 µs pulse-width, with 5 min interval. In specified experiments, picrotoxin (PTX, Sigma-Aldrich; dissolved in dimethyl sulfoxide (DMSO) to working concentration of 100 mM) was added to the ACSF (final concentration: 10 μM PTX, 0.01% DMSO) right after the baseline, to reduce GABA A receptor-mediated inhibition and evoke robust LTP. In this case, a second 30-min baseline was recorded to allow confirmation that PTX did not affect baseline responses. Right after LTP induction, the recording solution was switched back to regular ACSF.
Multi-electrode array data analysis. Raw data were extracted using MC_Rack software (Multi Channel Systems, Germany) in replayer mode. Although all electrodes of the MEA served as recording channels, we focused our analyses on the single channel adjacent to the stimulation electrode (in the anterograde direction of the perforant path). To obtain fEPSP slope values, the region of interest was set from peak-to-peak and the software calculated the average slope within 10-90% of this region. Paired-pulse ratios were calculated by dividing the slope of fEPSP 2 by the slope of fEPSP 1 . In LTP experiments, all data points were normalized to the average baseline slope.
Subsequently, tonic inhibition was measured in the same DG granule cells by adding 50 µM bicuculline methbromide (BIC) to the ACSF superfusate. The value of tonic current was calculated as the difference in baseline current before and after the addition of BIC at holding potential −60 mV, as described previously 69 . During acquisition, all data were low-pass filtered at 2 kHz and sampled at 10 kHz.
Patch-clamp data analysis. As previously described 69 , off-line amplitude and frequency analysis of mIPSCs was performed using MiniAnalysis software 6.07 (Synaptosoft, USA). Mean tonic current values were obtained from Gaussian fits to all-point amplitude histograms (not shown) and the difference between peak values of two simulated Gaussians was used as a measure of the baseline tonic current.

Statistics.
All results were plotted in graphs and analysed statistically using GraphPad Prism 5.01 (GraphPad Software, USA), unless otherwise specified. Where parametric statistical tests are used, the D' Agostino-Pearson normality test was first applied to confirm normal distribution of the data. In all tests, p < 0.05 was considered significant. Data are presented as mean ± standard error of the mean (SEM) and n refers to the number of animals tested.
For input/output curves, paired-pulse ratios and LTP induction, (two-way) repeated measures analysis of variance (RM-ANOVA) was used with hippocampal region as between-subjects factor, intensity/interval/time as within-subjects factors, and Bonferroni's post hoc test for multiple comparisons.
To analyse curve fits, the extra sum-of-squares F test was used. The ventral/dorsal ratios were analysed by two independent statistical methods: (1) unpaired t test with Welch's correction to compare the average ratios as calculated from all possible ratios (see Fig. 3), and (2) resampling by bootstrapping, using a custom-written script in the open-source statistical programming language R (R Foundation for Statistical Computing, Austria, http://www.r-project.org) (see Supplementary Fig. S1). The resampling procedure was as follows: for both conditions (with and without PTX), we randomly sampled six pairs from ventral and dorsal measurements, which were used to obtain ventral/dorsal ratios (thereby modelling an experiment with paired design and 6 subjects). The ratio obtained from the condition without PTX was then subtracted from the ratio with PTX. This procedure was repeated 10 6 times to generate a distribution of ratio differences (with -without PTX). In addition, we obtained maximum likelihood estimates for the parameters of a normal distribution (Supplementary Fig. S1).
To compare the cumulative probabilities of mIPSC inter-event intervals and amplitudes, statistical significance was determined by two-tailed Kolmogorov-Smirnov (KS) test using MiniAnalysis software. Tonic inhibition between dorsal and ventral slices was compared statistically by unpaired t test. Equal numbers of recordings from dorsal and ventral hippocampus were made from each mouse (but not more than 3 pairs). In total, for each hippocampal segment, we measured mIPSCs in 17 neurons from 6 mice, and assessed tonic inhibition in 14 neurons from 6 mice. Data Availability. The datasets generated and analysed in this study are available from the corresponding author on reasonable request.