Spatial Relational Memory Requires Hippocampal Adult Neurogenesis

A mouse is swimming in a circular pool, searching for a hidden platform just beneath the surface. It has found this platform before. It knows roughly where to go. But pull the release point to a different edge of the pool, change the angle of approach, and suddenly this mouse is lost — circling, doubling back, finding nothing. The mouse next to it, with a fully intact hippocampus, solves the new version immediately. The only difference between them is thirty days of doxycycline in the drinking water, and a handful of neurons that never got to grow. That experiment, designed by David Dupret and colleagues at the University of Bordeaux, is the cleanest causal test the field of adult neurogenesis has ever seen. What it revealed was not that new neurons matter for all spatial memory, but that they matter specifically for the most demanding kind. Here is the background tension. The dentate gyrus — a curved, tightly packed subregion of the hippocampus — keeps producing new granule neurons throughout adult life. This is genuinely unusual. Most of the mammalian brain stops building neurons shortly after birth. The dentate gyrus never does. For decades, researchers have noticed that conditions which improve spatial learning, like environmental enrichment, also increase neurogenesis; conditions that impair spatial learning, like prenatal stress, decrease it. A positive correlation between neurogenesis rate and water-maze performance turned up in aging rats. Spatial learning itself appeared to promote the survival of newborn cells. The correlations pointed in one direction. But when researchers tried to establish causation by actually blocking neurogenesis and watching what happened, the results were all over the map. Some studies found memory deficits while others found nothing. The methods used to kill newborn neurons, including high-energy irradiation or a DNA-methylating agent called methylazoxymethanol acetate, were blunt instruments. They damaged far more than just new cells. The behavioral tasks varied so much in how hard they pushed the hippocampus that comparing results across labs was nearly impossible. Dupret and colleagues saw both problems clearly and built a solution to both at once. The genetic tool they designed is worth understanding because its elegance is what makes the findings trustworthy. The strategy rests on three facts: the protein nestin marks neural precursor cells specifically; Bax is a pro-apoptotic protein that, when overexpressed, triggers cell death by multimerizing at the mitochondrial membrane; and the reverse tetracycline transactivator, or rtTA, is inactive by default and only switches on target genes when doxycycline is present. Put those together and you have an inducible kill switch. The nestin promoter drives rtTA expression, so only nestin-positive precursor cells can activate the system. Doxycycline in the drinking water turns it on. A bidirectional Tet-responsive construct then drives two tagged Bax fusion proteins — EYFPBax and ECFPBax — allowing the team to visualize multimerization using a technique called FRET, which detects when two fluorescent proteins are close enough to transfer energy between them. Corrected FRET — the raw FRET signal minus the bleed-through from each fluorescent channel — confirmed that the Bax proteins were indeed forming multimers at the mitochondrial membrane. At a dose of two milligrams per milliliter in drinking water, the effects on neurogenesis were substantial and regionally specific. Cell proliferation, measured by phospho-histone H3 staining, dropped significantly. Twelve-day-old newborn cells labeled with bromodeoxyuridine fell dramatically, with a p-value below 0.001. Cell death increased selectively in the subgranular zone, the narrow band where newborn neurons reside, while the surrounding granule cell layers were unaffected. The subventricular zone — the brain's other major site of adult neurogenesis — showed no change. Neither did the olfactory bulb, the cerebellum, the cortex, or the striatum. Blood vessel staining was unchanged. Microglial counts were unchanged. This was a scalpel, not a hammer. With a validated tool in hand, the team ran the behavioral experiments. Mice received doxycycline for at least six weeks before testing. The water maze they used was 150 centimeters in diameter, with water at 20 degrees Celsius and the platform hidden just below the surface — a standard setup. The critical manipulation was the starting position. In the variable-start procedure, the platform stayed in the same location, but mice were released from a different point on the edge of the pool on every trial. To find the platform reliably from any angle, an animal needs a genuine spatial map — a flexible, viewpoint-independent representation built from the relationships among distal cues. In the constant-start procedure, the release point never changed, allowing animals to solve the task through a fixed route or egocentric habit, no map required. The results split cleanly along that line. In the variable-start, relational version, neurogenesis-ablated mice were profoundly impaired, with a group effect that had a p-value below 0.002, and a group-by-time interaction with a p-value below 0.008. In the probe trial, when the platform was removed, control mice showed strong preference for the target quadrant — their t-statistic was 8.92 against chance. The ablated mice showed no preference at all, which was not different from random search. In the constant-start version, the ablated mice kept up with controls across all training days. That is the key contrast. It is not that neurogenesis-inhibited mice cannot learn water-maze tasks; they can. What they cannot do is build the flexible relational map that lets them generalize knowledge to a new viewpoint. The team made this concrete with one additional test: after constant-start training, they released mice from a novel start position. Control mice handled it smoothly — their performance was indistinguishable from their final training day. Neurogenesis-ablated mice collapsed. Their performance dropped sharply compared both to their own prior performance and to controls, with t-statistics around negative 6.2 in both comparisons. Before drawing conclusions, consider the skeptical questions. Were these mice anxious? Slow? Globally impaired? The data say no. Novelty-induced locomotor activity — a simple form of spatial habituation — was unchanged, with no group effect and no group-by-time interaction. Performance on the visible platform task confirmed intact motor function and vision. A positive correlation between escape latency and swim path length ruled out passive floating as an explanation for longer times. Contextual fear conditioning, a hippocampus-dependent associative task, was also intact, showing no significant difference between groups in contextual freezing. The team verified that their fear conditioning procedure actually engages the hippocampus by showing that acute inactivation of the dorsal hippocampus with lidocaine reduced freezing by approximately thirty-five percent. The hippocampus is involved — adult neurogenesis just isn't required for this simpler version of hippocampal learning. So why would new neurons be specifically required for relational memory? Dupret and colleagues don't claim to have answered this definitively, but they lay out the mechanistic hypotheses their result speaks to. Adult-born granule cells show enhanced synaptic plasticity compared to mature cells, including facilitated long-term potentiation in the dentate gyrus. Work by Kee and colleagues has shown that adult-generated cells are preferentially incorporated into spatial memory networks. Beyond plasticity, there is the pattern separation account: the dentate gyrus is proposed to orthogonalize similar inputs — to keep overlapping spatial events from blurring into each other. Relational memory demands exactly this because holding multiple position-cue relationships in mind without interference is the core computational challenge. Dupret and colleagues cite Leutgeb and colleagues for showing that small environmental changes drive strong decorrelation among dentate gyrus place cells, and McHugh and colleagues for behavioral evidence implicating dentate gyrus circuits in rapid pattern separation. Computational work by Aimone and colleagues and by Becker suggests that ongoing neurogenesis helps ensure each new event gets a unique encoding. What the paper stops short of claiming is which of these mechanisms is actually the operative one. The deficit they found reflects the loss of an ongoing contribution, not total ablation — residual neurogenesis persisted even after months of doxycycline treatment. The causal link they established is real and specific. Pinning down the exact cellular computation that adult-born neurons perform in the service of relational memory is the work that follows. What this study gave the field was something it had been missing: a clean, inducible, cell-type-specific tool that produced a clean, specific behavioral answer. New neurons in the adult dentate gyrus are necessary for the kind of spatial memory that requires holding relationships flexibly, applying a map from any angle, and generalizing knowledge beyond the conditions in which it was learned. The simpler tasks — recognizing a familiar place, freezing in a dangerous context, following a learned route — proceed without them. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.