William T. Greenough, Neal J. Cohen
& Janice M. Juraska
The authors are at the Department of Psychology and
the Beckman Institute, University of Illinois, 405 N. Mathews
Ave., Urbana, Illinois 61801,
USA. Greenough@uiuc.edu
Two studies examine how experience regulates neurogenesis in the adult
rodent hippocampus. Although their conclusions appear contradictory, they
may in fact be reconcilable.
It has been known for many years that in rodents, neurons continue to be
produced in the dentate gyrus of the hippocampus throughout adult life1,
2. This finding generated considerable interest when it was first
reported, not least because of its possible implications for brain repair.
This initial excitement was tempered by the failure to detect neurogenesis
in adult macaques3, but within the last two years the field
has been revived by reports of neurogenesis in the dentate gyrus of marmosets4, macaques (M.Fallah,
E.Fuchs, P.Tanapat,
A.J.Reeves, E.Gould,
Soc. Neurosci. Abstr.24, 796.9,
1998) and even humans5. Moreover, a recent study6 reported that the survival of newly formed neurons in mice can be
increased by exposure to a more complex environment, suggesting that neuronal
replacement in adults may be regulated by experience.
Two papers in this issue, by Gould et al. (
pages 260−265) and by van Praag et al. (
pages 266−270), add to this growing body of evidence, in particular
by demonstrating that both proliferation and survival of newly formed neurons
can be affected by experience. Both studies seek to identify the aspects of
experience that are essential to produce these effects, using learning tasks
that either require or do not require the hippocampus. Yet, although they
both use similar techniques, the two studies arrive at seemingly incongruent
conclusions.
Gould et al. examined neuronal survival in the dentate gyrus after
adult rats were trained in four different learning tasks. The first task was
'trace' associative eyeblink conditioning, in which a brief time interval
separates the conditioned stimulus (CS; a burst of white noise) from the unconditioned
stimulus (US; a shock to the eye). Performance on this task is impaired by
damage to the hippocampal formation. They compared this with 'delay' conditioning,
which is similar to trace conditioning except that the US overlaps and coterminates
with the CS; this task can still be learned after hippocampal lesions. In
a second experiment, they compared 'spatial' learning in a water maze, in
which the rat must learn to use distant cues to find the hidden platform,
with a 'local cue' task, in which the rat learns to recognize the platform
directly. As above, learning the first task requires an intact hippocampus,
whereas learning the second task does not. In both experiments, the authors
found that the survival rate of labeled neurons was more than two times greater
in animals that had learned the hippocampus-dependent task than in those learning
a similar but hippocampus-independent task.
Van Praag et al. examined both proliferation and survival of dentate
gyrus neurons in response to a variety of experiences, including spatial navigation
in a water maze as well as swimming, wheel running and housing in an 'enriched'
laboratory environment. Consistent with earlier results from the same group6, the enriched environment enhanced the survival of newly formed
neurons. The authors now find that running also enhances proliferation and,
to a limited extent, survival, whereas—in contrast to the results of
Gould et al.—navigational learning in the water maze produced
no effect at all.
What are we to make of these apparently contradictory results? The most
trivial possibility (which cannot be excluded) is that the results reflect
either different measuring methods or the use of different subjects; Gould
et al. used male rats, whereas van Praag et al. used female mice.
However, there are also several more interesting possibilities that are suggested
by differences in the experimental design of the two studies (
Fig. 1).
Figure 1. Time course of the water-maze experiments.
In both studies, dividing cells were labeled by BrdU injection, and animals
were subsequently sacrificed on the days indicated (S) and examined for labeled
neurons in the dentate gyrus. Some animals were trained in a water maze (learning
is manifested as reduced latency). Control animals were compared with animals
that had undergone spatial learning. (a) van Praag et al. administered
BrdU daily between days 1 and 12 and counted labeled neurons after one day
(day 13, assumed to reflect proliferation) or four weeks (day 43, assumed
to reflect survival). Mice were trained daily for 30 days beginning at the
start of the labeling period, with the platform location reversed at day 24.
Most of the initial learning occurred early in the labeling period, and learning
of the reversed platform condition occurred two weeks after labeling.
(b) Gould et al. adminstered BrdU for one day only, and counted
neurons on day 11 or 17 (reflecting survival in both cases). Training began
one week after labeling, and continued for one week. The discrepancy between
the two studies could be explained if we assume that newly formed neurons
go though a transient period of sensitivity to the survival-promoting effect
of learning (shown by the curve at the top of each panel). In the van Praag
et al. study, learning (and presumably elevated hippocampal activity)
would occur mainly before and after the sensitive period, whereas in the Gould
et al. study, it would occur during this period.
It is important to distinguish between effects on the formation of new
neurons and effects on their subsequent survival. Gould et al., by
administering the label one week before training, examined only survival effects.
In contrast, van Praag et al. tested the effect of experience both
during and after the labeling period; by counting labeled neurons immediately
after the labeling period, they measured the formation of new neurons, and
then by counting several weeks later, they also measured the survival of these
newly formed neurons.
Focusing first on survival, the results of Gould et al. suggest
that hippocampal learning is important. New neurons may be more sensitive
than more mature neurons to the effects of activity, and it is possible that
the period of maximum sensitivity may begin shortly after the neuron is formed.
Gould et al. suggest that the initial postproliferative period is critical
to the survival of newly generated neurons because this is when their axons
emerge from the dentate gyrus and begin to contact target cells in hippocampal
subfield CA3.
If there is a period of sensitivity of this sort, then the results of the
studies may be reconcilable after all. Van Praag et al. did not observe
any effect of water-maze learning on either proliferation or survival. Their
water-maze procedure, however, produced very rapid learning, such that the
mice were already approaching their asymptotic level of performance before
the labeling was complete. Thus, much of the learning may have been completed
before most of the neurons entered their period of maximum sensitivity. In
contrast, Gould et al. administered water-maze training one week after
the neurons were already labeled. These newly labeled neurons might be among
those most affected by experience, because the experience occurred during
their period of maximum sensitivity. The trace conditioning in the Gould
et al. study may likewise have been given at the time of maximum effectiveness
relative to the sensitivity of the newly formed neurons. A further difference
between the two studies is that the animals used by van Praag et al.
learned with fewer trials than those used by Gould et al.; it is possible
that the more difficult task in the latter study may have placed a greater
load on the hippocampus and therefore produced a stronger survival-promoting
effect.
The major positive finding of the van Praag et al. study, which
the other study does not contradict, is that wheel running leads to enhanced
labeling immediately after the BrdU incorporation phase. The authors' interpretation
is that exercise increases the proliferation of neuronal precursor cells.
It is curious that the animals in the enriched environment cages, which appear
from the figure to include similar exercise wheels, show no increase in proliferation.
Rather, they show an increase in subsequent neuronal survival that is considerably
greater than the survival effect found in the exercise group. There is evidence
that BrdU administration may have toxic effects on cell proliferation7, and it is possible that exercise—which has been shown to
increase vascularization and presumably blood flow8, as well
as the expression of trophic factors9, in some brain areas—could
offset this toxicity.
It is interesting to speculate about the possible adaptive significance
of this proliferative effect. Intense exercise in a natural environment may
be associated with a need for increased navigation skills. It might therefore
be beneficial to produce more neurons in anticipation of greater hippocampal
use during a period of active exploration. One should be cautious, however,
about accepting such interpretations, particularly given that some exogenous
effects on brain plasticity seem difficult to explain in adaptive terms. For
instance, the number of dendritic spines in area CA1 is increased10,
and long-term potentiation more readily induced11, during proestrus
when estrogen and progesterone levels are high. Yet during this period, performance
on several hippocampus-dependent tasks is slightly impaired, whereas performance
on similar but hippocampus-independent tasks is unaffected12,
13.
In other words, plasticity may be manifested even when it is not needed or
used.
Nevertheless, given that the hippocampus has this unique capability of
regulating the production of new neurons from a continuously generated population
of precursors, it seems important to consider the possible implications for
hippocampal function. It also seems important to ask what is unique about
the hippocampus, in other words why other parts of the brain do not seem to
exploit the possibility of continuous neurogenesis. One proposal comes from
analyzing the different ways in which the brain must process and store information.
Studies of information storage in model neural networks have shown that adding
new learning sequentially to the network can result in 'catastrophic interference'14. That is, the changes induced by encoding new information into
the network can obscure previously stored information. McClelland et al.
15 proposed that the hippocampus protects against this by
forming a very rapid and relatively short-lived form of representation that
permits reactivation of representations in the cerebral cortex, so that those
long-term representations in cortex can be learned in an interleaved fashion.
In this way, although new information arrives sequentially, the reactivation
of all related information in cortical long-term memory, mediated by the connection
to hippocampus, results in the whole set of representations being rehearsed
together. But, if the hippocampus protects the cerebral cortex from catastrophic
interference, in the way just described, what protects the hippocampus itself?
McClelland et al.15 say rather little about this, suggesting
only that the hippocampal representation is a sparse one, so that there is
relatively little interference.
On the other hand, what if the hippocampus stored its memories differently
from the cerebral cortex, by adding neurons that can deal with new information
and deleting those that encode obsolete information, rather than changing
connections? It is the changing of connections that produces catastrophic
interference. If new memories got new neurons with new connections, catastrophic
interference might be avoided. The hippocampus could aid the cerebral cortex
in developing functional interleaving of memories and throw out the information
that was needed to do this, once successful cortical storage was achieved.
If such a mechanism operates in the hippocampus, why does it not exist
in the cortex too? McCloskey and Cohen14 and McClelland
et al.15 suggested that the connectivity changes that produce
catastrophic interference are the same changes that provide the ability to
generalize from one set of observations to another, and to generate abstract
representations of information. That is, the strength of a fully distributed
representation system, in which an integrated set of connections stores all
the knowledge in a given domain (as is thought to occur in cortex), is its
ability to support generalization and abstraction. The flip side is that such
a system cannot learn sequentially without risking catastrophic interference.
Adding a 'front end' mechanism—the hippocampal temporary organizer—might
offer the virtues of distributed representation while protecting against the
negative consequences.
Kempermann, G., Kuhn, H. G. & Gage, F. H. Nature386, 493−495 (1997). | Article | PubMed | ISI | ChemPort |
Pollard, D. R., Baran, M. M. & Bachvarova, R. J. Embryol. Exp. Morphol.35, 169−178 (1976). | PubMed | ISI | ChemPort |
Isaacs, K. R., Anderson, B. J., Alcantara, A. A., Black, J. E. & Greenough, W. T. J. Cereb. Blood Flow Metab.12, 110−119 (1992). | PubMed | ISI | ChemPort |
Neeper, S. A., Gomez-Pinilla, F., Choi, J. & Cotman, C. Nature373, 109 (1995). | Article | PubMed | ISI | ChemPort |
Woolley, C. S., Gould, E., Frankfurt, M. & McEwen, B. S. J. Neurosci.10, 4035−4039 (1990). | PubMed | ISI | ChemPort |
Warren, S. G., Humphreys, A. G., Juraska, J. M. & Greenough, W. T. Brain Res.703, 26−30 (1995). | Article | PubMed | ISI | ChemPort |
Warren, S. G. & Juraska, J. M. Behav. Neurosci. 111, 259−266 (1997). | Article | PubMed | ISI | ChemPort |
Markus, E. J. & Zecevic, M. Psychobiology25, 246−252 (1997). | ISI |
McCloskey, M. & Cohen, N. J. in The Psychology of Learning and Motivation: Advances in Research and Theory vol. 24 (ed. Bower, G. H.) 109−165 (Academic, New York, 1989).
McClelland, J. L., McNaughton, B. L. & O'Reilly, R. C. Psychol. Rev.102, 419−457 (1995). | Article | PubMed | ISI | ChemPort |
