Cell cycle regulation in the postmitotic neuron: oxymoron or new biology?

Key Points

• The review examines the current state of knowledge concerning the regulation of the cell cycle in the CNS during development and in the adult. Several key points emerge from the assembled evidence.

• Neurons must constantly suppress the cell cycle once they complete neurogenesis. If they fail to do so they will re-enter a cell cycle that they will not be able to complete.

• These 'cycling' neurons seem to be consigned to one of two fates depending on their stage of maturation: in young differentiating neurons, re-entry into a cell cycle leads to death within hours, which can be blocked by inhibition of the cell cycle. In adult neurons, re-entry into a cell cycle leads to the expression of cell cycle proteins and replication of some or all of the genome, but for many months the neurons appear incapable of either completing the cycle or dying.

• The linkage between cell cycle and cell death appears to lie at the root of many human neurodegenerative diseases.

• The neuron uses a multi-tiered strategy to suppress the cell cycle: there is strong evidence for genomic rearrangements, cell cycle proteins are found in unexpected locations and proteins switch from functioning as cell cycle promoters to acting as cell cycle suppressors.

• Many proteins originally identified for their role in cell cycle progression seem to perform essential functions in the neuronal cytoplasm and, in particular, at the synapse.

Abstract

Adult CNS neurons are typically described as permanently postmitotic but there is probably nothing permanent about the neuronal cell cycle arrest. Rather, it appears that these highly differentiated cells must constantly keep their cell cycle in check. Relaxation of this vigilance leads to the initiation of a cell cycle and entrance into an altered and vulnerable state, often leading to death. There is evidence that neurons which are at risk of neurodegeneration are also at risk of re-initiating a cell cycle process that involves the expression of cell cycle proteins and DNA replication. Failure of cell cycle regulation might be a root cause of several neurodegenerative disorders and a final common pathway for others.

Main

Most neuroscientists are comfortable with the axiom that CNS neurons are 'permanently postmitotic'. The intricate cascades of kinases and transcription factors that regulate the cell cycle in a HeLa or lymphocytic cell line (Box 1) seem removed from the day-to-day biochemistry of an adult neuron. Indeed, for decades developmental neurobiologists have shown that the bulk of CNS neurogenesis takes place in a tightly packed layer of nuclei lining the lumen of the neural tube (the ventricular zone, VZ) and later in the closely apposed region known as the subventricular zone (SVZ). Patches of adult neural stem cells can be found lining the lateral ventricles — to various degrees in different vertebrates — but these cells represent the persistence of a mitotic progenitor population rather than the division of a fully differentiated adult nerve cell. Indeed, the concept of a neuronal birthday carries with it the inescapable conclusion that once a neuron leaves the VZ (or SVZ) it will never divide again.

The absence of cell division is believed to be a core feature of neuronal identity. The depth of this belief is such that, on the face of it, cell cycle regulation in the adult neuron hardly seems to be a problem at all — instead, it seems to be more of an oxymoron. This review aims to stimulate the idea that cell cycle regulation is actually a constant problem for mature neurons. We summarize an increasing body of evidence which illustrates that, far from being permanently postmitotic, adult neurons must continuously hold their cell cycle in check. This perspective stems from the finding that cell cycle control can fail in 'postmitotic' neurons. The price of failure is high and the death of neurons in many diseases may be an inexorable consequence of the re-initiation of the cell cycle in an adult neuron.

The neuronal cell cycle

The CNS begins as a microscopic sheet of a few hundred ectodermal cells (the neural plate) but rapidly develops into a macroscopic structure with as many as a hundred billion neurons in the adult human. This enormous expansion in cell number means that the proper development of the brain depends on a well-regulated pattern of cell division. This raises two questions: what molecular switches regulate the rate and number of cell divisions during growth of the nervous system, and how does the mature postmitotic neuron regulate the cell cycle machinery so that it can refrain from division for many years?

The kinetics of neurogenesis in the neural tube have been studied most carefully in the mouse neocortex. Here, each neuronal lineage apparently undergoes 11 divisions before ceasing neuron production^1,2,3,4,5. The VZ itself seems to be highly stratified. Mitosis and cytokinesis occur at the ventricular surface, whereas other cell cycle activities, including DNA synthesis, occur at some distance from the ventricle (Fig. 1). The proposed stratification of cell cycle phases is validated by the finding that the DNA precursors ³H-thymidine^6,7 or 5-bromo-2′-deoxyuridine⁸ (BrdU, a thymidine analogue) label cells mainly in the superficial regions of the VZ when injected shortly before the animal is killed (Fig. 1). A secondary germinative zone, the SVZ, develops later just beyond the VZ. The functional differences between the ventricular and subventricular proliferative regions are not entirely clear. Both are regions in which neurons undergo their final cell division (or birthday). Furthermore, each region generates neurons as well as glia, yet there seem to be many molecular and cytological differences between the two regions^9,10. By the end of the developmental period, the VZ has been depleted of all mitotic cells whereas the SVZ persists (to a greater or lesser extent in different vertebrates) at the lumen of the lateral ventricles and harbours stem cell precursor populations that give rise to neurons in the adult. The kinetics and function of these adult-generated cells is hotly debated^11,12,13,14. However, because they are generated by stem cells and not from the division of cells that are already part of the adult neuronal circuitry, they do not contravene the basic tenet of our review: a mature, differentiated nerve cell does not divide.

Figure 1: The developing cerebral cortex.

A cross-section through the wall of the neural tube is shown. The morphological zones are noted on the left. Dividing neuroepithelial cells are shown in blue, radial glia in green and mature cortical neurons in yellow. Various features of the cell cycle that are important for the development of the cerebral cortex are shown on the right. The proposed stratification of the cell cycle phases (G1, S, G2 and M) in the ventricular zone (VZ) is indicated. Note that the more superficial subventricular zone has continued mitotic activity but no proposed stratification. The mature cortex is generated by each successive wave of immigrants from the VZ resulting in an inside-out pattern of layering with the first generated (early-born) neurons residing in the deeper layers and the last generated cells (late-born) residing more superficially. The proteins listed on the right are examples of proteins discussed in the text shown in their approximate location in the normal CNS. ³H-T, ³H-thymidine.

The molecular signals that stimulate cell division during neurogenesis are not entirely known, but there is evidence that sonic hedgehog (SHH) signals from certain cell layers to stimulate mitotic activity in both the VZ and in a specialized germinal region of the cerebellum known as the external granule cell layer. The actions of SHH are complex; it functions in patterning, cell fate determination and survival as well as proliferation. SHH seems to regulate cell division directly, via the transcription factors GLI and N-MYC¹⁵ (for a fuller discussion, see Ref. 16). The mechanism by which the period of neuronal cell division in the VZ or SVZ is brought to an end is equally uncertain. Inherent in the above description is the conclusion that every neuron has a birthday — a last time in embryogenesis at which ³H-thymidine or BrdU will label its genomic DNA. This final cell division signals the beginning of maturation, but also marks the point at which the nerve cell must put in place the mechanisms that will ensure a permanent mitotic arrest. Despite the importance of these mechanisms, virtually nothing is known about them. The tumour suppressor retinoblastoma (RB) seems to be crucial, as null alleles of this gene cause neurons to die after re-entering a cell cycle^17,18,19,20. Cyclin-dependent kinase inhibitors (CKIs) also seem to be involved in cell cycle arrest, as the loss of CKI family members results in alterations in cell cycle kinetics²¹. Studies in the retina and other tissues indicate that two subclasses of CKIs — the Ink4 class (including p16^Ink4a, p15^Ink4b, p18^Ink4c and p19^Ink4d) and the Cip/Kip class (including p21^Cip1, p27^Kip1 and p57^Kip2) — work coordinately to ensure cell cycle exit²², although neither class is necessary or sufficient to do so.

Loose ends. There is surprisingly little information that links the biochemical and molecular description of the cell cycle with the largely cytological description of neurogenesis and migration. Cell cycle proteins are found in brain lysates at early developmental stages²¹; however, there are few immunocytochemical studies of these proteins^23,24. The model shown in Fig. 1 predicts that the nuclei of the VZ are stratified by cell cycle phase, and therefore that cell cycle proteins should be expressed in layers. Indeed, the M phase marker phosphohistone H3 is found in mitotic cells at the ventricular surface and BrdU, which labels cells in S phase, labels cells in the expected regions of the upper VZ^20,25. Most other cell cycle proteins are not so neatly distributed. For example, proliferating cell nuclear antigen (PCNA), another S phase marker²⁵, and cyclin E, a late G1 phase marker²⁶, appear in cells in all strata of the VZ. For the CKIs, the picture is just as complicated²². Cyclins and cyclin-dependent kinases (CDKs) are regulated by many factors that control their synthesis, degradation and localization. Their broad distribution in the VZ indicates that either the predicted stratification of cell cycle phase is incorrect or that the regulation of these proteins in dividing neuronal precursors differs from the simple models described in most textbooks.

Table 1 Mouse knockouts of cell cycle proteins with demonstrated CNS effects

Another unexpected finding is that DNA repair enzymes are active immediately after the completion of neurogenesis. Mutations in genes involved in DNA double-strand break (DSB) repair, including Xrcc4²⁷, ligase IV (Ref. 28) and the DNA repair polymerase, Polβ ^29,30, result in embryonic lethal phenotypes. Moreover, neuronal death in ligase IV knockouts can be suppressed by the introduction of a second mutation in ataxia telangiectasia mutated (Atm)³¹, an early component of the DSB detection and response system. The implication of these findings is that DSBs are normal events in early neuronal differentiation. The function of these proposed breaks is unknown^32,33,34, but the correlation of human CNS disorders with defects in DSB repair genes underscores their importance in the development of the brain^35,36.

While we might predict that the mutation of cell cycle proteins would have devastating effects on development and neurogenesis, a final loose end comes from the observation that homozygous null embryos for virtually every one of the CDKs, cyclins and CKIs (including CDK2, CDK4 and CDK6, as well as cyclins A1, B2, D1, D2, D3, E1 and E2) survive most or all of the way through neurogenesis (Table 1). The only exception to this rule are embryos lacking cyclin A2 (Ref. 37) or cyclin B1 (Ref. 38) (the knockout of CDK1 has not been reported). The survival of these mice may be due to functional redundancy, as mice with mutations in two or three of these proteins die earlier. However, even the cyclin D triple mutant (Ccnd1^−/−;Ccnd2^−/−;Ccnd3^−/−) progresses to approximately embryonic day 16.5 (Ref. 39), which is after most neurogenesis has occurred, suggesting that there is more to neuronal cell cycle regulation than cyclins and CDKs alone. Similarly, RB-deficient embryos die in the third trimester, but a normally patterned brain and spinal cord develop^17,18,19. Finally, the loss of the CKI p27^Kip1 or p19^Ink4d alters cell cycle kinetics during neurogenesis^40,41, and the balance of different cell types that differentiate can be affected, but cell division eventually stops. Indeed, the lasting morphological changes observed in these mice are relatively subtle, given the predicted role of these proteins in orchestrating neuronal exit from the cell cycle (for a fuller description of such mutations in the context of the entire developmental process, see Ref. 42). A tentative summary of how these gene products affect the three major phases of CNS neuronal development is outlined in Fig. 2.

Figure 2: Cell cycle proteins implicated in each of the three phases of neuronal development.

Of the cell cycle proteins, mutations in only two of the cyclins (A2 and B1) are known to interrupt the process of neurogenesis (left column); however, as a direct effect of these proteins on the behaviour of cells within the ventricular zone (VZ) has never been shown, it may be that these mutations block development before the nervous system can emerge. DNA synthesis is indicated by the deoxyribonucleotide triphosphate (dNTP) to DNA symbol on the left of the first column. Migration (centre column) describes the stage in which the neuron ceases new mitotic activity and begins the early differentiation steps required to emigrate from the VZ and sub ventricular zone. Maturation (right column) is the cytological and biochemical 'coming of age' of the neuron once it has reached its adult location. The inclusion of the transcription factor E2F1 and retinoblastoma (RB) in this stage is by inference only, as knocking down these proteins in the adult has not yet been shown to have an effect on maturation. CDK1, cyclin-dependent kinase 1; ORC, origin recognition complex.

Failure of cell cycle arrest in development

As described above, CNS neurons do not divide after they emigrate from the VZ or SVZ. What would happen if a neuron lost control of its cell cycle and re-entered cell division? Although it seems paradoxical, evidence suggests that the neuron would probably die. We refer to this process as cycle-related neuronal death (CRND) — a purposefully cautious term (see below).

The first experimental evidence for CRND came from two independent lines of experimentation in which RB protein was inactivated by the large transformation (T) antigen of the SV40 virus. Inactivation of RB leads to the release of the transcription factor E2F1 and the subsequent upregulation of many cell cycle genes. When T antigen is introduced into most cultured cells, they are released from normal proliferation control and become transformed. However, when T antigen is introduced into postmitotic neurons, they degenerate^43,44,45,46. Cell division begins and BrdU is incorporated into DNA, but M phase is not initiated and the neuron dies shortly after⁴⁴. As might be predicted, this process is E2F1-dependent⁴⁷.

There is also massive apoptotic neuronal death in mice with homozygous null alleles of Rb^17,18,19. In these animals, CRND occurs after the migrating neuroblasts leave the ventricular region. BrdU injections a few hours before sacrifice have revealed that neurons in normally postmitotic regions of the maturing brain and spinal cord are engaged in ectopic DNA synthesis^17,18,19. Most of the neuronal death in Rb knockouts is dependent on both E2F1 and another transcription factor, p53 (Ref. 48, 49). Detailed examination of the mutant embryos has revealed a second, less appreciated phenotype: before their death, neurons are both morphologically and biochemically immature, suggesting that RB is also required for neuronal differentiation²⁰.

This link between cell cycle suppression and final neuronal differentiation is found for many cell cycle genes, and seems to work both ways. Early investigations that explored the death response of cultured sympathetic neurons after withdrawal of nerve growth factor (NGF) revealed an unexpected increase in the levels of cyclin D mRNA⁵⁰. This indicated that loss of trophic support during development might lead to cell cycle initiation as part of the death process. Research on models of target-related cell death in the mouse cerebellum⁵¹ has revealed that neuronal death is preceded by re-expression of cell cycle markers and DNA synthesis in vivo. These correlations are strong evidence that a neuron must regulate its cell cycle or risk death during postmitotic maturation.

The use of in vitro models of CRND has been invaluable for establishing the causative nature of cell cycle induction in the process of cell death. Drugs that block cell cycle advance can prevent the death of both PC12 cells and sympathetic neurons after either NGF withdrawal or DNA damage^{52,53,54,55,56}. Similarly, various insults including trophic factor withdrawal, excitotoxicity and toxic concentrations of amyloid-β can drive a lethal cell cycle in cultured CNS neurons and, critically, blocking the cell cycle can prevent neuronal death^57,58,59,60. In sympathetic neurons, trophic factor withdrawal or DNA damage leads to an E2F1-dependent upregulation of B-myb and C-myb (transcription factors that are believed to have a role in S phase progression)⁶¹. This results in the expression of BIM (a BCL2 homology three domain-only molecule)⁶², which in turn leads to cell death. In activity-deprived cerebellar granule cells, phosphorylation of the proapoptotic molecule BAD by CDK1 appears to have a similar role^63,64.

Thus, both in vivo and in vitro, several lines of evidence converge on the following hypothesis: 'once a CNS neuron leaves the VZ, its cell cycle must be actively held in check. Relaxation of that vigilance leads to cell cycle initiation and death, which can follow within hours'.

Loose ends. At first glance Rb-deficient embryos seem to tell a simple story of failed cell cycle suppression leading quickly to cell death. However, it is not clear exactly how RB deficiency leads to neuron loss. First, the phenotype of these mice varies among brain regions, indicating that there is regional heterogeneity in RB dependence. Second, the lack of RB function in a neuron is not, by itself, sufficient to cause neuronal death. Rb^−/− neurons can survive in the mixed environment of an Rb^−/−↔Rb^+/+ chimeric brain^65,66,67, and conditional knockouts of Rb^68,69also support this idea. This is not simply a near-neighbour effect, as the CNS phenotype is rescued in Rb^−/−↔euploid Rb^+/+; tetraploid chimeras^69,70, in which Rb^−/− embryos develop in a conceptus that includes wild-type (Rb^+/+) extra-embryonic tissues. In the brains of the resulting animals, BrdU incorporation reveals that mutant nerve cells can begin a cell cycle but not die. Thus, failure of cell cycle suppression is not sufficient to lead to CRND by itself, although it might be necessary. Dissociation of the loss of cell cycle suppression from the process of neuronal death suggests a caveat to the final clause of the hypothesis — that death follows quickly after cycle initiation — a concept that is developed more fully by considering the situation in the adult brain.

Failure of cell cycle arrest in the adult

Our hypothesis predicts that the prohibition of neuronal cell division is life-long. Might examples of late-onset neurodegenerative disease be accompanied by loss of neuronal cell cycle control? This concept was first proposed to explain the presence of a unique species of phosphorylated tau protein, usually found only in dividing cells, in the neurons of patients who had died with Alzheimer's disease (AD)^71,72. A number of laboratories have reported the re-expression of various cell cycle proteins in neurons from patients with AD. The proteins include cyclins A⁷³, B^73,74,75,76, D^74,76,77,78 and E^73,78, as well as CDKs^71,77,79,80, PCNA^73,74, Ki67 (Refs 73, 81 and CKIs of both the Ink and Cip/Kip families^77,79,82,83. CDK1 may be particularly important in this group because of the tissue culture data mentioned previously^63,64, and the recent evidence of genetic linkage to AD^84,85. There are also reports of cell cycle protein re-expression in amyotrophic lateral sclerosis^86,87,88, ataxia telangiectasia⁸⁹, Parkinson's disease^90,91,92,93, stroke⁹⁴ and other neurodegenerative conditions^95,96. Where quantification has been carried out, it seems that around 10% of at-risk neurons re-express these proteins.

These findings represent correlations between elevated protein expression levels and disease, not proof of CRND. In fact, if the immunocytochemistry is offering a correct picture, it seems likely that cell cycle proteins from different phases are co-expressed and that their location is frequently cytoplasmic. In this context, two studies^75,76 are particularly noteworthy. One shows that cells that can be immunostained with antibodies to cell cycle proteins, when measured as a percentage of neurons in a region, are as prevalent during early stages of AD as they are during end-stage illness⁷⁶. The other study investigated whether these proteins initiate a true cell cycle process, complete with DNA replication, by probing the interphase nuclei of neurons in the brains of patients with AD using fluorescent in situ hybridization (FISH) probes against unique genomic loci⁷⁵. In at-risk regions of the brain, around 4% of the neurons had three or four spots of hybridization rather than two. Thus, the cell cycle protein expression in AD correlates with DNA replication, suggesting that at least the first phases of a true cell cycle have begun in the at-risk neurons. One clear implication of this is that in the adult, the presence of cell cycle proteins or even DNA replication on its own cannot be taken as evidence of adult neurogenesis. These events could easily represent the beginning of a cell death process with very much the opposite result.

Loose ends. Cell cycle regulation in the adult neuron is an area of study in which the number of loose ends vastly exceeds the clearly-established parts of the story. First, the cells never enter M phase. Condensed chromosomes are not found and there is no visible evidence of spindle formation. This makes it difficult to confidently characterize the process as a cell cycle as it is never completed. Second, where it has been quantified, there are too many cycling nerve cells^74,75,76. The average course of AD is 10 years from first symptoms to death. If, as the immunostaining suggests, 5–10% of the neurons are dying at any one moment, and if a typical apoptotic process takes roughly 12 hours, then half of the affected population should be dead in a week or less and 95% should be dead in less than a month. This is clearly not the case. The implication is that death by cell cycle in adult neurons must be a very slow process requiring in the order of 6–12 months. Although this protracted time period is unexpected, mouse models of late onset human neurodegenerative diseases indicate that, if anything, 6–12 months might be an underestimate of the true length of the process as discussed in the next examples.

Transgenic mouse models of human inherited diseases, including ataxia telangiectasia^97,98,99 and AD^{100,101,102,103,104,105,106}, are noteworthy because, with few exceptions¹⁰⁷, an otherwise excellent phenocopy falls short owing to the absence of neurodegeneration. The gene that is mutated in ataxia telangiectasia (Atm) signals the presence of DSBs and arrests the cell cycle to allow for DNA repair. In all reported mouse knockouts, the sterility, immune deficiency and radiosensitivity associated with the human condition are replicated faithfully, but their gross behaviour is normal and there is no detectable loss of Purkinje cells¹⁰⁸. A similar effect is found in mouse models of AD: amyloid-β plaques appear along with behavioural abnormalities, but there is no apparent loss of neurons¹⁰⁹. Remarkably, however, initiation of ectopic cell cycling is preserved. In Atm-knockout mice, cell cycle initiation can be found in Purkinje cells and striatal neurons⁸⁹. Similarly, in the R1.40 yeast artificial chromosome transgenic line, an AD model, cycle anomalies begin at 6 months after birth in regions analogous to those in which the human disease begins, progressively appearing in the brainstem and cortical regions as in AD¹¹⁰. Two groups examined the APP23 mouse, another AD model, but disagreed on the presence of neuronal cell cycle events^110,111. The difference may be in the age of the mice examined; however, the controversy remains unresolved. With the exception of this one report, a long delay between cell cycle initiation and cell death is apparent in these AD mice. Although R1.40 animals can live for two years after cell cycle events first appear, there is no apparent decrease in neuronal density and no documented loss of neurons.

The combined studies in humans and mice suggest a modification of the hypothesis articulated above. 'Once a CNS neuron leaves the VZ or SVZ, its cell cycle must be actively held in check. Relaxation of that vigilance leads to cell cycle initiation and death, which can follow within hours in newly generated neurons. Once neurons are fully mature, however, neuronal death by cycle re-initiation can take from months to years, and might require an additional stimulus to make the transition from cycle to death'.

This distinction between young and old neurons is illustrated in Fig. 3. The dashed arrows in this diagram reflect the possible need for an extra stimulus to trigger death in a cycling adult neuron.

Figure 3: Mature and immature neurons differ in their response to cell cycle initiation.

Healthy (yellow) immature neurons that are deprived of trophic factors or are exposed to other stimuli begin a cell cycle process that rapidly leads to cell death. Mature neurons, however, seem to require a two step process for cell death to occur in response to cell cycle re-entry. Under stress, for example, a healthy adult neuron can enter a cycle and proceed all or part of the way through S phase, but the cycle then stops just before M phase. These cells are in an unusual cell biological state (green), having twice the normal DNA complement (4n), but otherwise they seem normal. The dashed arrows leading through an intermediate stage marked by a question mark indicate the need for a second step that moves the nerve cell from this unusual state to death (grey cell).

What is a cell cycle protein?

Proteins identified as cell cycle proteins turn up in unlikely locations at unlikely times, to serve unlikely functions in nerve cells that, having left the VZ and SVZ, are supposed to have no further relationship with the cell cycle. One example of this is the Rb gene itself. Neurons of the embryonic Rb-deficient nervous system not only fail to survive, they also fail to differentiate²⁰. That would appear to define an embryonic time during which the protein functions, but RB — along with its binding partner E2F1 — are both found in the cytoplasm and dendrites of the supposedly postmitotic neuron and both respond rapidly to exogenous stressors by changing their normally nuclear localization to a predominantly cytoplasmic one^112,113,114. CKIs also have an important non-cycle role in differentiation^{115,116,117,118,119}, which is independent of their role in cell cycle suppression^117,119.

A further example of this cycle/differentiation duality is the 'atypical' cyclin-dependent kinase, CDK5 (Box 2). CDK5 has both pro-differentiation and anti-cell cycle activities during development¹²⁰. In the adult, however, the kinase 'diversifies'. In many if not most CNS synapses, CDK5 forms complexes with¹²¹ and affects the function of ^{122,123,124,125,126,127}various synaptic proteins.

Perhaps the most surprising example of the promiscuous function of cell cycle proteins is that members of the origin recognition complex (ORC) are found in synaptic fractions of adult neurons¹²⁸. This is remarkable because the ORCs are a family of proteins best known for binding to specific sites in the genome during the process of mitotic DNA replication. The synaptic function of these proteins is unknown, although in other cell types they are reported to form complexes with proteins other than those of the ORC¹²⁹.

Finally, a substantial fraction of neurons in the adult E2f1-knockout mouse are engaged in a strange cell cycle-like process (L. Wang and K.H., unpublished observations). They express high levels of cyclin A (cytoplasmic) and PCNA (nuclear), and by FISH criteria they have undergone DNA replication. Taken together, the data suggest that proteins best known for their role in cell cycle progression have separate functions in the fully differentiated, mitotically inactive neuron.

Are adult cycling neurons 'undead'?

As described above, neurons under stress can engage a cell-cycle-like process that leads to the constitutive expression of proteins normally found only in cycling cells and the acquisition of hyper-diploid DNA. Despite these oddities, cycling cells can persist in the brain for long periods. The dendrites of these neurons show only minimal atrophy⁸⁹, but the behavioural anomalies found in the mouse models of both AD and ataxia telangiectasia hint at functional impairments. The question posed by this situation is: what is the nature of the state in which these nerve cells find themselves? Data from embryonic studies indicate that cell cycle initiation in a post-VZ neuron is a first step towards cell death. However, data from adult studies suggest that the processes of cell cycle initiation and cell death are separable. Similar situations, of neurons on a path to death being blocked by genetic or trophic means, have been reviewed recently for several species and have raised many important issues¹³⁰.

Speculatively, it may be possible to draw an analogy between the arrested cell cycle state of these neurons and a cellular state defined in the imaginal discs of Drosophila melanogaster. The state was first proposed¹³¹ to describe cells that are triggered to die by an insult such as X-rays but blocked from completing the process by expression of an anti-apoptotic protein. The authors describe the resulting cells as 'undead' and have shown that they have some unusual properties^131,132,133. For example, they send signals that serve as a mitogen to neighbouring cells. If the neighbouring cells are wild type, they divide in an apparent attempt to fill the gap in the tissue expected to be left by the dying cells.

Could the cycle-positive neurons in AD and ataxia telangiectasia brains be analogous to these undead cells, triggered to die but blocked from completing the process? If so, these cells might be part of the problem rather than innocent victims of the disease process. If in their undead state they are sending mitogenic signals to their neuronal neighbours, they could be pressuring otherwise healthy neurons to initiate a cell cycle (Fig. 4). If this is true, their stability and persistence could represent a danger to the health of the brain. As they do not die, their continued presence might weaken those around them and yet, because re-entry into a cell cycle does not seem to result in cell death, not kill them. This would create a paradox for those who seek therapeutic interventions for neurodegenerative conditions such as AD. The prediction would be that the most effective therapies would block new neurons from entering a cell cycle while simultaneously ridding the brain of the undead cells.

Figure 4: Speculative model of how a 'cycling' neuron might be a threat to the health of the CNS.

Analogous to the concept of 'undead' neurons in the fruitfly imaginal disc, the figure illustrates a situation where cycling but living neurons pose a problem for the remaining cells in the neighbouring brain region. If, as in the fly, the undead (green) cells are capable of releasing a mitogenic signal (arrows), then a situation arises in which the victim may become the assailant. The transformation of a mature neuron into this unusual state would actually enhance the local 'mitotic pressure', possibly leading other neurons to enter this 'undead' state or to enter an apoptotic process (grey nucleus) and die (grey cell).

Conclusions and future directions

In many ways, descriptions of adult CNS neurons as having 'permanently exited the cell cycle' or having entered 'a permanent state of G0' are terms of convenience, which may lack true functional meaning. In fact, such terms might lull us into a false sense of security — a belief that the issue of cell cycle regulation in neurons is an oxymoron rather than an important biological question. We believe that the evidence reviewed here challenges us to rethink this position.

For billions of years, the prevailing entities (which ultimately became cells) were those that divided fastest and most efficiently. Metazoans evolved when aggregates of these early cells were able to realize gains from cooperation and specialization that exceeded those they could eke out by living on their own. It is worth considering that by this time the 'instinct' to divide was buried deep in the fundamental biology of every cell. It may even have preceded the emergence of a genome. Thus, the dividing instinct might be as powerful today as it was a billion years ago. The challenge faced by a metazoan, therefore, is to suppress the urge to cycle among its own constituents in order to enforce a regulated body plan and allow the extreme specialization (differentiation) of its members. Indeed, suppressing the cell cycle is arguably the most important step in any differentiation programme.

Neurons are among the most specialized of cells, and organisms that benefit from incorporating them into their structure seem to have largely solved the problem of reining in their drive to divide; the nature of the solution(s), however, remains largely unknown. But we may wish to rephrase questions such as what factors force a postmitotic neuron to re-enter a cell cycle in AD or what fail-safe mechanisms of cell cycle suppression are lost in AD such that pockets of neurons can break free of their postmitotic state and return to an earlier mode of existence. For neurons of the CNS it seems that anti-cycling kinases, cell cycle promoters turned into cell cycle suppressors, and DSBs during or immediately after the final cell division are all components of a strategy by which the metazoan developmental plan imposes cell cycle arrest on its neurons. This multi-tiered strategy involves not only changes in protein localization and altered biochemical networks, but perhaps rearrangements of DNA structure as well. The evolution of multicellular organisms appears to have been a stern mitotic master.

But evolution is also a miserly process. It tends to adapt old tools for new uses rather than fashion new tools. In neurons, this is reflected by the additional roles of cell cycle proteins such as E2F1, RB, CDKs, cyclins and ORC proteins in dendrites and axons. Under normal conditions these functions exemplify efficient design. Under stress, however, they appear to represent an Achilles' heel for the neuron. While some may start off as protective (for example, p27 and CDK5), many can be dangerous to the health of the cell if they end up in the wrong cellular location.

If cell cycle regulation is truly a constant issue for an adult neuron then an entire menu of action items seems spread before us. One important initiative would be to drop, whenever possible, the reference to proteins as cell cycle proteins. Each protein whose first known function was to advance or retard the progression of the cell cycle appears to have alternative identities in neurons and other differentiated cells. Restricting our focus to their role in cell cycle regulation may blind us to the full repertoire of their activities. We also need to remind ourselves that the mechanics of cell cycle regulation can be distinct in different cell types. Textbooks tell us that the E2F1 protein drives the cell cycle by upregulating the synthesis of key proteins needed for cell cycle progression, as is indeed the case in cultured fibroblasts and COS cells; however, E2f1-knockout animals are born and develop normally, but die in middle age with tumours and hyperplasias in a variety of exocrine tissues^134,135. Clearly, in some cells E2F1 functions as a tumour suppressor. Finally, it seems apparent that, just as in real estate, the three keys to activity are location, location and location. We need to define where these multifunctional proteins are in neurons at different junctures of the developmental process and during the response to insult or injury. The behaviour of CDK5 in CRND suggests that its function in the nucleus is different from that in the cytoplasm. This is not a new insight, but it is part of a readjustment to the notion that we can avoid concerning ourselves with the regulation of the cell cycle in all highly differentiated cells — especially neurons. Indeed, to reach a fuller understanding of how neurons remain mitotically silent for decades, there is a great deal of new biology yet to learn.

Box 1 | Basics of the cell cycle

Cell cycle phases

G1 (Gap or Growth 1) phase. The period of growth preceding any commitment to division. G0 is a specialized form of G1. Highly differentiated cells, which are unlikely to divide unless provoked, are said to be in G0. During late G1 the cell commits to divide, typically marked by the phosphorylation of the tumour suppressor gene retinoblastoma (Rb), and the activation of the transcription factor E2F1, which in association with its partner DP1 binds to the promoters of various cell cycle genes.

S phase. The period of DNA synthesis. The process of replication of the entire genome is completed, resulting in the doubling of the DNA content of the cell.

G2 (Gap or Growth 2) phase. A period during which the mechanical components that will organize the chromosomes and physically divide the cell are assembled.

M (Mitosis) phase. An exquisitely complex cytological event that accurately divides the duplicated sets of chromosomes and coordinates the events of fission to produce two cells from one. Many events take place during M phase including the dephosphorylation of RB. Upon completion of M phase, both daughter cells re-enter G1 and a new cell cycle is set to begin.

Cell cycle proteins

Proteins that drive or inhibit the cell cycle are shown in grey or pink boxes, respectively.

Cyclins. Activator proteins that are up- or downregulated depending on the phase of the cell cycle.

Cyclin-dependent kinases (CDKs). Serine/threonine kinases that require the binding of a cyclin (or related protein) for full activity. Their range of substrates is not fully defined, but interfering with their activity arrests or slows the cycle.

Cyclin-dependent kinase inhibitors (CKIs). Small peptides that block cyclin/CDK activity either by forming an inactive complex or by acting as a competitive CDK ligand.

DNA replication proteins. DNA polymerases and associated proteins such as proliferating cell nuclear antigen (PCNA) and mini-chromosome maintenance (MCM) proteins, as well as proteins that assure that each origin of replication initiates replication only once per cycle. These include origin recognition complex (ORC) proteins, CDT1 and its suppressor, geminin.

Checkpoint proteins. Members of a network of proteins that monitor DNA integrity and arrest the cell cycle until DNA damage can be repaired^136,137,138.

Box 2 | An atypical CDK

Cyclin-dependent kinase 5 (CDK5) is structurally similar to other CDKs, but is said to have no role in the normal cell cycle. It binds cyclin D, but its activity is not initiated by this partnership. Rather, two proteins known as p35 and p39 function as cyclin equivalents and activate CDK5. Mice that are genetically deficient in either CDK5 or both activator proteins have serious defects in neuronal migration and die before birth^139,140,141. Intermediate filament expression in CDK5-deficient neurons in vivo and in vitro is retarded or blocked¹²⁰. Early filament types such as nestin and βIII-tubulin remain expressed in regions of Cdk5^−/− cortex from which they are lost in the wild type, and later differentiation markers such as microtubule-associated protein 2c are seen only at very low levels¹²⁰.

In addition to its pro-differentiation functions, CDK5 also appears to act as a cell cycle inhibitor. Both in vivo and in vitro, Cdk5^−/− neurons cannot restrain their cell cycle and die rapidly after the re-expression of cell cycle proteins and the incorporation of 5-bromo-2′-deoxyuridine¹²⁰. In vitro models of cycle-related neuronal death suggest that the 'anti-cycle' activity of CDK5 requires nuclear localization (Cicero, S.A. et al. unpublished observations). Whereas in other cell death models the localization of CDK5 to the nucleus is protective¹⁴², it would appear that CDK5 is very much involved in cell cycle regulation in neurons.