Autumn senescence, like other forms of leaf senescence, is a type of programmed cell death, i.e. the leaf cells die in an organized, predetermined way controlled by the nucleus. On the other hand, there are fundamental differences between the leaf senescence program and apoptotic processes, which have been studied in great detail in animal systems and more recently in plants (Noodén et al., 1997; Kuriyama and Fukuda, 2002). In fact, it has even been speculated that the processes are mutually antagonistic (Ougham et al., 2005) since an appropriately executed senescence program avoids the pathological consequences of cell death (Hörtensteiner, 2004). Senescence has adaptive value because of the associated remobilization of nutrients, especially nitrogen, and, to a lesser extent, phosphorus, sulfur, and other elements (Himelblau and Amasino, 2001; Hörtensteiner and Feller, 2002). The timing of autumn senescence can be regarded as the result of a trade-off between the conflicting requirements for optimizing the nitrogen and carbon status of the plant. Trees entering senescence early will efficiently remobilize nitrogen at the expense of photosynthetic yield, while trees entering senescence late will gain more photosynthates, but in some years their leaves will die before the remobilization of their nutrients is complete. Nitrogen status influences the onset of autumn senescence. For instance, alders (Alnus subsp.) that host a nitrogen-fixing symbiont (Frankia) shed their leaves while they are still green, and gardeners in temperate regions know that nitrogen-rich fertilization should be avoided since it delays both autumn senescence and the development of hardiness, thereby compromising winter survival (especially of woody perennials).

Autumn senescence has not been well characterized at the cellular and molecular levels, but more information is available regarding leaf senescence in annuals (Buchanan-Wollaston, 1997; Quirino et al., 2000). The chloroplasts, present in green leaves, differentiate into gerontoplasts that lack stacked thylakoid membranes but are rich in electron-dense lipid bodies, plastoglobuli, which have an especially high content of carotenoids and carotenoid esters (Tevini and Steinmuller, 1985). Nutrient remobilization requires the breakdown of macromolecules and conversion of the breakdown products into transportable compounds. Many genes coding for catabolic enzymes (proteases, lipases, nucleases, etc.) are among those that are induced during senescence (Bhalerao et al., 2003; Buchanan-Wollaston et al., 2003; Andersson et al., 2004; Guo et al., 2004). When the photosynthetic capacity of the leaves is lost, the energy required for the remobilization must be provided by mitochondrial respiration and, consequently, nuclear genes coding for components of the mitochondrial electron transport chain are not down-regulated during autumn senescence, in contrast to genes coding for components of the photosynthetic apparatus (Buchanan-Wollaston, 1997; Andersson et al., 2004). Although the leaves do not perform photosynthesis, phloem transport out of the leaf is needed to export the nutrients so, from a transport perspective, senescing leaves are still source leaves. Eventually, however, phloem transport stops, a protective layer is formed on the inner side of the petiole, a separation or abscission layer is formed proximal to it, and when the cell walls in the separation layer are gradually loosened, the leaf eventually falls (Roberts et al., 2002). It is well known that autumn senescence in most trees is triggered by reductions in the photoperiod, a signal that is more reliable than temperature as a harbinger of the first strong frost, allowing autumn senescence calendars to be constructed that are quite constant over the years (e.g. www.great-lakes.net/tourism/fallcolor.html; www.arkansas.com/calendar/fall_foliage_pg1.asp). The pivotal role of phytochromes in the process has also been well documented (Olsen et al., 1997; Chen et al., 2002), but the extent to which low temperatures, perhaps coupled to light intensities inducing photooxidative stress, modulate the process is not clear. It is common knowledge that cold autumns result in intense leaf coloration, but the physiological basis of this color change is not well understood.

We have started a project to elucidate the genetic basis of autumn senescence in aspen (Populus tremula), and as part of the project we have identified many genes that are expressed in autumn leaves (Bhalerao et al., 2003) and constructed a transcriptional timetable for autumn senescence (Andersson et al., 2004). However, studies of gene expression alone do not always provide sufficient information for understanding the cellular events that occur during the senescence process, especially since not only anabolism, but also gene expression, ceases during the later stages of autumn senescence (Andersson et al., 2004). Consequently, catabolic processes leading to the degradation of cellular components and, eventually, cell death should perhaps be studied at the cell biological level rather than the transcriptional level.

By constructing the cellular timetable of autumn senescence for an aspen tree, we believe that we have obtained novel insights into some of the major unresolved questions concerning autumn senescence. We have shown that the entry into phase 2 was triggered by the photoperiod and that temperature (and light) had little or no effect on the onset of senescence. The autumn senescence process, once initiated by environmental factors, seems to be a tightly controlled developmental program that is not significantly altered by environmental factors. Chlorophyll was degraded at a fairly constant rate, although weather conditions fluctuated (and differed between years). If photooxidation of pigments had been a major determinant, the degradation rate would have been increased by cold clear days (like September 19 and 20), for example, which was not the case (Kukavica and Jovanovic, 2004). Chlorophyll was degraded at a constant rate, and carotenoid catabolites started to appear after September 12, before gerontoplast formation was observed. Photooxidative stress, on the other hand, seemed to be the major determinant of anthocyanin accumulation, and anthocyanin levels changed dynamically depending on the weather. The stimulation of autumn colors by cold weather seemed therefore to be solely due to effects on anthocyanin accumulation. Anthocyanin accumulation seemed to correlate better with excess light (high light and cold) than with either high light or cold alone, consistent with the hypothesis that anthocyanins have a photoprotective role. The large fluctuations in anthocyanin content showed that anthocyanins were rapidly metabolized right up to leaf abscission. Anthocyanin catabolism may be light regulated or, alternatively, regulated anabolism in aspen leaves could compete with a seemingly unregulated catabolism, similar to the regulation of the xanthophyll cycle. Interestingly, leaves in phase 1 did not seem to be as competent to respond to the stimuli that later led to anthocyanin accumulation. September 1 and 2 were as cold as September 19, and the light was brighter, but no anthocyanin accumulation was induced at this time. These observations, together with the fact that damaged PSII centers were also repaired during the first week of September, demonstrate that the leaves at this point were not yet ready to enter the senescence process. This is consistent with our previous suggestion that a peak in transcriptional activity in the first week of September coincides with a developmental switch reprogramming the leaf to senescence (Andersson et al., 2004).

On September 11, the degradation process started (phase 2). This date for entry into phase 2 was the same over several years, with different weather conditions, and is therefore likely to be controlled by the photoperiod. A senescence signal triggered the cells to degrade their photosystems (and chlorophyll), thereby releasing carotenoids, some of which (lutein and β-carotene) were degraded, whereas others (particularly neoxanthin and violaxanthin) were esterified and accumulated in the plastoglobuli. This process seemed, within each plastid, to be very rapid, since the F_v/F_m ratios stayed high and constant until very late in the process. If large proportions of PSII centers present in the chloroplasts had been undergoing degradation processes, this would probably have been manifested by reductions in the overall F_v/F_m values. Apparently, therefore, although total photosynthetic capacity decreases along with the chlorophyll degradation, the photosynthetic parameters of the remaining PSII centers stay constant until very late in the process. We interpret these data as indicating that the transition from a photosynthetically fully active chloroplast to a gerontoplast is very rapid, and these data are consistent with a model in which all of the chloroplasts in a given cell undergo the transition simultaneously, while a neighboring cell could be in a different developmental stage. Changes in the chlorophyll a/b ratio showed that the degradation of antenna chlorophylls at the end of phase 2 is slower than that of chlorophyll a in the reaction centers. This resembles somewhat the situation in some stay-green mutants, where the remaining chlorophyll in otherwise senescing leaves is bound to light-harvesting complex II (Hilditch et al., 1986). The adaptive benefit of keeping the antenna and dismantling the reaction centers is not clear, but has been demonstrated during leaf senescence in other species (Wolf, 1956; Lichtenthaler, 1987). In parallel with the chloroplast degeneration, expression of the photosynthetic genes was strongly repressed (Andersson et al., 2004). After entry into phase 2, the cytoplasm started to be degraded, eventually making it difficult to distinguish cytoplasm from the vacuole. We could detect small vesicles, perhaps similar to the vesicles reported by Otegui et al. (2005), nuclei gradually disappeared, and the cells seem to largely stop synthesizing protein at this point; at least we have not been able to extract high-quality RNA from the leaves after September 24 (Andersson et al., 2004).

During the leaf senescence process, most of the nitrogen and phosphorus was retrieved. The relative content of carbon was not changed (Fig. 7A), but since the leaf dry weight decreased (Fig. 2B) some carbon may also be retained. This massive remobilization requires, of course, the involvement of many catabolic enzymes, and genes encoding various types of proteases, lipases, and enzymes involved in nucleotide metabolism, glyconeogenesis, and nitrogen remobilization (inter alia) are either induced, or continue to be expressed, in the autumn leaves (Bhalerao et al., 2003; Andersson et al., 2004; Terce-Laforgue et al., 2004).

Senescence is an energy-requiring process in which ATP is needed for degradation, recycling, and transporting nutrients out of the cell. As photosynthesis declines in phase 2, mitochondrial respiration needs to take over to provide the cell with the energy required for these processes. Thus, a significant fraction of the carbon lost from the leaf is likely to be due to respiration and not to be retrieved. We did not measure respiration during the process, but it is clear that mitochondria stay active throughout senescence since intact mitochondria could be seen in electron microscopy images late into the process. Genes encoding components of mitochondrial electron transport also continue to be expressed throughout the period (Andersson et al., 2004). By monitoring changes in the levels of key metabolites (ATP/ADP ratios and pool size, starch, and various soluble sugars), we found evidence indicating that a shift in the main energy sources of the cell from chloroplasts to mitochondria probably occurred around September 19.

On September 30, the degradation process was largely complete and the leaves had entered phase 3. It is likely that this transition coincides with the formation of the protection and separation layers in the petiole, which blocks further phloem transport out of the leaf. The changes after this date were minor, and since leaves started to be shed and individual leaves in which the process was somewhat retarded were more likely to be withheld, this may explain the weakness of some trends in the data and suggests that our separation between phase 3 and phase 4 may be artificial. Some cells seemed at this stage to be empty (and probably dead), whereas many contained nuclei, mitochondria, and a small number of gerontoplasts. Vacuoles and cytoplasm could not be distinguished from each other. The dynamic changes in variables, such as the levels of xanthophyll cycle pigments and anthocyanins, and the presence of a few active photosystems during phase 3 suggest that a significant fraction of the cells were still alive and metabolically active. Many of these cells contained chlorophyll-containing plastids, and the residual chlorophyll level was approximately 5% of the phase 1 levels, but only a small fraction of this remaining chlorophyll seemed to be photosynthetically active, and this activity was lost during phase 3. The residual chlorophyll is likely to be localized in the plastoglobuli, where the high amounts of carotenoids could serve as efficient quenchers of excess excitation energy.

The sudden shift in the nitrogen isotope ratio at the very end of phase 2 (September 29) was an unexpected finding. Changes in the nitrogen isotope ratios of senescing pine needles have previously been reported, and three hypotheses have been put forward to explain them (Näsholm, 1994). Since two of the hypotheses (discrimination against δ¹⁵N in the process of nitrogen translocation and the possibility that nitrogen pools that are not degraded, like cell wall proteins, have relatively high δ¹⁵N values) are incompatible with such a dramatic shift in the ratio; the third hypothesis, the emission of nitrogen-containing volatile compounds (low in δ¹⁵N), may be the best explanation for our findings. Emission of volatiles from senescing leaves has been reported (Husted et al., 1996), and our finding that the sudden change in isotope ratio coincided with the transition from phase 2 to phase 3, and probably with the termination of phloem transport out of the leaves, is compatible with a hypothesis that nitrogen-containing compounds may accumulate in the leaf to toxic levels, since protein catabolism probably continues so ammonia may be released to reduce the amount of excess nitrogen. It is possible that this sudden change in δ¹⁵N is a signature of the end of the remobilization process in autumn leaves.

There are two possible explanations for the retention of some cells and organelles throughout phase 3. We believe that, by the end of phase 2, the tree has gone through the essential steps in the senescence process, enabling nutrient retrieval to occur, and that a separation layer has formed, so very little further remobilization is possible. However, if there are leaves that are still able to export some material, catabolic activities will be required for the remobilization. The accumulation of plastids, probably an indicator of living cells, closer to the veins is compatible with the hypothesis that, during phase 3 (and 4), the cells located far away from the veins may die, but that cells closer to the veins could participate in the degradation of macromolecules. The tree may be prepared for the winter by the end of phase 2, but if no frost hard enough to kill the cells occurs, the tree gets a bonus phase in which a small fraction of the remaining nutrients could perhaps be remobilized. The second possible explanation is that if the leaf cells have functionally separated from the trunk, their fate will not influence the rest of the tree so they may live or die without any particular adaptive benefit. In any case, these data are compatible with the concept that leaf senescence and cell death are to some extent antagonistic processes.

Our data show that anthocyanin accumulation is likely to have a role in photoprotection, as suggested by several authors (Feild et al., 2001; Hoch et al., 2003). In the ecological literature, there has recently been speculation about a possible connection between autumn senescence and insect herbivory (Archetti, 2000; Hagen et al., 2003), assuming that autumn colors may be an honest, or dishonest, signal to insects about the status of the tree. Strong arguments against this, from a physiological perspective, have been put forward by Ougham et al. (2005) and, given the fact that accumulation and degradation of pigments in autumn leaves can be explained by obvious physiological and developmental factors, we think that there is unlikely to be a direct link between autumn colors and herbivore preference, and instead the phenomena are highly correlated because both are highly dependent on the nitrogen status of the tree.

Is there a point of no return in autumn senescence of aspen leaves? To our knowledge, this question has not been experimentally addressed, but considering the events that we have followed here, one could speculate that the reprogramming of gene expression late in phase 1 could be an irreversible process or, alternatively, that the leaves could re-green if the whole tree is exposed to a longer photoperiod. However, it is doubtful that such a capability would have adaptive significance in nature, since trees will never experience an increase in the photoperiod during the autumn. However, we now have the tools to also address this question.

By detailed studies of a tree during autumn senescence, we believe that we have obtained novel insights into the fundamental questions posed in the introduction. The conversion of chloroplasts to gerontoplasts is driven by developmental factors, the nutrients are largely remobilized over a period of 18 d (Sept 11 to Sept 29, phase 2), and, in the middle of this phase, mitochondria take over as the main providers of energy for the process. During this phase, plastids are converted to gerontoplasts and many cells die, but some survive and retain their gerontoplasts, mitochondria, and perhaps also nuclei until the leaves abscise, up to 1 week later.