Control of Starch Granule Numbers in Arabidopsis Chloroplasts 1[W][OA]

The aim of this work was to investigate starch granule numbers in Arabidopsis ( Arabidopsis thaliana ) leaves. Lack of quantitative information on the extent of genetic, temporal, developmental, and environmental variation in granule numbers is an important limitation in understanding control of starch degradation and the mechanism of granule initiation. Two methods were developed for reliable estimation of numbers of granules per chloroplast. First, direct measurements were made on large series of consecutive sections of mesophyll tissue obtained by focused ion beam-scanning electron microscopy. Second, average numbers were calculated from the starch contents of leaves and chloroplasts and estimates of granule mass based on granule dimensions. Examination of wild-type plants and accumulation and regulation of chloroplast ( arc ) mutants with few, large chloroplasts provided the following new insights. There is wide variation in chloroplast volumes in cells of wild-type leaves. Granule numbers per chloroplast are correlated with chloroplast volume, i.e. large chloroplasts have more granules than small chloroplasts. Mature leaves of wild-type plants and arc mutants have approximately the same number of granules per unit volume of stroma, regardless of the size and number of chloroplasts per cell. Granule numbers per unit volume of stroma are also relatively constant in immature leaves but are greater than in mature leaves. Granule initiation occurs as chloroplasts divide in immature leaves, but relatively little initiation occurs in mature leaves. Changes in leaf starch content over the diurnal cycle are largely brought about by changes in the volume of a ﬁxed number of granules.

Chloroplasts in Arabidopsis (Arabidopsis thaliana) mesophyll cells are generally stated to contain about five starch granules at the end of the light period (Zeeman et al., 2002(Zeeman et al., , 2007, but nothing is known about how this number is determined or the extent of genetic, temporal, developmental, and environmental variation in the number. This is an important limitation in understanding control of starch degradation at night, a process essential for the normal growth of the plant (Gibon et al., 2006;Smith and Stitt, 2007;Stitt et al., 2007;Usadel et al., 2008). The surface area and volume of starch granules in the chloroplast are relevant to the control of starch degradation in two ways. First, surface area can potentially limit the rate of degradation during the night. This limitation is not a major determinant of the rate of degradation in wild-type Arabidopsis leaves in controlled conditions; degradation is near linear through most of the night and consumes almost all of the starch reserves by dawn. If degradation were limited by surface area, the rate would decline with time through the night. However, reductions in starch granule numbers may result in limitation of starch degradation. The starch synthase4 (ss4) mutant of Arabidopsis has only one starch granule per chloroplast. It has a low rate of starch degradation and a low growth rate, consistent with the idea that the reduced surface area provided by one rather than five starch granules restricts the rate of degradation and hence the availability of carbon for growth at night (Roldán et al., 2007).
Second, recent experiments implicate starch granules in the mechanism that sets the rate of starch degradation at the start of the night. When Arabidopsis plants grown in 12 h of light and 12 h of dark were subjected to darkness after only 8 h of light, the rate of starch degradation was one-half that on previous nights, so that starch reserves were not exhausted until dawn (Graf et al., 2010). The capacity to adjust the rate in this precise way requires measurement of both the time until dawn (probably provided by the circadian clock) and the amount of starch in the leaf at the onset of darkness. Although the way in which starch content is measured is not known, the presence of starch granules is essential for correct adjustment of the rate of starch degradation. In mutants that accumulate a soluble form of starch (phytoglycogen) in place of starch granules, reserves are degraded in an exponential manner and are exhausted well before dawn (Delatte et al., 2005).
Understanding variation in starch granule numbers in mesophyll chloroplasts is also important in the context of starch granule initiation. Almost nothing is known about how granules are initiated. It is reasonable to assume that initiation accompanies the three rounds of chloroplast division that occur during the expansion phase of mesophyll cell development, and it may also accompany the very large increase in starch content that typically occurs during each light period in mature leaves. However, no quantitative information is available on the times and locations of granule initiation in mesophyll cells. Such information would provide new possibilities for studying the nature and control of initiation.
To shed light on the control of starch granule number in Arabidopsis leaves, we developed two new methods for robust and reproducible estimation of the numbers of granules per chloroplast. The first involves sequential sectioning and imaging of leaves embedded in resin blocks by focused ion beam-scanning electron microscopy (FIB-SEM). The second involves measurement of the starch content of whole leaves and of known numbers of isolated chloroplasts, plus estimation of granule volumes from SEM images. We applied these methods to leaves at different points in the daynight cycle, of different ages, under different growth conditions, and with different chloroplast sizes and numbers per cell.

Choice of Methods for Estimating Starch Granule Numbers per Chloroplast
Most observations and measurements of starch granules in leaves have been made on a transmission electron microscope (TEM) or light microscope images of sections through mesophyll cells. These methods are not suitable for accurate estimation of granule numbers per chloroplast. A single section through a chloroplast is likely to underestimate granule numbers. Serial sectioning through multiple chloroplasts followed by three-dimensional (3D) image reconstruction would reveal the exact number of granules per chloroplast, but this is a time-consuming and technically challenging process unsuitable for straightforward comparison of multiple replicates and conditions. Flow cytometry has been used to estimate the number of starch granules per chloroplast in beans (Phaseolus spp.; Yang et al., 2002). Suspensions of ruptured chloroplasts separated into two peaks of particles, assumed to represent chloroplast membranes and starch granules. However, no direct evidence was provided for these peak assignations. In other studies, intact chloroplasts, ruptured chloroplasts, and isolated thylakoids migrated as three separate peaks (Ashcroft et al., 1986;Petit et al., 1989;Bergounioux et al., 1992;Schröder and Petit, 1992). Because rupture of a significant fraction of originally intact chloroplasts is almost inevitable during flow cy-tometry, the assumptions about peak identity made by Yang et al. (2002) may be incorrect. In addition, the small size of leaf starch granules (1-2 mm in diameter) is likely to prevent clean separation from other small particulate matter in chloroplast preparations.
We developed two methods for estimating starch granule numbers per chloroplast. In method 1, good resolution of granules within chloroplasts is obtained by imaging the surface of a resin block containing leaf material in a SEM. Removal of successive thin sections (typically approximately 50 nm) from the surface of the block with an FIB then allows a 3D image of the leaf tissue to be built up (FIB-SEM; see Armer et al., 2009;Bushby et al., 2011). Typical single images for individual chloroplasts are shown in Figure 1, and example images from which data were derived are shown in Supplemental Figure S1. The method has major advantages of speed, accuracy, and convenience over conventional serial sectioning and TEM. Because the block face is imaged rather than the removed material, there is no distortion of the image surface and alignment of the images is relatively simple. The risk of missing or damaged sections is eliminated, making it possible to collect large numbers of sequential images through a volume of many cubic micrometers. The process is automated, and data collection and processing for multiple samples, each containing numerous chloroplasts, can be performed in a few days. The processed 3D image enables accurate measurements of granule and chloroplast volume as well as granule numbers and analysis of the biological variability of all of these parameters within a cell or a region of the leaf. Major disadvantages are the requirement for specialized and expensive equipment and computing time, the fact that the small tissue samples may not be representative of the whole leaf and the fact that differential shrinkage of cell compartments can occur during sample preparation. We typically imaged blocks of tissue of approximately 50 3 10 3 mm 3 . This volume is comparable with that of a spongy mesophyll cell in a fully expanded Arabidopsis leaf (Wuyts et al., 2010). Loss of cell volume during sample preparation for microscopy is primarily because of vacuolar shrinkage, but chloroplast volume may decrease by about 30% (Winter et al., 1993(Winter et al., , 1994. This is not a major problem when making comparisons between chloroplasts, but it means that estimates of the relationship between starch volume and chloroplast volume contain systematic errors. Method 2 involves preparation of intact chloroplasts from a leaf sample, measurement of the starch content of a known number of chloroplasts, then estimation of the average volume of starch granules in preparations of starch from replicate leaf samples. These data are used to calculate the weight of starch per chloroplast and thus the number of starch granules, using published values for starch density. Calculation of granule volumes employs measurements of granule thickness made on SEM images of granules cracked by freezing a starch slurry in liquid nitrogen then grinding it in a mortar, and a method of estimating granule radii from measurements made on SEM images of intact granules (Supplemental Fig. S2). Method 2 does not require specialized equipment other than a standard SEM, and it can be applied rapidly to many samples. It provides values averaged over numerous leaves. However, it is much less direct than method 1, increasing the possibility of systematic errors, and it assumes that granule density is constant over the day-night cycle. It provides no information about chloroplast volumes or variation in granule numbers per chloroplast within leaves or cells, or differences between mature and developing leaves.
One potential concern for both methods is whether the full range of granule sizes is sampled without bias. In method 1, resolution may be insufficient to detect the smallest granules, leading to underestimation of granule numbers at the end of the night. In method 2, granules may be isolated from tissues other than the mesophyll, and there may be losses of particularly small or large granules. We compared the range of granule sizes detected by the two methods, using replicate samples from a single batch of plants (Table I). Both methods showed that granules isolated at the end of the day were more than 2-fold greater in diameter than those isolated at the end of the night. There was no statistically significant difference between the end-of-night values obtained by the two methods. The end-of-day value obtained by method 1 was 13% higher than that obtained by method 2, and the variance was also greater. Supplemental Figure  S3 compares these data with values obtained from measurements on thin sections imaged by light microscopy. Light microscopy gave values comparable to methods 1 and 2 for granules at the end of the night but lower values for granules at the end of the day.

Granule Numbers over the Light-Dark Cycle
Methods 1 and 2 were used to determine the average number of granules per chloroplast in mature, nonflowering rosettes at the end of the day when starch content is maximal and at the end of the night when starch content is minimal (Table I). At the end of the day, method 1 gave an average value of 6.8, and method 2 gave an average value of 5.5. Given the major differences between the methods themselves and the samples to which they are applied, the agreement between the methods is good. At the end of the night, method 1 gave an average value of 4.0 and method 2 gave an average value of 5.8. Thus, values from method 1 are consistent with a decline in total granule number per chloroplast overnight (from 6.8 to 4.0), whereas values from method 2 suggest that there is no change in granule number per chloroplast overnight. This discrepancy may be in part because of a failure to detect all of the very small granules present at the end of the night in the sections viewed in method 1.

Relationship between Starch Granule Number and Leaf Starch Content
To examine the relationship between granule numbers and leaf starch content, method 2 was applied to leaves with a wide range of starch contents, obtained by harvesting mature, nonflowering rosettes at different times of day and from different light and daylength regimes. Starch content in plants grown with 12 h of light and 12 h of dark increased about 40-fold over the light period (Table II). However, there was no statistically significant change in the number of starch granules per chloroplast over this period. Chloroplasts at the start, middle, and end of the light period contained on average between five and six starch granules. The starch contents of plants harvested in the middle of the light period from different daylength and light intensity regimes showed a 2-fold range of variation, but average granule number per chloroplast varied by only 12% across all of the treatments (Table II). When data from time of day, daylength, and light intensity treatments were considered together, it was apparent that average granule number per chloroplast was essentially constant over a wide range of values of leaf starch content (Table  II). This was reflected in a lack of correlation between the granule number per chloroplast and either chloroplast starch content or granule volume (Fig. 2, A and B) but a strong, positive relationship between granule volume and chloroplast starch content (Fig. 2C). Thus, changes in starch content through the day and in response to different light intensities and daylengths appear to be brought about primarily by changes in the volumes of a constant number of starch granules in each chloroplast, rather than changes in the numbers of granules per chloroplast.

Relationship between Starch Granule Size and Number and Chloroplast Volume in Wild-Type Leaves
Using method 1, we examined the extent of variation in chloroplast volumes and starch granule numbers per chloroplast within individual cells, in both mature (fully expanded) and immature leaves at the end of the day. For mature leaves, we harvested leaf 6 from 25-d-old plants grown with 12 h of light and 12 h of dark (growth stage 1.14; Boyes et al., 2001), where leaf 1 is the first true leaf formed. For immature leaves, we harvested leaves   14, 15, and 16 from the same plants. The extent of starch turnover in mature and immature leaves was similar. In a typical experiment, starch content decreased overnight by 12 mg g 21 fresh weight in a pool of mature leaves (leaves 5-8) and by 9 mg g 21 fresh weight in a pool of immature leaves (leaves 13-16; data not shown).
There was a huge range of variation in chloroplast size in both immature and mature leaves. Measurements of 50 chloroplasts in mesophyll cells from three different plants gave volumes between 38 and 240 mm 23 for mature leaves and volumes between 11 and 129 mm 23 for immature leaves. Within plants, the variation in chloroplast volume was 3.5-, 6.1-, and 2.7-fold for mature leaves and 3.8-, 8.1-, and 6.7-fold for immature leaves. There was also a very large variation in numbers of starch granules per chloroplast. There were between three and 14 granules per chloroplast in mature leaves and between four and 21 granules per chloroplast in immature leaves.
For both mature and immature leaves at the end of the day, chloroplast volume was positively correlated with the number of starch granules and the total volume of starch in the chloroplast (Fig. 3, A and B). Immature and mature leaves differed considerably with respect to the relationship between chloroplast volume and number of starch granules (Fig. 3B) and chloroplast volume and volume of individual granules (Fig. 3C). For a given chloroplast volume, immature leaves had more and smaller starch granules than mature leaves (Fig. 3). On average, immature leaves had more granules per chloroplast, smaller granules (Figs. 3, C and D, and 4), and less stromal volume per starch granule (Fig. 3B) than mature leaves. There was no strong relationship between granule number per chloroplast and the average volume of individual granules for either mature or immature leaves. (Fig. 4). However, the two ages of leaf had similar relationships between chloroplast volume and the total volume of starch per chloroplast (Fig. 3A). In both cases, starch occupied an average of 15% of the chloroplast volume.
For mature leaves, we examined differences between the end of the day and the end of the night in the relationship between chloroplast volume and starch granule numbers and volumes. Average chloroplast volume declined by about 45% during the night. About one-third of this decline was accounted for by starch degradation. The general relationship between chloroplast volume and starch granule volume seen at the end of the day was lost, so that starch contents and the volumes of individual granules were similar in large and small chloroplasts at the end of the night (Fig. 5, A and B). These data imply that volumes of individual granules underwent larger changes over the day-night cycle in large chloroplasts than in small chloroplasts. The average percentage of chloroplast volume occupied by starch was much lower at the end of the night than at the end of the day (Fig. 5A). However, the general relationship between chloroplast volume and the number of starch granules per chloroplast was maintained during the night (Fig. 5C).

Relationship between Starch Granule Number and Chloroplast Volume in Chloroplast Division Mutants
To shed further light on the control of starch granule numbers, we examined starch in six Arabidopsis chloroplast division mutants that have fewer, larger chloroplasts per mesophyll cell than wild-type plants. Whereas wild-type plants have 80 to 120 chloroplasts per mesophyll cell, the accumulation and regulation of chloroplast (arc) mutants used in this study have between one and about 30 chloroplasts per mesophyll cell (Table III). The ARC genes identified thus far encode components of the chloroplast division apparatus; failure of chloroplasts to divide at the normal rate during mesophyll cell expansion in the mutants  Table II. Seven different combinations of harvest time and light regime are represented (Table II) results in fewer chloroplasts per cell in mature leaves. There has been no systematic study of the effect of arc mutations on starch content or starch granule numbers. However, there are indications that some of the ARC proteins may influence starch accumulation. Arabidopsis arc3, arc5, arc6, and arc10 mutants were reported to have increased starch accumulation, whereas plants overexpressing FtsZ1 (ARC10) have no starch granules (Austin and Webber, 2005;El-Kafafi et al., 2008). Overexpression of FtsZ in potato (Solanum tuberosum) tubers resulted in smaller plastids apparently containing fewer, larger starch granules (de Pater et al., 2006) and rice (Oryza sativa) endosperm misexpressing FtsZ proteins contains granules of abnormal sizes and shapes (Yun and Kawagoe, 2010). Some other mutant and transgenic plants defective in chloroplast division are also reported to have altered starch granule numbers (e.g. giant chloroplast1; Maple et al., 2004).
To minimize any effects of genetic background on starch content and granule numbers, we compared the original arc mutants with their ecotype Landsberg erecta (Ler) or Wassilewskija (Ws) parental lines, and for four of the six ARC genes we also compared T-DNA insertion mutant and wild-type plants in the Columbia (Col) background (for details, see Table III and Supplemental Fig. S4). We checked that each mutant line had the expected reduction in chloroplast number per cell (Supplemental Fig. S5).
There were no statistically significant and consistent differences in starch content between mature nonflowering rosettes of wild-type plants and any of the arc mutants (Fig. 6A) at the end of either the day or the night. There were small differences in starch content between the arc mutants and wild-type plants at the end of the night, but no one genotype was statistically significantly different from all the others (data not shown). Significant differences were in some cases dependent on the genetic  background. For example, although arc3 and Ler wildtype had significantly different starch levels at the end of night, there was no difference between arc3-2 and Col wild type at this point. There were also no consistent differences in starch granule size between wild-type plants and any of the arc mutants (Supplemental Fig.  S6). The granule shape of all the arc mutants was similar to that of wild-type plants (Supplemental Fig. S5).
As expected, all of the arc mutants had greater chloroplast cross-sectional areas than wild-type plants. They also had greater numbers of granules per chloroplast. Mutants with fewer, larger chloroplasts had more starch granules per chloroplast than mutants with more, smaller chloroplasts (Fig. 6B). However, there was a remarkable conservation in the stromal area per granule across all of the arc mutants and wild-type lines; there were no consistent, statistically significant differences between mutant and wild-type plants in number of granules per mm 2 of chloroplast cross-section, regardless of chloroplast size or number per cell (Fig. 6B).

Comparison of Methods for Estimation of Granule Numbers
We have developed two new techniques that between them provide radically new information about the dynamics of starch granule numbers in leaves. Sectioning by FIB-SEM allows rapid and accurate imaging of whole chloroplasts and the starch granules they contain. FIB-SEM can be applied specifically to photosynthetic mesophyll cells and to defined regions of the leaf. Method 2, involving isolation of chloroplasts and starch granules, provides an average figure for starch granule number per chloroplast in the rosette as a whole. Both methods also have drawbacks, but their strengths are complementary.
Our results show that previous estimates of granule size and number made on single sections of leaves are likely to have been inaccurate and may have given rise to misleading conclusions. For example, based on TEM images, granule number was proposed to be about two per chloroplast in Arabidopsis leaves grown in 370 mL L 21 CO 2 and about three per chloroplast in leaves grown in 700 mL L 21 CO 2 (Teng et al., 2009). It is likely that these values are serious underestimates; hence, no firm conclusions can be drawn about the effects of elevated CO 2 on granule numbers.
We summarize our principal findings below and discuss their impact on our understanding of the control of starch degradation and starch granule initiation.

Remarkable Heterogeneity of Chloroplast Size
We observed a very wide range of chloroplast volumes in the mesophyll cells of wild-type plants. For mature and immature leaves, FIB-SEM revealed an average 4-and 6-fold variation, respectively, in chloroplast volume within a tissue volume equivalent to a single mature mesophyll cell. We are not aware of previous reports of this remarkable local heterogeneity of chloroplast volume in wild-type leaves, although some chloroplast division mutants have heterogeneous populations of chloroplasts because of unequal chloroplast division (e.g. Maple et al., 2002;Fujiwara et al., 2004).
There was also a marked increase in chloroplast size through leaf development. The three immature leaves we examined by FIB-SEM had statistically significantly different average chloroplast volumes ( Fig. 4; data not shown), indicating that chloroplast volume is changing rapidly through this developmental stage.
The origins of chloroplast heterogeneity in wild-type plants are unclear. Possibilities include nonsynchronous division of chloroplasts, so that different chloroplasts are at different stages of expansion after division, or irregularities of the division plane so that daughter chloroplasts are not of identical size. Resolution of this question will require deeper understanding of signals that trigger chloroplast division (Okazaki et al., 2009) and further analyses of changes in chloroplast numbers and volumes through mesophyll cell development. FIB-SEM offers new opportunities for robust, quantitative analyses of this sort.

Starch Metabolism Is Independent of Chloroplast Size
In general, starch metabolism is independent of chloroplast size for a given leaf age. The relationship among chloroplast volume, the volume of starch per chloroplast, and the size of starch granules at the end of the day was approximately the same for large and small chloroplasts. Strikingly, all of the arc mutants were very similar to wild-type plants with respect to patterns of starch turnover and starch granule sizes, even though they differed hugely from wild-type plants and from each other in chloroplast volume.
The relationship between chloroplast size and starch metabolism changes radically as mesophyll cells mature. Although starch occupied approximately 15% of the volume of chloroplasts of both mature and immature leaves at the end of the day, and daily starch turnover was comparable, the chloroplasts of immature leaves contained more, smaller granules than those of mature leaves. Thus, starch granules in immature leaves present a larger surface for synthesis and degradation than in mature leaves, and the rate of deposition/loss of material is slower per unit surface area.

Granule Number per Chloroplast Is Strongly Related to Chloroplast Volume
There is a strong, positive relationship between chloroplast volume and numbers of granules per chloroplast. The number of granules per unit of chloroplast stroma was similar across the very large range of chloroplast sizes found in mature wild-type and arc mutant leaves. This finding implies that the frequency of granule initiation may depend on the concentrations of factors in the stroma that are themselves rather invariant. The possible nature of such factors is discussed further below. The precise relationship between chloroplast volume and numbers of starch granules per chloroplast changes as the leaf matures. There are more granules per unit of stroma in immature than in mature leaves even though the amount of starch per unit of stroma is very similar in the two sorts of leaf.
The discovery that granule number is related to chloroplast volume for a given leaf age arose from measurements on many individual chloroplasts by FIB-SEM. This insight could not have been obtained by microscopy on individual sections or by our method 2. Previous microscopy studies and results from method 2 both indicate that granule number per chloroplast is relatively invariant. Results from method 2 showed a strong conservation of average granule number per chloroplast across a wide range of leaf starch contents and granule sizes. One might thus conclude that each chloroplast has the same, fixed number of granule initiation sites or primers (see also "Discussion" below). However, method 1 shows that such a conclusion would be incorrect. It reveals considerable variation in numbers of granules per chloroplast, dependent on chloroplast volume.

Granule Initiation Is Infrequent in Mature Leaves
Taken as a whole, our data indicate that the diurnal change in starch content per chloroplast in mature At1g75010 ARC3 arc3 (Ler) a : EMS b , Pyke and Leech (1992) Figure S4. e Mutants have one greatly enlarged chloroplast and some smaller chloroplasts in each cell.
f Mutants have a heterogeneous population of chloroplasts, 40% to 50% wild-type size, 50% to 60% larger than wild type.
leaves is brought about primarily by the growth during the day and shrinkage at night of a fixed number of granules, regardless of the rate of starch accumulation. Thus, loss of granules at night and appearance of new granules during the day are both infrequent events. It seems likely that individual granules are long lived and that initiation of new granules is rare. The relationship between granule surface area and starch content is simpler than would be the case were new granules initiated as chloroplast starch content increases. This relationship may be of significance in the unknown mechanism by which starch content is measured and used to set an appropriate rate of degradation at the onset of darkness. Elucidation of the precise nature of the relationship will require more detailed analyses of the changes in shape of individual granules through the day-night cycle.
Our conclusions are at odds with a previous proposal about diurnal patterns of change in granule size and number in Arabidopsis leaves. Based on measurements of leaf starch content and size of isolated granules (using a laser diffraction particle size analyzer), Howitt et al. (2006) reported that the distribution of starch granule sizes was the same at the end of the day and the end of the night. They concluded that some granules are quickly and completely degraded during the night, but others are not subject to degradation. This is not the case in our plant material. Most granules underwent a large reduction in volume during the night, and there was relatively little loss of individual granules.
In contrast to the situation in mature leaves, it is clear that granule initiation must occur throughout the period of expansion as the leaf matures. In Arabidopsis seedlings, starch granules appear in chloroplasts at the tip of the first true leaf between 6 and 8 d after germination (Sakamoto et al., 2009). We observed that chloroplasts in immature leaves, early in the expansion phase, contained on average only 1.5 times more starch granules than the chloroplasts of mature leaves. Because chloroplasts undergo at least three rounds of division during expansion of mesophyll cells (Marrison et al., 1999), division must therefore be accompanied by initiation of new granules. Further work will be required to establish whether the initiation of new granules is specifically and exclusively linked to chloroplast division events. Initiation of new granules may be triggered by chloroplast expansion, hence an increased stromal volume, following divisions. Alternatively, granule loss and initiation may simply be more frequent events on a daily basis in developing than in mature leaves.  Boyes et al., 2001). Leaves were harvested 9 h into a 12-h light period. For the bottom graph, asterisked values for mutants are statistically significantly different from the wild-type values (Student's t test, P , 0.05).

How Are Granules Initiated?
Two different types of mechanisms have been proposed to explain how granules are initiated. The first requires a specific, biologically generated primer. In the second, the process is spontaneous, without a specific primer requirement. Suggestions for specific primers originally centered on reversibly glucosylated proteins, thought to be analogous to the glycogen-priming glycogenins of yeasts and animals (for discussion, see Langeveld et al., 2002). However, it is more likely that reversibly glucosylated proteins are involved in cell wall synthesis (e.g. D'Hulst and Mérida, 2010;Mortimer et al., 2010;Rautengarten et al., 2011). A recent proposal for a priming mechanism involves the soluble starch synthase SS4, which is necessary for normal numbers of starch granules per chloroplast; Arabidopsis ss4 mutants have only one granule per chloroplast (Roldán et al., 2007;Szydlowski et al., 2009;D'Hulst and Mérida, 2010). Structural components of the chloroplast division apparatus might also act as primers or nucleation loci for starch granules (but see below).
The view that starch granule initiation may be a spontaneous event, dependent on biological and physicochemical conditions in the chloroplast, comes from consideration of the behavior of solutions of glucans and proteins in vitro. Under appropriate conditions, glucans can undergo phase separation followed by nucleation and the growth of a crystalline structure (Lorén et al., 2001;Ziegler et al., 2005;Biliaderis, 2009). The rate and frequency of formation of stable crystalline nuclei in the chloroplast stroma would depend on factors including glucan concentration, size, and structure as well as background conditions such as the nature and concentration of proteins. Although granule initiation via this mechanism would not require a specific, biologically generated primer, its frequency would be biologically determined and would thus be subject to genetic and developmental variation.
We could not confirm earlier reports of altered starch accumulation and granule number in arc3, arc5, arc6, and arc10 mutants (Austin and Webber, 2005;El-Kafafi et al., 2008); thus, there is at present no good evidence that components of the chloroplast division apparatus act as primers for granule initiation. Our observations do not allow us to distinguish between other possible priming mechanisms. However, efforts to discover this mechanism will be informed by our observation of the remarkable constancy of granule number per unit volume of stroma and facilitated by our methods for estimation of granule numbers.
Segregating populations of seeds carrying T-DNA insertions were obtained from the Nottingham Arabidopsis Stock Centre: SALK_057144 and SALK_012892 in At1g75010, SAIL_71_D11 in At3g19720, SAIL_693_G04 in At5g42480, and SALK_073878 in At5g55280. Oligonucleotide primers and methods used to select plants homozygous for the insertions are shown in Supplemental Table S1.
Unless otherwise stated, plants were grown in compost in a controlledenvironment room at 20°C and 75% relative humidity, with a 12-h photoperiod at 180 mmol quanta m 22 s 21 throughout the growing period.

Chloroplast Isolation
Intact chloroplasts were isolated mechanically at 4°C using the Sigma Chloroplast Isolation Kit. Approximately 4 g of leaves (whole rosettes, typically 25 d old) was washed, cut into 1-cm 2 pieces, and macerated in a blender (Multi-Moulinette) with 6 mL g 21 chloroplast isolation buffer (CIB) plus 0.1% (w/v) bovine serum albumin (BSA) using four 1-s bursts. After filtration (100-mm mesh), the chloroplast suspension was centrifuged at 200g for 3 min. The supernatant was centrifuged at 1,000g for 7 min, and the resulting pellet was resuspended in 2 mL of CIB plus BSA and centrifuged on a two-step Percoll gradient (80% and 40% Percoll in CIB plus BSA) at 1,500g for 15 min. Intact chloroplasts were collected from between the Percoll layers. Chloroplast intactness was estimated from assay of NADP-dependent glyceraldehyde-3phosphate dehydrogenase activity before and after rupturing the chloroplasts by osmotic shock (Batz et al., 1995).

Measurement of Starch Content and Iodine Staining
Leaves were harvested directly into liquid nitrogen and ground to a powder in a ball mill. Extraction in dilute perchloric acid, starch solubilization, enzymatic hydrolysis, and assay of Glc were as described previously (Critchley et al., 2001;Smith and Zeeman, 2006).

Purification of Starch Granules
Whole rosettes were harvested directly into liquid nitrogen at the end of the light period. Further processes were at 4°C. Frozen material was ground in a mortar in 4 volumes of isolation medium: 100 mM MOPS, pH 7.2, 5 mM EDTA, 2.6 mM Na metabisulfite, 0.05% (v/v) Triton X-100, and 5 mM dithiothreitol. The homogenate was filtered through two layers of Miracloth, and the filtrate was centrifuged at 20,000g for 10 min. The pellet was resuspended in 5 mL of isolation medium and centrifuged into 20 mL 90% of Percoll, 10% isolation medium in a 30-mL Corex tube at 5,000g for 10 min. The pellet was resuspended in 2 mL of isolation medium and centrifuged as above but into 10 mL of Percoll solution in a 15-mL tube. The pellet was washed three times with water by resuspension and centrifugation, then resuspended in 1.5 mL of acetone and centrifuged for 5 min at 20,000g at 4°C. The pellet was air dried for 4 h and stored at 220°C.

SEM
Starch granules were sprinkled onto double-sided, carbonated, sticky stills (Leit-tabs) attached to SEM stubs (both from Agar Scientific) then sputter coated with gold for 75 s at 10 mA in argon and imaged at 3 kV in a FEI XL30 FEG SEM (FEI). Measurements of granule radii were made with ImageJ software (http://rsbweb.nih.gov/ij/). Equations used to calculate the radius and thus derive the granule volume are in Supplemental Figure S2.

FIB-SEM: Method 1
Small pieces (2 mm 2 ) were rapidly cut with a razor blade from the lamina close to the middle of a leaf and immediately fixed in 2.5% (v/v) glutaraldehyde in 0.05 M Na cacodylate, pH 7.3. Samples were vacuum-infiltrated and left overnight in fresh fixative at 4°C, washed in 0.05 M Na cacodylate, and postfixed with 1% (w/v) osmium tetroxide in 0.05 M Na cacodylate for 1 h at 4°C. Samples were dehydrated in an ethanol series and infiltrated with LR White resin by successive changes of resin/ethanol mixes over 3 d at room temperature, with the last infiltration overnight in 100% resin. The resin was polymerized at 60°C for 16 h. FIB-SEM was performed on an FEI Quanta 3D FEG SEM by small modifications of the method of Bushby et al. (2011). Backscattered electron imaging was typically at an accelerating voltage of 5 keV with a 1,000-mm aperture, a current of 4 nA, a dwell time of 30 ms, an acquisition time per image of 28 s, and a resolution of 1,024 3 884 pixels at 50 nm pixel 21 . Milling was typically at 30 kV and 7 nA, with a slice thickness of 49 nm and a milling time per slice of 7 to 12 s. A typical volume of tissue 50 3 30 3 30 mm was imaged in a single run. Total run time for each block was 6 to 11 h.

Granule Number Estimation: Method 2
Method 2 consisted of the following steps. First, a suspension of intact chloroplasts was isolated, and an estimation of chloroplast number per unit volume was conducted using a double-celled modified Fuchs-Rosenthal hemocytometer (Agar Scientific Ltd.) according to Kouri et al. (2003). Starch content per unit volume of the chloroplast suspension was measured. After these steps, the average mass of starch per chloroplast was calculated. Isolation and SEM imaging of starch granules from a representative sample of leaves were conducted. Granule radii were estimated using imaging software. Starch granule volumes were calculated (see Supplemental Fig.  S2). The mass of individual starch granules was calculated by multiplying the volume by starch density (1.5 g cm 23 ; Dengate et al., 1978;Munck et al., 1988). The number of starch granules per chloroplast was calculated by dividing the mass of starch per chloroplast by the mass of individual starch granules.

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure S1. Seventeen sequential sections through a mature leaf at the end of the day, obtained by FIB-SEM.
Supplemental Figure S2. Estimation of granule volume.
Supplemental Figure S3. Starch granule size distributions obtained by different methods of measurement.
Supplemental Figure S4. Analysis of T-DNA insertion lines.
Supplemental Figure S5. Phenotypes of the arc mutants and their wild types.
Supplemental Figure S6. Starch granule size in the arc mutants.
Supplemental Table S1. Primers used in this study.