Live cell imaging of meiosis in Arabidopsis thaliana - a landmark system

Meiosis is essential for sexual reproduction and key to the generation of genetic diversity. To reveal the robustness of meiocyte differentiation and progression through meiosis, we have here established a live cell imaging setup to follow the dynamics of individual male meiocytes in Arabidopsis. Our method is based on the concomitant visualization of microtubules and a meiotic cohesion subunit that allowed following five cellular parameters: cell shape, nucleus position, nucleolus position, chromatin condensation and microtubule array. We find that the states of these parameters are not randomly associated and identify 11 states, referred to as landmarks, that occur much more frequently than closely related states, indicating that they are convergent points of meiotic progression. With this, the here-presented landmark system represents a novel method to analyze meiosis not only allowing a high-temporal dissection but also providing new criteria to evaluate mutants or environmental effects on meiosis.


Introduction 39
Meiosis is essential for sexual reproduction by reducing chromosome number to 40 eventually generate gametes with half the genomic DNA content as the parental 41 plant. Moreover, meiosis is central to the formation of genetic diversity by generating 42 recombination between the homologous chromosomes (homologs) and by randomly 43 selecting either the maternal or paternal homologs into a new set of chromosomes in 44 the gametes. Hence, understanding the molecular mechanisms underlying 45 recombination as well as chromosome distribution and subsequent modulation of 46 meiosis are also of key interest for breeding (Crismani et al., 2013;Hand and 47 Koltunow, 2014; Lambing and Heckmann, 2018). 48 Entry into meiosis is tightly regulated in all organisms. In plants, it involves the 49 reprogramming of somatic fate since plants, in contrast to animals, do not have a 50 germline that is set aside early during embryo development (Schmidt et al., 2015). 51 Designated meiocytes adopt a characteristic shape that radically changes during the 52 course of meiosis ultimately resulting in the formation of spores. These spores then 53 differentiate into gametophytes that produce the gametes, which will fuse during 54 fertilization. 55 In recent years, our understanding of meiosis in plants has been fostered by 56 genetic approaches, mostly in the model plants Arabidopsis thaliana, Zea mays and 57 including those that control entry and progression through meiosis (Lambing et  Arabidopsis sporogenesis and gametogenesis, albeit without resolving specific 89 meiotic stages (Ingouff et al., 2017). 90 Here we set out to develop a live cell imaging system for meiosis in 91 Arabidopsis. To this end, we have generated an easy applicable microscopic set up, 92 difficult to isolate, we did not explore this possibility further. Next, imaging can be 109 carried out in the context of an entire organism, e.g. in C. elegans (Mullen and 110 Wignall, 2017; Rosu and Cohen-Fix, 2017) with the benefit of perturbing the analyzed 111 cells as little as possible by preparation procedures. However, this set up is limited to 112 small organisms and/or short observation times due to size restrictions and the 113 problem of movement of the sample, e.g. the elongated stem that carries the flowers 114 in Arabidopsis results in a high degree of instability of the specimen during image 115 acquisition. Hence, such a set up was also excluded. Finally, live cell imaging can be 116 performed on isolated organs or tissues that are typically easy to obtain and that 117 and Villeneuve, 2017) and Drosophila melanogaster (Głuszek et al., 2015). As 120 conventional confocal laser scanning microscopes can reach cells up to a depth of 121 70-100 µm, they are suited to observe the meiocytes in Arabidopsis that are covered 122 by three cell layers in the anthers. Imaging of isolated organs has already been 123 successfully applied to the analysis of organogenesis in the shoot apical meristem 124 (SAM) of Arabidopsis (Hamant et al., 2014). Since shoots could be maintained for 125 several days without obvious perturbations of development, we decided to adapt and 126 optimize this approach for our purposes. 127 First, we selected inflorescences and removed all but one young flower 128 primordium presumably containing meiotic stages as indicated by its round shape 129 and an approximate diameter of 0.4-0.6 mm ( Figure 1B), corresponding to stage 9 of 130 flower development (Smyth et al., 1990). Next, the upper sepal was removed giving 131 access to two of the six anthers since the petals are shorter than the anthers at this 132 floral stage. Finally, the bud along with the pedicel and a few millimeters of the stem 133 was embedded into Arabidopsis Apex Culture Medium (ACM) and stabilized with a 134 drop of agarose ( Figure 1A,B). In agreement with the previous analysis of the SAM,135 we found that the flower buds stayed alive on the ACM medium for up to seven days 136 during which flowers grew and developed normally ( Figure 1C). 137 Imaging was performed with an up-right confocal laser scanning microscopy 138 equipped with a water immersion objective. The entire flower bud was submerged in 139 water and the objective was brought into direct contact with the sample ( Figure 1A). 140 During image acquisition the temperature was kept constant at 21°C. The flower is anchored into the medium with the anthers exposed at the top (B3). C) Magnification of the sample from B3. The two exposed anthers are highlighted in yellow; petals are in white, the three remaining sepals in blue, and the tip of the stigma in pink. D) Flower buds could be kept alive and growing for up to 1 week.
A generic set up for imaging of cell divisions includes a reporter that highlights 144 DNA/chromatin coupled with a marker for cytoskeletal components, usually 145 microtubules, so that chromosome and spindle behavior can be visualized (Nannas 146 et al., 2016;Peirson et al., 1997). Since fusions of histones with fluorescent proteins 147 have often been applied for this purpose, we first scanned through previously 148 generated transgenic lines expressing different histone variants fused to fluorescent 149 proteins, such H2B. However, while these labeled histones clearly marked DNA in 150 somatic cells, the signal was often fuzzy in meiosis. Moreover, since all or most cells 151 in an anther produced these fusion proteins, the identification of meiocytes was 152 sometimes difficult, especially at early stages of meiosis when chromosomes are not 153 condensed and meiocytes cannot easily be recognized by their size and shape. 154 Therefore, we aimed for a meiosis-specific gene and generated a GFP fusion to 155 REC8, the alpha kleisin subunit of the cohesin complex, also known as SYN1 or DIF1 156 in Arabidopsis (Bai et al., 1999). 157 Cohesin is key for chromosome segregation and its step-wise removal allows 158 the segregation of homologous chromosomes in meiosis I, followed by separation of 159 sister chromatids in meiosis II. In addition, cohesin is required for recombination and 160 repair of DNA double-strand breaks resulting in a highly pleiotropic phenotype that 161 leads to almost complete sterility of rec8 mutant plants (Bai et al., 1999). Expression 162 of our genomic PRO REC8 :REC8:GFP reporter in a homozygous rec8 mutant 163 background completely restored fertility of these plants and analysis of chromosome 164 spreads confirmed that chromosome segregation is indistinguishable from the 165 wildtype (Supplement 1). 166 REC8 replaces RAD21 in meiosis and is hence highly specific for meiocytes in 167 all species analyzed so far (Nasmyth, 2001). Consistent with previous immuno-168 localization studies, we found that the GFP signal of our functional reporter line was 169 only present in meiocytes providing a straightforward way to identify microspore 170 mother cells (Figure 2). 171 Moreover, the REC8 reporter allowed us to estimate the sensitivity of our 172 imaging procedure. While REC8 is removed from chromosomes arms at the end of 173 meiosis I to allow the resolution of cross-overs, a small fraction remains at the 174 centromeres to maintain sister chromatid cohesion. The detection of the centromeric 175 fraction of REC8 has been challenging by immuno-localization studies. When we 8 followed the first meiotic division, we observed the remaining REC8 at centromeres 177 indicating that our live cell imaging system is highly sensitive (Figure 2, Sup. movie 1). 178 Next, we combined the PRO REC8      Analyzing a first set of movies, gave rise to the hypothesis that some of the 247 parameter states are connected, e.g. the nucleolus apparently dissolves only after 248 the nucleus has moved to one side of the meiocyte and returned to a central position. 249 To assess the nature of these associations, we analyzed a subset of cells (n=169 250 from 35 anthers) assigning a combination of numbers that represents each 251 parameter state at every time point when a frame was taken, e.g. 1-1-2-2-1 describes 252 a meiocyte that is rectangular in shape, has a centrally located nucleus with a 253 centrally located nucleolus, with not condensed, yet not paired chromosomes and an 254 evenly distributed microtubule array ( Figure 3A). We subsequently analyzed 10,671 255 time points resulting from the first set of movies, that allowed us to judge 256 which parameter states occur together and in which frequency ( Figure 3B).  A) Schematic representation of the different states of the five parameters analyzed during meiotic progression. B) Heatmap of the correlation between the different states of the parameters. The darker the blue color, the tighter is the correlation and the higher is the frequency of co-appearance of two parameter states. Numbers refer to the scheme in A. C) Table illustrating the different parameter states at the moment of nuclear envelope breakdown. Even if the breakdown can be found (low number of observations), there is high variability of the combinations of parameter states that depict this moment. Hence, the neighboring scores are below zero, precluding the inclusion of the nuclear envelope breakdown as a landmark in this analysis.
combinations of the different parameter states only 101 were actually present in our 261 data set (Supplement 4) and their frequencies were distributed in a very dispersed 262 range (from 0.01% to 21.14% of the total number of observations). In the following, 263 we call a combination of all five parameter states a cellular state. 264 We realized at this point that an assessment of the cellular states, (e.g. 265 concluding that the frequency of a state appearance in the dataset relates to its 266 importance) is highly biased by the duration of the respective state, i.e. combinations 267 of parameters that depict long phases such as pachytene are present in higher 268 number of time points than combinations depicting short phases, e.g. metaphase I. 269 Hence, to identify significantly distinct cellular states from the observed data, we 270 defined a local or neighboring score, which quantifies the occurrence of a certain 271 cellular state compared to its neighboring states. 272 A neighboring state was defined as a cellular state that is one transition away 273 (-1 or +1) for at least one, but at most two, parameter states compared to the cellular 274 state analyzed. With this, 2-2-3-4-4, for example, is a neighbor of 2-2-3-4-3 and of 3-275 2-3-4-3, but not of the cellular state 2-2-3-4-2 and not 3-2-3-3-3 (Supplement 4). 276 Notably, we only took states into account that were actually observed. The 277 neighboring score was then compared with the subset of neighboring states, to find 278 the predominant state among the surrounding states, and is defined as: 279 where counts refers to the number of times a certain state is observed in the data, 281 and std refers to the standard deviation. This analysis revealed 11 clearly distinct 282 cellular states that differed from their neighbors with a score higher than one, 283 denoting that they occurred at least one standard deviation more frequent than the 284 mean of the neighboring stages (Figure 4).   Meiosis represented as a progression of parameter state combinations, here called cellular states. Each circle signifies an observed cellular state and the arrows are observed transitions between these states. The size of circles depicts the frequency of appearance of each cellular state while the color presents their neighboring score. Cellular states that have a score higher than 1 are defined as landmarks and were assigned a name (A1-A11). Landmarks are highlighted by outlined circles and their names written in the center. The intensity of the line color of the arrows specifies which are the predominant paths taken by a male meiocyte undergoing meiosis. Notably the arrows indicate progression from one state to the following one only when the transition was seen within 15 minutes interval time, therefore the presence of non-connected circles.

Figure 5. Landmark scheme
Illustration of the 11 here identified meiotic landmarks of male meiosis, A1-A11, and the combination of the parameter states that represent them. The first column provides a microscopy picture of meiocytes depicting each stage. The state of each parameter is separately shown in the following columns, the right-most column (Overlay) present their combination. On the right side, the classical stages of meiosis are assigned to each Landmark.  The break-down of the nuclear envelope in diplotene is an important hallmark of 318 meiosis (Wijnker and Schnittger, 2013). We also could clearly observe the 319 breakdown in our live cell imaging system although, due to its rapid progression, it 320 was only captured in 22 out 10,671 analyzed time points with a sampling interval of 321 one frame every 10 minutes ( Figure 3C and Supplementary movie 3). None-the-less, 322 the nuclear envelope break-down is not included in a landmark state since it 323 appeared to be only loosely connected with the other parameter states, e.g. the cell 324 shape can be oval or round, and the chromatin can be at different condensation 325 levels when the nuclear envelope breaks down ( Figure 3C). Thus, although very 326 distinct when looking at microtubule conformation (i.e. state 7, collapse of pre-spindle 327 Figure 3A), a clearly defined landmark state corresponding to nuclear envelope 328 breakdown was not reached with the parameters analyzed. 329 We could also clearly observe other short-lived phases such as diakinesis, 330 anaphase I, prophase II and anaphase II. However, due to their unexpected high 331 variation in terms of association with the here analyzed parameter states, these 332 phases, like nuclear envelope breakdown were also not designated as landmarks.

Correlation between meiocyte and tapetum differentiation 341
Our sample preparation, which keeps anthers intact, also provided the possibility to 342 follow the differentiation of the tissues surrounding the meiocytes, especially the 343 tapetum cells. These are in direct contact with the meiocytes and are thought to 344 nourish and support the meiocytes and spores (Pacini et al., 1985). A key feature of 345 tapetum cells in many plants species, including Arabidopsis, is that they become To estimate the duration of meiosis and approximate the landmarks to 375 classical stages, the transition states between two landmarks were added to the time 376 estimate of the preceding landmark. This also led to the re-assignment of diakinesis 377 to A7, anaphase I to A8, prophase II to A9 and anaphase II to A10 (Figure 4). While 378 long movies with more than 30 hours containing all meiotic stages could occasionally 379 be obtained, they were rarely fully informative due to loss of the focal plane by Our meiotic description is based on five morphological criteria of male meiocytes that 443 we could distinguish with our reporter genes, i.e., cell shape, position of the nucleus, 444 position of the nucleolus, REC8 status and information about chromatin state, and 445 microtubule array. Importantly, we found that these cellular parameters have two 446 aspects, which make them suitable for a classification system. First, they change in 447 the course of meiosis in a unidirectional manner, e.g., cell shape changes from 448 rectangular over trapezoidal and oval to circular. We never found an example where 449 a meiocyte skipped one of these cell shape changes or changed back from a later 450 stage to an earlier stage. Second, these parameters are linked with each other and 451 build a matrix. For instance, nest-like microtubule array was never found to be 452 associated with a rectangular cell shape of the meiocyte ( Figure 3B). 453 morphological states, called landmarks A1-A11. These differ from each other by at 455 least one characteristic of the parameter states, and always occur in the same order 456 in any cell progressing through meiosis. The pathway taken by an individual meiocyte 457 to reach each landmark could differ slightly, presumably due to biological variation, 458 and is described by the network of the transition states (Figure 4). It is an interesting 459 question to what degree this developmental plasticity depends on meiotic genes 460 and/or enviromental factors such as temperature. 461 The 11 landmarks together with their transitions could be assigned to the 462 classical phases of meiosis ( Figure 4). However, it has to be noted that the alignment 463 of our landmarks with the classically defined stages remains fuzzy for certain phases.  Table  501 S1. All seeds were surface-sterilized with chloride gas, sown on 1% agar plates (half-502 dish was filled with autoclaved water and placed under a W-plan-Apochromat 553 40X/1.0 DIC objective (Carl Zeiss AG, Oberkochen, Germany). GFP was excited at λ 554 488 nm, and detected at λ between 498-550 nm. RFP was excited at λ 561 nm and 555 detected at λ between 578-650 nm. Autofluorescence from chloroplasts was 556 highlighted in blue using excitation at λ 488, and detection at λ between 680-750 nm. 557 Time lapses were acquired as series of Z-stacks (6 planes, 50 µm distance). Interval 558 time was varying from a max of 15 to a min of 3 minutes depending on sample 559 conditions. The functions "Autofocus" and "Automatized positions" were used to 560 acquire images. Room temperature and sample temperature were controlled and 561 stabilized at 18 °C and 21°C respectively. 562 Image drift was corrected by the Stack Reg plugin (Rigid Body option) for Fiji. Cell 563 numbers were assigned manually. 564

Quantitative analysis of live cell imaging data 565
Data set description 566 The landmark system is based on the analysis of a subset of data on male meiocytes 567 from WT plants carrying the KINGBIRD reporter constructs. A subset of the 568 analyzable male meiocyte was described at every timepoint by assigning manually a 569 value for each of the five parameters assessed. A total of 169 meiocytes from 35 570 anthers were annotated, leading to a total of 18,531 data points spanning more than 571 3,269 hours. For 7,860 observations one or more of the parameters could not be 572 annotated with a well-defined state, with 5,893 observations not having a single 573 parameter recognizable. The resulting dataset, consisting in 10,671 time points, was 574 used to determine the co-occurrences of parameter states and the landmarks. 575 A second dataset was annotated solely using the landmark system for the 576 comparison of the calculation of the time course. The timing of image acquisition and 577 the landmark attribution for each cell at each time point was done manually. 578 Landmarks were described using numbers from 0 to 12 with a trailing "s" when the 579 landmark was appearing for the first time. An "n" was assigned to the cell when not 580 visible at a certain time point. All the starting points appearing later than 15 min after 581 the previous recorded landmark were discarded. Files in the CSV format were 582 created for each anther. 583 All analysis of the two datasets were done using the Python programming 584 language (Version 3.6, Python Software Foundation, https://www.python.org). 585

Landmark extraction : data preprocessing 586
The manually created data set contains a description of the state of each of the 5 587 parameters that were recorded in individual cells at 15 minute intervals. In some 588 cases, the time between consecutive measurements was more than 15 minutes. In 589 these cases, we inserted an unmeasured data point ('n') for each cell parameter such 590 that time between measurements was equal and at most 15 minutes for each 591 recorded time course. This was done to ensure that unmeasured periods are noted 592 as unmeasured properly, and to decrease the risk to assign any unrealistic transitions. 593 The combination of the state of each of the 5 cellular parameters makes up the cell 594 state. Transitions from one cell state to another occur when one or more parameter 595 transition to a new state. 596

Landmark extraction: Cell state co-occurrence 597
To create the co-occurrence heat map in Figure 4, we counted the number of times a 598 combination of two parameter states occurred in the same cell at the same time point. 599 Since some time courses were measured with different temporal resolution (e.g. 10 600 minute intervals versus 15 minute intervals), we first resampled the data points from 601 all time courses to have the same time between measurements. Co-occurrence 602 counts were normalized by the total number of counts in the column, including the 603 counts where the state of the 2 nd parameter could not be measured. This means that 604 in the upper triangular part of the matrix, the counts are normalized for the total 605 occurrence of the first parameter (all parameter sections per row together sum up to 606 1) while in the lower triangular part of the matrix for the second parameter, each 607 column sums up to 1. 608

Bootstrapping 609
To assess the robustness of the selected landmarks and thus our theoretical 610 framework, we performed a bootstrapping procedure on our data set. The total set of 611 observations was randomly sampled with replacement to obtain a data set 1.5 times 612 the size of the original data set. Scores for each state in this data set were calculated 613 using the procedure described in the previous paragraph. This process was repeated 614 1000 times to obtain estimates for the mean value, standard deviation and quantiles 615 of the score of each cellular state. 616