Control of plastid inheritance by environmental and genetic factors

The genomes of cytoplasmic organelles (mitochondria and plastids) are maternally inherited in most eukaryotes, thus excluding organellar genomes from the benefits of sexual reproduction and recombination. The mechanisms underlying maternal inheritance are largely unknown. Here we demonstrate that two independently acting mechanisms ensure maternal inheritance of the plastid (chloroplast) genome. Conducting large-scale genetic screens for paternal plastid transmission, we discovered that mild chilling stress during male gametogenesis leads to increased entry of paternal plastids into sperm cells and strongly increased paternal plastid transmission. We further show that the inheritance of paternal plastid genomes is controlled by the activity of a genome-degrading exonuclease during pollen maturation. Our data reveal that (1) maternal inheritance breaks down under specific environmental conditions, (2) an organelle exclusion mechanism and a genome degradation mechanism act in concert to prevent paternal transmission of plastid genes and (3) plastid inheritance is determined by complex gene–environment interactions.

The genomes of cytoplasmic organelles (mitochondria and plastids) are maternally inherited in most eukaryotes, thus excluding organellar genomes from the benefits of sexual reproduction and recombination. The mechanisms underlying maternal inheritance are largely unknown. Here we demonstrate that two independently acting mechanisms ensure maternal inheritance of the plastid (chloroplast) genome. Conducting large-scale genetic screens for paternal plastid transmission, we discovered that mild chilling stress during male gametogenesis leads to increased entry of paternal plastids into sperm cells and strongly increased paternal plastid transmission. We further show that the inheritance of paternal plastid genomes is controlled by the activity of a genome-degrading exonuclease during pollen maturation. Our data reveal that (1) maternal inheritance breaks down under specific environmental conditions, (2) an organelle exclusion mechanism and a genome degradation mechanism act in concert to prevent paternal transmission of plastid genes and (3) plastid inheritance is determined by complex gene-environment interactions.
Cytoplasmic genomes are maternally inherited in most eukaryotes 1,2 . It is generally believed that the uniparental inheritance of organelles and their genomes makes them asexually reproducing genetic systems [3][4][5] . Lack of sexual recombination is expected to lead to the eventual mutational meltdown of organellar genomes, a phenomenon widely known as Muller's ratchet [6][7][8] . This is due to the accumulation of deleterious mutations that cannot be separated from (only rarely occurring) beneficial mutations and can be considered as a case of 'genetic hitchhiking' 9 . While there must be evolutionary forces that explain the strong prevalence of uniparental inheritance, there must also be compensatory mechanisms that allow organellar genomes to escape mutational meltdown.
In plants, the two organellar genomes (plastids and mitochondria) have lower mutation rates than nuclear genomes 10,11 . Low mutation rates reduce genetic hitchhiking and thus slow down Muller's ratchet 8 . Recently, early mutation surveillance and induction of recombination repair (dependent on the MSH1 protein) has been proposed as part of a mechanistic explanation for the low mutation rates in plant organelles 12 . However, while low mutation rates can slow down Muller's ratchet, they cannot entirely stop it from turning. Also, there are exceptional plant taxa, such as Plantago, Silene and Pelargonium, where organellar genome mutation rates are strongly elevated [13][14][15][16] . Together, these observations and considerations suggest that additional mechanisms are probably needed to prevent mutational meltdown in plants.
Besides low mutation rates, episodes of biparental inheritance of organelles could also potentially counteract Muller's ratchet. Biparental transmission of organelles that can fuse (such as plant mitochondria) would allow them to participate in sex and recombination, thus providing the ability to generate genomes with favourable combinations of mutations and reduce genetic hitchhiking. Although fusion of seed plant plastids is only rarely observed [17][18][19] , biparental inheritance could help to resolve cytonuclear conflicts 20 , enable plastid genome capture 21,22 and contribute to adaptive evolution 23,24 .
Although maternal inheritance is the general rule, stable biparental inheritance has arisen several times independently in the evolution of both animals and plants [25][26][27][28][29] . For example, the seed plants Medicago Article https://doi.org/10.1038/s41477-022-01323-7 observation under the stereomicroscope. To distinguish paternal transmission events from (occasionally appearing) spontaneous spectinomycin-resistant cells 39 , all detected green sectors were examined by UV microscopy to test for expression of the fluorescent reporter protein GFP in plastids (Fig. 1c).

Chilling stress leads to biparental inheritance of plastids
Screening of progeny obtained by fertilization with pollen from high light-stressed, drought-stressed and heat-stressed plants revealed that all of these crosses showed similar levels of very low paternal leakage as the unstressed control crosses 31 (Table 1, and Extended Data Tables 1  and 2). By contrast, seedlings obtained from crosses with pollen from chilling-stressed plants displayed strongly elevated levels of biparental inheritance ( Fig. 1d and Table 1). Out of 479,230 progeny assayed, 1,151 seedlings inherited paternal plastid genomes, corresponding to a biparental transmission frequency of 0.24% (Table 1). This frequency represents a >150-fold increase in the rate of paternal plastid transmission compared with the unstressed control ( Table 1).
Regeneration of green sectors into plants in the presence of spectinomycin resulted in uniformly green plants (referred to as PPI-C lines for paternal plastid inheritance under chilling stress) that were analysed by Southern blotting to directly confirm the presence of the paternal plastid genomes. Restriction fragment length polymorphism (RFLP) analysis revealed that all lines contained the paternal plastid DNA (Fig. 1e). Confocal laser-scanning microscopy confirmed the plastids as the sites of GFP fluorescence (Fig. 1f).

Paternal plastid transmission in the absence of selection
The measured frequency of biparental plastid inheritance under chilling stress should be high enough to identify paternal transmission events in the absence of antibiotic selection, by randomly analysing a sufficiently large number of seedlings grown under non-selective conditions. When 8,334 seedlings grown on spectinomycin-free medium were screened for GFP expression, 9 seedlings showed clear GFP fluorescence and 4 had a fluorescent shoot apical meristem (Extended Data Fig. 2). This result confirms the high rate of biparental transmission measured in the spectinomycin selection experiment under chilling stress (Table 1) and, importantly, demonstrates that the selection process does not influence the detection of paternal transmission events.
Subsequent transfer of seedlings displaying fluorescent sectors and/or meristems to spectinomycin-containing medium resulted in bleaching of cells that contain only maternal plastids 40 and thus directly visualized the presence of inherited paternal plastids (Extended Data Fig. 2c).

Cytological basis of paternal transmission upon chilling
Given that low temperature greatly alters the inheritance mode of plastid genomes, we set out to investigate the effect of chilling stress on the fate of paternal plastids in gametogenesis. During male gametogenesis, a highly asymmetric cell division takes place in pollen mitosis I (PMI; Fig. 2a). Previous studies suggested that all (or the vast majority of) paternal plastids are excluded from the smaller generative cell (GC), resulting in maternal plastid inheritance 33,34 . To test whether chilling stress has a direct impact on this process, we examined the effect of low temperature on paternal plastid segregation during pollen maturation. To this end, two stages in pollen development were investigated by confocal laser-scanning microscopy using a transplastomic line that expresses the DsRed fluorescent reporter inside plastids (Fig. 2b): the early binucleate pollen (EBP, the product of PMI) and the early pollen tube stage (EPT; Fig. 2a). Detection of the DsRed reporter indicated that pollen plastids are capable of active transgene expression (Fig. 2c).
The newly formed GC in the EBP is bound by a callose cell wall (CCW) that transiently forms after PMI and can be visualized by aniline blue staining (Fig. 2a). Growth at low temperature (10 °C) delayed floral and pollen development, reduced pollen viability and occasionally and Pelargonium show frequent biparental transmission 27,30 . Furthermore, even in plant species with largely maternal inheritance such as tobacco (Nicotiana tabacum) and Arabidopsis thaliana, the occasional transmission of plastids through pollen ('paternal leakage') has been documented, albeit at very low frequencies 31,32 . The reasons and the selective forces that have shaped uniparental organelle inheritance (while still allowing occasional paternal leakage), as well as the forces driving the evolutionary switches in the mode of organellar inheritance, are completely unknown.
Several cytological mechanisms that contribute to maternal inheritance of plastids have been described on the basis of microscopic observations. These include (1) exclusion of plastids from the generative cell in pollen mitosis I (PMI), (2) degradation of plastids and/or their DNA during pollen maturation and (3) elimination of paternal plastids during fertilization 33,34 . The molecular mechanisms involved in these processes are not well understood. In Drosophila, a nuclease (endonuclease G) degrades the paternal mitochondrial DNA upon fertilization 35 . In plants, two unidentified nuclear loci were associated with plastid inheritance patterns in Pelargonium 36 but so far, not a single gene involved in determining the mode of cytoplasmic inheritance in seed plants has been identified. Also, the possible impact of environmental factors on the mode of plastid inheritance has not been explored.
As biparental inheritance is expected to profoundly affect organellar genome stability and evolution, we decided to study the determinants of cytoplasmic inheritance. To this end, we set out to identify environmental and genetic factors that determine plastid inheritance in the model plant tobacco.

Screens for environmental influence on plastid inheritance
To analyse the effect of environmental factors on organellar genome inheritance, we employed a sensitive genetic screening procedure to detect and quantify events of paternal plastid entry into the zygote (Fig. 1a). Screening is based on incorporation of the antibiotic resistance gene aadA (conferring resistance to spectinomycin) and the reporter gene gfp (encoding the green fluorescent protein, GFP) into the plastid genome of the model plant tobacco (Nicotiana tabacum 31,37 ; Fig. 1b). Spectinomycin is a specific inhibitor of protein biosynthesis in the chloroplast and its application to wild-type seedlings results in growth arrest and pigment loss. Biparental plastid transmission can be detected by pollinating wild-type plants (as maternal recipients) with pollen from plants containing the aadA marker gene in their plastid genome (as transplastomic father). Entry of paternal plastids into the zygote can be readily visualized by germinating seeds in the presence of spectinomycin because cells that receive paternal plastid genomes are resistant to the drug and remain green, thus resulting in green-white variegated seedlings (Fig. 1a,c). Using this experimental system, we applied mild abiotic stress conditions that commonly occur in nature: heat, drought, chilling and light stress. As maternal inheritance is usually established by organelle exclusion and/or organelle degradation during male gametogenesis 33,38 , the stress conditions were specifically applied during pollen development (Fig. 1a). To this end, plants raised under standard growth conditions were transferred to high light stress at 1,000 µE m −2 s −1 , drought stress (by withholding water until visible wilting occurred), heat stress at 35 °C or chilling stress at 10 °C after they had set flower buds (Extended Data Fig. 1). Following completion of pollen development, mature pollen from stressed plants was used to fertilize flowers of unstressed plants. Large-scale crosses were performed for all four stress conditions and for standard conditions in the absence of environmental stress 31 . For each stress condition, between 472,000 and 727,000 seeds were produced (Table 1).
Seeds were germinated in the presence of spectinomycin (to which only the plastid genome of the father confers resistance), and seedlings with green sectors were identified by visual inspection and Article https://doi.org/10.1038/s41477-022-01323-7 cause aberrant cell division patterns (Extended Data Fig. 3). Confocal microscopic analysis of EBP at ambient versus low temperature revealed that, at 10 °C, the shape of the GC is irregular and plastids are more frequently included in the cytoplasm of the GC (Fig. 2c,e and Extended Data Table 3; for three-dimensional reconstructions of representative pollen, see Supplementary Videos 1 and 2).
To test whether the plastids persist in later stages of pollen development, GCs were analysed in the pollen tube upon germination (EPT; Fig. 2a substantially increased paternal plastid transmission at low temperature ( Fig. 2e and Extended Data Tables 1-3).
It is important to note that the observed increase in the entry of paternal plastids into the GC is insufficient to fully explain the >150-fold increase in chilling-induced paternal plastid transmission (cf. Figs. 1d and 2e). Thus, in addition to cell division and organelle distribution, chilling probably affects other cellular processes that are relevant to organelle inheritance. As low temperature also reduces the activities of all enzymes in the cell, we therefore considered candidate enzymes that could be involved in plastid inheritance and whose reduced activity at 10 °C can potentially explain the increase in paternal transmission that remains unaccounted for by organelle distribution in PMI alone.

Control of plastid inheritance by the exonuclease DPD1
During male gametogenesis, plastid DNA is actively degraded, and the exonuclease AtDPD1 plays a key role in this process in Arabidopsis thaliana 41 . Thus far, AtDPD1 has not been shown to affect maternal inheritance 41 and instead, has been suggested to facilitate phosphate mobilization from plastid DNA under phosphorus starvation conditions 42 . Having seen that low temperature promotes inclusion of plastids in the GC, we hypothesized that escape from genome degradation can also enhance the rate of paternal plastid genome transmission. To test this idea, we set out to produce a dpd1 knockout mutant in tobacco by genome editing. Tobacco (N. tabacum) is an allotetraploid species comprising the genomes of two diploid species, N. sylvestris and N. tomentosiformis, thus necessitating generation of a tetra-allelic knockout ( Fig. 3a,b and Extended Data Fig. 4). Targeting of the dpd1 alleles in both subgenomes by CRISPR/Cas9 editing with appropriately designed sgRNAs (single guide RNAs) allowed isolation of a dpd1 null mutant (see Methods, Fig. 3a,b and Extended Data Fig. 4).
Consistent with previous findings reported in A. thaliana, the tobacco dpd1 mutant does not display a noticeable vegetative growth phenotype. Plant growth, reproductive transition and floral development in the dpd1 mutant were comparable to wild-type plants (Extended Data Fig. 5a). However, a reduction in pollen viability was observed in the dpd1 mutant (Extended Data Fig. 5b,c). Confocal laser-scanning microscopy and quantitative PCR analysis revealed retention of plastid DNA in mature pollen grains of the dpd1 mutant (Fig. 3c,d and Extended Data Fig. 5d). Large-scale inheritance assays showed that the dpd1 mutant displayed greatly elevated levels of paternal plastid transmission into the progeny (Fig. 3e,f and Table 1), thus identifying the rate of plastid genome degradation as a genetic factor determining the mode of plastid inheritance.

Synergistic gene-environment control of plastid transmission
Next, we wanted to examine whether the combined action of genes and environment confers even higher levels of biparental plastid inheritance. To this end, we comparatively assessed the effects of chilling stress on the transmission frequency of paternal plastids in the wild-type and the dpd1 genetic backgrounds. Indeed, pollen development at low temperature in the dpd1 mutant resulted in an approximately tenfold increase in paternal plastid transmission compared with the effect of the dpd1 genotype alone or that of the environment alone (Table  1 and Fig. 3e,f). Paternal plastid transmission reached frequencies between 2.2% and 3.2% (in two independent experiments; Fig. 3e and Table 1). These data show that genes and environment act synergistically in the control of plastid inheritance, and their combined action can result in substantial levels of biparental inheritance (Fig. 3e). It is also noteworthy that the environmental factor low temperature and the genetic factor dpd1 are not entirely independent in their effects on plastid inheritance, as revealed by the negative interaction between them (Fig. 3e).

Paternally transmitted plastids readily enter the germline
Biparental inheritance events in which the inherited paternal plastids are present in the shoot apical meristem (Extended Data Fig. 2b,c and The open arrow points to a plastid (magenta) included in the GC in pollen developed at 10 °C. This experiment was repeated four times and representative images are shown. Scale bar, 10 µm. d, Time-lapse confocal images of in vitro germinating EPT expressing DsRed in plastids. The GCN (green) is visualized by SYTO11 staining. The GC membrane is indicated by a dotted line. Arrowheads and asterisks indicate plastids (magenta) located inside or outside the GC, respectively. This experiment was repeated ten times and representative images are shown. Scale bar, 5 µm. e, Proportion of pollen grains with plastids included in the GC. Each circle shows the proportion of plastid-containing GCs in an imaging session/replicate (Extended Data Table 3). Differences between experimental groups are represented by parameter values (β) expressed as log 10 of the fold change, and were estimated using a binomial model (Model 2, n rep.total = 18 imaging sessions, 301 grains analysed; Extended Data Tables 1 and 2): β chilling represents the effect of the chilling treatment (P = 6.01 × 10 −5 ); β experiment represents differences between the EPT (P = 0.0288) and EBP experiments (Extended Data Table 2). Mean estimates and CI95 per experimental group are shown in black horizontal lines and coloured boxes, respectively. Parameters were tested using simultaneous two-tailed Wald z-tests. *P < 0.05; ***P < 0.001; α = 0.05.
Article https://doi.org/10.1038/s41477-022-01323-7 Fig. 4a,b) are particularly relevant in that from there, they enter the germline (upon transition of the apical meristem into a floral meristem) and become heritable. Therefore, large-scale screens were conducted to quantitatively assess the entry of paternal plastids into shoot and root apical meristems in the wild-type and the dpd1 genetic backgrounds under ambient temperature or chilling stress. Presence of paternal plastids was evidenced by GFP fluorescence in the meristem (Extended Data Fig. 2b,c) and continued plant growth in the presence of spectinomycin (Fig. 4a). The presence of paternal plastids in the shoot apical meristem was observed at high frequency, in almost half of the biparental transmission events in dpd1 under chilling stress (cf. Table 1 and Fig. 4b).
To determine whether the inherited paternal plastid genomes can be stably maintained across generations, plants with paternal plastids present in the shoot apical meristem were grown in the greenhouse and self-pollinated for seed production. Uniform resistance of the progeny to spectinomycin (Fig. 4c) demonstrated homoplasmy of the (paternally acquired) plastid genome, which is maternally inherited into the next generation in the absence of stress. Together, these data show that paternally inherited plastids readily enter the germline and are passed on to the next generation, suggesting that the discovered phenomenon has evolutionary significance for adaptation 43 and, potentially, speciation 44 .

Discussion
In the course of this work, we have discovered two distinct factors that jointly determine plastid inheritance in plants. While the environmental factor temperature affects plastid distribution during the asymmetric cell division occurring in PMI, the genetic factor DPD1 affects organellar genome degradation during male gametogenesis (Fig. 5).
Surprisingly, the possibility that plastid inheritance is influenced by the environment has not been considered previously. Here we discovered that maternal inheritance breaks down when male gametogenesis occurs at low temperature. We identified the underlying mechanism by showing that chilling promotes inclusion of paternal plastids in the GC. Chilling stress has been shown to result in destabilization of the cytoskeleton during gametogenesis 45 . The occurrence of altered cell division patterns in PMI upon plant exposure to low temperatures (Extended Data Fig. 3b) suggests that compromised cytoskeletal function is responsible for incomplete plastid exclusion from the GC.
The second elimination mechanism uncovered by our work acts at the level of organellar genome stability. The exonuclease DPD1 degrades organellar genomes in maturing pollen, thus providing a fail-safe mechanism that genetically inactivates those plastids that escaped the organelle exclusion mechanism operating in PMI. Consequently, when both mechanisms (plastid exclusion and genome degradation) are compromised, substantial levels of biparental inheritance ensue (Figs. 3e and 5). Together, our findings suggest that plastid inheritance is a complex trait that is affected by both genetic and environmental factors.
Although most organisms inherit their plastids and/or mitochondria maternally, biparental inheritance has evolved multiple times independently in both animals and plants 46 . It will be interesting to investigate species that exhibit biparental inheritance of plastids and/ or mitochondria, and determine the expression patterns of genes for cytoskeletal proteins (especially those that have been implicated in organelle movement [47][48][49] ) and the activity of organelle-targeted nucleases during male gametogenesis. DPD1 homologues are present in both angiosperms and gymnosperms 42 , and the cytoskeletal components are also generally highly conserved. These candidate genes and their expression patterns could also explain the variation in the rates of biparental plastid inheritance in natural populations 20 , and the switches in inheritance modes seen over evolutionary timescales 2 .
Given the high conservation of the male gametogenesis programme in seed plants, we speculate that the mechanisms identified in our work are likely to be relevant to all seed plant species. However, it is important to note that the mechanisms reported here are primarily operating during PMI. It seems possible that maternal factors and/ or post-fertilization mechanisms also contribute to the control of cytoplasmic inheritance in plants.
Our work also provides simple methods to alter organellar inheritance patterns. In the light of our findings, shifts to biparental inheritance can be achieved either temporarily (by applying chilling stress) or permanently (by genetic interventions, for example, DPD1 gene editing). Changing the mode of inheritance of organelles has important applications in plant breeding. In most crops, both plastids and mitochondria are maternally inherited, making it notoriously difficult to separate the influence of the plastid from that of the mitochondrial genome on important phenotypic traits such as growth, yield and stress tolerance 23,43,50,51 . Induction of biparental transmission will allow separation of plastid from mitochondrial effects by taking advantage of the random segregation and sorting of the two organelle types in the progeny.
Finally, our finding that mild chilling stress triggers substantial rates of biparental plastid inheritance has far-reaching consequences for our understanding of organellar genome evolution. Low temperatures (10 °C) are ubiquitous in nature, hence our findings provide a possibility for organelles to participate frequently in sexual reproduction. Assuming that our findings reported here extend to mitochondrial inheritance, the paternally transmitted mitochondria would be able to fuse with their maternal counterparts, leading to genome recombination 52,53 . This could efficiently counteract Muller's ratchet by generating new variation in mitochondrial genomes and facilitating the combination of favourable mutations.
In contrast to plant mitochondria, fusion and genome recombination in plastids appear to be rare [17][18][19] . However, temperature-induced biparental inheritance also provides a mechanism for plastids to  Tables 1 and 2). Black horizontal bars show mean rates per genotype/treatment combination. Rates per experimental group were estimated (coloured horizontal lines) with CI95s (coloured boxes). Dashed lines depict the basal plastid paternal transmission (grey) and the theoretical maximum (black). Effect estimates were tested by simultaneous two-tailed Wald z-tests: dpd1 genotype (P = 6.22 × 10 −35 ), chilling treatment (P = 2.20 × 10 −126 ) and the interaction between both factors (P = 3.17 × 10 −10 ) were significant. ***P < 0.001, α = 0.05. f, Visualization of paternal plastid transmission by spectinomycin selection (Experiments 2 and 3;  spread through pollen between species and populations, a phenomenon known as chloroplast capture 22,54 . Interestingly, accumulating evidence suggests that the plastid genome plays roles in tolerance to low temperatures 55,56 , thus potentially providing a direct link between chilling-induced biparental plastid inheritance, selection and adaptation.
In summary, our data reported here demonstrate that plastid inheritance is controlled at two distinct steps in male gametogenesis and determined by synergistic gene-environment interaction. Our finding that uniparental inheritance of plastids breaks down under conditions of mild environmental stress indicates that organellar genomes experience episodes of biparental inheritance, and casts   considerable doubt on the long-held tenet that organelles are asexual genetic systems.

Plant material and growth conditions
Tobacco plants (Nicotiana tabacum cv. Petit Havana) were grown in standard greenhouse conditions at ~300 µE m −2 s −1 light intensity under a 16 h light/8 h dark regime (day temperature: ~25 °C, night temperature: ~20 °C). Transplastomic lines (with a wild-type nuclear background; WT ptGFP ) used as pollen donor harbour a gfp expression cassette and the selectable marker aadA encoding an enzyme conferring spectinomycin resistance 31 . Transplastomic lines (WT ptDsRed ) used for confocal microscopic studies harbour a DsRed expression cassette (driven by the ribosomal RNA operon promoter and a chimeric 5′ untranslated region from the chloroplast psbA and clpP genes) and the spectinomycin resistance marker aadA 57 . Seed germination and plant cultivation in vitro were performed on agar-solidified synthetic medium containing 3% (w/v) sucrose 58 . Abiotic stress treatments were performed in controlled-environment chambers.

Crossing experiments and screening for paternal plastid transmission
Homoplasmic transplastomic plants were raised under standard greenhouse conditions and transferred to the designated abiotic stress condition in controlled-environment chambers upon onset of flower development (Fig. 1a). To ensure that male gametophyte development occurred under stress, flower buds ≥10 mm in length were removed upon transfer (Extended Data Fig. 1). High light stress was applied by shifting plants to 1,000 µE m −2 s −1 light intensity. Drought stress was applied by withholding water until visible loss of turgor and wilting occurred. Heat stress was applied by shifting plants to 35 °C, and chilling stress was performed by shifting plants to 10 °C. Upon pollen maturation, large-scale crosses were conducted by manual pollination 31 , using pollen from transplastomic plants (WT ptGFP or T 1 dpd1 ptGFP ) grown in standard greenhouse conditions or under abiotic stress. Greenhouse-grown plants with wild-type plastids were used as maternal recipients: male-sterile Nt-nms plants in Experiment 1 31 , and emasculated wild-type plants in Experiments 2 and 3. The resulting progeny was assayed for paternal plastid transmission by germinating seeds on synthetic medium in the presence of spectinomycin (500 µg ml −1 ), at a density of approximately 500 seedlings per 180-mm-diameter Petri dish or 120 mm square plate.  Seedling phenotypes were analysed by visual screening under a stereomicroscope (Zeiss), followed by inspection of detected green (antibiotic-resistant) sectors by UV and confocal microscopy to test for GFP fluorescence. Paternal plastid inheritance (PPI) lines were transferred to soil, grown to maturity in the greenhouse and selfed for seed production. Seeds were assayed for stable paternal plastid inheritance by germination on agar-solidified synthetic medium with spectinomycin (500 µg ml −1 ).

Tissue culture and plant regeneration
To regenerate plants that contain paternal plastids, green (spectinomycin-resistant) sectors were excised from cotyledons or primary leaves and regenerated on agar-solidified plant regeneration medium 58 containing 3% (w/v) sucrose, 0.1 mg ml −1 1-naphtaleneacetic acid (NAA), 1.0 mg ml −1 6-benzylaminopurine (BAP) and 500 µg ml −1 spectinomycin. To eliminate spontaneous spectinomycin-resistant mutants, tissue samples were exposed to double selection on medium containing spectinomycin and streptomycin (500 µg ml −1 each). Spontaneous spectinomycin-resistant cells bleach out on this medium, whereas cells with aadA-expressing transgenic plastids remain green and continue to grow 59,60 . To obtain seeds, regenerated homoplasmic lines with paternal plastids were rooted and propagated on synthetic medium in the presence of spectinomycin (500 µg ml −1 ).

Light microscopy, UV microscopy and confocal laser-scanning microscopy
Green sectors potentially harbouring paternal plastids were identified in germinating seedlings by visual inspection and/or light microscopy using a stereomicroscope (Stemmi 2000-C; Zeiss). GFP fluorescence was detected with an MZ FLIII fluorescence stereomicroscope (Leica) using filters GFP2 (excitation filter: BP 480/40 nm, barrier filter: LP 510 nm) and GFP3 (excitation filter: BP 470/40 nm, barrier filter: BP 525/50 nm). Subcellular localization of GFP fluorescence was determined by UV microscopy with an Axioskop 2 (Zeiss; excitation filter: BP 450/90 nm, barrier filter: BP 515/65 nm), or by confocal laser-scanning microscopy (TCS SP8; Leica) using an argon laser for excitation (at 488 nm), a 500-510 nm filter for detection of GFP fluorescence and a 610-700 nm filter for detection of chlorophyll fluorescence. Subcellular localization of DsRed fluorescence was determined by confocal laser-scanning microscopy (TCS SP8; Leica) using a diode-pumped solid-state laser for excitation (at 561 nm), and a 575-605 nm filter for detection of DsRed fluorescence.

Aniline blue staining
To visualize the transient CCW, EBP grains were extracted from flower buds with size of 15-18 mm (plants grown at 20-25 °C) or 32-35 mm (plants grown at 10 °C). The CCW was stained by decolorized aniline blue solution (0.1% (w/v) aniline blue (Sigma) in 0.1 M K 3 PO 4 (pH 11.0)) at room temperature for 10 min. Stained samples were imaged by confocal laser-scanning microscopy (TCS SP8; Leica) using an argon laser for excitation at 405 nm, and a 494-544 nm filter for detection of fluorescence emission. Z-stack imaging was performed with a Leica TCS SP8 using the Galvo flow system and optimized settings with a step size of ~0.34 µm.

SYTO11 staining of in vitro germinating pollen tubes
To visualize the nuclear DNA of the generative cell within pollen tubes, two to three anthers were detached from flowers at anthesis and collected in a 2 ml Eppendorf tube. Mature pollen grains were released from the stamen by vortexing. Subsequently, the stamens were discarded and 100 µl pollen germination buffer (1.6 mM boric acid, 3.0 mM calcium nitrate, 1.0 mM potassium nitrate, 0.8 mM magnesium sulfate, 10% sucrose, pH 7.4) supplemented with 1 µM SYTO11 stain (Thermo Fisher) were added to the pollen. Incubation was done in the dark at room temperature for 120-180 min. Stained pollen tubes were imaged by confocal laser-scanning microscopy (TCS SP8; Leica) using a 488 nm argon laser for excitation and a 500-550 nm filter for detection of the SYTO11 signal.

Pollen viability assay
To determine the effect of temperature on pollen viability, anthers were detached from flowers at anthesis and collected in a 5 ml Eppendorf tube. Pollen grains were harvested by vortexing with pollen harvesting buffer (100 mM Na 3 PO 4 (pH 7.0) and 1 mM Na 2 EDTA (pH 8.0)), followed by centrifugation at 1,000 × g for 5 min. The pollen pellet was then resuspended and stained in pollen viability solution (100 mM Na 3 PO 4 (pH 7.0), 1 mM Na 2 EDTA (pH 8.0), 10 µM propidium iodide (Abcam) and 20 µg ml −1 fluorescein diacetate Sigma) at room temperature for 30 min. Stained samples were imaged by confocal laser-scanning microscopy (TCS SP8; Leica). Fluorescein diacetate was excited with a 488 nm argon laser and the emission signal was collected at 500-550 nm. Propidium iodide was excited with a 561 nm argon laser and its emission signal was collected at 600-650 nm.

DAPI staining
To visualize nuclear and organellar DNA, uninucleate microspores (UNM) and mature pollen grains (MPG) were extracted from flower buds with sizes of approximately 10 mm (for UNM) or from flowers at anthesis (for MPG). DNA was stained with 4,6-diamidino-2-phenylindole (DAPI) staining solution (100 mM Na 3 PO 4 (pH 7.0), 0.1% (v/v) Triton X-100, 1 mM Na 2 EDTA (pH 8.0) and 1 µg ml −1 DAPI) at room temperature for 30 min. Stained samples were imaged by confocal laser-scanning microscopy (TCS SP8; Leica) using a 405 nm laser diode for excitation and a 430-495 nm filter for detection of fluorescence emission. For the pollen squash method, stained pollen grains were placed on a glass slide and gently squashed by tapping on the cover slip 41 . Squashed pollen were imaged by confocal microscopy (Leica TCS SP8), with z-stack imaging using the Galvo flow system and optimized settings with a step size of ~0.30 µm.

NtDPD1 gene identification and sgRNA design for genome editing
The protein sequence of DPD1 from Arabidopsis thaliana (AT5G26940) was used as query in a search against the genomic scaffolds in the available draft genome of N. tabacum 61 . Two homologues were identified and named NtDPD1S and NtDPD1T (scaffolds Nitab4.5_0002715 and Nitab4.5_0014337, respectively). The transcript structures of both loci were deduced from the sequences present in the associated transcriptome (Nitab4.5_0002715g0070.1 and Nitab4.5_0014337g0020.1). Both NtDPD1 genomic sequences were divided into 2 kb fragments and used as input for sgRNA generation using CRISPOR 62

Construction of a plant transformation vector for genome editing of NtDPD1
Plasmid pEG001 was assembled for constitutive expression of cas9 in plant cells. pEG001 is a derivative of plasmid pJF1031 described previously 63 . The backbone of pJF1031 was excised by digestion with the restriction enzymes SpeI and XbaI and purification of the ~14.5 kb fragment obtained, resulting in the removal of the promoter driving cas9. The ~1.5 kb promoter of the HPL gene of A. thaliana (hydroxyperoxide lyase, AT4G15440) was amplified from the pORE E2 plasmid 64 using primers oEG126 and oEG127 (Extended Data Table 4), and integrated into the pJF1031 backbone through a Gibson assembly reaction 65 (New England Biolabs), resulting in plasmid pEG001. For gene editing at the NtDPD1 loci, vector pEG037 was constructed. Primers oEG315 and oEG320 (Extended Data Table 4) were used to amplify Article https://doi.org/10.1038/s41477-022-01323-7 a fragment of plasmid pJF1046 63 containing part of an sgRNA scaffold, the U6 promoter and the U6 terminator. The primers introduce the targeting sequences of sgRNA L or R, respectively, as well as BsaI sites at both ends of the fragment. The PCR product was cloned into pEG001 through a simultaneous BsaI restriction and ligation reaction 66 . The resulting binary vector pEG037 for mutagenesis of NtDPD1 contains a hygromycin resistance marker for selection in planta and a kanamycin marker for selection in bacteria. Sequences of all oligonucleotides used in this study are provided in Extended Data Table 4.

Generation of a quadruple dpd1 knockout line
Transplastomic plants with a wild-type nuclear background (WT ptGFP ) were supertransformed with Agrobacterium tumefaciens strain GV2260 harbouring vector pEG037 using the leaf disc infiltration method. Selected hygromycin-resistant plant lines were maintained in medium containing hygromycin (15 mg l −1 ) and cefotaxime (250 mg l −1 ) for continued selection of transgenic plant cells and elimination of residual Agrobacterium cells, respectively. Extended Data Fig. 3 illustrates the PCR-based genotyping (Reactions 1-3) of transgenic lines and the screening procedure that led to the isolation of the double dpd1 mutant (tetra-allelic). Due to the allotetraploidy of the N. tabacum genome, the screening process was tailored to identify plants with four knockout DPD1 homeoalleles in the diploid state (that is, two S and two T alleles; Fig. 4a). PCR and sequencing were employed to identify plants with somatic Cas9 activity and desirable knockout mutations in dpd1 (Extended Data Fig. 4). Additional regeneration steps in tissue culture helped with purification of desired mutation events and reduction of the allele complexity arising from constitutive Cas9 expression.
Polymorphisms between the S and T loci allowed classification of the amplified sequences as S or T homeoalleles after sequencing (Fig. 3a,b and Extended Data Fig. 4). Reaction 1 (Extended Data Fig. 4b) generates a short product only when a large deletion was generated between targeting sites L and R. Presence of this product implies that the plant line possesses an active Cas9, and at least one mutant allele can confidently be identified by this PCR. The primers used in Reaction 2 (Extended Data Fig. 4b) flank the site targeted by sgRNA L. This reaction was performed to identify the other three mutant alleles by direct Sanger sequencing of amplified PCR products from ~320 T 0 samples analysed. Occurrence of small InDels often caused Reaction 2 to produce electropherograms showing several overlapping peak profiles. Whenever profiles of up to three overlapping sequences were obtained, deconvolution of sequence strings was attempted by visual inspection. At the L site, Cas9 frequently caused small InDels (from −4 to +1 bp) and produced recognizable shifted peak patterns compared to the known S and T wild-type sequences, which could be deconvoluted even when three sequences were overlapping. After three rounds of PCR screening and two rounds of regeneration in tissue culture, a double dpd1 mutant plant line was isolated (for simplicity, referred to as dpd1). Sanger sequencing confirmed that dpd1 possesses four distinct mutant alleles at the L site (Fig. 3a,b and Extended Data Fig. 4). One of the sequences was a large deletion, and the three other sequences were frameshift mutations.
The T 0 dpd1 plant was transferred to the greenhouse where T 1 seeds were obtained by selfing. Presence of the mutations at the L site was confirmed by genotyping T 1 seedlings (performing Reactions 1 and 2; Extended Data Fig. 4b). Allele segregation in the T 1 generation facilitated the individual sequencing of the four alleles after PCR and gel purification, as Reaction 2 yields products of different size for the S and T loci. Mutations at the sgRNA R site were analysed in T 1 seedlings by Reaction 3 (Extended Data Fig. 4b), applying visual deconvolution when necessary. The linkage between mutations in the L and R target sites was established by observing co-segregation of genotyped alleles in the T 1 generation (23 seedlings analysed). No other alleles were identified in the final dpd1 line, suggesting that the four mutant alleles detected in the T 0 generation had become fixed.

Isolation of nucleic acids and PCR
For DNA gel blot analysis, total plant DNA was isolated using a cetyltrimethylammoniumbromide (CTAB)-based protocol 67 . Extracted DNA samples were digested with the restriction enzymes XhoI and EcoRV, separated by gel electrophoresis in 0.8% agarose gels and blotted onto Hybond N nylon membranes (GE Healthcare) using standard protocols. For hybridization, α[ 32 P]dCTP-labelled probes were generated by random priming (Multiprime DNA labelling kit, GE Healthcare). A PCR product covering part of the 16S rRNA gene (amplified by primers P16Srrn-F and P16Srrn-R, and purified by agarose gel electrophoresis) was used as probe for RFLP analysis. Hybridizations were carried out at 65-68 °C in rapid hybridization buffer (GE Healthcare) following the manufacturer's instructions.
For genotyping reactions, genomic DNA was extracted from leaf tissue with the Extract-N-Amp kit (Sigma-Aldrich). One µl was used as template for PCR. For cloning and for the initial PCR-based screening of dpd1 mutations, Phusion DNA polymerase (Thermo Fisher) was used. In all other PCR reactions, DreamTaq DNA polymerase (Thermo Fisher) was used. PCR products were column-purified before sequencing (NucleoSpin Gel and PCR Clean-up, Macherey-Nagel).

Quantification of plastid DNA in pollen
Total DNA (comprising both nuclear and organellar DNA) of enriched pollen fractions was prepared using a published protocol 41 . Tobacco pollen grains were collected by rubbing dehiscent anthers from three opened flowers (WT ptGFP and dpd1 ptGFP plants, respectively) onto the wall of a 1.5 ml Eppendorf tube, followed by addition of 100 µl of distilled water for resuspension. Due to the procedure used for pollen harvest, some level of contamination by other anther cell types cannot be avoided, hence the samples should be considered as enriched pollen preparations (rather than pure pollen fractions). The enriched pollen suspension was incubated at 95 °C for 5 min, centrifuged at 16,000 x g for 5 min and then subjected to real-time PCR to determine the relative amounts of nuclear and plastid DNA. 18S rDNA (nuclear gene) and aadA (transgene present in the plastid genome of WT ptGFP and dpd1 ptGFP ) were used as proxies for nuclear and plastid DNA abundance, respectively. Primers oCK68 and oCK69 were used for 18S rDNA amplification, and primers oKPC579 and oKPC580 for aadA amplification (Extended Data Table 4). Quantitative real-time PCR was conducted using the LightCycler 480 Real-Time PCR system and LightCycler SYBR Green reaction mixtures (Roche). The relative amount of plastid DNA in the enriched pollen fractions was determined by the abundance of the aadA amplicon normalized to the abundance of the 18S rDNA. Relative quantification was performed using the ΔΔCt method 68 .

Modelling and statistics
The quantitative effects of genotype and stress treatments on paternal plastid transmission, plastid inclusion in the generative cell and pollen survival (after chilling stress, or when comparing WT ptGFP and dpd1 ptGFP mutant) were estimated using generalized linear models. Models were constructed with the maximum-likelihood method in R v3.5.3 (https:// www.R-project.org/). Genotype, treatment and experiment (stage of visualization in pollen; independent experiments to determine plastid transmission) were proposed a priori as explanatory variables. Five models were selected for data analysis (Models 1-5).
The datasets for plastid inclusion in the generative cell (for Model 2), pollen survival under chilling stress (Model 4) and pollen survival in the wild-type and the dpd1 genotypes (Model 5) were modelled as proportions of binary outcomes using the binomial distribution. The total amount of pollen grains analysed per unit of replication was provided in the 'weights' argument of glm(), as recommended 69 . The two paternal plastid transmission datasets (one representing Experiment 1, Article https://doi.org/10.1038/s41477-022-01323-7 the other including Experiments 1, 2 and 3) were modelled using the negative binomial distribution (for Models 1 and 3), after evidence of overdispersion was found in preliminary Poisson versions of these models. The unit of replication in Experiments 1, 2 and 3 is the 'harvest': a batch of seeds obtained from a set of maternal recipients (5-30 plants) fertilized by pollen donors from a single treatment (3-8 plants). For modelling of proportions and rates, the count of seedlings with sectors per unit of replication was set as the response variable, while the total amount of seedlings analysed was provided as offset. 'log' was used as the link function for all models.
For each dataset, a starting model was generated that includes effect parameters for proposed explanatory variables and interactions. All effect parameters were modelled relative to specific levels of the explanatory variable ('contrast to treatment'). Models were selected according to minimal AIC (Akaike's Information Criterion), as they are considered the most parsimonious 70 : a version corrected for small sample sizes (AICc) was used as recommended previously 71 . From the starting and following models, parameters were removed sequentially and only if they led to a reduction in AICc. Overdispersion of the preliminary Poisson models was checked by performing a likelihood ratio test (LRT): this change resulted in a large reduction in AICc. The final models with minimal AICc (one per dataset) were designated Models 1-5. Goodness of fit was tested for these models using LRTs comparing (1) each selected model against a 'saturated' model, where the test provides a general measure of whether the model fits the data well, or (2) the selected model against a competing model to measure relative goodness of fit. Models 1-5 were confirmed to fit as well as the saturated model. Parameter estimates in the selected models were evaluated with simultaneous Wald's z-tests (two-tailed). No corrections for multiple comparisons were applied for z-tests.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
Data supporting the findings of this work are available within the paper and its Supplementary Information files. Sequences from Arabidopsis (AT5G26940.1 and AT4G15440) are available through TAIR (https:// www.arabidopsis.org/). Genomic sequences from Nicotiana (scaffolds Nitab4.5_0002715 and Nitab4.5_0014337) and transcripts (Nitab4.5_0002715g0070.1 and Nitab4.5_0014337g0020.1) are available at Sol Genomics Network (https://solgenomics.net/). Source data are provided with this paper.

Code availability
The R code used for assembling the statistical models is available on GitHub at https://github.com/egonzalezduran/ modelsplastidinheritance.   Tables 1 and 2). Each replicate consisted of a single harvest of pollen from 2-3 flowers of a single plant, which were stained and imaged in the same session (delimited by dashed lines). During data analysis, it was found that there was significant variation between replicates (represented by replicate parameters with P < 0.05 in Extended Data Table 2). To better represent this variation, the graph shows the proportions of viable pollen in the 5-9 image fields (coloured circles) photographed in each replicate. The mean estimates per genotype are indicated with a black horizontal bar. β dpd1 represents the effect of the dpd1 genotype on pollen survival (P = 3.00 × 10 −9 ) compared to the wild type at normal temperature conditions (20-25 °C), expressed as log 10 of the fold change (Extended Data Table 2). The parameter estimate was tested by a two-tailed Wald z-test. ***, P < 0.001; α = 0.05. d, Confocal images of WT ptGFP and dpd1 ptGFP mature pollen stained with DAPI. Images were taken after applying the pollen squash method to release the mature pollen from the anthers. Z-stack overlaid images with maximized projection are displayed (total width of z-slices of WT ptGFP and dpd1 ptGFP is 2.15 µm and 1.78 µm, respectively). Arrowheads indicate the vegetative nucleus (VN) and the generative nucleus (GN). Arrows indicate organellar DNA released from squashed mature dpd1 ptGFP pollen. This experiment was repeated two times and representative images are shown. Scale bar, 10 µm.
Article https://doi.org/10.1038/s41477-022-01323-7 Extended Data Table 1 | Specification of generalized linear models for analysis of paternal plastid transmission and plastid inclusion  are the models analyzed in the manuscript. Models within a model family (Model Fam.) that are designated by a lowercase m are intermediates in the modelling process (that is, have parameters that were eliminated or added during the AICc minimization process). All models have 'log' as link function. †Dispersion parameter θ estimated from the data. The standard error of the estimate is given in parentheses. ‡ Includes the intercept, parameters for the effects of variable, and interactions. In negative binomial models, it also includes the dispersion parameter θ. The extra parameters (+) or the missing parameters (-) that the respective model includes compared to the selected model are indicated in parentheses. T, Temperature; G, Genotype; E, Experiment (in the case of Model 2, E refers to the experiments performed at different stages of development, whereas in the case of Model 3, E refers to the three independent experiments conducted to determine the rate of paternal plastid transmission). §LRT compares two models by checking whether λ LR (difference in residual deviances) surpasses the critical value χ 2 0.05 with Δdf degrees of freedom, where Δdf is equivalent to the difference in the number of parameters between the two models. For all LRTs: α = 0.05; ns, P > 0.05; **, P < 0.01; ***, P < 0.001. Null hypotheses are formulated depending on the comparisons. LRTs are single-tailed by definition, and no corrections for multiple comparisons are performed. || LRT between a candidate model and the saturated model built from the same data (also called deviance test) is a general measure of goodness of fit of the candidate model. The saturated model has as many parameters as observations, and it fits maximally to the data by definition (residual deviance is zero). Thus, λ LR is equal to the residual deviance of the candidate model. The null hypothesis is that the candidate model fits the data as well as the saturated model. Non-rejection of the null hypothesis is interpreted as evidence that the tested model fits well the data. ¶ LRT between a negative binomial model and its nested Poisson is a test for data overdispersion (when the conditional variance is higher than the conditional mean). The null hypothesis of the test is that the Poisson and negative binomial models fit to the data equally well. A rejected null hypothesis indicates the negative binomial model fits better to the data. $ Model Family 3 includes the calculation of the interaction between genotype and temperature across all experiments, as well as the interactions involving Experiment 2. By contrast, triple interaction terms of the form GxTxE could not be calculated, as well as the GxE and TxE interactions involving Experiment 3. The independent experiments of paternal plastid transmission did not always contain experimental groups of all levels of genotype and temperature, thus leaving those interactions undefined for the dataset.

Extended Data Table 2 | Parameter estimates obtained from modeling
Parameter values correspond to fold changes expressed in log 10 . Wald z-tests (z-statistic) are two-tailed with no corrections for multiple comparisons.

Parameter (β)
Base level for contrast