Maize germinal cell initials accommodate hypoxia and precociously express meiotic genes

Summary

In flowering plants, anthers are the site of de novo germinal cell specification, male meiosis, and pollen development. Atypically, anthers lack a meristem. Instead, both germinal and somatic cell types differentiate from floral stem cells packed into anther lobes. To better understand anther cell fate specification and to provide a resource for the reproductive biology community, we isolated cohorts of germinal and somatic initials from maize anthers within 36 hours of fate acquisition, identifying 815 specific and 1714 significantly enriched germinal transcripts, plus 2439 specific and 2112 significantly enriched somatic transcripts. To clarify transcripts involved in cell differentiation, we contrasted these profiles to anther primordia prior to fate specification and to msca1 anthers arrested in the first step of fate specification and hence lacking normal cell types. The refined cell-specific profiles demonstrate that both germinal and somatic cell populations differentiate quickly and express unique transcription factor sets; a subset of transcript localizations were validated by in situ hybridization. Surprisingly, germinal initials starting five days of mitotic divisions were significantly enriched in >100 transcripts classified in meiotic processes including recombination and synapsis, along with gene sets involved in RNA metabolism, redox homeostasis, and cytoplasmic ATP generation. Enrichment of meiotic-specific genes in germinal initials challenges current dogma that the mitotic to meiotic transition occurs later in development during pre-meiotic S phase. Expression of cytoplasmic energy generation genes suggests that male germinal cells accommodate hypoxia by diverting carbon away from mitochondrial respiration into alternative pathways that avoid producing reactive oxygen species (ROS).

Introduction

Despite copious attention dedicated to uncovering the developmental mechanism(s) underlying germinal fate acquisition, very little is known about the initial transcriptome changes that differentiate reproductive cells from pluripotent precursor cells and from the somatic cells constituting the bulk of fertile organs. In anthers, the male sexual organ of flowering plants, a defined cohort of germinal cells is specified de novo early in flower development and mature as a group (Scott et al., 2004); consequently, large populations of equivalently staged germinal cells are available for isolation and transcriptome analysis.

Recent advances in analysis of anther development include realization that the original lineage model to establish the germinal and interior cell types is incorrect. This model posited the existence of a single enlarged founder cell in the corner of each anther lobe, which was proposed to undergo an asymmetric cell division to establish a subepidermal somatic cell, founder of a somatic lineage, and an interior primary sporogeneous cell, founder of the germinal lineage (Sanders et al., 1999). With advances in microscopy and analysis of additional mutants, it is now clear that for both dicots (Feng, et al., 2013) and monocots (Kelliher and Walbot, 2012) there is a population of equivalent L2-d cells in immature anther lobes and that specification of the germinal cell population is a process that involves multiple lobe cells adopting that fate simultaneously (Figure 1).

Figure 1.

Staging of maize and Arabidopsis anthers. Shown in the upper block are aspects of maize anther development organized by anther length (mm) and in the lower block information about Arabidopsis is presented. The first line presents stages by days post-primordium, then there is a lineage diagram to illustrate the ontogeny of each of the five final lobe cell types, and next a cartoon depicting the successive stages of development. The color code is as follows: gold = L1-d cells; pale yellow = L2-d cells, white = connective; red = archesporial cells; pale blue = endothecial cells; dark blue = secondary parietal cells; pale green = middle layer cells; green = tapetal cells. The number of cells belonging to each cell layer at several stages is written below the cartoons. In the bottom panel, landmark events in Arabidopsis anther development are organized by anther length and days post-primordium.

The mechanism of germinal specification was recently defined for the male lineage. Maize anther primordia have four lobes, each consisting of an epidermal layer (Layer1-derived or L1-d cells) and pluripotent internal cells derived from the second layer of the meristem (L2-d cells) (Figures 1 and 2a) (Goldberg et al., 1993). At the primordium stage both the presumptive epidermal and L2-d internal cells are competent to differentiate as pre-meiotic cells; thus despite their origin from different meristem layers the cells are all considered pluripotent initials at this stage (Kelliher and Walbot, 2012). In maize anther primordia, rapid proliferation of the floral stem cell pool generates a low oxygen environment in the central most column of L2-d cells, triggering germinal fate acquisition mediated through the MSCA1 glutaredoxin (Kelliher and Walbot, 2012). The germinal initials, or archesporial (AR) cells, rapidly enlarge and secrete the MAC1 protein, a mobile ligand that sets somatic fate in the enveloping monolayer (Wang et al., 2012) by triggering a periclinal division to establish a bilayer consisting of the subepidermal endothecial (EN) cells and the secondary parietal layer (SPL) cells adjacent to the AR (Kelliher and Walbot, 2011) (Figures 1 and 2b). Progression from pluripotent progenitors to the patterned anther with a defined germinal / somatic boundary starts in the central zone (observed lengthwise) and proceeds towards the anther base and tip; these specification events occur over about 36 hours. After this, the somatic layer cells proliferate anticlinally for ~3 days and then the SPL cells undergo an additional periclinal division event, giving the organ a four-layer, structured wall and providing the germinal cells with nutrition and developmental cues. Meanwhile, the AR mature as a cohort: each cell conducts 3–4 clonal mitotic divisions and enlarges again before the group synchronously enters meiosis, about one week after specification.

Figure 2.

Experimental design. (a) 3D cartoon of a 0.30 mm anther. (b) Developmental progression of an anther lobe, from primordia to post cell fate specification. Fate acquisition initiates in the widest part of the anther and proceeds towards base and tip to establish the AR column. L2-d cells experience hypoxia causing a signal relayed through the glutaredoxin MSCA1 to result in AR fate acquisition, first observed by cell enlargement in 0.16 mm anthers and completing a full column within 18 hours by 0.22 mm (Kelliher and Walbot, 2012). Differentiating AR cells secrete the MAC1 protein ligand, which binds to receptors in surrounding L2-d initiating a signal cascade that results in a periclinal division to generate the endothecium and secondary parietal layer (SPL), as the anther elongates from 0.20 – 0.28 mm. These events precede a 7-day mitotic amplification period during which each AR cell undergoes 3–4 rounds of division prior to meiotic initiation in 1.5 mm anthers. (c) Confocal reconstructions illustrate target tissues and hybridization strategy with balanced dye swap. Laser capture microdissection (LCM) was used to isolate AR and somatic cells and their transcriptomes were directly compared to each other and indirectly to the 0.15 mm undifferentiated anther. As newly specified AR have undergone at most one mitosis since specification, gene expression changes are likely fundamental to setting germinal and somatic fate from common stem cell progenitors. (d) VENN diagram comparisons of LCM-dissected tissues contrasted with anther primordia. Counts denote presence / absence in indicated samples. Among the 28075 transcripts shared between AR and somatic tissues, 3826 (1569+1953+145+ 159) were significantly differentially expressed. (e) Experiment design of WT versus msca1 comparison: two biological reps were run in duplicate with balanced dye swap. (f) VENN diagram results of indirect comparisons between msca1, WT, and cell-specific gene sets.

Transcriptome profiling has illuminated aspects of anther development at pre-meiotic, meiotic, and post-meiotic stages in rice (Aya et al., 2011; Deveshwar et al., 2011; Lu et al., 2006; Wei et al., 2010), maize (Ma et al., 2006; Ma et al., 2007; Ma et al., 2008; Skibbe et al., 2009; Nan et al., 2011), and Arabidopsis (Wellmer et al., 2004; Mandaokar et al., 2006; Feng et al., 2012; Ma et al., 2012). Mixed stages of meiotic cells (Tang et al., 2010; Schmidt et al., 2011) and post-meiotic haploid male gametophytes (Honys and Twell, 2004; Ma et al., 2008; Calarco et al., 2012) have been evaluated, however, no plant or animal study has analyzed just-committed pre-meiotic cells directly following their differentiation from pluripotent somatic stem cells. Maize AR cell expansion signifying fate acquisition initiates in late day one of anther development (Kelliher and Walbot, 2012), yet the only flowering plant marker for the AR, Sporocyteless of Arabidopsis, is not expressed until long after germinal specification (Schiefthaler et al., 1999; Yang et al., 1999) and has no ortholog in maize. In this study we profiled anther germinal and somatic cell transcriptomes within 24 – 36 hours of specification, contrasting their expression profiles with that of their pluripotent precursors and to msca1 mutant anthers lacking normal germinal and somatic cell types. We identify hundreds of AR and somatic markers, with dozens confirmed by qRT-PCR and sixteen by in situ hybridization.

Results

Overview of maize anther ontogeny and comparison to Arabidopsis anther stages

Figure 1 provides an orientation to developmental events in maize anthers, charting them against days of development and anther length. The earliest stages are the focus of this study: anthers prior to AR specification (<0.15 mm), anthers during the instructive period with AR triggering the periclinal division of the L2-d cells (0.18 – 0.25 mm), and completion of AR, EN, and SPL fate specification events by 0.30 mm. The relationship of these early events to the final anatomy of an anther lobe is illustrated in a lineage diagram and by a set of cartoons (Figure 1).

Cell counts at maize anther stages are available (Figure 1), however, while precisely the same early stages are present in Arabidopsis, these anthers are very tiny and cell counts at the relevant stages have not yet been published. Using data on Arabidopsis floral (Smyth et al., 1990) and anther progression (Sanders et al., 1999), the sizes of Arabidopsis anthers could be assigned for key stages (Figure 1); the major difference between maize and Arabidopsis anther development is the intercalation of two phases of rapid mitotic proliferation after cell fate specification events, during the 0.28–0.55 mm period and the 0.7 – 1.0 mm maize stage (Figure 1). All of these maize cell divisions are anticlinal, contributing to anther growth in length and girth. Prolonged periods of anticlinal-only divisions do not exist in Arabidopsis anthers, and time compression results in more rapid ontogeny. For example, PMC reach the start of meiotic prophase about 5 days faster in Arabidopsis than in maize (Figure 1), and EN and SPL specification is 16 h longer in maize, likely because there are already many more cells in the much longer maize anther. Both time compression and the absence of cohorts of similar-aged flowers on an inflorescence contribute to the difficulty of recovering Arabidopsis anthers from the early developmental stages we focus on in this study.

Germinal and somatic differentiation is accompanied by the expression of thousands of transcripts not detected in anther primordia

Germinal and somatic cell RNA samples were isolated in biological duplicate by laser capture microdissection (LCM) from mid-day three maize anthers (Figure 2c); hybridization analysis on a 44K Agilent microarray in technical duplicate showed that thousands of transcriptome changes had already occurred in both cell populations compared to 0.15 mm anther primordia (Figure 2d). As was found with more advanced stages (Ma et al., 2008), anthers have a complex transcriptome: both undifferentiated 0.15 mm anthers and LCM-dissected cell types from 0.35 mm anthers express over 25,000 transcripts, more than half of the annotated maize gene set (Figure 2d).

We first addressed transcripts absent in primordia (OFF) that are detectable (ON) after fate specification. 4344 such transcripts define 0.35 mm differentiating anthers compared to primordia (408 AR-only, 847 somatic-only, and 3089 present in both cell types), and 1280 of these were above the median and therefore represent abundant stage-specific markers (Table 1). Additionally, 1999 transcripts present in primordia were absent in one tissue and 2180 were absent from both. In the direct comparison of LCM-dissected samples, 815 transcripts were exclusive to the AR and 2439 were detected only in somatic tissues (Figure 2d).

Table 1.

Counts of differentially expressed transcripts, sorted for expression intensity by quartile (columns) and log-fold change between samples (rows). (top panel) Germinal cell probes. (bottom panel) Somatic cell probes.

Germinal Cells		Expression intensity by quartile
		1st Q	2nd Q	3rd Q	4th Q	Total
	ON/OFF	758	53	4	0	815
Differential log ratio	0.58 < 1.0	1	126	364	757	1248
	1.0 – 1.5	0	11	87	223	321
	1.5 – 2.0	0	0	38	58	96
	2.0 – 3.0	0	0	8	36	44
	> 3.0	0	0	0	5	5
	Total	759	190	501	1079	2529

Somatic Cells		Expression intensity by quartile
		1st Q	2nd Q	3rd Q	4th Q	Total
	ON/OFF	2097	309	28	5	2439
Differential log ratio	0.58 < 1.0	0	144	421	534	1099
	1.0 – 1.5	0	12	235	260	507
	1.5 – 2.0	0	0	79	183	262
	2.0 – 3.0	0	0	6	199	205
	> 3.0	0	0	0	39	39
	Total	2097	465	769	1220	4551

Given the high reproducibility of the array platform employed (r² > 0.92 among replicates, Ma et al., 2007), differential expression analysis was also evaluated using log fold change >0.58, p <0.05, with a False Discovery Rate of 0.13% in classifying a transcript as significantly enriched. As shown in Table 1, 1714 AR and 2112 somatic transcripts were differentially expressed using these classification criteria; 92.1% of these cell-enriched transcripts were also found in primordia but during differentiation are significantly altered in cell-type specific abundance. Collectively these ON/OFF and differentially expressed transcriptome data illustrate significant reprogramming of the transcriptional palette accompanying reproductive and somatic fate acquisition.

Validation of cell-type specificity by comparison to the msca1 mutant

As a biological validation of the classification of anther germinal and somatic differentiation genes we directly compared the transcriptomes of 0.18 - 0.22 mm WT and msca1 whole anthers, then contrasted these profiles with those from the LCM samples. In the absence of the encoded glutaredoxin (Albertsen et al., 2009), msca1 anthers fail to specify AR cells. Because the germinal cells are required to trigger a periclinal division and then differentiation of the single layered sheath of subepidermal L2-d cells into endothecium and SPL, the mutant lacks interior anther cell types. Surprisingly, the anther organ continues to grow with a normal gross morphology, however, all of the constituent cell types are aberrant. One or more strands of vascular tissue replace the column of AR in each lobe, and no somatic L2-d anther cell types differentiate (Kelliher and Walbot, 2012); even the L1-d mis-differentiates because guard cells form in the epidermis (Chaubal et al., 2003). Previous transcriptome profiling experiments (Ma et al., 2008) indicated that msca1 anthers express leaf-stem characteristics and should be considered strictly somatic organs. At the 0.2 mm stage, msca1 anthers are just starting to morphologically differentiate from wild type, developing proto-vascular bundles in the center of the L2-d stem cell pool, while the surrounding L2-d cells appear nearly normal at this stage (Figure 2e) (Kelliher and Walbot, 2012). Of the 2529 transcripts classified as germinal-specific or –enriched after AR differentiation, 0.2 mm WT anthers express 1900 of them (75%), while 0.2 mm msca1 anthers only express 1766 (70%). For the somatic transcript set, the overlap is 3364 (74%) genes in WT and 3158 in msca1 (69%) (Figure 2f). Thus the majority of transcripts specific to germinal or somatic tissues are found in both 0.2 mm WT and msca1. This molecular observation along with continued growth of anthers with reasonably normal morphology suggests that some anther developmental programs can be initiated and maintained in the absence of AR differentiation.

Although the transcriptome profiles in msca1 and WT are qualitatively similar, there is moderate to strong down-regulation of the majority of cell-specific genes: Among the 1900 germinal markers found in 0.2 mm WT, 134 (7%) are absent in msca1 and 171 (9%) are significantly down-regulated (including 6 of the top 7 AR markers by expression differential) (Figure 2f). A further 1128 (59%) are expressed at levels that fall below 90% that of WT. For the somatic gene set, 1034 (31%) of the 3364 somatic marker transcripts found in 0.2 mm WT are absent in msca1 or expressed at < 90% of WT levels. These down-regulated transcripts are present in primordia and subsequently activated during germinal and somatic fate acquisition, which fails in msca1. The fact that AR cells differentiate first then set somatic fate during this 0.2 mm stage is reflected in the numerous AR markers mis-expressed in msca1 compared to relatively modest disruption to the somatic marker set at this early developmental stage when periclinal division of the L2-d is only partially completed (Figure 2f).

In situ hybridization analysis confirms AR and somatic expression patterns

For additional validation of transcript specificity we filtered the germinal and AR lists for high enrichment (log fold change >2) and abundance (expression >median), identifying 49 germinal (Table S1) and 244 somatic (Table S2) candidate cell-type specific markers. Initial screening by qRT-PCR confirmed 52/59 (88%) transcripts as significantly enriched in the expected tissue (Tables S3 and S4). A subset was selected for RNA in situ hybridization on 0.30 – 0.35 mm anthers, and 16/19 (84%) were confirmed as germinal- or somatic- specific markers while 3 gave no signal (Figure 3). The eight probes hybridizing to AR cells are the earliest germinal markers reported in flowering plants, and the first ever in monocots (Figure 3a-p). These include transcripts encoding Proteophosphoglycan4, a protein putatively secreted into the unique AR cell wall (Figure 3a), and Bax inhibitor-1, which is regulated by hypoxia and serves to suppress cell death during stress by promoting cytoplasmic homeostasis (Ihara-Ohori et al., 2007) (Figure 3g). Additional markers include the glutaredoxin encoded by Msca1, critical for hypoxia-mediated cell fate specification of the AR cell cohort (Figure 3k), a homolog of leafbladeless1 involved in second strand small RNA (sRNA) biosynthesis (Timmermans et al., 1998) (Figure 3m), and an RNA/DNA-binding winged helix transcription factor localizing to both AR and SPL cell types (see Appendix S1). Successful hybridizations also confirmed eight somatic markers (Figure 3q-af), including RNA-directed RNA polymerase 6 (RDR6) (Fig 3q), Argonaute10a (Figure 3s), and a MADS box transcription factor (TF) (Figure 3u). The beta-amylase is specific to the bipotent SPL (Figure 3aa) and hence represents a marker for this cell type.

Figure 3.

In situ hybridizations of anther lobe germinal (a-r) and somatic (s-aj) cell-specific candidate markers performed on 0.30 – 0.35 mm anthers, with cartoons diagramming cell layer positions within the lobe. In the image panels, arrows indicate cell-type specific hybridization patterns. (a) Proteophosphoglycan (ppg4) has no defined role, but it is the third most highly enriched AR transcript. (c) ARGONATE 4b is putatively involved in RNA-directed DNA methylation. The metabolic gene pyridine disulfide oxidoreductase (e), the suppressor of cell death gene Bax1-inhibitor (g), and the glyoxylate cycle enzyme aconitate hydratase (i) are all AR-enriched, while the antisense transcript of Bax1-inhibitor is common to the AR and SPL (h). (k) Msca1, critical for hypoxia-mediated AR cell specification, shows post-differentiation restriction to the germinal cells. (m) Leafbladeless1 is involved in second strand sRNA synthesis, and is required for correct abaxial-adaxial leaf patterning and anther lobing; this probe is to a member of the gene family. (o) A winged-helix transcription factor is enriched in both AR and SPL cells. (q-af) Somatic probes. (q) RNA-DIRECTED RNA POLYMERASE 6 is involved in second strand sRNA synthesis, while ARGONATE 10a (s) is involved in RNAi. The MADS-BOX 4 transcription factor (u) and a serine / threonine protein kinase (w) are highly enriched in SPL, endothecial and epidermal layers. (y,z) The phosphatase PTPLA exhibits patterned sense and antisense probe hybridizations. (aa,ab) The beta-amylase transcript is specific to the SPL, while the sense probe hybridized to both AR and SPL cells and lightly to the endothecium, indicative of antisense transcription at this locus. (ac,ad) The MADS-box transcription factor hybridizes to SPL and endothecium while the sense probe detected antisense transcription in AR and, to a lesser extent, SPL cells. (ae,af) Finally, a transcript encoding an X1-like transcription repressor was confirmed as a somatic marker, while the sense probe indicated epidermal antisense transcription.

Many genes expressed in maize anthers also produce a long antisense transcript (Skibbe et al., 2009), and 5/19 (26%) sense probes hybridized in a cell-specific pattern (Figure 3h,z,ab,ad,af). Sense transcripts accumulating in either AR or somatic cells are commonly matched by antisense expression in the other cell type, suggesting that sense-antisense patterning facilitates rapid differentiation from pluripotent progenitors. It is also striking that the Bax inhibitor-1 sense probe yielded a stronger signal than the antisense probe, and specifically accumulated in AR and SPL cells. Given the central importance of hypoxia as a developmental cue and physiological condition in maize anthers, the regulation of hypoxia-mediated gene expression may be more complex than anticipated.

Precocious expression of meiotic process genes

After specification, the ~10 AR initials proliferate over the next five days, reaching a final population of ~160 AR per lobe by the 1 mm stage. Already the largest cells within anther lobes, after 1 mm the AR differentiate as gigantic pollen mother cells (PMC). During the maturation from AR to PMC, cells are proposed to acquire meiotic competence during a several day pre-meiotic period (Boateng et al., 2008; Nan et al., 2011). Similarly, animal germinal cells pause prior to meiosis, and during pre-meiotic S-phase are thought to remodel chromatin utilizing many of the components of the switch mechanisms employed in single cell organisms (Klar, 2007). To test whether pre-meiotic cells were previewing future events at the transcriptional level, we compared our lists of germinal-specific and germinal-enriched transcripts with a list of 297 genes defined by maize meiotic-specific expression. Surprisingly, just-specified maize AR cells were specifically enriched for 34.3% (102/297) of these transcripts (Nan et al., 2011), along with 14 others that either have defined meiotic roles or are close homologs of meiotic proteins (Table 2 and Table S5). This category includes genes responsible for chromosomal pairing, synapsis, and recombination, including DYAD/SWI1 (Mercier et al., 2003; Boateng et al., 2008), AFD1 (Golubovskaya et al., 2006) PHS1 (Ronceret et al., 2009), DMC1 (Couteau et al., 1999), and homologs of RAD51 and ZYP1 (Mikhailova et al., 2006; Osman et al., 2006). These are genes defined by their functional specificity to meiosis. The AR set also includes AGO5a/b, which are OsMEL1 homologs regulating meiotic chromosome condensation (Nonomura, et al., 2007). Therefore, germinal cells transcribe many genes encoding meiotic factors about one week and 3 mitoses prior to canonical “pre-meiosis” (Tang et al., 2010). These results challenge current dogma that the meiotic decision point is during PMC maturation (Klar, et al. 2007; Nan, et al., 2011), pre-meiotic S, or interphase. It is possible that precocious expression of meiotic factors permits gradual dilution of mitotic chromatin components during the mitotic proliferation period, a hypothesis recently proffered for the mouse germ-line (Hackett et al., 2013). An alternative explanation is that AR cells are storing meiotic transcripts for translation at later developmental stages.

Table 2.

AR-enriched or -specific transcripts with defined roles in meiosis. This set also includes genes that have been characterized developmentally as critical to the maize mitotic / meiotic transition (Nan et al., 2011). See Table S5 for the complete list.

AR-enriched	Role	Probe ID	Protein ID	Log-fold change	AR avg intensity	Somatic avg intensity	Ratio
SWI1 / DYAD	recombination	28072	GRMZM2G300786	2.044	461.7	103	4.48
DMC1 (RecA-like)	recombination / repair	30014	TC313913	1.691	238.5	57.2	4.17
XPB1	DNA repair	38195	TC312637	1.276	1401.7	546.4	2.57
SKP1-like	recombination	42998	GRMZM2G032562	1.095	163.7	77.7	2.11
AFD1	synapsis	21701	GRMZM2G059037	1.045	160.5	80.1	2.00
aldose 1-epimerase	meiosis	575	GRMZM2G103287	1.022	3484.2	1634.4	2.13
RAD51/RHP55	DNA repair	40178	GRMZM2G058954	0.968	186.9	96.3	1.94
3'-5' exonuclease	DNA repair	14753	GRMZM2G111436	0.959	104.6	53.5	1.96
PHS1	Chromosome pairing	37135	GRMZM2G100103	0.928	126.7	67.5	1.88
RAD51/RHP55	DNA repair	5048	GRMZM2G058954	0.8	2155.7	1230.8	1.75
MSH2	mismatch repair	31864	GRMZM2G056075	0.78	2523.6	1413.3	1.79
RecA-like	recombination / repair	21435	GRMZM2G700757	0.752	1379.6	773.2	1.78
ZYP1	synapsis	17004	TC283445	0.751	521.7	301.9	1.73
Endonuclease III	base-excision repair	12222	GRMZM2G113228	0.751	376.4	224.2	1.68
aldose 1-epimerase	meiosis	14667	GRMZM2G103287	0.714	251.7	154.3	1.63
SNF2-containing	DNA repair	15367	TC309700	0.701	100.6	58.9	1.71
DNA helicase	NHEJ, DSB repair	22116	GRMZM2G137968	0.626	731.7	454.4	1.61
RecA-like	recombination / repair	39122	GRMZM2G700757	0.588	166.9	112.1	1.49
Cyclin	meiotic checkpoint	17125	TC296255	ON/OFF	149	0	N/A
EMENI	recombination	42970	GRMZM5G856297	ON/OFF	49.8	0	N/A

Supporting this latter hypothesis, we identified numerous germinal transcripts encoding RNP complex components (6 germinal-enriched versus 0 somatic-enriched), RNA helicases (14 versus 2), PUF/Pumilio translational repressors (5 versus 0), and a suite of ribosomal genes (97 versus 14) and translation initiation and elongation factors (18 versus 6). Collectively these transcripts encoding RNA binding or translational control proteins account for 16.3% of AR cell GO terms, compared to 3.2% for the somatic cells (Figure 4 and Table S6). What could be the functions of copious RNP factors in germinal cells? Maize meiocytes constitute just 1.5% of anther cells but contain 20% of anther RNA (Nan et al., 2011), much of which will contribute to haploid cell cytoplasm (either as stored RNA or translated protein) following meiosis. Additionally, AR cells have conspicuous nucleoli (Batygina, 2005), sites of ribonucleoprotein (RNP) complex biogenesis and function (Luehrsen et al., 1994). RNP-based mRNA protection and storage is a well-established aspect of animal germ-lines, particularly in eggs (Extavour and Akam, 2003). While RNPs have been described in plants (Luehrsen et al., 1994), reproductive roles remain undefined in both male and female plant germinal cells.

Figure 4.

Distribution of GO terms within AR and somatic sets – including 2529 and 4551 genes, respectively. The AR cells are enriched in the categories of RNA binding and RNP biogenesis (4.2% of all terms versus 1.8% for the somatic cells), and translation (including ribosomal proteins, accounting for 12.1% of all terms versus 1.4% for the somatic cells). Somatic tissues were enriched for catalytic activity (10.7% of all terms versus 4.1% for the AR cells), cell communication (4.0% of all terms versus 1.8% for the AR cells), and transcriptional regulation including DNA polymerase II subunits and DNA binding transcription factors (9.2% of all terms versus 4.0% for the AR cells).

In addition to the abundance of RNA binding proteins, anther germinal cells are enriched for transcripts encoding five of the 18 maize ARGONAUTES (Qian et al., 2011), including AGO4b / AGO105 (Figure 2c) and AGO121, both of which cluster with Arabidopsis AGO4/6/9 implicated in RNA-directed DNA methylation (RdDM) (Table 3) (Havecker et al., 2010) (Appendix S2). Together with germinal enrichment of IDN2 (Ausin et al., 2009) and DRD1 responsible for non-CG methylation (Chan et al., 2006), a cadre of genes responsible for RdDM are enriched in germinal cells, setting up a potential contrast with animals germ-lines that exhibit demethylation and imprint erasure during transit amplification (Hackett et al., 2013). Finally, anther somatic cells are enriched in transcripts encoding AGO1 and AGO10 family members (Table 3), which participate in post-transcriptional gene silencing (PTGS) (Ji et al., 2011), suggesting that this mechanism facilitates somatic but not germinal differentiation. We also found evidence of germinal enrichment of the recently described, grass-specific AGO18 family (Table 3) (Qian et al., 2011). To date, no functions have been ascribed to this class, but they are among the most highly enriched AGOs in maize AR cells suggesting that there could be additional types of sRNA metabolism yet to be discovered in pre-meiotic cells.

Table 3.

Array data showing germinal probes coding for proteins involved in sRNA pathways or functions. This includes those genes involved in RNA-directed DNA methylation (RdDM) and post-transcriptional gene silencing (PTGS). Upper panel, AR-enriched transcripts. Lower panel, somatic-enriched transcripts. AGO105 and AGO121 are ChromDB (http://www.chromdb.org/) identifiers; AGO105 is also known as AGO4b. AGO18a belongs to a monocot-specific clade (Qian et al., 2011). The AR cells also are enriched for AGO5a and AGO5b, homologs of rice MEL1 / AGO5 involved in chromosomal condensation during meiosis (Nonomura et al., 2007). The somatic tissues express transcripts for AGO1 and AGO10 family proteins involved in miRNA-based mRNA cleavage or translational repression (PTGS). We speculate that these may function in fine-tuning cell fate specification in the somatic layers by targeting specific transcripts for degradation, similar to their role in establishing leaf polarity in Arabidopsis (Timmermans, et al., 1998). In agreement with this hypothesis, the somatic set includes DCL4 involved in tasiRNA biogenesis (Xie et al., 2005).

AR-enriched, annotation	Role	Probe ID	Protein ID	Log-fold change	AR avg intensity	Somatic avg intensity	Ratio AR:S
LBL1-like	sRNA biosynthesis	6692	TC299943	3.09	1342	153.6	8.74
AGO18a	unknown	24515	GRMZM2G105250	3.013	1306.2	137.2	9.52
AGO18a	unknown	16609	GRMZM2G105250	1.749	180.4	54.2	3.33
AGO105	unknown	41571	GRMZM2G589579	1.289	714.8	307.9	2.32
IDN2	RdDM	36073	GRMZM2G096367	1.286	130.2	53	2.46
RDR1	sRNA biosynthesis	4610	GRMZM2G481730	0.945	376.4	170.7	2.21
AGO5a	Meiotic chromosome condensation	9040	GRMZM2G461936	0.923	4673.9	2446.2	1.91
AGO5b	Meiotic chromosome condensation	5117	GRMZM2G059033	0.792	182.3	95.1	1.92
AGO105	unknown	28189	GRMZM2G589579	0.772	132.8	74.4	1.78
AGO121	unknown	21908	GRMZM2G432075	0.745	449.4	240.9	1.87
DRD1	RdDM	41186	GRMZM2G393742	0.678	99.6	60.8	1.64

Somatic- enriched	Role	Probe ID	Protein ID	Log-fold change	AR avg intensity	Somatic avg intensity	Ratio S:AR
AGO1a	PTGS	2763	TC280574	1.474	1440.4	4798.7	3.33
AGO10a	PTGS	28292	TC301470	1.172	3017.5	6536	2.17
AGO1d	PTGS	13266	AW497968	1.015	5033.1	10809.7	2.15
AGO1b	PTGS	45085	CD989466	0.712	80.7	132.9	1.65
ROS1	Methylation / Imprinting	21640	CO466307	0.634	172.8	267.6	1.55
AGO1b	PTGS	43416	CO447420	0.616	95.1	146.3	1.54
AGO1b	PTGS	3023	TC298532	0.607	1036.7	1719.4	1.66
DCL4	RdDM	9560	CD438700	ON/OFF	0	55.7	N/A
DCL4	RdDM	36905	CD438700	ON/OFF	0	51.9	N/A

Acclimations to hypoxia in AR cells

Previously, we demonstrated experimentally that cellular redox status triggers germinal specification from pluripotent meristematic cell progenitors (Kelliher and Walbot, 2012) within maize anther lobes. We hypothesized that, in the context of the expanding lobes growing primarily by cell proliferation, the hypoxic physiology of germinal cells would curtail mitochondrial electron transport, thus minimizing generation of reactive oxygen species (ROS), a major cause of DNA damage. We now report the discovery that germinal cells express genes ensuring multiple routes for ATP and reducing power generation without respiration. First, germinal cells are significantly enriched in phosphoenolpyruvate (PEP) carboxylase kinase, which is regulated by hypoxia (Saze et al., 2001) and phosphorylates PEP carboxylase, also AR-enriched, to activate cytosolic ATP production (Nimmo, 2003) (Figure 5; see Table S7 for the complete list of alternative metabolic genes). The germinal set also includes pyruvate dehydrogenase kinase2 diverting pyruvate away from the citric acid cycle (TCA) (Randle, 1986) towards other AR-specific or -enriched enzymes that convert it to ethanol, such as pyruvate decarboxylase and alcohol dehydrogenase, or dehydrogenases utilizing lactate or malate to regenerate NAD⁺ (Figure 5a,b and Table S7). These pathways are also well-characterized in flooded maize roots where they contribute to survival during hypoxia (Bailey-Serres and Voesenek, 2008), and in the oversized meristems of mutants that exhibit excessive cellular proliferation (Helliwell et al., 2001). Additionally enriched in germinal cells were four of the six components of the glyoxylate shunt, a pathway used in germinating seeds converting lipids to acetyl CoA and subsequently to sugar (Figure 5c and Table S7).

Figure 5.

Alternative metabolic pathways adapted from Pathway Tools / maizecyc (http://pathway.gramene.org/gramene/maizecyc.shtml). This shows transcripts for enzymes that are AR-specific or -enriched highlighted in bronze. Both of these pathways divert carbon derived from glycolysis away from the citric acid cycle and use it to regenerate NAD+ for continued use in cytosolic glycolytic ATP production. (a) Production of ethanol and NAD⁺ from pyruvate by the sequential action of pyruvate decarboxylase and alcohol dehydrogenase. (b) Production of oxaloacetate and NAD⁺ by malate dehydrogenase. Transcripts for lactate dehydrogenase are also enriched in AR cells and the enzyme performs a similar role using lactate as substrate. (c) The cytosolic glyoxylate shunt pathway converting fatty acids to sugar. Probes corresponding to enzymes in italics were not on the array.

Preferential expression of superoxide dismutase and thioredoxins is indicative of a priority on ROS scavenging in the reproductive cohort (Table S7). The emphasis on non-mitochondrial ATP production and ROS clearance highlights the importance of genome integrity to reproductive success, and indicates that hypoxia is not only a mechanism of AR fate specification, but also a persistent physiological feature of early developing anther tissues. Non-respiratory energy metabolism requires much more carbon input than respiration. Classic physiological analysis established that at flowering, plants direct most photosynthate to developing floral buds (Rolland et al., 2006), and that within flower primordia, immature anthers receive the majority of transported sugar through the meiosis stage of ontogeny (Clement et al., 1996). This rich supply of carbohydrates should sustain non-respiratory energy metabolism in hypoxic AR cells, despite the lower recovery of ATP and reducing equivalents from alternative metabolic pathways compared to oxidation.

Distinctive suites of transcription factors expressed in AR and somatic lobe cells

In our AR to somatic comparison, 25 TFs are identified as germinal, including 9 in the ON/OFF category and 16 that are significantly enriched. The two most highly enriched genes are a MADS box TF and an AP2 type TF (Table S8). 126 somatic TFs were found, including 90 significantly enriched and 36 specific, of which 19 are first expressed after cell fate specification in the soma (Table S9). The high diversity likely reflects pooling of three cell types (epidermis, endothecium, and SPL). The most abundant TFs by expression intensity are bZIP (10 genes), MADS box (15), and homeodomain TFs (2), while Myb (16), WRKY (10), and bHLH (15) are well-represented families. No mutants of these TFs are known to impact anther growth or cause male sterility, suggesting these may have an earlier role in plant development or act redundantly in anthers. With this list of TFs, gene knockout strategies could be designed to examine TF roles in the context of anther development avoiding loss-of-function in other organs.

Discussion

A fundamental difference between the kingdoms is the existence of a germ-line in animals and its absence in plants and fungi. In most animal phyla, germ-line stem cells are sequestered during early embryogenesis and dedicated to continuous gamete production in adulthood (Leatherman et al., 2010). Animal male gonads resemble an assembly line, with diploid germ-line stem cells attached to a somatic hub at one end of the tissue initiating a continuum of developmental stages ending with gametes. As a result cells at a given stage of pre-meiotic development are never present in sufficient quantity to facilitate isolation for molecular analyses. The closest attempt to delineate different stages of animal pre-meiotic transcriptomes utilized fluorescent activated cell sorting (FACS) to isolate germinal and somatic cells from fetal mouse gonads before and after sex-determination (Jameson et al., 2012). This sophisticated study is a rich resource of information on pre-meiotic development, yet it is limited by the fact that the sorted germinal cells are already partially differentiated by lineage, low abundance transcripts are likely lost during sorting, and leaky transgene expression results in heterogeneity in the somatic populations. In contrast to animals, most plants and fungi produce cohorts of germinal cells late in life. Pluripotent anther stem cells within growing lobes rapidly acquire a germinal or somatic fate, exhausting the entire stem cell reservoir (Kelliher and Walbot, 2012) (Figure 1). We took advantage of these facts to isolate a virtually pure sample of germinal and a mixed population of endothecium, epidermal and SPL somatic cells within 24–36 hours of their differentiation from stem cells.

A major conclusion of our profiling is that germinal and somatic cell differentiation is rapid and involves de novo and differentially regulated expression of thousands of genes. Both AR and the surrounding somatic layers express suites of unique TFs that distinguish these cell populations from each other and from their pluripotent progenitors in anther primordia. The utility of LCM to enrich for maize anther cell types was amply demonstrated, and the transcriptome data served as the foundation for in situ hybridization assays that generated eight AR and eight somatic marker genes, including the first marker for the transient SPL cell type (beta amylase).

The transcriptome data indicated that in addition to hypoxia being a physiological trigger for differentiation it is a continuing condition of anther germinal cells. The AR cells express a suite of transcripts for cytoplasmic proteins that ensure ATP and NAD+ production in the absence of oxygen. Cytoplasmic ATP generation is less efficient than mitochondrial respiration but has the benefit of eliminating the major source of ROS as well as freeing up carbon for anabolism. As plants possess abundant sugar from photosynthesis and the anthers are the most well-fed organs in flowers, hypoxic metabolism suffices to sustain rapid cell division required to build a maize anther of ~50,000 cells from a primordium containing about one hundred precursor cells nine days earlier (Kelliher and Walbot, 2011; Kelliher and Walbot, 2012). It is striking that mammalian stem cells also exist in a hypoxic niche (Morrison et al., 2000; Gustafsson et al., 2005; Keith and Simon, 2007), despite the less robust tolerance of low oxygen compared to plant cells. Reducing ROS to protect genomic integrity may be of greater value than maintaining normal respiratory energy output. Stem cell maintenance in a hypoxic niche may also reflect the evolutionary history of multicellular organisms, which evolved when life was still restricted to aqueous environments before formation of an ozone shield sufficient to ameliorate ultraviolet radiation. That complex animals and plants evolved in water means that their developmental mechanisms were well suited for hypoxia, because the partial pressure of oxygen in liquid is about 1/10,000 that of air.

A second parallel between higher animals and flowering plants is the involvement of unusual RNA metabolism in their reproductive cell types. Although the mechanism(s) underlying germinal fate specification vary widely among different animal phyla and plants (Extavour and Akam, 2003; Ewen-Campen et al., 2009), copious RNA-binding proteins are a common feature of pre-meiotic cells (Schmidt et al., 2011). We propose there is evolutionary convergence on factors that sequester mRNAs for precisely timed translation at distinct pre-meiotic, meiotic, or later stages. These factors include RNP granules in animal germinal cells during germ-line specification. There is no information on whether RNPs exist in plant germinal cells, but we have provided evidence of precocious meiotic transcription as well as germinal enrichment of dozens of RNA binding genes, leading to the hypothesis that many of the meiotic transcripts are stored during AR proliferation until meiosis. To test this hypothesis it would be informative to track RNA and protein expression of these factors during the ontogeny of the germinal lineage in anthers. Similarly, the abundance of ribosomal components in AR cells suggests that germinal cells acquire the ability to boost translational capacity or build functionally distinct ribosomes -- two possibilities that merit future attention.

Grass anthers contain a unique class of non-coding sRNAs termed phased siRNAs (phasiRNAs) that are similar in size to trans-acting siRNAs (tasiRNAs). Although their role in reproduction remains undefined, they are functionally important because forward genetics identified a long non-coding (lncRNA), subsequently processed to 21-nt phasiRNAs, required for rice male fertility (Ding et al., 2012a; Ding et al., 2012b). tasiRNAs and phasiRNA loci are both transcribed by RNA polymerase II, however, tasiRNAs are complementary to and target mRNAs via PTGS, while phasiRNAs are not complementary to mRNAs or to repetitive elements in the genome. It has been established that precursor phasiRNA transcripts are converted to double stranded templates by RDR6 and then cleaved into phased 21- or 24- nucleotide siRNAs by a Dicer ribonuclease (Johnson et al., 2009; Song et al., 2012). We propose that early anther 21-nt phasiRNA production is somatic-specific, because RDR6 (Figure 2s) and Dicer-like4 (DCL4) (Table 3) responsible for synthesis and cleavage of the 21-nt size class (Xie et al., 2005) are both specific to the somatic tissues. The finding that germinal cells are enriched in several distinct Argonaute RNA species, particularly of the grass-specific AGO18 family, is suggestive of yet to be defined sRNA metabolism unique to AR cells. Because RDR6 is somatic-specific we speculate that somatic to germinal transmission of sRNAs such as phasiRNAs may occur. Now that we have pinpointed the timing and distribution of transcripts encoding these factors in sRNA metabolism in somatic and germinal cells, future research can focus on the function of sRNAs in plant pre-meiotic development and meiotic preparation in the surrounding somatic cells.

Anther germinal cells are specified de novo from somatic stem cell progenitors within each flower, and they mature as a cohort. We previously reported on the developmental progression of these archesporial cells over ~7 days: after specification they grow in diameter from 15 to 25 microns, proceed through 3–4 mitoses, and then enlarge to a 45-micron diameter during canonical pre-meiosis (Kelliher and Walbot, 2011). During the maturation of AR cells into PMC, the cell cycles are aligned permitting a synchronous start to meiosis. Here we have identified numerous transcript types that characterize the first 36 hours of germinal development post-specification. RNA management is a critical feature, in agreement with other plant and animal germinal cell types; however, RNA binding proteins do not appear to play a role in germinal specification. We previously showed that hypoxia provides a positional cue, triggering germinal fate in the center of the stem cell pool during rapid early anther growth, and we now know that germinal cells accommodate this physiological condition by supporting alternative metabolism, while they protect their genome with efficient ROS scavenging. Another fundamental discovery was that meiotic fate acquisition involves gene expression priming, as AR cells precociously express recombinases and other meiotic genes nearly a week before meiosis. This suggests that meiotic preparation requires more than a just-in-time cell cycle switch during PMC maturation. We hypothesize that sRNA mediates important epigenetic changes during the mitotic proliferation period – making the enrichment of Argonautes, RdDM complex components and other RNA binding proteins in AR initials more intriguing. With these ideas in mind, future investigations can address these questions: Are meiosis-associated transcripts translated during AR mitoses – and if so, are the proteins loaded onto chromatin? If not, are the mRNAs degraded or stored for later translation? To what extent does precocious expression of meiosis genes occur in egg cells -- is this phenomenon a particular characteristic of male reproductive development? Intriguingly, grass anthers are highly enriched (and perhaps exclusively express) the 24-nt phasiRNAs. Do sRNAs guide the early stages of germinal development? What role(s) do they play in meiotic preparation, recombination, and / or resetting the epigenetic landscape for gametogenesis?

Experimental Procedures

W23 bz2 (deficient in vacuolar anthocyanin accumulation) inbred line plants and the msca1 mutant introgressed into W23 were greenhouse-grown in Stanford, CA as described previously (Kelliher and Walbot, 2011; Kelliher and Walbot, 2012). Anther length is a reliable indicator of developmental stage. From fertile plants, AR and somatic cells were isolated by LCM and total RNA was extracted from two biological replicates as described previously (Kelliher and Walbot, 2011) and tested for RNA quality on the 2010 Agilent Bioanalyzer (Santa Clara, CA). We used 0.30 – 0.35 mm anthers for the LCM collections measured at dissection from the central spike of 2.5 cm tassels. The 0.15 WT collection pooled anthers from 0.14 to 0.16 mm in length, while the 0.2 mm WT and msca1 anther collections were pooled from anthers 0.18 to 0.22 mm long. Total RNA was extracted, DNase-treated and tested on the Bioanalyzer before being amplified and hybridized in a 4x44K format two color microarray (Agilent Part number: G2519F; Design ID: 016047) in a balanced dye swap design as described previously (Kelliher and Walbot, 2012).

Slides were scanned and data were processed as described previously (Nan et al, 2011). The resulting median foreground values for the red and green channels were normalized in two steps using the limma package in R: “within arrays” using the lowess method and “between arrays” using the quartile method. Probes with expression values greater than 3.0 standard deviations above the average foreground of the array’s negative controls were considered “ON”, resulting in an estimated false discovery rate of 0.13%. A few probes with < 75% of the replicate measurements scored as “ON” were excluded from further analysis. Significance for differential expression was set at ~1.5-fold (log2 ~ 0.58) with a p-value ≤ 0.05. As confirmation, normalized intensities averaged across replicates were compared.

qRT-PCR was performed as described previously (Kelliher and Walbot, 2012). In situ hybridizations were performed with probe transcribed using the DIG RNA Labeling Kit (T7/SP6) (Roche, http://www.roche-applied-science.com). Sense and antisense probes were synthesized from PCR fragments amplified from cDNA clones obtained from the University of Arizona maize cDNA collection (http://maizecdna.org/). RNA in situ hybridizations were performed on 0.30 – 0.35 mm anthers taken from spikelets residing on the central spike of ~2.5 cm tassels, as described previously (Schiefthaler et al, 1999).

Supplementary Material

Supp TableS1-S9-AppendixS1-S2

Table S1. Highly abundant (expression >median), enriched (log fold change >2) AR-specific markers from the LCM tissue comparison.

Table S2. Highly abundant (expression >median), enriched (log fold change >2) somatic-specific markers from the LCM tissue comparison.

Table S3. Confirmation of cell-type specificity of AR markers by qRT-PCR.

Table S4. Confirmation of cell-type specificity of somatic markers by qRT-PCR.

Table S5. Germinal transcripts involved in meiosis.

Table S6. Germinal transcripts for genes involved in RNA-binding, ribosome biogenesis, or translation.

Table S7. Germinal transcripts for genes involved in alternative metabolism and ROS scavenging.

Table S8. Germinal transcripts for genes involved in transcriptional regulation.

Table S9. Somatic transcripts for genes involved in transcriptional regulation.

Appendix S1. Supplementary discussion of RNA-directed DNA methylation.

Appendix S2. Supplementary discussion of an AR / SPL marker.