The SUMO ligase MMS21 profoundly influences maize development through its impact on genome activity and stability

The post-translational addition of SUMO plays essential roles in numerous eukaryotic processes including cell division, transcription, chromatin organization, DNA repair, and stress defense through its selective conjugation to numerous targets. One prominent plant SUMO ligase is METHYL METHANESULFONATE-SENSITIVE (MMS)-21/HIGH-PLOIDY (HPY)-2/NON-SMC-ELEMENT (NSE)-2, which has been connected genetically to development and endoreduplication. Here, we describe the potential functions of MMS21 through a collection of UniformMu and CRISPR/Cas9 mutants in maize (Zea mays) that display either seed lethality or substantially compromised pollen germination and seed/vegetative development. RNA-seq analyses of leaves, embryos, and endosperm from mms21 plants revealed a substantial dysregulation of the maize transcriptome, including the ectopic expression of seed storage protein mRNAs in leaves and altered accumulation of mRNAs associated with DNA repair and chromatin dynamics. Interaction studies demonstrated that MMS21 associates in the nucleus with the NSE4 and STRUCTURAL MAINTENANCE OF CHROMOSOMES (SMC)-5 components of the chromatin organizer SMC5/6 complex, with in vitro assays confirming that MMS21 will SUMOylate SMC5. Comet assays measuring genome integrity, sensitivity to DNA-damaging agents, and protein versus mRNA abundance comparisons implicated MMS21 in chromatin stability and transcriptional controls on proteome balance. Taken together, we propose that MMS21-directed SUMOylation of the SMC5/6 complex and other targets enables proper gene expression by influencing chromatin structure.


Introduction
Plants like other cellular organisms exploit a plethora of post-translational modifications to expand the functionality of their proteomes, including controls on enzymatic activity, subcellular location, interaction with other effectors, and ultimately on the turnover rates of the affected proteins. One reversible modification that is emerging as a key regulator involves attachment of the~100-amino-acid protein SMALL UBIQUITIN-LIKE MODIFIER (SUMO), which is structurally related to ubiquitin and likewise becomes covalently linked via an isopeptide bond to accessible lysines within its targets [1,2].
SUMOylation is driven by an ATP-dependent enzymatic cascade involving the sequential action of a SUMO-activating enzyme (or E1) and a SUMO-conjugating enzyme (or E2) that prepares SUMO for addition, and in most cases, a SUMO-protein ligase (E3) that identifies appropriate substrates and encourages transfer from a thioester-linked E2-SUMO donor [1,2]. While some substrates become modified with a single SUMO, others become iteratively modified with multiple SUMOs attached either at multiple lysines within the substrate or to previously bound SUMOs connected internally through SUMO-SUMO isopeptide linkages. The conjugated SUMOs can also be ubiquitylated by a family of SUMO-targeted ubiquitin ligases, thus merging the influence of these two ligation systems. SUMO attachment is often reversible through a collection of deSUMOylating enzymes that specifically cleave the isopeptide bond between the SUMO moiety and target lysines [25].
To date, three classes of SUMO E3s have been characterized in plants: SAP AND MIZ1 DOMAIN-CONTAINING LIGASE (SIZ)-1, PROTEIN INHIBITOR OF ACTIVATED STAT-LIKE (PIAL)-1/2, and METHYL METHANESULFONATE (MMS)-21/HIGH-PLOIDY (HPY)-2/NON-SMC-ELEMENT (NSE)-2, referred to here as MMS21 [1,2]. They share a SP-RING domain that binds the E2-SUMO intermediate [26], along with a variety of other motifs that presumably identify specific substrates and/or help anchor the E3 to appropriate surfaces/complexes, including DNA and methylated histones. While SUMOylation by the SIZ1 E3 has been connected to a wide range of cellular events and substrates in plants, especially those related to stress defense [5,14,21,23,27], the function(s) of the other E3s are currently unclear.
MMS21, in particular, has garnered interest given its high conservation among eukaryotes and its potential action in a variety of nuclear processes. Arabidopsis mms21 null mutants are viable but develop stunted roots, altered apical meristems, dwarfed rosettes, and higher chromosome ploidy numbers in both somatic tissue and male gametes, suggestive of roles in the cell cycle and endoreduplication [12,28,29], while the null mms21 mutants in rice also have stunted vegetative growth [19]. More recently, MMS21 was linked to the DNA damage response and chromatin structure potentially through the respective modification of BRAHMA, a conserved ATPase component within the SWI/SNF chromatin-remodeling complex, and the cell-cycle check point protein DPa, [18,30,31]. In other organisms, MMS21 has been shown to be a crucial component of the nuclear STRUCTURAL MAINTENANCE OF CHROMOSOMES-5/6 (SMC5/6) complex, which is an evolutionarily-conserved ATPase that influences chromatin compaction and is required for recombinational DNA repair, replication fork restart, ribosomal DNA and telomere maintenance, and genome stability [32]. In accord, MMS21 has been found associated with Arabidopsis SMC5/6 complex [13], with one possible target being the integral NSE4 subunit [5].
Here, we further investigated the functions of MMS21 in maize (Zea mays) using a library of UniformMu transposon-insertion and CLUSTERED REGULARLY INTERSPACED SHORT PALINDROMIC REPEATS (CRISPR)/Cas9-induced mutations. These analyses revealed that MMS21 is essential in maize with critical roles in root, shoot, pollen, and seed development. While the exact mechanism(s) are not yet clear, defects in DNA repair, and misregulated proteome/transcriptome balance were evident for the mms21 germplasm. Interaction studies, RNA-seq analyses, and in vitro SUMOylation assays collectively linked MMS21 to the NSE4 and SMC5 subunits of the SMC5/6 complex, suggesting that MMS21-directed SUMOylation of the SMC5/6 complex and possibly other targets are essential for proper chromatin function and subsequent maize development.

Results
The Mms21 locus is essential for normal maize development From tBLAST scans of the maize genome using Arabidopsis MMS21 (HPY2) as the query, we identified a single maize Mms21 locus within the B73 background (GRMZM2G022065), which is located on the long arm of chromosome 6. It spans 4.3 kbp over 7 exons (Fig 1A), and encodes a 245-amino-acid protein with 49% sequence identity to that from Arabidopsis, and 66% and 90% identity to those from the more closely related species, rice and sorghum, respectively. Sequence alignment of maize MMS21 with its plant, yeast, and human orthologs revealed substantial homology throughout the protein, especially within the SP-RING domain that binds the E2-SUMO donor [26]. Quantitative reverse-transcribed (qRT)-PCR showed that the Mms21 gene is widely expressed throughout maize, including silk, ovaries, and developing embryos and endosperm, with its highest mRNA levels found in pollen ( Fig 1B). Total RNA was isolated from shoots collected 10-DAS. The expression level of Mms21 in W22 was arbitrarily set as 1 using expression of Act1 as an internal standard (n = 4 technical replicates, ±SD). (F) Amino acid sequence alignment of MMS21 proteins from plants, yeast and humans, along with the position of the mutations found in the maize allele collection. Black and grey shading indicate identical and similar amino acids, respectively. The amino acid length of the wild-type proteins is shown at the end of the sequence. The region encompassing the SP-RING domain is located by the blue line. The coding sequences expected to be absent in the mms21-1 and mms21-2 mutants are located by the dashed red lines. Our efforts to characterize Mms21 genetically began with a search for compromising Mutator (Mu) insertion alleles within the UniformMu population generated with the W22 inbred [33], which led to the identification of the mms21-1 allele (mu1068022). Based on genomic PCR and DNA sequence analyses of RT-PCR products, the Mu element inserted at the front of intron 4, which induced mis-splicing of the Mms21 transcript and ultimately fusion of the 3 rd and 5 th exons (Fig 1C, 1D and 1F). If translated, this mutation should eliminate 37 amino acids upstream of the SP-RING region. Searches of the UniformMu population for seed defects (see below), based on bulk segregation analysis by the Mu-seq method [34], identified a second mms21 allele-mms21-2. Its Mu insertion site was confirmed by sequencing PCR products generated with TIR6 and Mms21 gene-specific primer pairs. Here, a 17-bp insertion was found within exon 6 after the Pro180 codon that introduced six additional amino acid codons followed by a stop codon; if translated, the mms21-2 protein would be missing most of the SP-RING domain and the C-terminal end (Fig 1C, 1D and 1F). While it remained possible that the 208-residue mms21-1 protein retained some of its activity, the 183-residue mms21-2 protein should be poorly functional given its compromised SP-RING domain. As judged by qRT-PCR analysis of homozygous plants, the mms21-1 and mms21-2 lines accumulated only 35% and 2%, respectively, of the Mms21 mRNA level found in W22 ( Fig 1E).
After backcrossing the mms21-1 and mms21-2 lines five times into the W22 inbred followed by self-pollination, homozygous plants were identified that displayed numerous growth defects as compared to their normal siblings. These defects segregated as recessive traits, with the phenotypic severity stronger for the mms21-2 allele in agreement with its more compromised gene architecture and expression. While viable, homozygous mms21-1 and mms21-2 seeds germinated poorly and the seedlings grew more slowly, ultimately developing into severely stunted plants (as judged by fresh and dry weights; S1A Fig) with shorter roots and fewer leaves at maturity (Fig 2A-2D). mms21-1/2 leaves were also significantly shorter and narrower but developed similarly-sized epidermal cells as compared to W22, suggesting that the growth defects arose from fewer cell numbers and not reduced cell expansion (S1B, S1C, and S1D Fig). To confirm that the phenotypes associated with the two mutant alleles are caused by the same gene, we performed genetic complementation tests using reciprocal crosses between the two mms21 alleles. As shown in Fig 2E, F1 progeny harboring both mutant alleles, as determined by genomic PCR, displayed the identical stunted shoot phenotype as single homozygous mms21-1 or mms21-2 plants.
Further analysis of floral organs revealed that the mms21 mutations also compromise reproduction. The mutant cobs grew poorly with fewer ovules. While tassels did appear, they were underdeveloped and the emerging anthers often failed to open, and even if opened, they shed much less pollen (Figs 2F, 2G and S2B). Silk emergence was also substantially delayed (S2A Fig), consistent with a delay in overall development. Germination assays on the small amount of mms21-1/2 pollen collected revealed a substantial block in pollen tube emergence (Fig 2H  and 2I), which when combined with other defects in cob and tassel development, likely underpinned the low fecundity seen for mms21-1/2 plants. Finally, the resulting mms21 seeds were slightly smaller in size, and often showed a pitted surface as compared to their normal siblings (Fig 2J and 2K). Approximately 25% of the seeds from self-pollinated mms21-1 and mms21-2 cobs acquired this appearance consistent with a recessive trait. Dissection of the mature seeds from self-crossed heterozygous mms21-1/2 cobs revealed reductions in embryo size and often an underfilled endosperm, which were stronger for the mms21-2 allele (Fig 2L-2N). This seed phenotype could be seen as early as 12-days-after pollination (DAP) for mms21-2, suggesting that MMS21 is important early in maize seed development (S3 Fig).
To further link the mutant phenotypes with the Mms21 locus and potentially isolate stronger alleles, we generated additional germplasm by CRISPR/Cas9-mediated mutagenesis of the Hi-II background. The resulting seven mms21-CR mutations harbored an array of defects around the two Protospacer Adjacent Motif (PAM) sequences designed within the CRISPR target sites just upstream of introns 1 and 2 for Cas9 cleavage (Figs 1A,1C and 3A). Included were a 1-bp deletion before the 2 nd PAM for mms21-CR1, a 14-bp deletion around the 2 nd PAM for mms21-CR2, an additional A nucleotide before the 2 nd PAM for mms21-CR3 which likely shifted the reading frame to begin at an ATG codon located at the front of exon 3, an additional A nucleotide before the 1 st PAM combined with 2-bp ΔCG deletion before the 2 nd PAM for mms21-CR4, 6-and 8-bp deletions before the 1 st and 2 nd PAMs, respectively, for mms21-CR5, a 3-bp ΔAGT deletion before the 1 st PAM for mms21-CR6, and a 33-bp deletion at the 1 st PAM for mms21-CR7 that removed 11 codons within exon 1 (Fig 1C and 1F). All these mutations altered the Mms21 coding region by deleting one to as many as 200 residues if translated ( Fig 1C).
Overall, our phenotypic analysis of the CRISPR/Cas9-derived collection after two backcrosses to the B73 inbred showed similar developmental consequences as seen for the Uni-formMu alleles, which included abnormal root, shoot and seed development, and pitted seeds with smaller embryos and dampened germination (Figs 3B and S4). Surprisingly, only the mms21-CR7 allele was fertile when homozygous (S4C and S4D Fig) and resembled the Mutator-derived lines, while most of the remaining mutants failed to produce viable offspring, thus requiring maintenance as heterozygotes. The strongest phenotypic mutants were the mms21-CR2 and mms21-CR5 alleles; they produced homozygous seeds, but these seeds failed to germinate (S4A and S4B Fig). The next strongest impact was seen for the mms21-CR1, mms21-CR3 and mms21-CR6 alleles that produced germinable seeds but the homozygous seedlings invariably died within approximately two weeks of growth under normal greenhouse conditions (Figs 3B and S4A-S4C). The mms21-CR3 mutant was the strongest allele among the three and stalled growth soon after radical emergence (Fig 3B). qRT-PCR analysis of the six mutants that germinated (i.e., all but mms21-CR2 and mms21-CR5), using primers that spanned the coding region for part of the SP-RING domain (see S1 Table), showed that, with the exception of mms21-CR7, all strongly dampened accumulation of the Mms21 mRNA as compared to their normal siblings (Fig 1E). The most intriguing homozygous-lethal allele was mms21-CR6 as its 3-bp deletion effectively suppressed accumulation of the Mms21 transcript even though its mRNA was in frame and missing only a single Val23 codon, suggesting that this region influences mRNA stability ( Fig 1F). This severity for mms21-CR6 was in contrast to the mms21-CR7 allele, which was missing a larger portion of Mms21 transcript but was still expressed at reasonable levels.

mms21 mutants have relatively normal SUMOylome profiles
As one strategy toward understanding how MMS21 affects maize, we assessed its overall impact on SUMOylation by subjecting total tissue extracts to immunoblot analysis with anti-SUMO antibodies [24]. As shown in Fig 4A, SUMO-conjugate profiles in mms21-1 and mms21-2 leaves were mostly indistinguishable from those seen in W22 both before and after a 30-min heat stress at 42˚C, which dramatically increases the pool of SUMO conjugates [24]. No species were absent in mms21-1/2 leaves with or without the heat stress, and at most, only a few new species at~60 and 37 kDa appeared. Similarly, we tested embryo and endosperm tissue harvested from seeds at 16 DAP ( Fig 4B). Again, little differences in the profiles and levels of SUMO conjugates and free SUMO were detected in the mutants, strongly suggesting that MMS21 modifies only a small subset of SUMO substrates in maize, consistent with similar studies with Arabidopsis [5].

mms21 mutants strongly alter the maize transcriptome
Alternatively, we analyzed the maize transcriptome anticipating that the phenotypes of the mms21 mutant collection were underpinned by robust, informative changes in gene expression. Here, we subjected total RNA isolated from mms21-1 and mms21-2 shoots harvested 10 days-after-sowing (DAS), and embryos and endosperm at 16 DAP to in-depth profiling by RNA-seq analysis, using equivalent samples from wild-type W22 for the comparisons. Each tissue was analyzed by three biological replicates (27 total samples) using paired-end reads of at least 150 bp, which ultimately resulted in~47 M reads on average per sample. Approximately 80% of the high-quality reads could be uniquely mapped onto the maize B73 reference genome (RefGen_V4.48), which resulted in 28,657, 27,062 and 24,977 unique transcripts in the shoot, embryo and endosperm samples, respectively. For every genotype/tissue group, the replicates showed strong correlations in mRNA abundances (0.98 to 0.99) based on log 2 -transformed transcript read counts, thus confirming the reliability of the data (S5 Fig), which was also supported by principal component analyses (PCA) of the transcriptomes in which the mutant samples invariably clustered together and away from those of W22 (Fig 5A).
To further confirm the reliability of the datasets, we compared the expression profiles of significantly impacted genes (adjusted p-value <0.05) by fold change (FC) in abundances for each transcript. As shown in Fig 5B, strong correlations were seen between the mms21-1 and mms21-2 datasets versus W22 with correlation values for the log 2 -transformed data of 0.94, 0.83 and 0.87 for the shoot, embryo and endosperm samples, respectively, implying that the mutant transcriptomes responded similarly. Between W22 and the mms21-1/2 mutants, 7,214 significant differentially expressed genes (DEGs; adjusted p-value < 0.05 based on multiple comparisons) were identified in shoots (3353 up-regulated transcripts and 3861 down-regulated), 5,864 in embryos (2,943 up-regulated transcripts and 2,921 down-regulated), and 1,320 in the endosperm (401 up-regulated transcripts and 919 down-regulated), which collectively indicated strong and pervasive changes in mRNA profiles for the mutants (Fig 5C). DEG comparisons among the biological replicates also revealed that the transcriptional responses of the mms21-1 and mms21-2 mutants were markedly similar. The impacted genes in common between the mms21-1 and mms21-2 samples represented a major fraction of the total DEGs in shoots but more minor fractions of the DEGs in embryos and endosperm ( Fig 5D).
As one approach to identify the functional classes of genes whose expression were altered by the mms21 mutations, we subjected the DEGs to Gene Ontology (GO) analysis using the Annotation Hub database [35]. Enrichment analysis detected numerous GO categories that spanned many aspects of plant growth and development, including 'sexual reproduction', 'cellular catabolic', 'proteasomes', 'nucleosomes', 'chromatin', 'vacuoles', 'ER', and 'peptidases' for the upregulated genes, and numerous photosynthesis-related categories for the downregulated genes (S6A- S6C Fig). This conclusion was corroborated by Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses [36], where 'sugar metabolism', 'protein processing in ER', 'photosynthesis', and 'photosynthesis-antenna proteins' emerged as enriched categories for the mms21 mutants (S6D Fig). Overall, our RNA-seq data implied that the MMS21 globally influences much of the maize transcriptome.
To hone in on the transcriptional response of specific genes, we analyzed the RNA-seq datasets by volcano plots that compared the FC and p-value of significance for the DEGs between W22 and the mms21-1/2 mutants (Figs 6A and S7). The exaggerated splay of the plots revealed that the abundances of many transcripts were significantly altered (24.9%, 22.4% and 5.4% of total transcripts in shoots, embryos and endosperm, respectively), some by as much as a thousand FC (log 2 FC >10). When we focused on the SUMOylation pathway, expression of almost all identified components were unaffected by the mutations with the exception of the Mms21 mRNA itself whose abundance was strongly reduced (Figs 6A and S7). These observations not only confirmed that the mutant alleles dampened Mms21 expression, but also implied that other aspects of SUMOylation were not upregulated as compensation. Given the likely connections between MMS21 and DNA repair, we also interrogated the volcano plots for genes associated with this process (Fig 6A). Notably, most DNA repair-associated mRNAs were modestly impacted (either up or down) using a FC cutoff �2 and an adjusted p-value <0.05. Intriguingly, the exceptions were mRNAs for Nse4a and Nse1 (annotated as embryo-defective 1379 in reference genome RefGen_V4.48) that encode components of the SMC5/6 complex involved in DNA repair and chromatin stability (Figs 6A and S7). While their mRNA abundance was robustly elevated by the mms21 mutations, those encoding the central SMC5 and SMC6 subunits and four other maize Nse4 paralogs were not altered, suggesting the upregulation of Nse1 and Nse4a is unique among Smc5/6 complex genes.
Additional upregulated loci in the mms21 backgrounds connected to genome integrity were Mre11b, which encodes a core component of the Mre11/Rad50/Nbs1 complex that functions in DNA repair and is critical for maintaining genome stability in Arabidopsis [37], and MICRORCHIDIA (Morc)-6, which encodes a relative of the Arabidopsis MORC transcriptional repressor family that associates with nuclear bodies and globally mediates transcriptional silencing [38,39] (Figs 6A and S7). mRNAs encoding the small subunits of ribonucleoside diphosphate reductase (RNR), RNRS-2 and RNRS-C, that participate in DNA repair via the synthesis of deoxyribonucleotides [40] were also significantly enriched in the mms21 datasets. In addition, genes encoding several protein chaperones (HEAT SHOCK PROTEIN (HSP)-70, HSP-23.5kDa, and the class-II small HSP17.5-kDa) were upregulated in some of the mms21-1/2 samples that could reflect enhanced proteotoxic stress.
The most interesting DEGs encoded zein storage proteins, which displayed a pronounced expression in mms21-1/2 shoots, a tissue that normally does not express these proteins. While this effect was not evident in embryos and endosperm, the latter of which naturally accumulates high levels of zeins during development [41], the response was substantial in leaves and, in fact, represented the most significantly impacted collection of mRNAs in the mms21-1/2 datasets (Figs 6A and S7). Of the 54 transcripts showing a �32 FC in abundance for mms21-2 versus W22 shoots, 37 encoded zeins. Interestingly, when we assayed two general classes of zein proteins (α-and γ-zein) by immunoblot analysis with class-specific antibodies [41], we failed to detect such proteins in leaves despite their ectopic mRNA accumulation (S8 Fig).
To confirm these changes in mRNA levels for Nse4a, Mre11b, and two representative zein genes (ZeinL2a and Zein3), we quantified their transcript abundances by qRT-PCR using our complete collection of mms21 mutants generated by UniformMu and CRISPR/Cas9-mediated mutagenesis. As shown in Fig 6B, the majority of these genes were uniformly upregulated in the mms21 mutants but often not to the same robust levels as those seen by RNA-seq. The only outlier was Nse4a in some of the CRISPR/Cas9 mutant alleles; the abundance of this mRNA varied widely as compared to the values obtained for its siblings, which might reflect variations in allele severity and the hybrid state of the CRISPR/Cas9 backgrounds.

mms21 mutations alter the transcriptome/proteome balance
To further assay the impact of MMS21 on protein accumulation, we compared the proteomes of mms21 mutant and normal siblings by shot-gun mass spectrometry (MS) of young seedlings [42]. Here, trypsinized total protein extracts from 10-DAS seedlings were subjected to reversed-phase separation followed by tandem MS, which allowed relative quantification for approximately 4,000 proteins from the MS1 scans based on the mean of three biological replicates each with two technical replicates. The proteome data were then normalized among samples using a list of 150 proteins relatively unaffected by the mutations [42], which was then validated by assaying the levels of abundantly detected histones which should be consistent across genotypes (Figs 7A, S9A and S9B).
When the normalized proteome data were displayed by volcano plots that assessed both log 2 FC in abundance versus p-values of significance, numerous proteins were significantly more or less abundant in the mms21 seedlings compared to their normal siblings based on a FC = 2 threshold and adjusted p-values <0.05. For the mms21-2 background, the values were 476 up-and 569 downregulated (26% mis-accumulated), while for the mms21-1 background, the values were 496 up and 640 downregulated (27% mis-accumulated) as compared to 9% of the proteins having different values (>2 FC, p-value <0.05) when comparing normal siblings of mms21-2 and mms21-1 to each other (Figs 7A, S9A and S9B). (We presume that most of the proteins assigned as differentially accumulating between the normal siblings represent noise inherent to MS data collection and analysis.) As with the immunoblot assays, we failed to detect any zein storage proteins by MS in the mms21 seedlings despite having elevated mRNAs. When the differentially accumulating list was subjected to GO analysis, a broad spectrum of protein functionalities was impacted in the mms21 backgrounds, such as 'cellular process', 'metabolic process' and 'catalytic activity', with little selective impact seen on specific subcategories, thus likely reflecting a general alteration in protein composition ( Fig 7B).
We then compared the differences in protein abundance for the mms21-2 seedlings versus normal siblings with those described above for the corresponding transcripts for 3,756 proteins with data available for both. Surprisingly, a modest but significant correlation was seen (R 2 = 0.42), with the proteins less abundant in the mutant also having less mRNA, while those proteins more abundant in the mutant also having more mRNA (Fig 7D). This correlation was even more robust (0.69) when we analyzed only those proteins (431 total) that were The volcano plot depicts protein abundance changes for 4,084 proteins detected from mms21-2 leaves as compared to those of its normal sibling. Each dot represents one protein that had detectable expression in both samples and was plotted based on its log 2 FC in abundance (mutant/normal siblings) and its -log 10 p-value of significance based on the three biological replicates, each with two technical replicates. The horizontal and vertical dashed lines mark a FC = 2 in protein abundance and a p-value = 0.05, respectively. Histone proteins used to confirm data normalization are shown as green. SUMO pathway components and DNA repair-associated proteins are highlight in blue and red respectively. (B) GO analysis of significantly regulated proteins in the mms21 mutant versus W22 seedlings. The vertical coordinates indicate the enriched GO terms, and the horizontal coordinates show the number of genes for each GO term when comparing differentially expressed genes common between mms21-1 and mms21-2 seedlings. Negative values indicate downregulation, positive values indicate upregulation. GO enrichment was performed using all the three sub-ontologies: 'biological process', 'molecular function', and 'cellular component'. (C) SMC5 protein abundance is reduced in strong mms21 mutant backgrounds. Relative protein abundances of SMC5 from the mms21-1, mms21-2 and mms21-CR1 mutants and their normal siblings. Each bar represents the mean of three biological replicates(±SD). (D) Positive correlation between transcriptome and proteome data in mms21 mutants. Scatter plot showing the relationship between changes in protein and mRNA abundances for the mms21-2 mutant versus normal sibling as determined by plotting the log 2 FC in mRNA abundance versus the log 2 FC in protein abundance. Red dots (154) and blue dots (277) highlight proteins that were more or less abundant (FC>2), respectively, in both mms21-1 and mms21-2 leaves as compared to their normal siblings. The black and dashed blue lines show the correlations for mms21-2 and for the combined data of mms21-1 and mms21-2 (red and blue dots). up or down-regulated in both the mms21-1 and mms21-2 backgrounds (Fig 7D). Collectively, the protein/mRNA correlations implied that the lack of MMS21 globally influenced the proteome balance primarily by altering the transcriptome balance.

Role of MMS21 in DNA repair
To more specifically address a possible connection between MMS21 and DNA repair [13,43], we compared mRNA abundances for a number of likely contributors as identified by sequence homology to known Arabidopsis factors. From the analysis of this collection by Z-scores, it became apparent that the mms21-1/2 lines had globally altered expression of DNA repair-associated genes, suggesting a dysregulation of the process. For example, of the 66 mRNA analyzed from embryos, many were upregulated in the mutant backgrounds, including those encoding a number of well-described DNA-damage repair factors such as PCNA, NSE1, NSE4a, MRE11b, BRCA1, and multiple subunits of the REPLICATION PROTEIN A (RPA) and RNR complexes, while others were downregulated, including those encoding the DNA mismatch repair protein MSH4 and DMC1 involved in meiotic recombination (Fig 8A and S2 Table). Similar responses were also seen for the few corresponding DNA repair proteins that we could detect by MS, but their measured changes were more muted with most failing to rise above/ below a FC = 2 cutoff (p-value<0.05). Only a few showed increases or decreases >2 fold in mms21-2 seedlings, including the DNA mismatch repair protein MLH1, the UVRB/UVC homolog, DRT family members, a Tudor/PWWP ortholog, Whirly1, and SMC5 (Fig 7A).
Observing a potential connection of MMS21 to DNA repair, we next checked the sensitivity of the mms21 mutants to DNA-damaging agents as assayed by the growth of emerging roots. Both mms21-1 and mms21-2 roots were strongly hypersensitive to MMS and mitomycin C, relatively unaffected by hydroxyurea, and slightly affected by bleomycin and zebularine but only for the stronger mms21-2 allele (Fig 8B). The hypersensitivity of mms21 to MMS in particular was consistent with the first discovery of this locus via a yeast MMS-sensitivity screen [44]. And finally, we measured the frequency of DNA strand breaks by comet assays [45]. Here, W22 and mms21 nuclei were isolated from seedling roots, embedded in agarose, and then their DNA was electrophoresed under alkaline conditions; increased DNA breaks were then observed by a greater comet tail length as the DNA migrated toward the anode. As can be seen in Fig 8C and 8D, DNA from the mms21 mutants had more breaks as compared to W22. Taken together with the RNA-seq data, we found that loss of MMS21 impacted DNA integrity and repair, along with substantially altered gene expression that included the ectopic expression of zein-encoding loci.
Prior studies with Arabidopsis mms21/hpy2/nse2 mutants implicated MMS21 in endoreduplication, with the mutant shoot nuclei often harboring excess whole genome duplications [12] and pollen having a high frequency of diploid male gametes [29]. When similarly assessed for DNA content by flow cytometry of nuclei, we detected a comparable distribution of 2N and 4N nuclei for 10-DAS mms21-1 leaves as compared to those from wild-type W22, suggesting little impact of MMS21 on cell division in maize somatic tissues (S10A and S10B Fig). Given that endoreduplication is a common feature of maize seeds, especially in the endosperm that undergoes multiple rounds of DNA replication before maturation [46], we then examined the ploidy levels of seed nuclei, using the mms21-2 allele which allowed us to visually discriminate homozygous mutant kernels from their wild-type siblings on the same cob (see S3 Fig).
As expected, we detected an expanded series of ploidy levels in whole seeds at 16-DAP, which included 2N, 4N and 8N nuclei in embryos, with the 3N, 6N and 12N nuclei likely representing triploid endosperm nuclei (S10C and S10D Fig). Again, a comparable ploidy distribution was seen for the mms21-2 seed tissues versus W22. Taken together, we were left to conclude Shown are RNA-seq heat maps generated by Z scores from mms21-1 and mms21-2 embryos as compared to those from W22 focused on the altered expression of 66 genes associated with DNA damage repair. Each column represents an individual biological replicate; rows represent specific genes of interest. The numerical Z scores in each box shown standard deviations away from the mean. Groups of genes with similar expression patterns were clustered by columns and rows. (B) mms21 seedlings are hypersensitive to DNA-damaging agents. mms21-1, mms21-2, and W22 seeds were germinated on sterile filter papers and then transplanted to solid growth medium supplemented with 20 ppm methyl methanesulfonate (MMS), 2 μM mitomycin-C (MMC), 1 mM hydroxyurea (HU), 100 nM bleocin (Bleo), or 10 μM zebularine (Zebu). After one-week, root growth, was measured and plotted relative to that seen with untreated W22 roots. Each bar represents the mean of three biological replicates (±SD). (C) Comet assays measuring by electrophoretic mobility the extent of DNA breaks. DNA from nuclei isolated from mms21-1, mms21-2, and W22 roots were subjected to agarose gel electrophoresis and then stained with propidium iodide. Mobility of the DNA was measured by the distance from the center of the nucleus to the edge of the comet tail. Scale bar = 50 μm. (D) Quantification of the comet assays in panel (C) by box plots based on the distribution of comet tail lengths for individual nuclei. The bottom and top of each box indicate the first (Q1) and third (Q3) quartiles, and the middle line reflects the median; the upper-limit equals Q3 plus 1.5 times interquartile range (IQR), and the lower-limit equals Q1 minus 1.5 times IQR. Each dot represents a single measurement (n = 50 cells).
https://doi.org/10.1371/journal.pgen.1009830.g008 that MMS21 has little impact in maize endoreduplication in contrast to that reported in Arabidopsis [12]. However, it is not yet known whether the ploidy levels of male gametes are affected.

MMS21 interacts with SUMO, the SCE1 E2, and the SMC5 and NSE4a
Subunits of the SMC5/6 complex We presumed that MMS21 influences DNA dynamics, and ultimately maize development, through SUMOylation of one or more targets. Given: (i) the known connections between MMS21 and the SMC5/6 complex in Arabidopsis and other organisms [13,43], (ii) proteomic indications that Arabidopsis NSE4a is a MMS21 substrate [5], and our discoveries here in maize that (iii) mms21 mutants are hypersensitive to DNA damage, and that (iv) the mRNAs encoding the NSE1 and NSE4a subunits of the SMC5/6 complex are selectively upregulated in mms21 backgrounds, led us to speculate that MMS21 SUMOylates one or more components of the SMC5/6 complex. A further connection was evident when directly quantifying SMC5 protein levels in mms21 seedlings by MS; as shown in Fig 7C, SMC5 levels were significantly lower in the strong mms21 mutants, mms21-2 and mms21-CR1, as compared to their normal siblings.
To provide further connections, we examined by yeast two-hybrid (Y2H) assays, whether MMS21 binds maize NSE4a and SMC5 using maize orthologs of the reported interactors BRAHMA and DPa as controls [30,31]. As shown in Fig 9A, MMS21 bound to one of its cognate E2s SCE1b (but not SCE1f) and to DPa, but only poorly to an N-terminal soluble fragment of BRAHMA as judged by growth on selection medium. None of the interactors bound SUMO1a. This association between MMS21 and SCE1b was lost when we used the expected polypeptides derived from the mms21-1 and mms21-2 alleles, implying that these aberrant forms poorly bind the SUMO-E2 intermediate. Intriguingly, MMS21 also bound strongly to SMC5 and NSE4a, with this association only weakly dampened when using the mms21-1 and mms21-2 protein variants (Fig 9A). Because MMS21 and NSE4a likely do not touch each other directly based on a general model of the SMC5/6 complex ( [47]; see S14D Fig), we hypothesize that SMC5 helps tether these two proteins.
To further validate the MMS21 interactions in planta, we applied bimolecular fluorescence complementation (BiFC) assays that transiently expressed maize SUMO1a, SCE1b, SCE1f, SMC5 and NSE4a as fusions with the N-terminal and C-terminal fragments of YFP in Nicotiana benthamiana epidermal cells: interactions were then scored by reconstituted YFP fluorescence. Even though we failed to detect interactions between MMS21 and SUMO1a by Y2H, strong BiFC signals was evident in the nucleus and cytoplasm of leaf cells co-expressing MMS21 and SUMO1a (Figs 9B and S11), confirming the expectation that MMS21 interacts with SUMO1a in planta. Likewise, we detected BiFC interactions between MMS21 and SCE1b and now weak interaction of MMS21 with SCE1f in both the nucleus and cytoplasm. Most interestingly, N. benthamiana cells co-expressing MMS21 with NSE4a or SMC5 also reconstituted YFP fluorescence but these signals were only evident in the nucleus, consistent with the known nuclear location of the SMC5/6 complex (Fig 9B). While the mms21-1 protein appeared to retain its affinity for NSE4a and SMC5 based on the BiFC signals, this affinity appeared less strong for the mms21-2 protein, potentially in agreement with its more compromised architecture.

SUMOylation of SMC5 by MMS21 in vitro
Our next objective was to demonstrate that MMS21 is a SUMO ligase that modifies SMC5 and/or NSE4. Here, we developed an in vitro SUMOylation system modified from Augustine et al. (2016), using recombinant maize proteins affinity purified via appended 6His tags. Ultimately, the system was built with SUMO1a, the SAE1/SAE2 E1 heterodimer, the SCE1b E2, and full-length MMS21 (S12 Fig). Specificity of the reactions for processed SUMO1a was confirmed by using the unprocessed SUMO1a precursor bearing its C-terminal extension that blocks conjugation (UP-SUMO1a), processed and active SUMO1a with its exposed glycine needed for the isopeptide bond, and a K0 variant in which all 7 SUMO lysines were substituted for arginines and thus unable to assemble SUMO-SUMO chains [24]. In reactions with just SUMO1a, E1 and E2, the kinetics of SUMOylation was then optimized for time, and ATP and E2 concentrations, using the conversion of free SUMO1a into higher molecular mass adducts for the output as assessed by immunoblot assays with anti-SUMO1 antibodies (S13A and S13B Fig). When we added MMS21 to the E1 and E2-containing reactions, robust SUMOylation was clearly evident, which reached saturation within 1 hr as opposed to overnight reactions missing MMS21 (S13A and S13C-S13D Fig). To confirm the correct enzymatic scheme, we tested the impact of pyrophosphate, which interferes with formation of the SUMO-adenylate intermediate [48]. In reactions containing 2 mM ATP, SUMOylation was strongly suppressed by pyrophosphate concentrations at 0.1 mM or above (S13E and S13F Fig).
Using this optimized SUMOylation system, we confirmed that MMS21 is a bona fide E3 that requires ATP, processed SUMO1a, and E1 and E2 activities to generate SUMO adducts ( Fig 10A). While MMS21 did not use unprocessed SUMO1a, conjugation was successful with the K0 variant although not to the same extent as the processed form, strongly suggesting that some conjugation reflected assembly of SUMO-SUMO chains. When the predicted mms21-1 and mms21-CR3 polypeptides harboring the SP-RING were used, conjugation was retained as expected if this domain is required for E2-SUMO binding (Fig 10B). reactions containing processed SUMO1a, the SAE1/SAE2 E1 heterodimer, the SCE1b E2, and ATP. A silver-stained gel of the reaction mixture is presented in S12A Fig. (B) The mms21-1 (1-1) mutant protein and a truncation missing the N-terminal region of MMS21(CR3) encompassing residues Met1-Ala68, retained SUMO ligase activity in vitro similar to wild-type MMS21. Reaction were for 1.5 hr as in panel (A). A silver-stained gel of the reaction mixture is presented in S12B Fig. (C The differing banding patterns for reactions containing wild-type MMS21 and the two mutants suggested that at least some of the SUMOylation by MMS21 reflected self-modification or direct transfer of SUMO1a from the E2 to MMS21. To confirm that MMS21 SUMOylates other proteins, we tested a V5-tagged version of MMS21 in vitro using both anti-SUMO1a and anti-V5 antibodies for conjugate detection. While modest SUMOylation of V5-MMS21 was seen in complete reactions using the anti-V5 antibodies, a strong and distinctive smear of conjugates was seen even at low E3 concentrations when using the anti-SUMO1a antibodies, implying that most conjugates were not MMS21-SUMO adducts (S14A-S14C Fig).
Our subsequent attempts to demonstrate that MMS21 will conjugate either NSE4a or SMC5 in vitro proved challenging due to the complicated architecture of the SMC5/6 complex that assembles around DNA into a topologically-closed, heterodimeric configuration (Garcia-Rodriquez et al., 2016; S14D Fig). This unique design is particularly evident for SMC5 and SMC6, which use their N-terminal and C-terminal regions to generate a folded Head contact region, and two internal coiled-coil regions that associate to generate a dimeric Arm, which itself is connected by an internal Hinge. The Arm is responsible for binding MMS21 at its Nterminus [49]. Unfortunately, we failed to recombinantly express full-length versions or several partial fragments of either protein without aggregation. However, using a column refolding strategy (see Materials and Methods), we successfully resolubilize full-length NSE4a and a 62-kDa ΔHead fragment of SMC5 missing both sections that comprise the Head region.
When tested in in vitro reactions, only the ΔHead fragment of SMC5 was successfully SUMOylated by MMS21. When added to complete reactions also containing processed SUMO1a, MMS21 modified the T7-tagged ΔHead fragment in the presence of ATP (Fig 10C). An adduct was detected at~100 kDa, which was consistent with the addition of a single SUMO moiety to the 80-kDa ΔHead polypeptide. As with general SUMOylation, both the mms21-1 and mms21-CR3 mutant proteins also directed SUMOylation of the T7-ΔHead fragment. By contrast, equivalent reactions containing full-length NSE4a tagged with V5 failed to generate even after prolonged incubations any new species with apparent molecular masses above the unmodified protein at 62 kDa ( Fig 10C). As the Head domains of SMC5 and SMC6 link NSE4 to the rest of the SMC5/6 complex (S14D Fig), this failure likely reflected a disrupted connection between MMS21 and NSE4.

Discussion
Despite the crucial importance of SUMOylation to plant growth, development, and defense against environmental challenges, the roles of the ligases that drive this post-translational modification and the identit(ies) of their targets remain largely obscure [1,5,50]. Here, we addressed these issues through genetic and molecular analyses of the SUMO E3 MMS21 using maize as a tractable model. Consistent with previous studies with its Arabidopsis MMS21 (HPY2) ortholog, we found through analysis of Mutator and CRISPR/Cas9 mutants of varying strengths, that MMS21 has a plethora of functions in maize. Included are roles during gametogenesis, root and shoot growth, and seed development that are likely underpinned by altered nuclear functions, including defects in DNA-damage repair, chromatin dynamics, and proper transcriptional maintenance of proteome balance. For the most part, strikingly similar phenotypes were observed for the maize collection of mms21 alleles as compared to Arabidopsis mms21 mutants, including substantially attenuated fertility and seed formation and compromised vegetative development [12,28,29].
Given the phenotypic severity of strong mms21 mutants, we focused our studies on weaker UniformMu-insertion alleles (mms21-1/2) that not only permitted the analysis of reproduction but also allowed the study of MMS21 throughout the maize life cycle. Particularly notable for these alleles were defects seen in anther maturation and pollen germination, and a delay in seed development that generated poorly-filled, shrunken seeds. The pollen defects could be related to a role for MMS21 in maintaining haploid ploidy levels in male gametes as recently described for Arabidopsis mms21 mutants [29]. While bulk analysis of SUMO conjugates assembled in planta revealed little change in SUMOylation in mms21 tissues, we found dramatic changes in the transcriptome, with a substantial percentage of mRNAs showing altered abundance in the mms21-1/2 backgrounds. Instead of finding specific categories of maize genes that were either significantly up-or downregulated, numerous loci were impacted from a wide range of GO categories, implying that MMS21 through its SUMOylation activity, impacts a large swath of the maize transcription. This transcriptional misregulation then translated into an imbalance of the maize proteome, with approximately 24% of the proteins displaying altered accumulation of greater or less than 2 fold even in these mild alleles.
In addition, substantially altered expression in the mms21-1/2 backgrounds was seen for several potentially informative loci. Included were mRNAs encoding the RNRS2-2 and RNRS-C subunits of the RNR complex important for ribonucleotide metabolism [40], MRE11b needed to maintain genome integrity [37], MORC6 that helps direct DNA silencing [38,39], and the NSE1 and NSE4a subunits of the SMC5/6 complex. All are intimately connected to DNA repair and chromatin dynamics, implying that MMS21-directed SUMOylation modifies these processes in maize which then strongly impacts transcription. Along with the cohesion and condensin complexes, the SMC5/6 complex is especially critical for providing compaction and elasticity to chromatin and the interconversion between euchromatic and heterochromatic states that globally influence gene expression and genome stability [32].
In agreement with a potential broad impact on maize mRNA abundance, our transcriptome profiling surprisingly revealed the ectopic accumulation of zein storage protein mRNAs in mms21-1/2 shoots. In fact, a large collection of zein mRNAs became the most highly upregulated set of transcripts in both mutant backgrounds, which was also confirmed by qRT-PCR analysis of our mms21-CR lines. A wealth of literature has shown that these mRNAs are expressed to high levels in endosperm (and likely embryos) as seeds develop with their encoded zein proteins then accumulating as dense aggregates in specialized protein bodies as seeds mature [41]; these proteins are ultimately consumed during germination to nourish that developing seedling. A similar upregulation was not seen in embryos and endosperm, presumably because these tissues already express zein transcripts to high levels. Also surprising was that this increase in zein mRNA abundances did not coincide with detectable levels of α-and γ-zein proteins in leaves, suggesting either that maize shoots were unable to translate the zein mRNAs out of context, or that shoots were not equipped to stably accumulate zein proteins without the concomitant assembly of appropriate storage compartments.
Collectively, both pervasively altered proteome/transcriptome balance and this unprecedented accumulation of zein mRNAs in mms21 shoots illustrate a global alteration in gene expression without MMS21. While the exact mechanism(s) remain unclear, we note that SUMOylation in yeast and mammalian cells has been intimately connected to transcriptional repression [51] and the maintenance of heterochromatin stability [52,53]. Moreover, the SUMO E2 and SUMO itself can often be directly mapped to repressed genes, suggesting that this mark is critical for enforcing heterochromatic, transcriptionally-repressed states [54][55][56]. Consequently, we imagine that the loss of MMS21-directed SUMOylation in the maize mms21-1/2 shoots causes a loss of heterochromatic organization around genes which are designed to be highly expressed but only under specific developmental conditions. This dysregulation, in turn, derepresses expression in other developmental contexts, thus allowing ectopic expression of genes such as zeins in non-appropriate tissues, or suppressing normally active genes. In support, we found that some highly up-or downregulated genes (FC �16 or -log 10 p-value �20) in the mms21-2 backgrounds (including those encoding zeins) were clustered within maize chromosomes (S15 Fig and S3 Table). These clusters could reflect heterochromatic regions under the global influence of MMS21-directed SUMOylation. Clearly, further structural analysis of chromatin surrounding zein and other MMS21-impacted genes and mapping chromatin regions that bear SUMO marks in different maize tissues should help clarify this possibility.
Beyond their influence on transcription globally, the mms21 mutations also impacted DNA damage repair, which is consistent with the role of MMS21 during this process in Arabidopsis and other organisms [13,18,57]. Of the many maize loci encoding factors connected to DNA repair, most had altered expression patterns in the mms21-1/2 backgrounds, with some upregulated and some downregulated. In accord, we found that mms21-1/2 roots were hypersensitive to a subset of DNA-damaging agents and housed nuclear DNA with increased DNA breaks as seen by comet assays. Surprisingly, while Arabidopsis mms21/hpy2 mutants displayed increased endoreduplication (Ishida et al., 2009), suggestive of defects in the endocycle, we did not observe a similar effect with the maize mms21-1 mutant for leaves, embryos and endosperm tissues, the latter of which is known to undergo numerous scheduled rounds of endoreduplication as seeds mature [46].
Central to understanding MMS21 is the identit(ies) of its substrates, which is complicated by the sheer size of the plant SUMOylome [3][4][5][6], and by the fact that SUMOylation can often impact multiple proteins in proximity without direct interaction via a "SUMO spray" mechanism [58]. Global proteomics on the Arabidopsis mms21-1 mutant implied that few proteins are actual MMS21 targets [5], which is supported in Arabidopsis [5,12] and maize (this report) by the similar profiles of SUMO conjugates seen by immunoblot analysis of total mms21 and wild-type lysates.
Previous studies with Arabidopsis implicated the SWI/SNF chromatin-remodeling complex subunit BRAHMA [30] and the cell-cycle check point protein DPa, [31] as MMS21 substrates. Here, we discovered through interaction and expression studies, a connection between maize MMS21 and the SMC5 and NSE4a subunits of the SMC5/6 complex involved in chromatin dynamics and DNA repair in Arabidopsis and other organisms [32,59,60]. As predicted, our BiFC studies showed that these interactions between SMC5/NSE4a and MMS21 occur in the nucleus, in agreement with the known location of the SMC5/6 complex [59,60]. NSE4a was previously suggested as an MMS21 substrate based on global SUMOylome profiling with Arabidopsis [5], while SMC5 had not yet been classified as a SUMO substrate in plants. Outside of plants, SMC5, NSE3, and NSE4 have all been shown to be MMS21 substrates, possibly via a SUMO spray mechanism [61,62].
Interestingly, analysis of our panel of UniformMu and CRISPR/Cas9 mms21 mutants suggested that both the interactions between SMC5 and MMS21 and the phenotypic effects of mms21 mutants are influenced by subtle alterations in the MMS21 protein sequence, especially near the N-terminal end where the single amino acid substitution in mms21-CR6 was found to be deleterious. The importance of this region is unclear, but it could harbor a contact used by MMS21 to associate with the rest of the SMC5/6 complex or other influential targets [49]. Interestingly, recent studies by Hays et al. [63] found a similar sensitivity of yeast MMS21 function by single amino acid changes near the N-terminus. While we do not yet know the in vivo SUMOylation status of SMC5 and NSE4a in the mms21 lines, we could show that MMS21 will selectively SUMOylate a partial fragment of SMC5 in vitro. A challenge in working with the SMC5/6 complex is its unusual topology which hinders in vitro approaches (S14D Fig), and the lack of genetic and molecular tools to easily dissect the process in maize. Work is now underway to overcome these hurdles. As SMC5 protein levels are lower in strong mms21 mutants, it is possible that SUMO addition by MMS21 protects it from turnover.
In conclusion, our genetic, phenotypic, molecular, transcriptome, and proteome studies identify MMS21 as an essential SUMO ligase for normal maize development. While the exact mechanism(s) remain unclear, we propose that this E3 helps maintain genome integrity against DNA damage and influences chromatin organization to allow faithful transcription throughout maize development, possibly by controlling chromatin compaction through modification of regulators such as BRAHMA, DPa, and subunits of the SMC5/6 complex. Consequently, MMS21-directed SUMOylation is likely key to DNA-damage repair and the heterochromatin/euchromatin transitions necessary for maintaining properly activated/ repressed gene states.

Description of mms21 mutants
The mms21-1 (mu1068022) and mms21-2 alleles in the W22 background were found within the UniformMu population [33], and were obtained from the Maize Genetics Cooperation Stock Center or provided directly from the University of Florida collection, respectively. The Mu insertion sites were verified by genomic PCR using insertion-specific primers together with the UniformMu border primer, TIR6. The mutant cDNAs were cloned directly from the homozygous lines and sequenced for further analysis. Descriptions of all oligonucleotides are listed in S1 Table. To generate the CRISPR/Cas9 lines (mms21-CR1 to 7) in the Hi-II background, guide RNA-binding sites 1 and 2 (GGACGCGCAAGCCCTAGTCG and GAGGGACGGAAAG TCCGATA) immediately upstream of a PAM were selected using the CRISPR Genome Analysis Tool (http://cbc.gdcb.iastate.edu/cgat/; [64] and the CRISPR RGEN Cas-Designer algorithm (http://www.rgenome.net/cas-designer/; [65,66], followed by secondary screens using MaizeGDB BLAST to avoid regions that might induce off-target effects. The final gRNA oligonucleotides were engineered with 5' overhangs containing TGTT and GTGT sequences for the sense strands of Guide 1 and Guide 2, respectively, and an AAAC sequence was appended to the 5' of all antisense oligonucleotides. Guide 1 oligonucleotides were annealed to one another and phosphorylated with T4 polynucleotide kinase before ligation into the BtgZ1-digested pENTR-gRNA1 [67], thereby placing their expression under the control of the rice RNA polymerase-III pU6.1 promoter. The resulting plasmid was digested with BsaI and ligated to phosphorylated Guide 2 oligonucleotides to generate the pENTR-gRNA1-ZmMms21 plasmid that would drive expression from another rice U6 gene promoter (pU6.2). LR clonase (Invitrogen) reactions recombined the guide RNA cassette into the pGW-Cas9 [68] gateway cassette to generate the final pGW-Cas9-gRNA-ZmMms21 construction, which was subsequently introduced into Agrobacterium tumefaciens strain EHA101. This plasmid was transformed into the Z. mays Hi-II line by the Plant Transformation Facility at Iowa State University (https://www. biotech.iastate.edu/biotechnology-service-facilities/plant-transformation-facility). The nature of the mms21-CR mutations was determined by PCR amplification of the genomic region encompassing the guide RNA target sites, followed by Sanger sequencing of ExoSAP (Applied Biosystems)-treated PCR amplicons to identify the exact lesions.

Plant growth and phenotypic analyses
The mms21-1 and mms21-2 mutants were backcrossed five times to the W22 ACR inbred line [33] before a self-cross to obtain homozygous plants, whereas the mms21-CR alleles were back-crossed twice to B73 and self-pollinated to generate segregating populations. The plants were grown to maturity using 3:1 Metromix 900 Turface MVP (Sungrow) in the greenhouse under a 16-hr light/8-hr dark photoperiod with 1000 W metal-halide lamps used to supplement natural light, and 24-27˚C and 22-25˚C day and night temperatures, respectively. Watering was performed as necessary with  Cal-Mag special supplemented with Sprint 330 iron chelate until flowering, at which point watering continued without amendments [24].
Heights were measured from the soil level to the youngest ligule beginning 14 DAS. Leaf length was measured from the auricle to the tip, and width was measured on fully mature plants at the halfway point of the fourth, sixth, or eighth leaves from the tassel. For fresh and dry weight measurements of roots and shoots, seeds were sown into~1.5 cm fine sand Turface Quick Dry layered atop coarse Turface MVP. Plants were watered twice daily; every fifth watering day used modified Hoagland solution [24]. To correct for delays in germination, tissues were harvested 7 or 14 d after first emergence of the coleoptile from the Turface growth substrate. For analysis of shoot and root growth, seedlings at 7 or 14 d after emergence were separated from Turface in a water bath and briefly dried with paper towels, and then roots were scanned and fresh weights were measured. Dry weights were measured after overnight drying at 55˚C. Root area was determined in ImageJ from scanned images by selecting the blue channel after RGB separation, inverting the lookup table, thresholding by MaxEntropy, and then measuring the area of the resulting image.
For the heat stress assays, plants were grown in controlled environment chambers (Conviron) at 28˚C under 16-hr light/8-hr dark photoperiod on soil containing 3:1 Metromix:Turface. Seedlings at the 3-leaf stage were transferred to a pre-equilibrated chamber set at 42˚C and 60% humidity; after 30 min, tissue was collected and immediately flash frozen to liquid nitrogen temperatures.

Pollen germination assays
Pollen was collected from flowering tassels from which the florets and anthers were gently removed. Pollens were incubated in solid germination medium containing 15% sucrose, 0.6% bacto-agar, 0.03% calcium nitrate, and 0.01% boric acid [69] for 3 hr at 28˚C in darkness. Germination was scored by protrusion of the pollen tubes seen microscopically.

RNA isolation, qRT-PCR, and cDNA cloning
Total RNA was isolated from 10-DAS shoots using the RNeasy Plant Mini Kit (Qiagen). RNA integrity was assessed after treatment with DNase I (ThermoFisher) by agarose gel electrophoresis and OD260/OD280 measurements. For RT-PCR, one μg of total RNA was subjected to the Superscript III First Strand Synthesis System (ThermoFisher) using an oligo (dT) 20 primer, followed by qRT-PCR using the Light Cycler 480 and SYBR Green I master mix (Roche Diagnostics) in combination with gene-specific primers (S1 Table). Relative transcript abundances were calculated by the 2 -ΔΔCT method [70], using amplification of the maize ACT1 gene as a reference. For Y2H, BiFC, and recombinant protein expression, open reading frames were PCR amplified from cDNA using Phusion High-Fidelity DNA Polymerase (Thermo-Fisher) and then recombined into pDONR221 via the Gateway BP clonase II reaction (Ther-moFisher). All clones were sequenced in their entirety to confirm the absence of secondary mutations.

RNA-seq analysis
Shoots were harvested 10 DAS from the mms21-1 and mms21-2 plants and their corresponding W22 inbred line grown in a 25˚C growth chamber, and immediately frozen in liquid nitrogen and stored in -80˚C. Endosperm and embryos were dissected from seeds at 16 DAP and frozen in liquid nitrogen. The mms21-2 seeds were identified visually at the time of harvesting, while the mms21-1 seeds were scored by genomic PCR after dissection of the cobs. Total RNA was extracted from three biological replicates, each of which was prepared from three seeds from the same segregating ear. Shoot and embryo RNAs were isolated using RNeasy Plant Mini Kit (Qiagen), while the endosperm RNAs were first enriched using TRIzol (Invitrogen) before isolation.
To prepare for RNA-seq analysis, the RNA samples were digested with DNase I at 37˚C for 10 min to remove contaminant DNA. RNA-seq library construction and sequencing were performed by BGI Americas Inc. (Hong Kong, China), using their DNBseq technology platform (PE150, 20 million reads). Roughly 7-Gb of raw data generated from each sample, after removing reads from the sequencing adaptors, were further screened to remove low-quality sequences using Trimmomatic v0.39 under the PE mode [71], which truncated the reads for base quality <15 within 4-base windows and kept only reads longer than 36 bases after trimming under the parameters: phred33, LEADING:3, TRAILING:3, SLIDINGWINDOW:4:15, and MINLEN:36). Trimmed reads were uniquely aligned to the B73 RefGen_v4.48 maize (Zea mays) genome assembly using STAR [72]. Counts for gene expression were obtained using HTSeq [73].
DEG analysis was conducted with the DESeq2 Bioconductor package version 1.28.1 in R 4.0, using the total uniquely mapped read count per gene parameter as input [74]. Up-or downregulated DEGs were selected from both mms21-1 and mms21-2 alleles based on adjusted p-values <0.05 and FCs �2, and were used as inputs to draw the Venn diagrams in R. PCA was calculated in R to visualize clustering of the datasets. Correlation analysis were performed using the log 2 -FC values provided by DESeq2, filtered using an adjusted p-value <0.05, and drawn with the R packages-ggpubr and ggplot2 [75,76]. Heatmaps were draw with pheatmap in the R package [77], based on the same datasets used for the correlation analyses. For gene set enrichment analysis (GSEA), all genes identified from RNA-Seq were first filtered to remove those that were not significant based on adjusted p-values <0.05. Filtered genes were ranked based on their calculated value using equation: -log 10 (p-value) � sign(log 2-FC) and imputed into the clusterProfiler package for GSEA in R [78]. GSEA was used to assess whether genes in biologically predefined sets occurred toward the top or bottom of a ranked list of all examined genes more than expected by chance. Significant gene sets were called as those with an adjusted p-value < 0.05 [79]. Volcano plots were generated in ggplot2 [76]. Mapping of DEGs on the maize chromosomes were determined by TBtools [80].

Proteomic profiling
Total maize proteome profiling was performed as described by McLoughlin et al. (2018) with minor modifications. Proteins were extracted into 50 mM HEPES (pH 7.5), 5 mM Na 2 EDTA, 2 mM dithiothreitol and 1 × plant protease inhibitor cocktail (Sigma-Aldrich), and precipitated with methanol-chloroform. Pellets were resuspended into 8 M urea, reduced with 20 mM dithiothreitol, and alkylated with 20 mM iodoacetamide for 1 h before digestion with 0.5 μg trypsin (Promega) overnight. The resulting peptides were desalted and concentrated with a 100-μl Bond Elut OMIX C18 pipette tip (Agilent Technologies) and resuspended in 20 μl 5% acetonitrile and 0.1% formic acid. The peptides were then analyzed with a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific) after reversed-phase nano-HPLC separation with a 25-cm analytical C18 resin column (Acclaim PepMap RSLC; Thermo Fisher Scientific) and a 5 to 95% acetonitrile gradient in 0.1% formic acid (FA) at a flow rate of 250 nL/min for 135 min. The mass spectrometer was operated in the data-dependent mode to automatically switch between full-scan MS and MS/MS acquisition. Data-dependent acquisitions were obtained using Xcalibur 4.0 software in positive-ion mode. MS1 spectra were measured at a resolution of 70,000 with an automatic gain control of 1 × 10 6 , a maximum ion time of 50 msec, and a mass range of 300-1,800 m/z. Up to 12 MS2 scans, with a charge state of 2 to 4, were triggered at a resolution of 17,500, an automatic gain control of 5 × 10 5 with a maximum ion time of 120 msec, a 1.6-m/z isolation window, and a normalized collision energy of 28. MS1 scans that triggered MS2 scans were dynamically excluded for 30 sec. Each biological replicate was analyzed by two technical replicates.
The resulting MS data sets were searched against the maize B73 proteome database (Zm-B73-REFERENCE-GRAMENE-4.0 from www.maizegdb.org) using Proteome Discoverer (version 2.5.0.400; Thermo Fisher Scientific) and a list of common protein contaminants [42]. Peptides were assigned by SEQUEST HT, allowing a maximum of two missed tryptic cleavages, a minimum peptide length of 6, a precursor mass tolerance of 10 ppm, and fragment mass tolerances of 0.02 Da. Carbamidomethylation of Cys and oxidation of Met were specified as static and dynamic modifications, respectively. Protein abundances reflected the average of two technical replicates if proteins were detected twice or used directly if the proteins were only detected in one technical replicate. Values among samples were normalized using the average values of 150 proteins considered least variable among the samples (using SD/average as ranking [42]). Only those proteins found in at least 2 biological replicates were included in the final datasets. Missing values for proteins not identified in all replicates were imputed in R using random draws from a Gaussian distribution centered around the minimal 1% of the observed values in that sample. To assess statistically significance differences among samples, the Limma algorithm in the "DEP" R package [81] was applied with the following filters: log 2 FC >1 (FC >2 on a linear scale) and adjusted p-value (corrected with Benjamini-Hochberg method) <0.05.

Genotoxic stress assays
Maize seeds were surface sterilized with 15% bleach containing 0.01% Triton X-100 for 15 min with shaking and then washed extensively with water. Sterilized seeds were first germinated on wet sterile filter paper for 3 d at room temperature and then transplanted into test tubes filled with Murashige and Skoog growth medium also containing 3% sucrose and 1% agar, which was supplemented with methyl methanesulfonate, mitomycin C, hydroxyurea, bleocin, or zebularine (Sigma-Aldrich). mms21-1 and mms21-2 seeds were germinated 2 d ahead of the W22 control to compensate for their slower germination. Root lengths were measured after 1 week of growth; lengths for the drug-treated samples were expressed relative to those observed under control conditions.

Comet assays
Comet assays were performed as described [45] with some modifications. Briefly, young roots were chopped with a razor blade in 400 mM Tris-HCl (pH 7.5). The nuclear suspensions were filtered through 40 μm cell strainers (Corning) to remove cell debris, mixed with an equal volume of 1% low melting point agarose prepared with PBS, and then spread on microscope slides cooled to 4˚C for 5 min to solidify the mixture. Embedded nuclear DNA was unwound at 4˚C for 5 min in electrophoresis buffer (1 mM Na 2 EDTA and 300 mM NaOH (pH> 13)), and then electrophoresed at 300 mA for 20 min at 4˚C. Slides were rinsed 3 times with 400 mM Tris-HCl (pH 7.5) before staining the DNA with 100 μg/ml propidium iodide. Fluorescence was visualized with a Leica SP8 laser scanning confocal microscope equipped with a 40X oil objective, using 488-nm for excitation and emission between 500 and 530 nm.

DNA flow cytometry analysis
Leaves at 10 DAS and whole seeds, endosperm and embryos dissected at 16 DAP were chopped with fine razor blades directly into ice cold Galbraith's buffer (45 mM MgCl 2 , 30 mM sodium citrate, 20 mM MOPS (pH 7.0), 0.1% Triton X-100, and 1% PVP-40). Nuclei were enriched by filtration through 30 μm CellTrics nylon filters (Sysmex), digested with RNase A for 10 min, and stained with 50 μg/mL of propidium iodide for 30 min in the dark. Flow cytometry was performed using an Accuri C6 cytometer (BD Biosciences) setup to measure at least 10,000 PI fluorescence events per sample, with emission detected using the FL2 optical filters (585/40 nm), and a minimum threshold intensity of 10 4 required for data collection.

Bimolecular fluorescence complementation (BiFC)
BiFC assay were accomplished using N. benthamiana leaves as described [83]. Sequence-confirmed coding sequences of maize genes, cloned into pDONR221 as above, were recombined in-frame via Gateway LR clonase II reactions with the N-or C-terminal halves of EYFP in the pSITE-N-EYFP-C1 or pSITE-C-EYFP-C1 vectors (ABRC Cat. No. CD3-1648 and CD3-1649), respectively. The resulting plasmids were introduced into A. tumefaciens strain GV3101 by freeze-thaw transformation, and positive clones were cultured overnight and resuspended to an OD 600 = 0.4 in 10 mM MES (pH 5.7), 10 mM MgCl 2 , and 200 μM acetosyringone. A. tumefaciens suspensions were incubated for 4 to 6 hr in darkness at room temperature before infiltration into 4 to 6-week-old N. benthamiana leaves. Leaf sections were excised 40 to 45 hr after infiltration and visualized with a Nikon A1+ confocal laser scanning microscope equipped with a 40x oil objectives (numerical apertures 1.30). Representative fluorescent and brightfield images were captured using Nikon Imaging Software Elements, using 488-nm light for GFP fluorescence excitation and 500-530 nm light for emission capture.

Recombinant protein expression and purification
Unless otherwise noted, all recombinant proteins were expressed as full-length versions with appended tags. The coding sequences of maize SUMO1a and MMS21, obtained by PCRamplification of B73 cDNA, were first cloned into pDONR221 as described above. Other maize SUMO components, including processed SUMO1a and the K0-SUMO1a(H89R) variant (both ending in its diGly motif), SCE1b-Myc, HA-SAE1, SAE2a-FLAG bearing the indicated tags were as described [24]. The constructions were further modified to include coding sequence for an N-terminal 6His tag to facilitate protein purifications and further subcloned into pRSFDuet-1 expression vector for expression in the E. coli strain BL21(DE3) pLysS (Promega). The cells were cultured at 37˚C to an OD 600 of 0.4~0.8 in of LB medium supplemented with 50 μg/mL kanamycin and 34 μg/mL chloramphenicol, followed by a 5-hr induction with 1 mM isopropyl-β-D-thiogalactopyranoside at 37˚C. Cells were harvested by centrifugation at 10,000 X g for 5 min at 4˚C, frozen in liquid nitrogen, and lysed in 15-20 mL BugBuster Master Mix (Millipore Sigma). The 6His-tagged proteins were affinity purified by nickel-nitrilotriacetic acid (Ni-NTA) agarose chromatography (QIAGEN). The bacterial cell lysates, supplemented with one tablet of Pierce Protease Inhibitor Tablets (ThermoFisher Scientific) and 10 mM imidazole, were applied to 5 mL of PBS-washed Ni-NTA beads (Bio-Rad) at 4˚C and incubated for 30 min with mixing. The Ni-NTA beads were washed once with 30 mL of NaH 2 PO 4 wash buffer (50 mM NaH 2 PO 4 (pH 7.8), 300 mM NaCl, and 20 mM imidazole), and once with 30 mL HEPES wash buffer (50 mM HEPES (pH 7.8), 300 mM NaCl, and 40 mM imidazole). Bound proteins were eluted twice with 1 mL of Tris elution buffer (50 mM Tris-HCl (pH 7.8), 100 mM NaCl, and 300 mM imidazole). Eluants were dialyzed at least twice for 3 hr at 4˚C against 1 L of dialysis buffer (40 mM HEPES (pH 7.5), 50 mM NaCl, 10 mM MgCl 2 , and 10% glycerol) to remove the imidazole.
To purify and solubilize NSE4a and the ΔHead (D291-E783) fragment of SMC5, IPTGinduced E. coli cells were first suspended in BugBuster Master Mix. The insoluble material was collected by centrifugation at 10,000 X g, solubilized into 20 mM Tris-HCl (pH 8.0), 0.5 M NaCl, 5 mM imidazole, 6 M guanidine-HCl and 1 mM 2-mercaptoethanol, loaded onto a Ni-NTA column equilibrated in the same buffer, and incubated at 4˚C with gentle shaking for 1 hr. The beads were washed with 10 volumes of Wash Solution-1 (20 mM Tris-HCl (pH 8.0), 0.5 M NaCl, 5 mM imidazole, 6 M guanidine-HCl, and 1 mM 2-mercaptoethanol), and 10 volumes of Wash Solution-2 (20 mM Tris-HCl (pH 8.0), 0.5 M NaCl, 20 mM imidazole,1 mM 2-mercaptoethanol, and 6 M urea). Refolding of the bound proteins were performed using a linear 6-0 M urea gradient, starting with the Wash Solution 2 and finishing with Wash Solution 2 minus urea. Bound proteins were eluted with 20 mM Tris-HCl (pH 8.0), 0.5 M NaCl, 500 mM imidazole and 1 mM 2-mercaptoethanol, and dialyzed twice at 4˚C against 1 L of dialysis buffer to remove the imidazole.

Accession numbers
The raw RNA-seq files are available at the NCBI Sequence Read Archive database under the submission number PRJNA685214 (https://www.ncbi.nlm.nih.gov/sra/PRJNA685214  siblings and mms21-2 normal siblings. Each dot represents one protein that had detectable expression in both samples and was plotted based on its log 2 FC in abundance (mutant/normal siblings) and its -log 10 p-value of significance based on the three biological replicates, each with two technical replicates. The horizontal and vertical dashed lines mark a FC = 2 in protein abundance and a p-value = 0.05, respectively. Histone proteins used to confirm data normalization are shown as green. SUMO pathway components and DNA repair-associated proteins are highlight in blue and red, respectively. (TIF) S10 Fig. The mms21 Mutations do not Substantially Impact the Ploidy Level of Maize. Nuclei were isolated from seedling leaves, whole seeds, endosperm and embryos, stained with propidium iodine, and subjected to FACS sorting to determine ploidy number based on fluorescence counts.