Many rice genes are differentially spliced between roots and shoots but cytokinin has minimal effect on splicing

Alternatively spliced genes produce multiple spliced isoforms, called transcript variants. In differential alternative splicing, transcript variant abundance differs across sample types. Differential alternative splicing is common in animal systems and influences cellular development in many processes, but its extent and significance is not as well known in plants. To investigate alternative splicing in plants, we examined RNA-Seq data from rice seedlings. The data included three biological replicates per sample type, approximately 30 million sequence alignments per replicate, and four sample types: roots and shoots treated with exogenous cytokinin delivered hydroponically or a mock treatment. Cytokinin treatment triggered expression changes in thousands of genes but had negligible effect on splicing patterns. However, many genes were alternatively spliced between mock-treated roots and shoots, indicating that our methods were sufficiently sensitive to detect differential splicing in a data set. Quantitative fragment analysis of reverse transcriptase-PCR products made from newly prepared rice samples confirmed nine of ten differential splicing events between rice roots and shoots. Differential alternative splicing typically changed the relative abundance of splice variants that co-occurred in a data set. Analysis of a similar (but less deeply sequenced) RNA-Seq data set from Arabidopsis showed the same pattern. In both the Arabidopsis and rice RNA-Seq data sets, most genes annotated as alternatively spliced had small minor variant frequencies. Of splicing choices with abundant support for minor forms, most alternative splicing events were located within the protein-coding sequence and maintained the annotated reading frame. A tool for visualizing protein annotations in the context of genomic sequence (ProtAnnot) together with a genome browser (Integrated Genome Browser) were used to visualize and assess effects of differential splicing on gene function. In general, differentially spliced regions coincided with conserved regions in the encoded proteins, indicating that differential alternative splicing is likely to affect protein function between root and shoot tissue in rice.

Differential alternative splicing is common in animal systems and influences cellular 23 development in many processes, but its extent and significance is not as well known in plants. To 24 investigate alternative splicing in plants, we examined RNA-Seq data from rice seedlings. The 25 data included three biological replicates per sample type, approximately 30 million sequence 26 alignments per replicate, and four sample types: roots and shoots treated with exogenous 27 cytokinin delivered hydroponically or a mock treatment. Cytokinin treatment triggered 28 expression changes in thousands of genes but had negligible effect on splicing patterns. 29 However, many genes were alternatively spliced between mock-treated roots and shoots, 30 indicating that our methods were sufficiently sensitive to detect differential splicing in a data set. 31 Quantitative fragment analysis of reverse transcriptase-PCR products made from newly prepared 32 rice samples confirmed nine of ten differential splicing events between rice roots and shoots. 33 Differential alternative splicing typically changed the relative abundance of splice variants that 34 co-occurred in a data set. Analysis of a similar (but less deeply sequenced) RNA-Seq data set 35 from Arabidopsis showed the same pattern. In both the Arabidopsis and rice RNA-Seq data sets, 36 most genes annotated as alternatively spliced had small minor variant frequencies. Of splicing 37 choices with abundant support for minor forms, most alternative splicing events were located 38 within the protein-coding sequence and maintained the annotated reading frame. A tool for 39 visualizing protein annotations in the context of genomic sequence (ProtAnnot) together with a 40 genome browser (Integrated Genome Browser) were used to visualize and assess effects of 41 differential splicing on gene function. In general, differentially spliced regions coincided with INTRODUCTION 52 Differential splicing of pre-mRNA transcripts, called alternative splicing, enables one gene to 53 produce multiple transcript variants encoding different functions. Alternative splicing is an 54 almost universal phenomenon in higher eukaryotes, occurring to varying degrees in every animal 55 and plant genome examined to date (Kalsotra and Cooper, 2011;Reddy et al., 2013). In animals, 56 differential expression of splice variants has been recruited as a regulatory mechanism in 57 multiple processes, such as sex determination in invertebrates and neuronal differentiation in 58 mammals (Kalsotra and Cooper, 2011;Salz, 2011;Barbosa-Morais et al., 2012). 59 In plants, less is known about the functional significance and patterns of alternative splicing. 60 However, several trends are apparent. Genes involved in circadian regulation are highly 61 alternatively spliced, often producing multiple splice variants that fluctuate in concert with 62 day/night cycling along with overall transcript abundance (Filichkin et al., 2015b). The serine 63 and arginine-rich (SR) family of RNA-binding, splicing regulatory proteins is greatly expanded 64 compared to mammals and includes many plant-specific forms (Kalyna and Barta, 2004; 65 Barbosa-Morais et al., 2006;Plass et al., 2008;Filichkin et al., 2015a). SR transcripts 66 themselves are also highly alternatively spliced in plants, with the relative abundance of these 67 alternative transcripts varying according to environmental stresses and hormones (Palusa et al.,68 2007; Gulledge et al., 2012;Filichkin et al., 2015a;Keller et al., 2016;Mei et al., 2017). 69 A growing body of evidence indicates that cell and tissue specific regulation of alternative 70 splicing occurs in plants, but its significance and extent is not well established (Vitulo et al., 71 2014;Li et al., 2016a;Sun et al., 2018). We previously found through analysis of RNA-Seq data 72 from Arabidopsis pollen that the relative abundance of splice variants was similar between 73 leaves and pollen, despite the differences between the two tissues (Loraine et al., 2013). 74 However, this latter analysis was limited by having just one biological replicate for pollen and 75 only two biological replicates for leaves. A more comprehensive analysis of multiple 76 Arabidopsis data sets found a high incidence of isoform switching, in which the identity of the 77 most prevalent variant differs between sample types (Vaneechoutte et al., 2017). However, this 78 splicing diversity may have arisen in part from the heterogeneity of the data sets used, which 79 were produced using rapidly changing (and improving) sequencing technologies at different 80 times by different groups. 81 In this study, we used a well-replicated RNA-Seq data set from rice to re-examine prevalence 82 of alternative splicing between tissues and hormone (cytokinin) treatment. This data set was 83 previously generated to investigate cytokinin regulation of gene expression in roots and shoots 84 from 10-day old rice seedlings (Raines et al., 2016). A parallel study produced an analogous data 85 set from Arabidopsis for comparison, but was less deeply sequenced (Zubo et al., 2017). Both 86 the rice and Arabidopsis RNA-Seq data sets included three biological replicates per sample type 87 and four sample types -roots and shoots treated with exogenous cytokinin or a mock, vehicle-88 only treatment. In both data sets, the treatment triggered differential expression of thousands of 89 genes, with roots affected to a greater degree than shoots. 90 For most alternatively spliced genes, regardless of whether or not they were differentially 91 spliced, the relative abundance of splicing forms was highly skewed, with most alternatively 92 spliced genes producing one major isoform. Nonetheless, there was a large minority of 93 alternatively spliced genes where minor isoforms were more abundant and therefore seemed 94 likely to affect gene function. We found that the relative abundance of splice variants for most 95 alternatively spliced genes was remarkably stable, with very few differentially spliced genes 96 between cytokinin treated and control samples. By contrast, many more genes were differentially 97 spliced between roots and shoot, and most differential splicing occurred within the protein-98 coding sequence. Moreover, nearly every differential splicing event detected merely change the 99 relative abundance of splice variants that co-occurred in the same sample. These results suggest 100 differential alternative splicing likely contributes to gene function diversification between roots 101 and shoots by moderating the relative abundance of splice co-expressed splice variants, but 102 alternative splicing plays little role in cytokinin signaling.

107
Rice and Arabidopsis samples were prepared and sequenced as described in (Raines et al., 108 2016;Zubo et al., 2017). Rice seedlings (Nipponbare) were grown hydroponically for ten days in 109 a growth chamber set to 14 hours light (28°C) and 8 hours of dark (23°C) with light intensity 700 110 mmol m -2 s -1 . Around six to ten seedlings were grown in the same pot, in four pots. On the tenth 111 day, culture media was replaced with new media containing 5 µM of the cytokinin 112 benzyladenine (BA) or 0.05 mN NaOH as a control. After 120 minutes, roots and shoots were 113 harvested separately. Roots and shoots from treatment or control pots were pooled to form three   and BowTie2 (Langmead and Salzberg, 2012) with maximum intron size set 137 to 5,000 bases. A command-line, Java program called "FindJunctions" was used to identify 138 exon-exon junctions from gapped read alignments in the RNA-Seq data. FindJunctions produces 139 BED format files containing junction features, and the score field of the BED file lists the 140 number of read alignments that supported the junction. Only reads that aligned to a unique 141 location in the genome were considered. Source code and compiled versions of FindJunctions are 142 available from https://bitbucket.org/lorainelab/findjunctions.

143
Identification of alternative splicing events and differential splicing 145 To date, there have been two major releases of O. sativa japonica gene models: the MSU7 146 gene set (Kawahara et al., 2013) and the RAP-Db gene set (Sakai et al., 2013). The two gene 147 model sets contain mostly the same data, but the MSU7 gene models appear to be the most 148 heavily used and annotated with Gene Ontology terms. For simplicity, and to take advantage of 149 available functional annotations, we used the MSU7 annotations here. For analysis of 150 Arabidopsis data, we used TAIR10 (Lamesch et al., 2012) and Araport11 (Cheng et al., 2017) 151 gene models.  For each alternatively spliced region representing two mutually exclusive splicing choices, 182 RNA-Seq read alignments that unambiguously supported one or the other splicing choice were 183 counted. For AS and DS events, only gapped reads that aligned across intron junctions were 184 counted. For RI events, gapped reads that aligned across the retained intron were counted as 185 support for the intron-removed (S) form, and un-gapped reads that overlapped at least 20 bases 186 within the intron were counted as support for the intron-retained (L) form. 187 For each alternatively spliced region in each biological replicate, the number of reads 188 supporting L or S, but not both, were used to calculate percent-spliced-in (PSI) as N/M*100, 189 where N was the number of reads supporting the L form and M was the number of reads that 190 supported S or L but not both. This is the same as the splicing index described in (Katz et al., 191 2010 (Benjamini and Hochberg, 1995), as implemented in the R 196 programming language "p.adjust" method. Alternative splicing events with FDR less than or 197 equal to 0.1 were considered differentially alternatively spliced.

RT-PCR and capillary gel electrophoresis analysis of alternative splicing
207 Differential alternative splicing detected by analysis of RNA-Seq was re-tested using the 208 reverse transcriptase, PCR-based fragment analysis method described in (Stamm et al., 2012). 209 Differentially spliced regions identified computationally were PCR-amplified using fluorescently 210 labeled primers and quantified using capillary gel electrophoresis. One benefit of the method is 211 that the results are expressed as relative abundances of splice variants within a sample, thus 212 eliminating the need to normalize using reference genes as in traditional qRT-PCR experiments 213 aimed at measuring overall gene expression. 214 For splicing validation, new rice seedlings equivalent to the mock-treated (control) samples 215 from the RNA-Seq experiment were grown and harvested. Seedlings were grown hydroponically 216 in pots containing either liquid media only or calcined clay granules watered with liquid media 217 as recommended in (Eddy et al., 2012). After twelve days, plants were removed from the pots 218 and roots and shoots were collected separately. Roots and shoots from the same pot were 219 combined to form paired biological replicates. Samples were flash frozen in nitrogen and stored 220 at -80°C prior to RNA extraction. Using the exon-intron overlap method described previously (English et al., 2010) and Fig. 1,   242 alternative splicing events within each gene were identified and annotated. Following annotation 243 of alternative splicing events, RNA-Seq read alignments from the rice and Arabidopsis libraries 244 described in (Raines et al., 2016;Zubo et al., 2017) were used to assess alternative splicing in 245 four sample types: roots and shoots from seedlings treated with the cytokinin compound 246 benzyladenine (BA) or with a mock, control treatment. For each alternative splicing event, the 247 number of sequence alignments unambiguously supporting each alternative was counted. These 248 counts were used to calculate percent-spliced-in (PSI), the percentage of read alignments 249 supporting the longer (L) isoform. 250 In the combined data from all libraries from the rice data set, 77% of AS events had at least 251 one read supporting each of the two splicing choices, and 19.8% had support for just one splicing 252 choice. Only 2.8% of AS events has no reads supporting either form; these corresponded to 253 genes with low or no expression in any of the sample types tested. Most genes annotated as 254 alternatively spliced had small minor variant frequencies, i.e., the less frequently observed forms 255 were supported by fewer than 20% of informative RNA-Seq sequences (Fig. 2). Nevertheless, 256 there was a large minority of alternative splicing events (around one third) where the minor, less-257 frequently observed form was more abundant and was supported by at least two out of ten 258 informative alignments. These alternatively spliced regions correspond to the middle, trough-like 259 region of Fig. 2. 260 261 Fig. 2. Distribution of percentspliced-in (PSI) for annotated splicing events in rice where each choice was supported by at least one RNA-Seq alignment. PSI was calculated as 100*L/(S+L), where L and S were the number of reads that supported the splicing choice that included (L) or excluded (S) the differentially spliced region. Read alignment counts from all twelve libraries were combined to obtain a global view of alternative splicing occurrence in rice seedlings. The Ushaped character of the distribution persisted whether lower or higher thresholds of informative reads were used. 262 diverse functions 264 We used Gene Ontology term enrichment to determine if the subset of genes in rice for 265 which alternatively spliced forms were unusually abundant exhibited enrichment with specific 266 functions or processes, e.g., circadian cycling, in which alternative splicing might play a 267 prominent regulatory role. We asked if some Gene Ontology terms were significantly enriched 268 with genes containing alternative splicing events in which the minor form frequency was 269 between 20 and 50%, corresponding to the central trough region of Fig. 2. Interestingly, we 270 found that these genes exhibited a diversity of gene functions, with no significant enrichment of 271 functional categories. Thus, alternative splicing in which minor forms are highly prevalent 272 appears to affect genes with many functions in rice.

274
Many rice genes are differentially spliced between roots and shoots but cytokinin hormone 275 application has minimal effect on splicing 276 In animals, differential splicing between cell or tissue types contributes to cellular 277 differentiation, especially in the nervous system (Naftelberg et al., 2015). Less is known about 278 the role of alternative splicing in regulating cellular differentiation and other processes in plants. 279 Rice shoots and roots are profoundly different tissues, but our previous analysis of this same data 280 set found that many of the same genes were expressed in both (Raines et al., 2016). This raises 281 the question of how these two different tissues are able to carry out their specialized roles, and 282 suggest the hypothesis that differential splicing could enable differential functions in genes 283 expressed in both tissues, as proposed in (Reddy et al., 2013). 284 Analysis of the effects of cytokinin treatment on this same data set from rice identified many 285 thousands of genes that were differentially expressed in response to cytokinin (Raines et al.,286 2016). However, little is known about the role of alternative splicing during cytokinin response, 287 except for one study in Arabidopsis that reported a shift in splicing of SR protein genes 288 following cytokinin hormone treatment (Palusa et al., 2007). Therefore, we examined differential 289 splicing in the rice RNA-Seq data set comparing root and shoot tissue with or without cytokinin. 290 First, we asked: When an alternatively spliced gene was expressed in two different sample 291 types, was the relative abundance of splice variants the same or different? To address this, we 292 examined correlation of PSI between roots and shoots or between BA-treated versus mock-293 treated samples (Fig. 3). We found that PSI was similar between treated and untreated samples, 294 as revealed by the tighter clustering of scatter plot points (Fig. 3A-B). This indicated that genes 295 that were alternatively spliced in BA-treated samples were also alternatively spliced in the 296 controls, and that the relative abundance of splice variants was similar. Thus, the cytokinin 297 hormone treatment had minimal effect on splicing. By contrast, there were many genes where the 298 relative abundance of splice variants was different between roots and shoots (Fig. 3C). 299 Consistent with Fig. 3, statistical testing of PSI differences between sample types identified 90 300 genes where PSI was significantly different between roots and shoots (FDR ≤ 0.1) but only four 301 and two genes where PSI was different between cytokinin-treated samples and controls in roots 302 and shoots, respectively (Supplementary Table S1). Thus, we observed limited but non-trivial 303 levels of differential alternative splicing between roots and shoots but minimal differential 304 alternative splicing between control and BA-treated samples. 305 306 Fig. 3. Scatter plots comparing percent-spliced-in (PSI) between sample types in rice for annotated splicing events.

307
PSI was calculated from RNA-Seq reads obtained from sequencing rice seedling shoots and roots grown 308 hydroponically and subjected to a two-hour treatment with BA, a cytokinin analog, or a mock-treatment (control).

314
Comparison of Arabidopsis differential splicing shows similar patterns to rice 315 To determine whether the observed patterns of differential splicing are similar in other plants, 316 we analyzed splicing in Arabidopsis roots and shoots that had also been treated with cytokinin 317 (Zubo et al., 2017). Due to the Arabidopsis libraries not being sequenced to the same depth as 318 the rice libraries, many more splicing events had little or no support. Using the same FDR 319 threshold as with the rice data set (FDR ≤ 0.1), we identified few differentially spliced regions 320 between shoots and roots (3) and none in the control to treatment comparisons (Supplementary   321   Table S2). However, PSI was distributed similarly to rice in that most alternatively spliced genes 322 expressed one major isoform (Supplementary Fig. S1A). In addition, scatter plots showing 323 average PSI in treated versus untreated samples showed a much tighter clustering of points as 324 compared to scatter plots comparing roots and shoots ( Supplementary Fig. S1B-D). Statistical 325 testing of PSI differences confirmed the cytokinin hormone treatment had minimal effect on 326 splicing in Arabidopsis. Thus, the general pattern of more differential splicing between tissue 327 types as compared to treatment with exogenous cytokinin appears conserved between rice and 328 Arabidopsis. Alternative splicing can occur anywhere in a gene, including UTR and protein-coding 333 regions. Most differential splicing between roots and shoots (67%) occurred within protein-334 coding regions (Table 1 and Supplementary Table S1), suggesting that differential splicing is 335 likely to affect gene function at the level of the protein product. In every instance of differential 336 alternative splicing, major and minor isoforms were both detected, with differential splicing 337 observed as a change in the relative abundance of the two forms. 338 Because three bases encode one amino acid, the lengths of spliced coding regions in a 339 transcript are multiples of three. Thus, when alternatively spliced regions occur in coding regions 340 and are not multiples of three, then inclusion of these regions in transcripts is likely to introduce 341 a frame shift, resulting in a premature stop codon and a truncated protein product. As shown in 342 Table 1, there was an enrichment of alternatively spliced regions in rice that were evenly 343 divisible by three in coding regions versus non-coding in all subsets of the data. These subsets 344 included all annotated alternatively spliced regions, regions where the minor form was unusually 345 prevalent (the trough region of Fig. 2), and differentially spliced regions. Thus, alternative 346 splicing within the coding regions of genes was biased against introducing frame shifts and 347 promoted protein remodeling rather than truncation. *P-value obtained from binomial test of the null hypothesis that the true probability of a differentially spliced region having a length divisible by three is 1 in 3 and an alternative hypothesis that the probability is greater than 1 in 3.

349
To further understand the effects of splicing on protein-coding sequences, we visualized 350 differentially spliced regions together with RNA-Seq alignments, coverage graphs, and inferred 351 junctions using genome browsers. Two genome browsers were used to visualize the data -352 Integrated Genome Browser  and ProtAnnot (Mall et al., 2016). the InterProScan Web service (Finn et al., 2017). 358 Of the 105 differentially spliced regions, 71 overlapped protein-coding sequence regions, 359 suggesting that in these cases, alternative splicing affected protein function. All but one (70/71) 360 of the differentially spliced regions embedded in coding regions overlapped a predicted 361 functional motif (e.g., a predicted transmembrane helix) or a region found by protein 362 classification systems (e.g., Pfam (Finn et al., 2016) or PANTHER (Thomas et al., 2003)) to be 363 conserved among members of the same protein family (Supplementary Table S1 and Fig. 4). (lower panels) shows a zoomed-in view of RNA-Seq coverage graphs from rice root (blue) and shoot 369 (green). Y-axis is the number of RNA-Seq aligned sequences with MSU7 gene models in black below. 370 371 RT-PCR with capillary gel electrophoresis confirmed differential splicing between rice 372 roots and shoots for nine of ten genes tested 373 We used a method based on capillary gel electrophoresis of fluorescently tagged PCR 374 products to assay alternative splicing of ten genes detected as differentially spliced between rice 375 roots and shoots (Stamm et al., 2012). New rice seedlings were grown under a close-to-identical 376 replication of the RNA-Seq experiment. Primers were designed to amplify differentially spliced 377 regions, including one primer that was conjugated to a fluorescent tag. Following PCR 378 amplification of cDNA prepared from the new rice samples, products were resolved on a 379 capillary-based sequencer and PSI calculated (Table 2). In nine out of ten genes, differential 380 alternative splicing was confirmed. In the one case where differential alternative splicing was not 381 confirmed, there were very few RNA-Seq read alignments covering the differentially spliced 382 region, suggesting this was likely a false positive result. The FDR cutoff used to detect 383 differential splicing in the RNA-Seq data was 0.1, corresponding to one in ten false discoveries, 384 in line with results from the microcapillary-based analysis.  This study profiled alternative splicing using a high coverage RNA-Seq data from 10-day 390 old, hydroponically-grown rice seedlings treated with a cytokinin hormone. A less-deeply 391 sequenced data set from similarly treated Arabidopsis seedlings provided comparison with 392 another plant species. We found that cytokinin treatment induced very few splicing changes 393 between treated and untreated controls. However, there were many differences in splicing 394 between untreated roots and shoots, and most of these changed the protein coding region of 395 genes. 396 Palusa et al. found that BA-treatment of Arabidopsis seedlings triggered splicing changes in 397 multiple SR genes (Palusa et al., 2007), encoding RNA-binding proteins whose counterparts in 398 metazoans regulated alternative splice site choice. Their study used PCR amplification of cDNA 399 followed by agarose gel electrophoresis to detect changes in splicing and focused on SR protein 400 genes only. Thus, we expected to observe changes in SR gene splicing due to the cytokinin 401 treatment, leading to changes in splicing for many downstream genes. However, no such 402 differential splicing was apparent in either RNA-Seq data set tested. It is possible that the 403 differences in methodology used to measure splicing changes between the two studies (RNA-Seq 404 versus visualization of PCR amplification of cDNA) could account for the differences in 405 observations. However, close examination of SR splicing genes in our dataset revealed no 406 significant differences. 407 One possible explanation for why the cytokinin treatment had minimal effect on splicing was 408 that the treatment itself was ineffective. However, differential expression analysis showed that 409 many genes were up-or down-regulated by the treatment in the two data sets tested -rice and 410 Arabidopsis (Raines et al., 2016). Fewer genes were detected as differentially expressed in the 411 Arabidopsis data set, most likely reflecting higher variability between biological replicates 412 combined with lower sequencing depth as compared to the rice data set. Nevertheless, known 413 cytokinin response genes were differentially regulated in both data sets, showing the cytokinin 414 treatment penetrated plant cells and induced stereotypical cytokinin signaling without also 415 triggering changes in splicing. 416 The relative lack of differential splicing between cytokinin-treated and mock-treated samples 417 suggests that cytokinin signaling does not employ alternative splicing as a regulatory mechanism 418 to the same degree as with other plant hormones, notably abscisic acid (ABA). ABA plays a role 419 in perception and response to stresses, especially desiccation stress (Maia et al., 2014). ABA also 420 plays a role in regulating splicing of SR45, an SR-like protein, and SR45 plays a role in 421 regulating downstream splicing of multiple genes (Cruz et al., 2014 (Argyros et al., 2008;Kieber and Schaller, 2018 Using the same methods and data set, we identified a relatively large number of genes in rice 433 (90) that were differentially spliced genes between shoots and roots, and we validated nine of ten 434 using fragment analysis of independently produced rice samples. This observation of differential 435 splicing between roots and shoots is important for two key reasons. First, it shows that our data 436 analysis methods can identify differential alternative splicing in a data set. In other words, the 437 roots versus shoots comparison provided a positive control for differential splicing detection. 438 Second, the detection of differential splicing between roots and shoots illuminates the function of 439 alternative splicing in plant cells. For most of the differentially spliced regions, both forms were 440 present, and the difference in relative abundance between forms was often slight, rarely more 441 than five or ten percent (Supplementary Table S1). Our data supports the growing body of 442 evidence that alternative splicing is cell, tissue, and stage specific in plants (Vitulo et al., 2014;443 Gupta et al., 2015;Li et al., 2016a;Sun et al., 2018), including in roots (Li et al., 2016b). It is 444 likely that alternative splicing plays a role in fine-tuning gene function to meet the needs of 445 different plant tissues or cell types where a gene is expressed. 446 We also examined the prevalence of alternative splicing, independent from differential 447 splicing. That is, we used RNA-Seq read alignments to assess how often annotated alternative 448 splice sites were used in our RNA-Set data sets. For most genes annotated as alternatively 449 spliced, the minor form frequency was typically low, accounting for less than 20% of sequence 450 read alignments across the differentially spliced region. Genes where minor form frequency 451 exceeded 20% exhibited a diversity of functions. Thus, many diverse processes in rice involved 452 alternatively spliced genes in which splice variants were expressed at levels likely to affect gene 453 function in different ways.

454
A major limitation of this study was that we limited our analysis to annotated splice forms 455 and did not attempt to form new transcript models based on the RNA-Seq data. This was done 456 mainly because the libraries used were not strand-specific and attempts to assemble transcripts 457 using transcript assembly tools led to incorrect fusion of neighboring genes and other artifacts 458 (not shown). Future studies will therefore benefit greatly from using better library preparation 459 protocols to simplify and streamline data analysis. Nonetheless, this analysis provides new 460 insight into the role of alternative splicing in plant tissues and hormone response.

461
In conclusion, by analyzing the number of reads that supported different splice variants, we 462 identified examples of differential splicing with confirmation by RT-PCR with capillary gel 463 electrophoresis. There were 90 genes differentially spliced between rice root and shoot tissues, 464 but only four between cytokinin-treated and non-treated samples. For most differential splicing 465 events, the protein-coding regions were affected, strongly suggesting that differential splicing is 466 playing a role in modulating gene function between roots and shoots.  Arabidopsis roots (y axis) compared to mock roots (x axis). (C) BA-treated Arabidopsis shoots (y axis) 486 compared to mock shoots (x axis). (D) Mock shoots (y axis) compared to mock roots (x axis). 487 488