A key role for UV sex chromosomes in the regulation of parthenogenesis in the brown alga Ectocarpus

Although evolutionary transitions from sexual to asexual reproduction are frequent in eukaryotes, the genetic bases of these shifts remain largely elusive. Here, we used classic quantitative trait analysis, combined with genomic and transcriptomic information to dissect the genetic basis of asexual, parthenogenetic reproduction in the brown alga Ectocarpus. We found that parthenogenesis is controlled by the sex locus, together with two additional autosomal loci, highlight the key role of the sex chromosome as a major regulator of asexual reproduction. Importantly, we identify several negative effects of parthenogenesis on male fitness, but also different fitness effects between parthenogenesis and life cycle generations, supporting the idea that parthenogenesis may be under both sexual selection and generation/ploidally-antagonistic selection. Overall, our data provide the first empirical illustration, to our knowledge, of a trade-off between the haploid and diploid stages of the life cycle, where distinct parthenogenesis alleles have opposing effects on sexual and asexual reproduction and may contribute to the maintenance of genetic variation. These types of fitness trade-offs have profound evolutionary implications in natural populations and may structure life history evolution in organisms with haploid-diploid life cycles.


INTRODUCTION 25
Although sexual reproduction, involving fusion of two gametes, is almost ubiquitous across 26 eukaryotes, transitions to asexual reproduction have arisen remarkably frequently [1]. 27 Parthenogenesis, which is widespread in all major eukaryotic lineages [2][3][4][5][6][7], involves the 28 development of an embryo from an unfertilized gamete, without contribution from males [1]. In plants, parthenogenesis is a component of apomixis, which is the asexual formation of seeds, 30 resulting in progeny that are genetically identical to the mother plant. In gametophytic apomixis, 31 the embryo sac develops either from a megaspore mother cell without a reduction in ploidy 32 (diplospory) or from a nearby nucellar cell (apospory) in a process termed apomeiosis. Apomeiosis 33 is then followed by parthenogenesis, which leads to the development of the diploid egg cell into an 34 embryo, in the absence of fertilization (reviewed in [8]). 35 The molecular mechanisms underlying parthenogenesis in plants and animals remain largely 36 elusive, although the factors triggering the transition to asexual reproduction have been more 37 intensively studied in plants than in animals, motivated by the potential use of asexual 38 multiplication in the production of crop plants for agriculture (e.g. [9,10]). In some apomictic plants, 39 inheritance of parthenogenesis is strictly linked to an apomeiosis locus (reviewed in [11]). In other 40 species the parthenogenesis locus segregates independently of apomeiosis [12][13][14]. For example, 41 apomixis in Hieracium is controlled by two loci termed LOSS OF APOMEIOSIS (LOA) and LOSS OF 42 PARTHENOGENESIS (LOP), involved respectively in apomeiosis and parthenogenesis, respectively 43 4 developed beyond the 10-cell stage at 16-days post release and as non-parthenogenic (P-), when 94 less than 4% of the gametes had developed at 16d after release ( Figure 1B, Table S1). 95 96 In several brown algal species, unfused male and female gametes show different 97 parthenogenetic capacity, and it is usually the female gametes that are capable of parthenogenesis 98 whereas male gametes are non-parthenogenic (e.g. [23,24]). To investigate if there was a link 99 between parthenogenetic capacity and sex, we crossed the female (EA1) P+ strain with the male 100 (RB1) P-strain described above ( Figure S1, Table S1). The diploid heterozygous zygote resulting 101 from this cross (strain Ec236) was used to generate a segregating family of 272 haploid 102 gametophytes. These 272 siblings were sexed using molecular markers [25] and their gametes 103 phenotyped for parthenogenetic capacity (see above). The segregating population was composed 104 of 144 females and 128 males, consistent with a 1:1 segregation pattern (chi2 test; p-value=0.33, 105 Table S2). Phenotypic assessment of the parthenogenetic capacity of the gametes released by each 106 gametophyte revealed a significant bias in the inheritance pattern, with 84 individuals presenting a 107 P-phenotype and 188 a P+ phenotype (Chi2 test; p-value=2.86x10 -10 ) ( Table S2, S3). Strikingly, all 108 female strains exhibited a P+ phenotype whereas 30% of the male strains were recombinants, i.e. 109 had a P+ phenotype (Table S2). This result indicated the presence of a parthenogenesis locus or loci 110 that was not fully linked to the sex locus, and suggested a complex relationship between gender 111 and parthenogenetic capacity. 112 again phenotyped for parthenogenetic capacity, and the results showed without exception that the 126 parthenogenetic phenotype was stably maintained across generations (Table S4). 127 To further investigate the inheritance of parthenogenetic capacity, a male P+ individual was 128 crossed with a P+ female ( Figure S1). A total of 23 gametophyte lines were produced from two 129 heterozygous sporophytes resulting from this cross. Phenotyping for sex and parthenogenesis 130 revealed that all gametophyte lines exhibited a P+ phenotype, regardless of the sex (Table S5). We 131 concluded that parthenogenesis is controlled by a genetic factor(s). 132 Generation of a genetic map for E. siliculosus 133 To produce a genetic map based on the EA1 x RB1 cross, a ddRAD-seq library was generated 134 using 152 lines of the segregating progeny ( Figure S1) and sequenced on an Illumina HiSeq 2500 135 platform. A total of 595 million raw reads were obtained, of which 508 million reads passed the 136 quality filters with a Q30 of 74.1%. A catalogue of 8648 SNP loci was generated using filtered reads 137 from the parental strains and the STACKS pipeline (version 1.44) [26]. Twenty-eight individuals were 138 removed due to excessive missing genotypes (see Methods) and highly distorted markers were also 139 removed. The final map constructed with 124 individuals contained 5594 markers distributed 140 across 31 linkage groups (LGs) and spanning 2947.5 centimorgans (cM). The average spacing 141 between two adjacent markers was 0.5 cM and the largest gap was 17.6 cM (on LG23). The lengths 142 gametes and that carried a non-synonymous polymorphism (Table S11, Figure 2D) further reduced 243 the list of candidate genes to 9/14 (U/V), 1 and 16 candidates (in P1, P2 and P3 respectively). 244 Parthenogenetic male gametes exhibit reduced fitness in sexual crosses 245 It is not clear why some strains of Ectocarpus exhibit male gamete parthenogenesis whilst others 246 do not. More specifically, bearing in mind that all strains tested so far exhibit parthenogenesis of 247 female gametes, why are male gametes not parthenogenetic in some lineages? To address this 248 question, we investigated if there were differences in fitness between P-and P+ male gametes for 249 parameters other than parthenogenetic growth. Specifically, we examined fertilisation success 250 (capacity to fuse with a female gamete) and growth of the resulting diploid sporophyte. 251 We tested several combinations of crosses between P-or P+ males and several females (Table Importantly, embryos arising from a P-male gamete grew significantly faster than embryos derived  255 from fusion with a male P+ gamete ( Figure 4B, 4C, Mann-Whitney u-test p<0.05). 256 The overall size of zygotes is expected to be correlated with zygotic and diploid fitness [34][35][36]. 257 We therefore hypothesised that if P-male gametes are larger, fusion with a female gamete would 258 generate larger (and therefore fitter) zygotes. Measurements of gamete size of P+ and P-strains 259 revealed significant differences in gamete size between different strains (Kruskal-Wallis test, 260 Chi2=3452.395, P<2.2e-16, Table S14, Figure 4D, Figure S2). However, there was no correlation 261 between the parthenogenetic capacity of male gametes and their size, suggesting that the 262 increased fitness of the zygotes was unlikely to be related to the size of the male gametes. 263 Taken together, these analyses indicate that P+ male gametes exhibit overall reduced fitness in 264 sexual crosses, both at the level of success of fusion with a female gamete and growth of the 265 resulting embryo. We found no link between the size of the male gamete and the capacity to 266 perform parthenogenesis, which excludes the possibility that the fitness decrease is due to the size 267 of the male gamete. 268

DISCUSSION 269
A key role for the sex chromosome in parthenogenesis 270 In this study, we uncover the genetic architecture of parthenogenesis in the brown alga E. 271 siliculosus and demonstrate that this trait is controlled by two major and one minor QTL loci that, 272 together, account for 44.8% of the phenotypic variation. The two main QTL loci were located in the 273 SDR on the sex chromosome and on LG18 respectively, and the minor QTL was also located in LG18. 274 Analysis of differential expression pattern and polymorphism for genes within the QTL intervals 275 allowed the establishment of a list of a total of 89 candidate parthenogenesis genes: 17/22 genes 276 within the sex chromosome QTL interval (in the U and V respectively), 11 genes within the P2 locus 277 and 56 within the interval of the minor P3 locus. Interestingly, within the major P2 QTL a strong 278 candidate gene coded for a membrane-localized ankyrin repeat-domain palmitoyltransferase (Ec-279 20_004890). In S. cerevisiae, genes belonging to the same family are involved in the gamete 280 pheromone response pathway, regulating the switching between vegetative and mating states 281 [37,38]. 282 Our results reveal a critical role for the sex chromosome in the control of parthenogenesis, with parthenogenesis was triggered in females regardless of the allele carried at the P2 or P3 locus. The 286 observed effects could be due to a conditional repressor of parthenogenesis in the male V-specific 287 region or an activator of parthenogenesis in the female U-specific region. However, a recent paper 288 on another brown alga Undaria pinnatifida described genetically male individuals that were capable 289 of producing oogonia and whose eggs were parthenogenic [39]. Similarly, several male L. pallida 290 lines from a South African population had unusual reproductive structures resembling small eggs, 291 which are also capable of parthenogenesis (Ingo Maier, pers. commun.). These results would 292 therefore be consistent with a repressor of parthenogenesis being present on the V-specific region 293 in these brown algae, that appears to be impaired in variant strains, or with an activator of 294 parthenogenesis downstream of the female cascade. 295 Male fitness effects of parthenogenetic capacity 296 Our results indicate that parthenogenetic capacity has a dramatic impact on the fitness of male 297 gametes. Specifically, P-male gametes are fitter than P+ male gametes for sexual reproduction and 298 this is reflected in significantly higher fertilisation success and higher growth rate of the resulting 299 zygote. Considering that P+ males would be expected to exhibit reduced fitness in sexually 300 reproducing populations, and the fact that females are phenotypically P+ regardless of the allele at 301 the P2 and P3 QTL, how can the P+ allele be preserved in the population? In other words, how is 302 the parthenogenesis polymorphism maintained? 303 Heterozygous advantage can maintain polymorphism in diploid organisms. For instance, most 304 obligate parthenogenetic vertebrates arise from hybridization between closely related species, 305 resulting in elevated individual heterozygosity relative to the parental genotypes [40][41][42]. This is 306 considered adaptive for colonizing new areas where high genetic diversity may provide the 307 necessary genetic tools to adjust to new conditions. In the case of Ectocarpus, fixing the P+ allele in 308 the female SDR and the P-allele in the male SDR would be a way to maintain the alleles polymorphic 309 in the sporophyte. Note however that this process would be applicable to the SDR QTL, and would 310 not necessarily explain the polymorphism maintained at the autosomal QTLs. 311 One interesting possibility is that parthenogenesis is a sexually antagonistic trait (or at least 312 differentially selected in males versus females), i.e., P+ alleles would be advantageous for females 313 because they would be capable of reproducing even in absence of gametes of the opposite sex, so 314 that P+ would be selected for in females, whereas P-increases male fitness because sporophytes 315 sired by a P-male can grow more rapidly. Polymorphism could therefore be maintained by 316 balancing selection [43][44][45]). Although we could not measure the effect of parthenogenetic would be consistent with the pervasiveness of the female P+ phenotype and the differences in 319 fitness between P+ and P-males. This phenomenon would be particularly relevant in spatially 320 heterogeneous and/or unpredictable environments, where the P+ or P-allele(s) in males would 321 alternatively selected for, depending on female density. In this scenario, parthenogenesis capacity 322 could be considered a bet-hedging strategy for males. 323 Temporal or spatial changes in population density are extremely common (e.g. Another potential mechanism for the maintenance of genetic variation is opposing selection 337 during the diploid and haploid stages of biphasic life cycles, also known as ploidally-antagonistic 338 selection [51]. Parthenogenesis could be considered an example of a trait under 339 ploidally/generation antagonistic selection because the P-allele transmitted by the male gamete is 340 advantageous to the diploid (sporophyte) generation (because zygotes grow faster if the father is 341 a P-) but detrimental to the haploid (partheno-sporophyte) generation (because if they do not find 342 a female gamete, males that carry a P-allele die). Ploidally-antagonistic selection has been 343 proposed to have a significant impact on major evolutionary dynamics, including the maintenance 344 of genetic variation ( [51][52][53] and the rate of adaptation [54]. Moreover, it appears that P+ and P-345 are under differential selective pressures in males (when populations reproduce sexually, P-should 346 be beneficial to males and P+ detrimental). Mathematical modelling [55] predicts that when 347 selection differs between the sexes (and in particular when the gametophyte-deleterious allele is 348 neutral or slightly beneficial in one of the sexes), being close or within the SDR expands the range 349 of parameters allowing generation-antagonistic mutations to spread. Note that conflict arising from 350 generation-antagonism or from differences in selection in gametophytes versus sporophyte Is parthenogenesis adaptive? 353 In the brown algae, the ancestral state appears to have been sexual reproduction through 354 fusion of strongly dimorphic gametes (oogamy) [56], that were incapable of parthenogenesis 355 (reviewed in [24]). This suggests that gamete parthenogenesis was superimposed on a sexual cycle, 356 having evolved secondarily possibly to ensure reproduction in conditions where populations have, 357 for instance, low population density. A challenge for understanding the adaptive nature of gamete 358 parthenogenesis in these organisms would be to identify the conditions under which it occurs in 359 nature. Brown algae exhibit a remarkable degree of reproductive plasticity during their life cycle 360 [21,57] and it is possible that this plasticity is related to capacity to adapt to new conditions, in 361 particular low population density or very fragmented habitats where finding a partner may be 362 A cross between a parthenogenetic female (strain EA1) and a non-parthenogenetic male (strain 388 RB1) was carried out using a standard genetic cross protocols [62] and a diploid heterozygous 389 sporophyte was isolated (Ec236) (Figure 1; Table S1). At maturity, the sporophyte (strain Ec236) 390 produced unilocular sporangia, i.e, reproductive structures where meiosis takes place (Figure 1). A 391 total of 272 unilocular sporangia were isolated, and one gametophyte was isolated from each 392 unilocular sporangium. 393 The 272 strains of the EA1 x RB1 derived segregating population were cultivated in autoclaved 394 sea water supplemented with half strength Provasoli solution [63] at 13°C, with a light dark cycle 395 of 12:12 (20 µmol photon m -2 s -1 ) using daylight-type fluorescent tubes [61]. All manipulations were 396 performed in a laminar flow hood under sterile conditions. We phenotyped the strains for 397 parthenogenetic capacity (P+ or P-) and for sex (male or female). Parthenogenetic capacity was 398 assessed by scoring the capacity of the gametes to develop into partheno-sporophytes in the 399 absence of fertilization. In order to assess phenotype stability, gametophytes were sub-cultivated 400 in different conditions for two weeks and then exposed to high intensity light to induce fertility. 401 Parthenogenetic capacity was measured using the released gametes (Table S3). We monitored 402 gamete germination every two days. In P+ strains, >96% of the gametes developed as partheno-403 sporophytes in the absence of fertilization whereas in P-strains, less than 4% of the gametes were 404 capable of parthenogenesis. To test the stability of the phenotype across generations, we cultivated 405 partheno-sporophytes and induced them to produce unilocular sporangia and release meio-spores 406 to obtain a new generation of gametophytes. The parthenogenetic capacity of gametes derived 407 from these second-generation gametophytes was then tested (Table S3). Note that this experiment 408 is feasible in P-males because a very small proportion (less than 4%) of their gametes are 409 nevertheless able to develop into mature partheno-sporophytes. 410 Each of the 272 gametophytes of the EA1 x RB1 segregating family was frozen in liquid nitrogen 411 in a well of a 96 well plate. After lyophilization, tissues were disrupted by grinding. DNA of each The ddRAD-seq library was constructed as in [64] using HhaI and SphI restriction enzymes datasets and its algorithm reduces data filtering and curation on the data, yielding more markers in 466 the final maps with less manual work. In order to obtain the expected AxB segregation type for this 467 haploid population, the pedigree file was constructed by setting the parents as haploid grand-468 parents and two dummy individuals were introduced for parents. The module ParentCall2 of LP3 469 took as input the pedigree and the vcf files to call parental genotypes. The module Identification and mapping of QTL were carried out using the R package R/qtl (version 1.39-5) 481 [71] and MapQTL version 5. Because parthenogenetic capacity was phenotyped as a binary trait 482 (either non-parthenogenetic 0 or parthenogenetic 1) non-parametrical statistics were used to 483 identify loci involved in parthenogenesis. In R/qtl, the scanone function was used with the "binary" 484 model to perform a non-parametrical interval mapping with the binary or Haley-Knott regression 485 methods. In MapQTL, the Kruskal-Wallis non-parametric method was used. To determine the 486 statistical significance of the major QTL signal, the LOD significant threshold was determined by 487 permutation. 488

Analysis of linkage disequilibrium 489
In order to determine an approximate interval around the QTL peaks for the candidate genes The small number of gametes released from Ectocarpus siliculosus strains did not allow RNA-seq 494 data to be obtained from this species. To analyse gene expression in P-(male) and P+ (female) 495 gametes, we therefore used two Ectocarpus species 1 strains belonging to the same Ectocarpus reference genome Ec32 predicted proteins 514 (http://bioinformatics.psb.ugent.be/orcae/overview/EctsiV2) (e-value cut-off = 10e-5) and the 515 orthology relationship between Ectocarpus species 1 and Ec32 (Ectocarpus species 7) was 516 established based on the best reciprocal blast hits. 517 Identification of candidate genes in the QTL intervals 518 We used two methods to identify putative candidate genes located in the QTL intervals. First, a 519 marker-by-marker method, by mapping the sequences of the markers located within each QTL 520 interval to the reference genome of the closely reference species strain Ec32 (Cock et al., 2010). 521 When a sequence successfully mapped to the Ec32 genome, a coordinate was recorded for the 522 marker, relative to its position on the physical map of Ec32. The linkage disequilibrium (see method 523 above) estimated for each linkage group was used to refine the number of genes non-randomly 524 associated with these markers, giving a first list of candidate genes within each QTL region. The 525 second method used the same approach but was based on the reference genome of the paternal 526 strain of the population (strain RB1). There were some differences between the two lists obtained 527 by the two methods, which are due to the following factors: (a) because the assembly of the RB1 528 genome was guided by the Ec32 reference genome and its annotation was based on Ec32 529 transcriptomic data, the RB1 genome potentially lacks some genes that would be due to loci such 530 as genes that are unique to the species E. siliculosus (RB1 strain) being omitted during the guided 531 assembly. Hence the list obtained with the first method (using the Ec32 genome) contains genes 532 that are absent from the RB1 genome; (b) while the two species are closely related, they are not 533 identical, and the E. siliculosus genetic map exhibited some rearrangements compared to Ec32 534 which placed some markers, along with associated genes, into the QTL intervals (these missing 535 markers were located elsewhere on the Ec32 genome). In summary, the list obtained with Ec32 536 genome contained some genes that are missing from the RB1 genome because of its imperfect 537 guided assembly and the list obtained with the RB1 genome contained some genes absent from 538 the corresponding intervals on Ec32 because of rearrangements. A final, conservative list of 539 candidate genes was obtained by merging the two lists in order not to omit any gene that were 540 potentially located within the intervals (Table S11). 541 SNP and indel detection method 542 Draft genomes sequences are available for the parent strains RB1 and EA1 [32]. Using Bowtie2, 543 we aligned the EA1 genome against the RB1 genome and generated an index with sorted positions. 544 The program samtools mpileup [76] was used to extract the QTL intervals and call variants between 545 the two genomes. The positions of variants between the two genomes were identified and filtered 546 based on mapping and sequence quality using bcftools [72]. The annotation file generated for the 547 RB1 genome was then used to select SNPs and indels located in exons of protein-coding genes for 548 further study (bcftool closest command). The effect of polymorphism on modification of protein 549 products was assessed manually using GenomeView [77], the RB1 genome annotation file (gff3) 550 and the vcf file for each QTL region. 551 GO term enrichment analysis 552 A Gene Ontology enrichment analysis was performed using two lists of genes: a predefined list 553 that corresponded to genes from all three QTL intervals and a reference list including all putative 554 genes in the mapped scaffolds based on the Ec32 reference genome and that had a GO term 555 annotation. The analysis was carried out with the package TopGO for R software (Adrian Alexa, Jörg 556 Rahnenführer, 2016, version 2.24.0) by comparing the two lists using a Fisher's exact test based on 557 gene counts. 558

Epistasis analysis 559
Epistasis analysis was carried out with the R package R/qtl (version 3.3.1). Two analyses were 560 performed, one with the full data set (female and male genotypes generated with RAD-seq method) 561 and the second with only the male individuals. For both analyses, the scantwo function from R/qtl 562 were used with the model "binary" as the phenotypes of the individuals is either 1 (P+) or 0 (P-). 563

Fitness measurements 564
Reproductive success was assessed in the segregating population by measuring the capacity of 565 male P+ and P-gametes to fuse with female gametes and by measuring the length of the 566 germinating sporophytes derived from these crosses. For this, we crossed males and females as 567 described in [62]. Briefly, we mixed the same amount of male and female gametes (app. 1x10 3 568 gametes) in a suspending drop, and the proportion of gametes that succeeded in fusing was 569 measured as in [78]. Two different P+ males (Ec236-34 and Ec236-245) and two different P-males 570 (Ec236-10 and Ec236-298) were crossed with five different females (Ec236-39; -203; -233; -284 and 571 Ec560) (Table S13). Between 50 and 150 cells (zygotes or unfertilised gametes) were counted for 572 each cross. The length of zygotes derived from a cross between the female strain Ec236-105 and 573 either the male P-strain Ec236-191 or the male P+ strain Ec236-154 was measured after 5h, 24h, 574 48h, 3 days and 4 days of development using Image J 1.46r [79] (13 zygotes for the P-male parent 575 and 14 zygotes for P+ male parent). For all datasets, the assumption of normality (Shapiro test) and 576 the homoscedasticity (Bartlett's test) were checked. The latter's assumptions were not met for zygote length, and consequently statistical significance differences at each time of development 578 was tested with a non-parametrical test (Mann Whitney U-test, α=5%). 579

Measurement of gamete size 580
Gamete size was measured for representative strains of each parthenogenetic phenotype found 581 in the segregating population (P+ and P-) (Table S3). Synchronous release of gametes was induced 582 by transferring each gametophyte to a humid chamber in the dark for approximately 14 hours at 583 13°C followed by the addition of fresh PES-supplemented NSW medium under strong light 584 irradiation. Gametes were concentrated by phototaxis using unidirectional light, and collected in 585 Eppendorf tubes. Gamete size was measured by impedance-based flow cytometry (Cell Lab 586 QuantaTM SC MPL, Beckman Coulter®). A Kruskal-Wallis test (α=5%) followed by a posthoc Dunn's 587 test for pairwise comparisons were performed using R software to compare female and male 588 gamete size (Table S14). Genes in QTL intervals were selected based on differential expression of their orthologs in P+ versus 831 P-in gametes, their differential expression between generation (gametophyte/partheno-832 sporophyte) and polymorphisms exhibited in exons and predicted to modify the protein product. 833 *SDR gametologue; X, sex-specific gene. 834     Table S10. Summary of the sequencing methods and raw data obtained. 899 Table S11. Predicted functions, expression patterns and polymorphisms of genes in the QTL 900 intervals. Expression data in transcript per million (TPM) for P-(male) versus P+ (female) gametes 901 were obtained from strains belonging to the Ectocarpus siliculosi group (Ectocarpus species 1). 902 Information about the type of polymorphism in the parental strains of E. siliculosus segregating 903 population (EA1 female and RB1 male) is also included. Genes represented in Figure 2 are 904 highlighted in bold. "-" means that there is no best reciprocal ortholog with detectable expression 905 in Ectocarpus species 1. Pseudogenes in the sex-determining region were removed except for those 906 which have a gametologue in the opposite SDR, and these are italicised. 907 Table S12. List of polymorphisms in coding sequence of genes located within the three 908 parthenogenesis QTL intervals. 909 Table S13. Fusion success of male P-versus P+ gametes with gametes of the opposite sex. The 910 total number of individuals corresponds to the total number of scored individuals (developing 911 either by parthenogenesis or derived from fusion of gametes). 912 LG1 LG3 LG4 LG5 LG6 LG7 LG8 LG9 LG10 LG11 LG12 LG13 LG14 LG15 LG16 LG17 LG19 LG20 LG21 LG22 LG23 LG24 LG25 LG26 LG27 LG28 LG29 LG30 LG2