The callus transformation protocol was adapted from Toki [16] and fully described in Supplemental File 1.

Single knockouts were obtained using plasmids kindly provided by Miao et al. [17], following the guidelines reported by the authors [17], pOs-sgRNA was used to clone a single gRNA targeting HTD1. Subsequently, using the Gateway cloning system the cassette was moved to the binary vector pH-Ubi-cas9-7 encoding for the Cas9 endonuclease and hygromycin resistance as plant selection marker.

Multiple knockouts were made using pRGEB32, provided by Yinong Yang (Addgene plasmid # 63142) [18].

Four gRNAs of 20 nt, one for each gene to be disrupted HTD1, GN1A, GS3 and GW2, were designed using the online tool CRISPR-P v2.0 (http://crispr.hzau.edu.cn/CRISPR2/). gRNAs proposed by the software were manually screened and selected based on their proximity to the start codon and avoiding potential off-targets.

Since Kabre was indicated to be an O. glaberrima accession both by the National Small Grain Collection (USDA) repository and in a previous publication [19], we initially designed the gRNA matching HTD1 using the O. glaberrima orthologous gene sequence (ORGLA04G0179300) identified by performing a blast-search with the O. sativa locus (OS04G0550600). Prior to proceed with plant transformation in order to ensure a perfect sequence match between the O. glaberrima annotated gene and Kabre, the gRNA targeting location and neighboring region were cloned and sequenced using primers Osp_1332 and Osp_1333 (S3 Table). Primers Osp1209 and Osp1210 were used to create the gRNA for targeting HTD1 (S3 Table).

As described in the results section, analysis of morphological traits and genotyping revealed that Kabre is most likely an O. sativa accession. For this reason, gRNAs used to target each of the three seed yield regulators GS3, GN1A and GW2 were designed on the O. sativa loci Os03g0407400, Os01g0197700 and Os02g0244100, respectively. PCR amplification and sequencing of the Kabre gRNA-targeting regions was conducted using specific primers (S3 Table) to confirm sequence identity and the absence of mismatches.

Primers Osp1367 to Osp1372 (S3 Table) are the gRNAs for each of the three genes with adapters linked in order to assembly the multiplex CRISPR construct following detailed instructions published by Xie et al. [24] (pRGEB32 vector). Osp1584 and Osp1585, matching Cas9 of pRGEB32, were used to assess the presence of the transgene in plants transformed with the multiplex construct. Atp5706 and Atp5718 instead were adopted to screen for the presence of the transgene in plants transformed with the construct for a single gene knock out (i.e. Kabre htd1) as published by Miao et al [17].

Genomic DNA (gDNA) was extracted following the protocol described in Supplemental File 1. All PCR reactions using gDNA were performed with GoTaq® (Promega) polymerase according to the company instructions.

Genotyping of Kabre was conducted in order to assess whether it has an O. sativa or O. glaberrima genetic background. We amplified by PCR three different species-specific genetic markers RM197, PROG1 and S1 as described in the results section. PCR was performed using Terra^™ PCR Direct Polymerase Mix (Takara) using a small piece of rice leaf directly in the reaction tube according to the Takara instruction manual. The following PCR condition were used: 98°C 2min, 35 cycles (98°C 10 sec, 60°C 15 sec, 68°C 30 sec), 68°C 10 min. Samples from O. sativa (IR64 and Gigante accessions) and O. glaberrima (CG14 and TOG5681 accessions) were used as control. All the primers used for this study are listed in S3 Table.

All the stages of tissue culture, except co-cultivation, were conducted under continuous cold fluorescent light at a temperature of 28° C. Once regenerants were ready for transplanting they were moved to soil and kept at 27°C in long day for about two months (18 hrs light). Later, plants were moved to short day conditions (14 hrs light) to induce flowering, after about two months rice seeds were harvested.

Five different Cas9 T1 negative plants derived from htd1-1 were selected and analyzed for the dwarf phenotype, composed by 3 and 2 plants having -7 and -17 bps homozygous deletion, respectively. Kabre mutants for seed traits were selected among the T2 generation. Seeds were measured for length and width using “SMARTgrain” a free program available online [20]. Six wild type, 10 gs3 and 10 gn1a single mutants, 18 gs3 gn1a double mutants and 11 gs3 gn1a gw2 triple mutant plants were analyzed taking 40 seeds from each plant; results are presented as average values (Fig 1). All selected mutant genotypes (e.g. single and double mutants) were equally composed in numbers by two different mutant alleles as described in the results section (S2 Fig). In contrast, T2 triple mutants were derived from a single T1 line and thus after segregation were comprised of 5 plants having either +1 insertion on gn1a and 6 plants carrying a -1 deletion on the same gene, while all the plants carried the same mutation on the two others loci gs3 and gw2 as reported in the Results section (S2 Fig). The statistical significance of differences was analyzed by variance (ANOVA) and post-hoc Tukey HSD.

CRISPR-Cas9 genome editing technology has great potential as a tool for plant improvement. However, one requirement for the use of this system is the ability to express the CRISPR guide RNA (gRNA) and Cas9 protein in plant cells. To investigate which African landraces can be used for Agrobacterium-mediated transformation, the regeneration capacity of embryogenic calli from different accessions was tested using a modified protocol published previously by Toki [16] (S1 File). All the African rice varieties that we tested were reported to be Oryza glaberrima rice species (Table 1).

Eight of the thirteen accessions that were selected for this study showed the ability to regenerate plantlets from calli cultured on hormone gradient plates (see examples in S1 Fig). The Kabre landrace from Northern Ghana was one of the most vigorously regenerating varieties and was selected for Agrobacterium-mediated transformation with CRISPR-Cas9 constructs. Notably, when grown in the greenhouse, Kabre plants showed phenotypes typical of landrace rice accessions but not modern cultivars, like elevated plant height, small seeds and small panicles. Kabre was thus a perfect candidate for our pilot experiment to show that by targeting specific “domestication” loci, fast improvement of landraces is possible.

Table 1 accessions were chosen based on their different geographical distribution across West Africa where O. glaberrima was domesticated. Accession identifiers can be used to retrieve online general information and features through the General Resources Information Network (GRIN–Global https://training.ars-grin.gov). TOG5681 and CG14 accessions were obtained from IRD (Montpellier, France). All the remaining accessions were obtained from the U.S.D.A. National Small Grains Collection (NSGC, Stuttgart, Arkansas, USA). In the last column those accessions that showed regeneration of plantlets from calli are indicated with “+”. The hormonal concentrations for which we obtained an optimal regeneration are reported in S2 Table.

African landraces are often lodging-sensitive due to their excessive height. In Asian rice, loss of function oshtd1 mutants were reported to exhibit dwarfism and over-proliferation of tillers [21], making plants less lodging sensitive and able to produce a higher number of panicles per plant. OsHTD1 (OS04G0550600), encodes a carotenoid cleavage dioxygenase. We performed a Genebank search to identify the O. glaberrima HTD1 (OgHTD1) ortholog. The OgHTD1 gene was identified on chromosome 4 (ORGLA04G0179300) sharing over 99% sequence homology with OsHTD1. For CRISPR-mediated mutagenesis a single gRNA matching the second exon of the open reading frame (575 bp downstream of the start codon) of OgHTD1 was designed (S2 Fig). Subsequently embryogenic Kabre calli were transformed with the pH-Ubi-cas9-7 editing construct [17] and the two htd1 mutants that were obtained contained bi-allelic mutations. Htd1-1 harbored deletions of 17 and 7 bases and htd1-2 deletions of 3 and 2 base pairs (S2 Fig).

Five Cas9 negative plants from the T1 progeny of htd1-1 were analyzed. These plants displayed excessive tiller production and dwarfism (Fig 1A) in agreement with what has been previously shown in O. sativa Nipponbare [21] thus indicating potential use of this mutation to reduce sensitivity to lodging in this African landrace. As compared to the wild type HTD1 locus which encodes for 609 amino acids, deletions of 17 and 7 base pairs would result in truncated proteins of 195 and 243 amino acids, respectively (S2 Fig). The obtained phenotypes suggest that CRISPR-mediated approaches could lead to important traits improvement in African rice landraces.

When carefully analyzing the htd1-1 and htd1-2 mutant phenotypes we observed that the mutant and wild-type Kabre plants showed some typical O. sativa characteristics, despite being registered as an O. glaberrima species. The mature panicle was not erect and open as in O. glaberrima but more like in O. sativa (Fig 2A). Moreover, the ligule, the membranous appendage on the adaxial surface of a leaf, at the junction between the leaf sheath and the leaf blade was typical of O. sativa [22] (Fig 2B). Therefore, we decided to investigate the genetic background of Kabre in more detail and performed a genotyping experiment using different markers discriminating O. sativa and O. glaberrima species (Fig 3). Firstly, SSR marker analysis using the marker RM197 was performed according to Chen et al. [23]. Furthermore, the deletion of the PROG1 gene and a deletion nearby this gene were investigated by PCR analysis since they are reported to be reliable species-specific O. glaberrima markers [8] [24]. Lastly, a marker specific to the S₁ locus corresponding to a postzygotic reproductive barrier–associated locus between O. sativa and O. glaberrima was also used for the analysis [25]. As reference controls for the two different species genomic DNA extracted from O. sativa cv. IR64 and cv. Gigante and from O. glaberrima accessions CG14 and TOG5681 were used. All these analyses suggest that Kabre is an African O. sativa landrace.

To further investigate possible yield improvements in African rice landraces by simultaneously stacking multiple agronomic traits selected during domestication in Asian rice, three genes, GRAIN NUMBER 1A (GN1A), GRAIN SIZE 3 (GS3) and GRAIN WEIGHT 2 (GW2) [12] were selected for mutagenesis. These loci are known to be major QTLs for negatively regulating plant yield. GN1A (OsCXK2), artificially selected during domestication, encodes a cytokinin oxidase/dehydrogenase that negatively controls panicle length and consequently spikelet number [26]. GS3 encodes a Gγ protein subunit containing a transmembrane domain spanning the plasma membrane and is implicated as a negative regulator of seed length and weight in several cultivars [27]. Lastly, GW2 encodes a RING-type E3 ubiquitin ligase and plays a role as negative regulator of grain girth and weight [28].

One gRNA for each of the three target genes was designed based on the O. sativa japonica genome sequence [11]. gRNAs showing ideal features in terms of target specificity and GC content were designed to match the first, fifth and second exon of the GN1A, GS3 and GW2 locus, respectively (S2 Fig). Subsequently the target sequences were verified by sequencing the gRNA matching regions for the three selected loci in the Kabre background to verify the absence of any sequence mismatch (S3 Table). The gRNA sequences were then cloned into the CRISPR multiplex vector pRGEB32 [24]. Following transformation, 76 plants were identified as Cas9 positive by PCR screening. Disrupted loci were verified by sequencing. Four classes of mutants were obtained in T₀: 3 gn1a and 19 gs3 single mutants, 38 gn1a gs3 double mutants, and 16 gn1a gs3 gw2 triple mutants. Among the 16 triple mutants, only one plant carried the gw2 mutation in homozygosity, all the others were heterozygous for gw2. All the other plants (single and double mutants) had the targeted loci mutated in homozygosity or carried bi-allelic mutations suggesting that the gRNA targeting GW2 had the lowest efficiency. In the T₁ generation, Cas9 negative plants were selected and grown for phenotypic analysis in the T₂ generation. In total, six wild type (WT), 10 gs3 and 10 gn1a single mutants, 18 gs3 gn1a double mutants and 11 gs3 gn1a gw2 triple mutant plants were analyzed. Of these lines we selected two independent mutant lines for further phenotypical analysis. In T₂ selected single mutant gs3 lines carried either a +1 insertion or a -2 deletion, predicted to result in a protein matching the wild type only for the first 150 amino acids and lacking in-frame stop codon thus likely resulting in an aberrant transcript and leading to non-sense mediated decay [29]. Single gn1a mutants analyzed instead carried either a -2 bps deletion or a +1 bp insertion, in both cases the resulting protein were predicted to match wild type only for the first 86 amino acids and contained premature stop codons. gn1a gs3 double mutant lines selected for phenotypic analysis of seed features either carried a -2 or -4 bps deletion in the gs3 locus, the latter possibly translating in an aberrant protein that matched the wild type for the first 150 amino acids, while both genotypes had homozygous -2 bps deletion for gn1a. Finally, Cas9 negative plants were selected among the segregant T₂ progeny obtained from a T₁ triple mutant carrying a -2 bps deletion at the gs3 locus, a biallelic +1/-1 bp mutation for gn1a and a -1 homozygous deletion on the gw2 locus. The latter was predicted to translate to a non-functional truncated protein of 89 amino acids.

The gn1a single mutants showed an important increase in panicle length of approximately 49% relative to wild type. Interestingly, combining the gn1a mutant with gs3 and gw2 partially attenuated this phenotype, even though the overall panicle length increase was maintained (Fig 1B). Grain length and width were measured with SMARTGrain software [20] (Fig 1C and 1D). Analysis of seed length showed that gs3, gs3 gn1a and gs3 gn1a gw2 mutants all produced longer seeds compared to wild type (see also Fig 1F). Among the four classes of mutants, seed width was slightly increased in gn1a and drastically increased in the triple mutant. Moreover, the gn1a gs3 gw2 mutant produced seeds with a significantly enhanced weight of 24% compared to wild type (Fig 1E). This analysis confirmed that, among the three genes, mutating GW2 in Kabre caused the most significant increase in seed width and weight (Fig 1D, 1E and 1G).