By the end of this century, the frequency of heat waves in the Sahelian region will increase, with temperatures exceeding the maximums observed over the past century (Battisti and Naylor, 2009; Russo et al., 2016). Global warming has already led to significant crop yield losses in Africa—up to 20% for pearl millet and 15% for sorghum (Sultan et al., 2019). This trend is likely to accelerate with a further 8% yield loss or more predicted in Africa by the 2050s (Roudier et al., 2011; Knox et al., 2012). The main staple food for communities in sub-Saharan Africa, pearl millet (Cenchrus americanus (L.) Morrone syn. Pennisetum glaucum (L.) R. Br.) is the top-ranked crop in terms of cultivated area in Nigeria (2,000,000 ha), Burkina Faso (1,183,792 ha), Mali (2,164,374 ha), and Senegal (1,023,065 ha) in 2020 (FAOStat, 2016) due to its high tolerance to extreme climate conditions. However, its production remains low, with an average grain yield of 900 kg/ha (Pucher et al., 2015). In the wake of drought periods in the 1970s and 1980s, West African breeding efforts have mostly been focused on developing short-cycle and non-photoperiodic varieties (Matlon, 1990; Niangado, 2001; Evenson and Gollin, 2003). Yet, West African smallholders rely on both early-and late-flowering types (De Rouw and Winkel, 1998; De Rouw, 2004). Despite the drought episodes, late-flowering landraces have been conserved or reintroduced from the 2000s with the trend towards increased rainfall (Lalou et al., 2019). In terms of performance, early-flowering landraces ensure minimum production in the event of a short raining season, while late-flowering landraces benefit from late rains but others specificities such as adaptation to different types of soil, different culinary uses and secondary uses, e.g., fodder, also explaining why farmers conserved both flowering regimes. It has been demonstrated that flowering time highly contributes to the landraces genomic vulnerability to future climatic conditions and thus constitute a major trait for pearl millet adaptation (Rhoné et al., 2020). The pearl millet flowering cycle ranges from very early (<40 days) to very late (>120 days, Haussmann et al., 2006; Upadhyaya et al., 2017). Variations in flowering time are associated with different climatic conditions of cropping and the pattern usually follows a latitudinal gradient (Haussmann et al., 2006; Pucher et al., 2015). While a landrace can be classified as early-or late-flowering, there is still a great deal of heterogeneity. For instance, within landrace full-sib families, differences between the earliest and the latest flowering accessions were reported to be 16 days at minimum and 39 days at maximum; a flowering variability that appears to be genetically controlled (Haussmann et al., 2007, 2012). Landrace variability has already been exploited by farmers to shorten flowering cycles, as a response to the 1970s and 1980s drought episodes, while conserving their varieties (Vigouroux et al., 2011; Dussert et al., 2015). Substantial effort has been made in the past decades to investigate genes associated with flowering cycle in pearl millet (Saïdou et al., 2009; Vigouroux et al., 2011; Lakis et al., 2012; Kumar et al., 2017; Diack et al., 2020). To date, only the role of PgPHYC gene has been really validated and some other QTLs were identified (Saïdou et al., 2009; Kumar et al., 2017). Which other genes are involved in the flowering variability of pearl millet landraces remains an open question.

Although the difference in flowering time is one of the primary characteristics between the two types, many differences in other important agronomic traits distinguish the two groups and lead to different uses by farmers as exposed earlier. Late-flowering landraces also have more productive tillers, high-tillering and small panicle, characteristics that can improve drought avoidance strategy (Van Oosterom et al., 2006; Yadav, 2008). Therefore, pearl millet landraces display both individual and populational variability, an asset that needs to be valorized to ensure productivity and stability with the upcoming increase in climate uncertainty in Africa.

We previously used the core collection from Senegal to identify genes involved in the agromorphological differentiation of the two flowering groups (Diack et al., 2020). This study identified a number of genetic markers associated with biomass, but did not identify genes clearly involved in flowering. Here, we adopt a different strategy, by sampling early-and late-flowering accessions that are grown in the same locations, in the groundnut growing area (13–15°N) where they can be cultivated in mixtures allowing gene flow between the two groups (Lalou et al., 2019). By doing so, we minimized the drift effect and maximize the probability of highlighting the regions involved in the differentiation between the two groups. With the objective to identify key genes underlying the phenotypic differentiation between the two flowering groups, we implemented two complementary approaches: (1) genome-wide selection scan to identify key variants showing the highest genetic differentiation between early-and late-flowering landraces, and (2) genome-wide association analysis (GWAS) to detect relevant variants segregating in both groups. This combination of approaches allowed us to identify genes that are strongly involved in flowering including PgPHYC and other promising candidates such as FSR12 and HAC1; but also genes related to yield which magnitude of the effects can vary according to the genetic background in which they evolve.

In Senegal, late-flowering landraces compared to early-flowering landraces are preferred for fodder as well as for roof and fence construction because of longer and more robust stems. This implied a genetic structuring and different selection pressure on these two types. Data reported here supported a clear genetic differentiation between EF and LF landraces. Meanwhile, a large fraction of genetic diversity (95%) is still segregating inside each flowering group. Genetic clustering was consistent with the flowering structure; other phenotypic findings were also significant. Short cycle landraces had longer and heavier panicles compared to long cycle landraces, which had longer stems and shorter panicles. This finding was consistent with those of previous studies (Marchais, 1982; Dancette, 1983; Zongo et al., 1988; Akanvou et al., 2012; Diack et al., 2020) and was in accordance with the farmer’s uses.

To identify key genes related to flowering time and relevant agronomic traits, two complementary approaches were used: (1) genome-wide selection scanning and (2) genome-wide association studies (GWAS). Conservative thresholds (PCAdapt: FDR of 0.001, BayeScan: prior odds of the model with selection to 10,000) were applied for both genome-scanning methods to pick up the strongest selection signature and identify key variants. GWAS models were used while taking the genetic structure into account. As in our case the genetic structure very closely matched the classification into EF and LF landraces, GWAS was expected to be more effective for detecting SNPs segregating within the flowering groups. For instance, SNPs detected by the genomic approaches were fixed in both groups with allelic frequencies >0.99 (Table 2), and thus could not be detected by the GWAS approaches, notably by the conservative CMLM method which seems to overcorrect the population structure. GWAS analysis was thus able to highlight other phenotypic traits and notably to detect a number of polymorphisms and genes associated with yield (Supplementary Tables 5, 6). Noteworthy, our combination of approaches appears to have been effective and relevant since we were able to detect genes that had been identified in previous studies with different samples and spatial scales [five genes identified by Diack et al. (2020) and Rhoné et al. (2020)], suggesting false discoveries were reduced and that our results could be mainstreamed to other pearl millet landraces.

In relation to flowering time, we highlighted fives relevant genes all found on chromosome 2: PhyC (PHYTOCRHOME C), FRS12 (FAR1-RELATED SEQUENCE 12), HEADING DATE 16 (Hd16)/EARLY FLOWERING 1 (EF1), HEADING DATE 3 (Hd3a) and OsHAC1/OsHAC703 genes. The PhyC gene plays an important role in flowering time induction via photoperiodic cues in several cereal species (Izawa et al., 2002; rice: Takano et al., 2005; Ishikawa et al., 2011; wheat: Chen et al., 2014). PhyC has already been found to be a major flowering time gene in pearl millet (Saïdou et al., 2009, 2014; Vigouroux et al., 2011; Diack et al., 2017). In our study, SNP chr2_11155563 with is 4,745 bp away from the polymorphisms found in those previous studies, was detected by genomic scan approaches and in association with stem length by LFMM and EMMA methods. This gene was also reported to be associated with spike length and stem diameter in an inbred line panel (Saïdou et al., 2009), and with other traits including stay green, panicle diameter, panicle harvest index and panicle length under both well-watered and drought stress conditions (Sehgal et al., 2015; Debieu et al., 2018). Sequencing the entire PhyC gene from West African inbred lines would be interesting to investigate haplotypic diversity and the effect of the different variants of flowering time.

This is the first time that the role of FRS12 in pearl millet flowering is highlighted. Loss-of-function of FRS12 in A. thaliana results in early-flowering plants with overly elongated hypocotyls (Ritter et al., 2017). In pearl millet, the FRS12 gene is located in a 530 kb region encompassing 12 genes detected in our study. This genomic region may represent an important quantitative trait locus (QTL) for flowering time. Although FRS12 is a good candidate, we cannot overlook the possibility that other genes in this specific region could be directly causative of the phenotype. The gene Pgl_GLEAN_10033790 is annotated as HEADING DATE 16 (Hd16)/EARLY FLOWERING 1 (EF1). The role of Hd16 on flowering has been studied in rice (Dai and Xue, 2010) and linked to adaptation to high latitude (Hori et al., 2013). Interestingly, plants with non-functional Hd16 have delayed flowering and heavy straw and panicle weight in paddy fields with low nitrogen input and no fertilizer application (Tanaka et al., 2019). [AA] pearl millet genotypes for the SNP chr2_4843975 showed a + 1.5 days delay in flowering and longer ([AA] 279 cm vs [CC] 240 cm) and thicker ([AA] 1.8 cm vs [CC] 1.6 cm) stems. Another strong candidate is the gene Pgl_GLEAN_10020525, which is annotated as OsHAC1/OsHAC703 (histone acetyl transferase). This gene was also detected as associated to the heading date in pearl millet by Rhoné et al. (2020). By aligning on the reference genome the SSR markers used in the QTL analysis from (Kumar et al., 2017), we were able to find that the HAC1 gene is part of QTL1:166 identified in that study as strongly involved in flowering. In Arabidopsis HAC1 mutants flower late due to increased FLOWERING LOCUS C (FLC) expression (Deng et al., 2007). In rice, OsHAC1 may be involved in the abscisic acid signaling pathway for responses to environmental stress during rice seedling growth, as well as to salt and drought stress (Liu et al., 2012; Fang et al., 2014). Kumar et al. (2017) study identified two additional QTLs on chromosome 3 linked to flowering, the QTL3/62 and the QTL3.100. The QTL3/62 encompasses the gene Pgl_GLEAN_10027181 annotated as CyclinT1;3 which was found to be associated with heading date in our study. The mean CMLM effect of the alternate allele of SNP chr3_295789863 is +1.53 days. The Arabidopsis thaliana homolog AT1G27630 may be involved in pathway linking circadian clock and cell cycles. The QTL3/100 encompasses the gene Pgl_GLEAN_10032681 that was detected in our PCAdapt analysis. This gene is annotated as an ethylene-responsive factor like protein (ERFL1) from the APETALA2 family.

The GWAS analysis detected SNPs linked to yield through association with panicle and grain characteristics. The annotation of homologous genes in Arabidopsis thaliana revealed a number of potentially interesting genes which magnitude of effect seems to be also a function of the genetic background, i.e., early-vs. late-flowering landraces. Fourteen of these genes are annotated as being related to responses to abiotic stimuli (salt stress, water deprivation, osmotic stress) or related to the immune response. A total of 13 genes are related to reproduction. One of them is Pgl_GLEAN_10001878 and it is annotated as LORELEI in Arabidospis thaliana (Tsukamoto et al., 2010). When looking at its effect, it seems that the presence of the alternate allele on SNP chr3_18486380, increases 1,000 seed weight (Supplementary Figure 6). A second gene was found to be associated with 1,000 seeds weight, Pgl_GLEAN_10007024. This gene annotated as a PGR5-like protein and it may be involved in photosynthesis (DalCorso et al., 2008) in Arabidopsis. A clear decrease of 1,000 seeds weight is noted when individuals carry the alternative allele on SNP chr7_55923381 (Supplementary Figure 6). We have detected four genes whose annotation was linked to root development. Three of them were detected by the GWAS approach through their association with panicle size. The gene Pgl_GLEAN_10017314 is annotated as a global transcriptor factor in Arabidopsis thaliana. The alternate allele of SNP chr4_48223622 increases panicle diameter at the homozygous state (Supplementary Figure 6). Another of these genes is Pgl_GLEAN_10003976. This gene is homolog to the ERULUS gene in Arabidopsis thaliana. ERULUS gene has been described as a core root hair regulator, involved in the establishment of a functional apical [Ca²⁺] gradient (Bai et al., 2014; Kwon et al., 2018). The establishment of this gradient seems to impact the pollen tube growth and fertilization (Schoenaers et al., 2017).

Two complementary genome-wide approaches, i.e., selection scanning and association analysis (GWAS) were used in this study with the main objective to identify key genes linked to flowering and agromorphologic traits in pearl millet, a major staple cereal in sub-Saharan Africa. To date, PhyC was the only clearly identified flowering-related gene. We were able to identify several other genes that appear to be as much as important, such as FSR12 and HAC1. HAC1 and two other genes appear to be part of QTLs identified in previous studies and deserve further investigation. At the same time, we were able to identify a large number of genes and variants that could contribute to the improvement of pearl millet yield, especially since their impact was demonstrated across flowering cycles.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.

AB, CB-S, ADF, NK, and YV designed the study. AB, CB, CB-S, PC, ADF, and BR performed the analyses. AD, MC, CM, and LZ generated genomic datasets. AB, CB-S, AC, ADF, ALF, KO, DM, MT, and YV managed the field studies and generated phenotypic datasets. ADF and CB-S wrote the paper with all authors’ contributions. All authors contributed to the article and approved the submitted version.

This project was supported by Agropolis Fondation under reference ID 1403–057 through the Investissements d’avenir programme (Labex Agro: ANR-10-LABX-0001-01) in the framework of I-SITE MUSE (ANR-16-IDEX-0006), the CERAO project (ANR-13-AGRO-002), while the research leading to these results received funding from the UK National Environment Research Council (NERC)/Department for International Development (DFID) Future Climate For Africa Program, under the AMMA-2050 project (grant numbers NE/M020002/1; NE/M019934/1). YV was also funded under the CRP on Dryland Cereals.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Tuning flowering time plays a key role for adaptation to future climate change. We show that PhyC, FRS12 and HAC1 are key genes in the pearl millet flowering cycle.