Major niche transitions in Pooideae correlate with variation in photoperiodic flowering and evolution of CCT domain genes

Long-day flowering evolved early in the grass subfamily Pooideae, with later, novel evolution of short-day flowering in Stipeae, probably associated with evolution of CCT domain gene expression.


Introduction
The ability of plants to coordinate flowering with favorable environmental conditions results in optimization of reproductive fitness through increased seed set and survival (Greenup et al., 2009).The exact timing of flowering is determined by several different external (e.g.photoperiod and temperature) and internal (e.g.age and hormone) signals that are integrated at the shoot apical meristem (SAM) throughout the lifetime of the plant.In the non-equatorial tropics, shortening photoperiods signal that the rainy season or monsoon is coming to an end, resulting in flowering of short-day grasses (Poaceae) such as rice (Oryza sativa) and maize (Zea mays) at the end of the greening period, prior to the extreme heat of summer (Naranjo et al., 2014;Mascheretti et al., 2015;Preston and Fjellheim, 2020).
In contrast, lengthening photoperiods during the impending warm season of temperate regions trigger flowering in longday plants such as the grasses wheat (Triticum aestivum) and barley (Hordeum vulgare), circumventing the negative effects of winter freezing (Nishida et al., 2013;Chen et al., 2014).Photoperiodicity in flowering is thus a good predictor of current plant distributions (Zhang et al., 2015;Preston and Fjellheim, 2020), but the evolutionary genetic basis of switches between long-and short-day responses is not well understood.
Similar to angiosperms as a whole (Hochuli and Feist-Burkhardt, 2013;Mannion et al., 2014), the grass family evolved when the terrestrial Earth was largely tropical (Burke et al., 2016;Gallaher et al., 2019;Schubert et al., 2019a), suggesting that the ancestor would either have flowered under short days or been daylength neutral (Preston and Fjellheim, 2020).Indeed, of the ~12 000 extant grass species (Soreng et al., 2015), the majority remain in the tropics, with only a couple of major subfamilies-Danthonioideae and Pooideae-dominating southern and northern temperate regions, respectively (Edwards and Smith, 2010;Visser et al., 2014;Schubert et al., 2020).Evidence suggests that the ability of an early Pooideae ancestor to respond to inductive photoperiods was contingent upon receiving a prolonged period of winter cold (vernalization) (McKeown et al., 2016), although data also suggest later modifications to this ancestral vernalization pathway (Woods et al., 2016).It is further hypothesized that the last common ancestor of Pooideae evolved from a daylength-neutral/shortto a long-day plant, the mechanisms underlying which are unknown (Preston and Fjellheim, 2020).
Comparative analyses across both long-and short-day angiosperms have revealed remarkable conservation in the photoperiod flowering pathway, suggesting that flowering in response to different daylengths evolved through fine-tuning of a shared ancestral pathway (Andrés and Coupland, 2012;Matsubara et al., 2014).Central in this pathway is the florigen FLOWERING LOCUS T (FT).FT and related proteins act as universal signals to integrate flowering pathways and promote reproduction.Crucial for perception of photoperiod are various light receptors, one of which is PHYTOCHROME C (PHYC).PHYC is a weak floral repressor in short days in rice (Takano et al., 2005), whereas it promotes flowering under long days in barley and Brachypodium distachyon (Nishida et al., 2013;Woods et al., 2014).Another gene family that has been implicated in fine-tuning flowering is the CCT [CO, CO-LIKE, and TIMING OF CAB EXPRESSION 1 (Robson et al., 2001)] domain gene family of transcription factors, with nine members in long-day barley (Pooideae) and 16 members in short-day rice (Oryzoideae) (Griffiths et al., 2003;Song et al., 2015).Examples of CCT domain-containing genes implicated in intraspecific variation in flowering responses are barley PHOTOPERIOD 1 (PPD1) and its ortholog PSEU-DORESPONSEREGULATOR 37 (PRR37) in rice, barley CO1 and CO2 and their ortholog HEADING DATE 1 (Hd1) in rice, CO9, and barley VERNALIZATION 2 (VRN2) and ortholog Grain number, plant height, and heading date 7 (Ghd7 or OsI) in rice (Komiya et al., 2008;Xue et al., 2008;Stracke et al., 2009;Takahashi et al., 2009;Lu et al., 2012;Koo et al., 2013;Wei et al., 2014;Zhang et al., 2015;McKeown et al., 2016;Zheng et al., 2016;Zhang et al., 2017;Shaw et al., 2020).
Like its CO ortholog in Arabidopsis thaliana (Brassicaceae), CO1 in barley and wheat is up-regulated in the afternoon by the PHYTOCHROME A and B (PHYA/B)-mediated circadian clock under both long-and short-day conditions (Campoli et al., 2012;Mulki and von Korff, 2016).In A. thaliana, photoperiod regulation through CO occurs at the protein level in the presence of light-induced stabilizing proteins, resulting in the up-regulation of FT to induce flowering only under long days (Yanovsky and Kay, 2002;Valverde et al., 2004;Hayama et al., 2017).Although it has not been confirmed that similar light-induced protein stabilization exists for CO1, genetic evidence from barley and wheat cultivars with nonfunctional PPD1 alleles has shown that this protein also promotes flowering under long-day conditions, concomitant with peak expression in the light (Campoli et al., 2012;Mulki and von Korff, 2016;Shaw et al., 2020).On the other hand, in the presence of functional PPD1 and VRN2 alleles, at least wheat CO1 is converted to a mild floral repressor under long days to prevent precocious pre-winter flowering, probably as a result of protein-protein interactions between PPD1, CO1, CO2, and possibly VRN2 (Shaw et al., 2020).
In rice, the CO1 ortholog Hd1 is also assumed to be regulated by light-and dark-dependent proteins, and also forms an Hd1/CO1-PRR37/PPD1-Ghd7/VRN2 protein complex under long days to repress flowering via repression of Early heading date 1 (Ehd1) and hence FT/Hd3a (Griffiths et al., 2003;Xue et al., 2008;Zhang et al., 2015;Fujino et al., 2019).Together with the fact that rice Hd1 and wheat CO2 promote and repress flowering under short days, respectively, these data support a role for changing CO-like protein interactions in transitions between short-day, day-neutral, and long-day flowering photoperiodism (Kitagawa et al., 2012;Song et al., 2015;Mulki and von Korff, 2016).
In addition to positively and negatively regulating FT (also named VRN3; Yan et al., 2006) in barley and wheat, CO1 and CO2 are involved in a regulatory feedback loop with VRN2 (Mulki and von Korff, 2016).VRN2 is a monocot-specific repressor of flowering that is negatively regulated by vernalization in the large 'core' Pooideae clade, comprising species such as wheat, ryegrasses (Lolium sp.), and oats (Avena sp.).However, VRN2 is not down-regulated in response to cold in other 'non-core' Pooideae clades, including vernalization-responsive B. distachyon (Woods et al., 2016).Under long days of the early autumn, winter barley VRN2 is strongly up-regulated in leaves by the action of CO1, CO2, and PPD1 (Distelfeld et al., 2009).Overexpression of CO1 and CO2 in spring barley results in up-regulation of VRN2, leading to delayed flowering in both long and short days (Mulki and von Korff, 2016).In turn, VRN2 negatively regulates CO1/2 and PPD1, thereby dampening its own expression (Mulki and von Korff, 2016).As winter approaches, low-temperature-induced expression of the flowering promoter VERNALIZATION 1 (VRN1; Oliver et al., 2009) results in the gradual repression of VRN2, and a concomitant increase in FT, partly mediated by CO1/2 and PPD1 (Song et al., 2015;Mulki and von Korff, 2016).
CO9 is a grass-specific paralog of VRN2/Ghd7 (Woods et al., 2016), and overexpression in rice suggests that it acts as a floral repressor (Kikuchi et al., 2012).Based on expression and functional analyses, this occurs under both short and long days, where transcript abundance peaks early after dawn (Kikuchi et al., 2012).Since barley VRN2 is expressed at its highest level towards the end of the light period (Trevaskis et al., 2006), and rice Ghd7 is expressed at high levels throughout the light period (Xue et al., 2008), these data suggest evolution of VRN2/ CO9 genes in terms of both photoperiodic and circadian regulation following both duplication and speciation events.
Here, we reconstruct the evolution of photoperiodic flowering in Pooideae to test the hypothesis that flowering in response to long days evolved early in the subfamily and hence facilitated a range shift into northern temperate regions.We show that ancestral Pooideae was probably long day responsive, and that a secondary transition back to tropical climates was coincident with a shift back to short-day flowering.To determine if this derived short-day responsiveness can be explained by changes in the (co-)expression of CCT domain genes, we assess relative transcript levels for long-and short-day lightdark cycles across time in exemplar long-and short-day flowering species.

Plant growth and experimental conditions
Forty-seven Pooideae species (13 core and 34 non-core) and the outgroup Ehrharta calycina from subfamily Oryzoideae (Supplementary Table S1) were selected to represent phylogenetic and geographic diversity across Pooideae.The plants were grown under different treatment conditions to score for long-day, short-day, or day-neutral flowering.Fifteen plants were grown per treatment.All seeds were stratified in moist soil (Gartnerjord, Tjerbo Torvfabrikk AS, Rakkestad, Norway) in complete darkness for 6 d, first under 4 °C for 5 d, followed by 1 d at room temperature.Seeds were then transferred to an open greenhouse in long days (16 h light:8 h dark) at 17 °C and grown for 4 weeks before the plants were randomized and assigned to one of four treatments: 17 °C short days (8 h light:16 h dark), 17 °C long days, 4 °C short days, or 4 °C long days for 12 weeks.We included vernalization in two of the treatments to see the effect of photoperiod even in vernalization-responsive species.All short-and long-day-grown plants were then maintained in short or long days, respectively, at 17 °C until flowering (calculated as days to heading) or termination of the experiment at 200 d.The experiment was repeated following the same conditions, except for reduction of the vernalization period to 8 weeks, switching the upper temperature to 20 °C, and termination of the experiment at 120 d.Light intensity under vernalization was 50 ± 5 µmol m −2 s −1 , and for all other conditions it was 150 ± 10 µmol m −2 s −1 .Light used in the experiment was produced by HQI lightning systems (LU400/XO/T/40 Philips Osram, General Electric, Hungary) giving a red/far-red ratio of 1.8 ± 0.2.Plants were watered and fertilized (water containing 4% Yara Kristalon Indigo and 3% YaraTera Calcium Nitrate, Yara Norway AS), adjusted to an electron conductivity of 1.5 as needed, and moved to a new position twice a week within the chamber.
To investigate more closely molecular responses to different photoperiods, two non-core Pooideae species in tribe Stipeae, long-day Oloptum miliaceum (USDA GRIN PI207772) and short-day Nassella pubiflora (USDA GRIN PI478575), and the long-day flowering Meliceae species Melica ciliata (Millennium Seed Bank 31675) were chosen for a follow up-experiment based on results from the first growth experiment.Ehrharta calycina (USDA GRIN PI284803 and PI578674) from the subfamily Oryzoideae was included as an outgroup.None of these species had an absolute vernalization requirement.Growth experiments were performed in two Conviron CMP6010 (Conviron, Winnipeg, Canada) growth chambers.Approximately 160 seeds of each of M. ciliata, N. pubiflora, and O. miliaceum, and 88 seeds of E. calycina were sown on moist filter paper and stratified under darkness for 4 d at 4 °C followed by 1 d at room temperature.Seeds were then planted in Metro-Mix 380, grown under long days at 20 °C for 4 weeks, and randomly assigned either to a long-day 20 °C or short-day 20 °C treatment until flowering, death, or termination of the experiment.For each plant, days to heading, number of leaves on the main stem at flowering, and tillers at flowering were recorded.The top fully expanded leaf of at least three plants without repeated measures were sampled at 2, 16, and 30 d after the initial 4 weeks growth under long days at 2, 8, 14, and 20 h post-dawn (ZT).

DNA extraction and sequencing
Genomic DNA was extracted from leaf material using the DNeasy Plant MiniKit (Qiagen, Valencia, CA, USA), following the manufacturer's protocol.We obtained sequences for three DNA plastid regions matK, ndhF, and rbcL, using custom Pooideae-specific primers (Schubert et al., 2019a; Supplementary Table S2).PCR was performed on a Tetrad 2 Thermal Cycler (Bio-Rad, Hercules, CA, USA) and a Mastercycler ep Gradient Thermal Cycler (Eppendorf, Hamburg, Germany) using JumpStart REDTaq ReadyMix (Sigma-Aldrich, St. Louis, MO, USA) and standard conditions with 58 °C annealing and 2 min extension.PCR products were Sanger-sequenced in both directions using the same primers as for PCR.Chromatograms and sequences were inspected in BioEdit (Hall, 1999), and automatic alignments generated with manual adjustments (Supplementary Datasets S1-S3).

Ancestral state reconstruction of flowering responses
Phylogenetic trees were generated for the concatenated matK, ndhF, and rbcL chloroplast dataset in MrBayes on XSEDE (Miller et al., 2010) implemented through the CIPRES Science Gateway v.3.3.The dataset was partitioned by gene, rooted with sequences from maize (Zea mays ssp.mays), and run twice for 10 million generations sampling every 1000 generations, with four chains, 25% burn-in, and other default parameters.The consensus tree was visualized in FigTree v1.4.3 (http://tree.bio.ed.ac.uk/software/figtree/) and edited in Adobe Illustrator CS6.To account for uncertainty in topology prior to ancestral state reconstruction, 200 rooted trees with branch lengths were collated from the two independent runs as input for BayesTraits v2 (Pagel et al., 2004;Pagel and Meade, 2006).BayesTraits was run using the Multistate function, and a one-rate/symmetrical model was chosen based on results of steppingstone estimation comparing symmetrical and asymmetrical state transition models.Markov chain Monte Carlo (MCMC) analyses were run with 10 million generations, sampling every 1000th generation, with a burn-in of 25%.Trait states for all internal nodes in the Bayesian consensus tree were inferred by calculating the means of posterior probability distributions for each node.

Scanning electron microscopy
To document if differences in time of transition reflect the flowering phenotype, we chose a representative subset of our focal species to investigate this at the level of SAM development under different photoperiods.We documented the developmental stage of O. miliaceum, N. pubiflora, and E. calycina SAMs across time points and treatments by dissecting meristems and subjecting them to SEM.At 2, 16, 27, and 41 d after onset of treatment, three SAMs from each species were fixed in formalin acetic acid (FAA) (50% ethanol, 5% glacial acetic acid, 10% of 37% formaldehyde) solution for 8-12 h.Following this, meristems were progressively transferred in five steps from 50% to 100% ethanol before critical point drying.Meristems were mounted on stubs, sputter coated with argon, and photographed using a JEOL 6060 SEM with an accelerating voltage of 25 kV.

RNA extraction, cDNA synthesis, and quantitative PCR
Leaves of M. ciliata, O. miliaceum, N. pubiflora, and E. calycina were flashfrozen in liquid nitrogen, stored at -80 °C, and later macerated for RNA extraction using TriReagent (Ambion, Thermo Fisher Scientific, Waltham, MA, USA) followed by removal of DNA by DNase treatment with the TURBO DNA-free kit (Ambion).cDNA was then synthesized from 500 ng of RNA using the iScript cDNA synthesis kit (Bio-Rad, Hercules, CA, USA).All procedures followed the manufacturer's instructions.
A CO9 ortholog from E. calycina and VRN2 orthologs from N. pubiflora and O. miliaceum were amplified in a standard PCR with cDNA pooled across time points and treatments for each species using previously published (Woods et al., 2016), as well as newly designed, primers (see Supplementary Table S2).Amplicons were ligated into pGEM-T (Promega, Madison, WI, USA), plasmids used to transform competent DH5α Escherichia.coli cells, and ~10 clones were sequenced per amplicon by the Advanced Genomes Technology Core at The University of Vermont.VRN2 from M. ciliata has previously been published (Woods et al., 2016).To identify orthologs of CO9 from O. miliaceum, N. pubiflora, and M. ciliata, as well as orthologs of PPD1, CO1, and PHYC from all Pooideae species, we generated transcriptomes for each species from leaves sampled in both conditions throughout a 24 h cycle.Briefly, leaves were flash-frozen in liquid nitrogen, and total RNA was extracted from homogenized tissue using the RNeasy Plant Mini Kit (Qiagen) including purification using the Invitrogen TURBO DNAfree kit (Thermo Fisher Scientific).Sequencing libraries with an insert size of 350 bp were constructed with the TruSeq Stranded mRNA Library Prep kit (Illumina, San Diego, CA, USA).Library preparation and paired-end sequencing was carried out by the Norwegian Sequencing Centre (NSC) at the University of Oslo on an Illumina HiSeq 4000 System (Illumina) with 150 bp reads.Read trimming and quality assessment of the transcriptomes followed Schubert et al. (2019b).The target sequences were identified through a BLAST search against the transcriptomes from the respective species using verified sequences from H. vulgare as queries (Supplementary Datasets S4-S7).
For each target gene and focal species, we designed primers for quantitative reverse transcription-PCR (RT-qPCR) (Supplementary Table S2).Primers for VRN3 and the housekeeping genes UBIQUITIN 5 (UBQ5) and ELONGATION FACTOR 1α (EF1α) were either previously published (Ream et al., 2014) or designed based on conserved regions in alignments of Lolium perenne, wheat, and Oryza brachyantha or rice, whereas VRN2 primers were constructed based on previously published alignments (McKeown et al., 2016).All new primers were designed using Primer3 (Rozen and Skaletsky, 2000), and the amplification efficiency of each primer pair was determined using a dilution series as previously described (Scoville et al., 2011).To quantify relative gene expression, target gene critical threshold c(T) values were normalized against the geometric mean of the two housekeeping genes after correction for primer efficiency with three technical and at least three biological replicates.

Western blot
An alignment was made of translated transcript sequences of CO9 and VRN2 from N. pubiflora, O. miliaceum, and M. ciliata as well as a selection of other grass species (Supplementary Dataset S8).Polyclonal antibodies were constructed for N. pubiflora and O. miliaceum (antigenic peptide sequence RRGMRCGVADLNRGC) and M. ciliata (a mix of the antigenic peptide sequences AGRRCGVAADLNLRC and VDQQEPAVIGGGGAC) to avoid cross-reactivity with VRN2.Leaf tissue was sampled from three biological replicates of each species subjected to long or short days at ZT2, 8, 14, and 20 one week after start of treatment, as previously described.Approximately 100 mg of tissue was ground in liquid nitrogen using a mortar and pestle, before adding 200 µl of DTE extraction buffer [3 mM DTT, 20 mM sucrose, 3 mM Na 2 CO 3 , 0.5% SDS, 1 mM EDTA, and 1:100 v/v of protease inhibitor cocktail (Sigma)].Each sample was mixed briefly by vortexing, sonicated for 2.5 min (5 s on, 5 s off for a total of 5 min), and centrifuged at 12 000 g for 20 min at 4 °C.The supernatant was then transferred to a fresh tube and centrifuged for another 15 min.A 50 µl aliquot of the supernatant was precipitated using 500 µl of 10% trichloroacetic acid (TCA), and centrifuged at maximum speed (20 000 g) for 10 min at 4 °C.The liquid was removed, and the pellet left to air-dry before being dissolved in 0.1% NaOH.The concentration of protein extract was measured using a Qubit protein assay kit after adding 1:1 volume of 2× Laemmli sample buffer containing 5% β-mercaptoethanol.Three technical replicates of protein extract were incubated at 75 °C for 10 min, put briefly on ice, and centrifuged at 12 000 g for 1 min at 4 °C.
A 25 µg aliquot of protein was applied to a 12% Mini-PROTEAN® TGX Stain-Free™ Precast Gel, using 3 µl of Precision Plus Protein Unstained Standard as a marker.The gel was run at 200 V for 40-45 min in 1× Tris/Glycine/SDS buffer (Bio-Rad), UV-activated for 1 min using GelDoc Stain Free gel application, and the proteins blotted onto a 0.2 µm polyvinylidene difluoride (PVDF) membrane using the Trans-Blot Turbo Mini 0.2 µm PVDF Transfer Packs (Bio-Rad) and Trans-Blot® Turbo™ Transfer System (Bio-Rad).The Turbo program was set at 25 V and 2.5 mA for 3 min, and the membrane was analyzed using GelDoc Stain Free Blot application for loading control and normalization as per the manufacturer's instructions.After being left to air-dry, the membrane was activated for 3 min with methanol and blocked in 2% dry milk solution in 1× TBS-T (500 mM NaCl, 20 mM Tris-HCl pH 7.5) for 1 h at room temperature.After washing twice for 10 min in TBS-T, the membrane was incubated with primary antibody diluted in blocking solution at 4 °C overnight.Dilutions were 0.25 µg ml -1 for N. pubiflora and 0.5 µg ml -1 for O. miliaceum.Following this, membranes were washed for 6 × 10 min in TBS-T followed by incubation for 1 h at room temperature with a 1:1000 dilution of mouse anti-rabbit horseradish peroxidase (HRP)-conjugated secondary antibody (SC-2357 Santa Cruz Biotechnology, Dallas, TX, USA).Subsequently, the membrane was washed 6 × 10 min in TBS-T and the signal was developed using Clarity™ Western ECL Substrate (Bio-Rad).After visualizing the signal using the GelDoc Chemi application, quantitation and analysis were performed using the Image Lab 6 software.

Statistical analyses
To capture both qualitative and quantitative variation in flowering behavior across species, we calculated both the proportion of individuals flowering per treatment and absolute dates to heading per treatment.In cases where flowering consistently occurred in the absence of vernalization, we used non-vernalized long-and short-day-treated plants to calculate the photoperiod response.However, when plants had an absolute requirement for vernalization to flower, we used vernalized long-and short-day-treated plants to calculate the photoperiod response.We classified species as long day responsive if the proportion of individuals flowering was significantly more (P<0.05as determined by a χ 2 test), and/or days to heading was significantly less (P<0.05determined by a two-tailed t-test) in long as compared with short days, and vice versa for short-dayresponsive species.
For the relative gene expression data, two-way ANOVAs were performed with expression of PHYC, PPD1, CO1, VRN2, and CO9 as dependent variables, and treatment and ZT time as independent variables.We removed the effect of sampling day (samples were taken at days 2, 16, and 30 after onset of treatment) by centering and standardizing expression data for all days using the 'scale' and 'center' functions in R.This was repeated for all genes and species, except for VRN3 that is expected to increase in expression only after receiving several upstream inductive signals; in this case, expression was analyzed over the three sampling days separately.Analyses were done using both raw and transformed data, and analyses where the residuals best fitted a normal distribution were chosen for further interpretation.To investigate the effect of photoperiod on expression at specific time points, we performed post-hoc contrasts for all species, genes, and time points.All ANOVAs and post-hoc tests were carried out in R (RCoreTeam, 2016) using the stats and emmeans (Lenth, 2021) packages.
Each western blot was run with one complete set of samples from both treatments from one species, with three technical replicates.Three biological replicates were run per species.One of the replicates of M. ciliata produced smeared bands and we were unable to quantify protein abundance.As values cannot be compared directly across different blots, we removed the effect of blotting gels by centering and standardizing the protein expression data per blot using the 'scale' and 'center' functions in R, before averaging over technical replicates, and the biological replicates for each time points in each treatment.Graphs of mRNA and protein abundance were plotted in R (RCoreTeam, 2016), using packages ggplot2 (Wickham, 2016), tidyverse (Wickham et al., 2019), ggalt (Rudis et al., 2017), and patchwork (Pedersen, 2019).We visually inspected the resulting graphs to find the diurnal expression pattern.
Nine species (Melica nutans, Festuca pratensis, Poa alpina, Dactylis glomerata, Anthoxanthum odoratum, Lolium perenne, Piptatherum aequiglume, Hordeum bulbosum, and Hordeum vulgare) flowered in adequate numbers only after vernalization.Of these, only M. nutans flowered in response to both photoperiods and was scored as long day responsive because flowering was faster in long compared with short days (t-test, P<0.05).All other Pooideae species were evaluated as long day responsive as significantly more individuals flowered in long than in short days (χ 2 test, P<0.05).As predicted, the outgroup species E. calycina (Oryzoideae) was classified as short day responsive as it flowered significantly more in short versus long days without vernalization (P<0.05).
To reconstruct the ancestral history of Pooideae photoperiodic flowering, we added several GenBank accessions to our new chloroplast dataset, resulting in alignment lengths of 1582 bp for matK, 1348 bp for ndhF, and 1104 bp for rbcL.Bayesian ancestral state reconstruction based on this concatenated dataset and flowering behaviors supported an early origin of long-day-induced flowering at or around the base of Pooideae (Fig. 2).In addition, at least four transitions to day-neutral flowering were inferred to occur in as many tribes across the tree, and one origin of short-day flowering was inferred near the base of Nassella (tribe Stipeae).We here ignore previous reports on day-neutral flowering in artificially selected crop cultivars (Dubcovsky et al., 2006;Beales et al., 2007;Faure et al., 2012;Campoli et al., 2013;Nishida et al., 2013;Turner et al., 2013;Pankin et al., 2014), as we focus on reconstructing the natural evolution of photoperiodic flowering.No transitions from short-to long-day photoperiodic flowering were inferred in Pooideae.The position of long-day Achnella caduca within the short-day Nassella tribe should be qualified by it being a hybrid between Nassella viridula and Achnatherum hymenoides.These data support the hypothesis that loss of long-day flowering (i.e.day neutrality) is easier than to gain than short-day flowering, or that there has been stronger selection pressure for the former.To complement our ancestral reconstruction with gene expression analyses, we conducted a second flowering time experiment under different photoperiods in exemplar species: outgroup E. calycina, long-day Pooideae M. ciliata and O. miliaceum, and short-day Pooideae N. pubiflora.Unexpectedly, E. calycina plants failed to flower under either long or short days in our follow-up experiment.The lack of adult vegetative or inflorescence meristems at day 27 in both photoperiods suggests that these plants failed to become competent to flower (Supplementary Fig. S1).This result was consistent with no detectable FT/VRN3 expression.
For M. ciliata and O. miliaceum, ANOVA verified the prediction that FT/VRN3 expression would be higher in long as compared with short days (P<0.001 and P<0.001, respectively, Fig. 3), consistent with clear spikelet meristems being visible by day 41 under long but not short days in O. miliaceum (Supplementary Fig. S1; data for M. ciliata not collected).In contrast, and in line with the observation of well-developed inflores-cences at day 41 in short but not long days (Supplementary Fig. S1), N. pubiflora showed significantly higher FT/VRN3 in short days (P<0.001) (Fig. 3).

Pooideae PHYC and PPD1 expression is generally conserved
After phylogenetically confirming orthology with other single-copy PHYC-and PPD1-like grass genes, we determined transcript levels for our focal Pooideae taxa, first to determine any differences between naturally occurring long-and shortday Pooideae, and second to provide context for expression of other CCT genes whose protein products potentially interact with PPD1.For PHYC, ANOVA showed no significant effect of photoperiod on expression for long-day O. miliaceum and M. ciliata or short-day N. pubiflora (Fig. 4).In contrast, photoperiod had a significant effect on expression levels of PPD1 for both O. miliaceum and M. ciliata (P<0.001 for both, Fig. 4).Post-hoc tests showed significantly higher expression in long as compared with short days at ZT2, ZT8, and ZT14 in O.  miliaceum (P<0.005,P<0.001, and P<0.05, respectively) and ZT8 and ZT14 in M. ciliata (P<0.005 and P<0.001, respectively).ANOVA showed no significant effect of photoperiod on PPD1 expression in N. pubiflora (Fig. 4); however, the posthoc test showed that expression was higher in long days at ZT14 (P<0.05).For all species, expression peaked in the dark in both photoperiods.

Evolution of CO1 and CO9 expression is consistent with derived short-day flowering in Stipeae
Previous authors have suggested that CO1 and CO2 were derived from a segmental duplication event at the base of grasses (Higgins et al., 2010).Since both Pooideae copies have been implicated as flowering promoters in the absence of a functional PPD1, or flowering repressors in the presence of PPD1, and CO1 is expressed more highly than CO2 at least in wheat, we chose CO1 for further analysis (Shaw et al., 2020).No effect of photoperiod on CO1 expression was identified in longday O. miliaceum and M. ciliata.However, ANOVA showed a significantly higher expression of CO1 in long versus short days in N. pubiflora (P<0.001), and post-hoc tests identified significant differences identified at ZT8,14,and 20 (P=0.05,P<0.05,and P<0.001,respectively).For all species, expression was at its lowest at ZT2 for both photoperiods and increased throughout the day.
It was previously reported that barley CO9 is more highly expressed under short versus long days, and peaks in expression during the light (Kikuchi et al., 2012).No data are currently available for the model species rice or B. distachyon.To determine if photoperiod regulation of CO9 is conserved across the BOP clade, and if changes in regulation are associated with the secondary shift to short-day Pooideae flowering, CO9 expression was profiled in all focal species (Fig. 4).ANOVA and post-hoc tests showed that CO9 expression in the short-day outgroup E. calycina and long-day M. ciliata was similar in abundance across photoperiods, with the peak of expression coinciding with the light period under both conditions (Fig. 4).This pattern for M. ciliata CO9 appeared to be confirmed at the protein level based on results of the western blot (Fig. 5; but see the Discussion for potential caveats) (E.calycina not tested).ANOVA did not identify a significant effect of photoperiods on expression for O. miliaceum and N. pubiflora CO9.However, in both species, variation in periodicity resulted in a peak of expression in the light for long days and dark for short days for mRNA (Fig. 4), with a significant difference of expression between long and short days at ZT2 (P=0.01,Fig. 4).Furthermore, whereas the peak of mRNA expression was in the light for long days and the dark for short days (Fig. 4), CO9 protein peaked in abundance during the light of both photoperiods and species (Fig. 5), potentially suggesting transcriptional instability or protein degradation in the dark.

VRN2 expression has evolved in both long-and shortday Stipeae
VRN2/Ghd7 is positively regulated by long days in rice and barley (Trevaskis et al., 2006;Xue et al., 2008).To determine if this long-day response is generally conserved, or has evolved in short-day Pooideae taxa, we assessed VRN2 expression in M. ciliata, O. miliaceum, and N. pubiflora (Fig. 4).Unfortunately, we were unable to amplify the rice Ghd7 ortholog from E. calycina, suggesting either low expression in leaf tissues under our experimental conditions or high levels of sequence divergence relative to rice.ANOVA showed a significant effect of photoperiod on expression of VRN2/Ghd7 in M. ciliata (P<0.0001) and expression was higher in long as compared with short days at all time points (P<0.01,P<0.01, P<0.001, and P<0.05, respectively, Fig. 4), peaking in the light in both photoperiods (Fig. 4).Contrary to prediction, O. miliaceum showed no significant difference in VRN2/Ghd7 transcript levels between photoperiods (Fig. 4), and expression during the light period in both long and short days.Finally, despite its relatively close relationship to O. miliaceum, and its short-day responsiveness, N. pubiflora VRN2/Ghd7 was expressed at a significantly higher level under long compared with short days (P<0.0001) at all time points (P<0.05,P<0.01, P<0.01 and P<0.001, respectively, Fig. 4).Interestingly, expression peaked during the dark for both photoperiods (Fig. 4).Beyond expression patterns within species averaged across days, interspecific comparisons of VRN2 demonstrated relatively weak expression for M. ciliata and O. miliaceum under both photoperiods, with much stronger expression observed for N. pubiflora VRN2 under long days by treatment day 16 (Supplementary Fig. S2).Assuming that VRN2 is a conserved repressor of flowering, these data are consistent with stronger long-day suppression of flowering in a short-day versus long-day species.

Variation in photoperiodic flowering correlates with major niche transitions in the BOP clade
Pooideae is the most dominant grass subfamily of the northern temperate, continental, and Arctic regions (Hartley, 1973).We hypothesized that one of the keys to this success was the use of lengthening days in the spring and summer as a cue to flower rapidly at the appropriate time within limited growing seasons.In line with predictions of this hypothesis, ancestral state reconstruction of photoperiodic flowering responses supports both the dominance of long-day-induced flowering in Pooideae, and its evolution relatively early in the subfamily, after it diverged from Bambusoideae (bamboos).An obvious caveat to our study is the lack of exhaustive sampling across Pooideae, at both the inter-and intraspecific level.However, we believe our attempt to capture accessions spanning geographic variation within the subfamily makes our findings robust to any sampling deficits.
An early origin of vernalization-mediated flowering was previously reconstructed for Pooideae (McKeown et al., 2016), which together with our data (Fig. 2) suggests that the dual photoperiod-temperature induction of flowering long known from winter Pooideae cereals (Heide, 1994) was a key step toward colonizing newly expanding temperate climates.Recent dating of the grasses (Burke et al., 2016;Gallaher et al., 2019;Schubert et al., 2019a) places the origin of Pooideae at the transition between the Cretaceous and Paleocene, 60-70 million years ago (Mya), at a time when mean temperatures were relatively high (Zachos et al., 2001) and seasonality in temperature relatively low (Archibald et al., 2013).Biogeographic studies suggest a Eurasian origin for Pooideae (Bouchenak-Khelladi et al., 2010), and a recent reconstruction of the ancestral niche of Pooideae suggests that its ancestor experienced frost (Schubert et al., 2019a), consistent with a cold micro-habitat origin, possibly in montane Eurasia.Together, these results imply that Pooideae was already to some degree adapted to the cool, seasonal northern climates that developed after the Eocene-Oligocene (E-O) boundary 34 Mya (Strömberg, 2011), and that the early origins of vernalization responsiveness and long-day flowering played crucial roles in the shift of Pooideae from tropical to temperate regions.
Equally as interesting was the evolution of short-day-responsive species within the Stipeae tribe that correlates with a shift back to the tropics (Fig. 2).Specifically, Nassella pubiflora, N. neesiana, and N. brachyphylla are all native to the South American Andes, although N. neesiana has been introduced to other parts of the world (www.gbif.org).On the other hand, N. cernua, N. lepida, and N. pulchra are endemic to California.Faster flowering under the short-versus long-day conditions of our experiment seems counter-intuitive to the fact that N. pulchra naturally flowers in June and July.However, we previously found that this species also has a strong vernalization response (McKeown et al., 2016).We thus suggest that vernalization responsiveness has adapted N. cernua, N. lepida, and N. pulchra to the northern warm temperate growth cycle by blocking flowering in the shortening days of warm autumns.Whereas long days alone would delay flowering, the coincidence of lengthening days after a winter cold spell allows some physiological release, resulting in eventual flowering in the summer.
In addition to flowering, many traits, such as abscission, dormancy, cold acclimation, senescence, growth, and metabolism, are under the control of photoperiod (Salisbury, 1981).Molecular crosstalk between the networks controlling these traits has the potential to constrain their evolution through antagonistic or adaptive pleiotropy.In our experiment, most species flowered under both long and short days, although it was usually faster or biased in one condition (Fig. 1).This is consistent with data found for other grass species (Preston and Fjellheim, 2020), and suggests that Pooideae have the molecular machinery to flower under both photoperiods.Given this interpretation, other internal or external constraints must be invoked to account for the strong partitioning in geographic space between the Pooideae and other grass subfamilies (Visser et al., 2014).One possible explanation is that competition prohibits the expansion of species with maladapted flowering phenotypes into areas already occupied by species with more favorable flowering responses (Sherry et al., 2007).If long-dayresponsive flowering evolved early in Pooideae species inhabiting a cold Eurasian montane micro-niche, it could have given the Pooideae a competitive advantage and been an important facilitator for the group's rapid expansion into the emerging and expanding temperate biomes that followed the E-O split.

Conservation of flowering time gene expression across Pooideae
The ability of many grasses to flower under both long and short days, but still be faster flowering under certain photoperiods, underscores the complexity of the flowering time gene network (Shaw et al., 2020;this study).In this regard, understanding what aspects of flowering control are conserved provides important context to determine how the pathways might have changed.In the case of the evolutionary transition to short-day flowering in Stipeae, we noted that expression of PHYC and PPD1 in our exemplar short-day flowering species N. pubiflora broadly matched the pattern found for long-day species (Fig. 4).
PHYC conveys photoperiod sensitivity to plants, with wildtype alleles promoting flowering in long-day barley, but repressing flowering in short-day rice (Takano et al., 2005;Nishida et al., 2013).These opposing roles are mediated through epistatic interactions with other flowering time genes, as exemplified by the fact that expression of barley HvPHYC actually delays flowering in a rice phyA/phyC background (Nishida et al., 2013).In our focal species, M. ciliata, O. miliaceum, and N. pubiflora, PHYC mRNA levels were similar under both long and short days, and, as in the case of barley, generally peaked after dusk (Nishida et al., 2013).
PPD1 is a downstream target of PHYC whose exact function is again affected by epistatic interactions with other flowering time genes (Zhang et al., 2019;Shaw et al., 2020).In wheat and barley, PPD1 is expressed under both long and short days during the light period, but only accelerates flowering under long days or in response to a flash of light during long nights (i.e.short days) (Nishida et al., 2013;Pearce et al., 2017).Although the PPD1 ortholog PRR37 delays flowering under long days in its native rice (Zhang et al., 2019), like PHYC its expression in long-day plants accelerates flowering, suggesting conservation of protein function (but see the effect of mutant alleles on day-length sensitivity) (Koo et al., 2013;Shaw et al., 2020).PPD1 transcript abundance in O. miliaceum, N. pubiflora, and M. ciliata peaked in the light in both treatments (Fig. 4), as in wheat and barley (Shaw et al., 2020;Gauley and Boden, 2021), and was higher under long versus short days.Given the roles of PHYC and PPD1 in photoperiodicity, it is not surprising that their expression patterns are conserved across long-and short-day grasses.On the other hand, it would be interesting to assess perturbations in their expression patterns that might explain loss of long-day photoperiodism in non-core Pooideae, such as high latitude Nardus stricta and the widespread Eurasian-North American Glyceria striata.

Evolution of VRN3 and CCT family gene expression in both short-and long-day flowering Pooideae
As expected based on similar work across a range of angiosperms (Andrés and Coupland, 2012), FT/VRN3 expression tracked the flowering behavior of our focal Pooideae grasses (Fig. 3).Among others, CCT domain-containing genes are known direct regulators of FT/VRN3 and often function in a photoperiod-dependent manner (Shen et al., 2020).These attributes make them good candidates to explain evolutionary transitions between long-day, short-day, and day-neutral flowering in Pooideae through the differential regulation of FT/ VRN3.
In the absence of a functional VRN2/Ghd7 allele, or when VRN2/Ghd7 transcripts are low, both CO1 and PPD1 have been shown to promote the expression of FT/VRN3 in grasses, leading to the acceleration of flowering (Campoli et al., 2012;Yang et al., 2014;Mulki and von Korff, 2016;Zhang et al., 2017).For long-day M. ciliata and O. miliaceum, CO1 was expressed in a similar manner to rice and sorghum in that its expression level was no different in long versus short days (Fig. 4).However, PPD1 was more highly expressed in long as compared with short days, whereas VRN2 expression was low under both photoperiods (Fig. 4).Assuming conservation of the model from wheat, barley, and rice, the lack of strong VRN2 transcription suggests that CO1-PPD1 will work as part of a floral activator complex under long-day conditions, consistent with long-day-regulated flowering in both M. ciliata and O. miliaceum.
In contrast to M. ciliata and O. miliaceum, CO1 and VRN2 transcripts were both high specifically under long days in the derived short-day flowering species N. pubiflora (Fig. 4; Supplementary Fig. S2).In wheat, barley, and rice, high levels of functional VRN2/Ghd7 form a repressor complex with Hd1/ CO1 and PRR37/PPD1 (Yang et al., 2014;Mulki and von Korff, 2016;Fujino et al., 2019).Thus, again assuming functional conservation of the CO1-PPD1-VRN2 complex, these data provide at least a partial mechanism for the evolution of short-day flowering in Stipeae, whereby the VRN2-PPD1-CO1 repressor complex is strengthened specifically under long days.
In addition to the CCT domain-containing genes CO1, VRN2, and PPD1, CO9 has been implicated as a repressor of flowering under both long and short days in barley, but no data are available for rice or sorghum (Kikuchi et al., 2012).Expression data from the short-day rice relative Ehrharta calycina and long-day M. ciliata revealed a conserved pattern of expression, with no difference between long and short days, and transcript levels peaking in the light under both photoperiods (Fig. 4).In contrast, CO9 expression peaked in the morning under long days and in the dark under short days for O. miliaceum and N. pubiflora, revealing a shift in the diurnal rhythm within Stipeae (Fig. 4).Since light is required to stabilize at least A. thaliana CO protein (Hayama et al., 2017), we compared mRNA with protein accumulation in all three noncore Pooideae species and found that the short-day dark peak for O. miliaceum and N. pubiflora CO9 appeared to deteriorate at the protein level.As a result, CO9 protein was higher under long versus short days, representing a second avenue by which the loss of short-day flowering repression could have evolved in N. pubiflora (Fig. 5).
A potential caveat to the protein data relates to the fact that the western blot band for the CO9 antibody was ~8 kDa larger than predicted based on the amino acid sequences derived from transcriptomes of the target species (Supplementary Fig. S3).This result might be interpreted as non-specific binding to off-target proteins.However, given that the results were consistent using two independent antibodies that were designed to avoid cross-targeting to other CO-like proteins, and were generally in line with the mRNA expression profiles, we feel this unlikely.Rather, we posit that the larger size indicates ubiquitination of the target CO9 proteins, which is a common mechanism of regulating flowering time proteins (Piñeiro and Jarillo, 2013).In particular, light-dark regulation of A. thaliana CO involves its ubiquitination (Liu et al., 2008).
Assuming correct interpretation of the protein data, less clear is the effect of high long-day CO9 protein expression in O. miliaceum that flowers more rapidly under long days (Fig. 5).One possible explanation is that short-day flowering evolved early in Stipeae, with the unique gain of short-day-specific VRN2 expression in O. miliaceum resulting in a novel block to flowering under short photoperiods.One argument against this is the fact that VRN2 levels are relatively low in O. miliaceum.Investigation into further Stipeae species and the use of functional approaches will be required to test these alternative hypotheses.

Conclusions
Daylength is used as a cue to promote or repress the reproductive transition in most plants, and the photoperiod pathway largely shares a common evolutionary basis (Andrés and Coupland, 2012).We have shown that a switch from short-to longday induction of flowering was probably a major evolutionary innovation allowing Pooideae grasses to establish and diversify within temperate climates.However, whereas transitions to daylength-neutral flowering are common and phylogenetically widespread, reversions to short-day flowering appear relatively difficult and/or uncommon.We suggest that changes in the diurnal and long-term regulation of CCT domain genes by photoperiod have been important drivers of ecologically important niche shifts.Together, these data highlight both the complexity and flexibility of flowering time evolution in plants and provide novel hypotheses that can be tested through further sampling and functional analyses.

Supplementary data
The following supplementary data are available at JXB online.
Table S1.Materials used in the study.

Fig. 1 .
Fig. 1.Flowering behavior for 31 of the 47 tested Pooideae species as well as Ehrharta calycina of the Oryzoideae.Comparisons were made between either long-or short-day treated plants (LD/SD) or between vernalized plants followed by long-or short-day treatments (VLD/VSD).(A) Barplots of proportion of individuals flowering under different photoperiods.(B) Boxplots of days to heading under different photoperiods.Three species are included in both (A) and (B) as they flowered in both compared treatments, but in inadequate numbers for t-tests (Melica altissima) or results were conflicting between comparisons of proportion of plants flowering and heading dates (Elymus caninus and Nardus stricta).Nassella pulchra and Nassella lepida flowered with only one individual in one of the treatments, and plots for heading date are not shown.The remaining 18 species failed to flower consistently enough to score.*P>0.05,**P>0.005,***P>0.001.Red color indicates long-day treatment and blue color indicates short-day treatment.

Fig. 2 .
Fig.2.Consensus Bayesian Pooideae tree showing Bayesian state reconstruction for photoperiodicity in flowering.Colored internal branches refer to best-supported [>0.50 posterior probability (PP), shown as numbers above branches] inferred character states: long day (red), short day (dark blue), and day neutral (green).Extant species with light blue branches did not flower, and the internal branches were inferred as ambiguous (PP=0.33 long day, 0.33 short day, and 0.33 day neutral).Tip branches are colored based on results of experiments (see Fig.1).The topology is supported by >0.95 PP except branches bearing a double backslash.Outgroups are Zea mays (Panicoideae), and Ehrharta calycina and Oryza sativa (Oryzoideae).

3.
Relative expression of VRN3 in long-or short-1 day treated plants of (A) Oloptum miliaceum, (B) Nassella pubiflora, and (C) Melica ciliata.Sampling time points are given as Zeitgeber time (ZT) indicating hours after dawn per sampling day.Error bars indicate standard error.White background represents time points that are in the light period in both treatments, light gray background represents time points that are in the dark in the short-day treatment and in the light in the long-day treatment, whereas dark gray background represents time points that are in the dark in both treatments.***P<0.001,**P<0.01,*P<0.05.

Fig. 4 .
Fig. 4. Relative expression of PHYC, PPD1, CO1, CO9, and VRN2 in long-and short-day-treated plants of (A, D, G, J, N) Oloptum miliaceum, (B, E, H, K, O) Nassella pubiflora, and (C, F, I, L, P) Melica ciliata.(M) Relative expression of CO9 in Ehrharta calycina.Sampling time points are given as Zeitgeber time (ZT) indicating hours after dawn.Error bars indicate the SE.A white background represents time points that are in the light period in both treatments, a light gray background represents time points that are in the dark in the short-day treatment and in the light in the long-day treatment, whereas a dark gray background represents time points that are in the dark in both treatments.***P<0.001,**P<0.01,*P<0.05.

Fig. 5 .
Fig. 5. Relative abundance of CO9 protein in long-or short-day-treated plants of (A).Oloptum miliaceum, (B) Nassella pubiflora, and (C) Melica ciliata.Sampling time points are given as Zeitgeber time (ZT) indicating hours after dawn.A white background represents time points that are in the light period in both treatments, a light gray background represents time points that are in the dark in the short-day treatment and in the light in the long-day treatment, whereas a dark gray background represents time points that are in the dark in both treatments.