C3–C4 intermediacy in grasses: organelle enrichment and distribution, glycine decarboxylase expression, and the rise of C2 photosynthesis

Highlight C2 photosynthesis in grasses is facilitated by organelle enrichment in tandem with enhanced levels of GDC in the carbon-concentrating cells consistent with changes in expression of a single GLDP gene.


Introduction
C 4 photosynthesis has independently evolved >60 times in angiosperms (R.F. Grass Phylogeny Working Group II, 2012). Within the angiosperms, the Poaceae represents the most prolific family of C 4 origins, with approximately twice as many origins as any other family (Sage, 2016). C 4 grasses also make up the greatest number of C 4 species, comprising ~60% of the 8000 estimated number of C 4 species (Still et al., 2003;Sage, 2016). Roughly a quarter of global net primary productivity on land is due to C 4 photosynthesis (Still et al., 2003), of which the vast majority is contributed by grasses (Sage et al., 1999). C 4 grasses also have great significance for humanity as they dominate the fraction of biomass entering the human food chain as grain (maize, sorghum, and millets), sugar (sugarcane), and fodder for animals, and efforts are underway to engineer the C 4 pathway into C 3 grass crops such as rice and wheat to exploit the superior productivity of C 4 photosynthesis (Peterhansel, 2011;von Caemmerer et al., 2012). Given these considerations, there is now a great interest in understanding how C 4 photosynthesis evolved in grasses, to understand both how this complex trait repeatedly arose, and how we might learn from the evolutionary examples to direct C 4 engineering in major crops (Hibberd et al., 2008).
Studies of C 4 evolution are informed by the presence of species exhibiting intermediate stages between fully expressed C 3 and C 4 life forms within a single evolutionary clade . Ideally, there should be C 3 -C 4 relatives from multiple independent lineages of C 4 photosynthesis to facilitate evaluation of evolutionary hypotheses using comparative approaches. Multiple independent clades provide the possibility to assess whether evolutionary trends are replicated, as they should be if C 4 photosynthesis evolved along common trajectories (Heckmann et al., 2013). To date, the majority (85%) of C 3 -C 4 species occur in eudicots, with the genus Flaveria standing as the major group used in studies of C 4 evolution (R.F. Sage et al., , 2014. Flaveria has over twice the number of intermediates (10) as any other evolutionary clade, and these form two distinct clades that each evolved C 3and C 4 -like species (McKown et al., 2005;Lyu et al., 2015). In addition to Flaveria, at least 11 other C 4 evolutionary lineages have been identified with C 3 -C 4 intermediates branching in sister positions to the C 4 line . Most of these have only one or two species, although in recent years there is evidence of three clades (Heliotropium, Anticharis, and Blepharis) potentially having more than five intermediates (Muhaidat et al., 2011;Khoshravesh et al., 2012;Fisher et al., 2015). All but three of the C 4 lineages with C 3 -C 4 intermediates are eudicots. Among monocots, the genus Neurachne contains a C 3 -C 4 intermediate that branches in a sister position to a C 4 (Christin et al., 2012). A recent study has reported C 3 , C 3 -C 4 , and C 4 photosynthetic genotypes in Alloteropsis semialata (Lundgren et al., 2015). The genus Steinchisma also contains C 3 -C 4 intermediates (Brown et al., 1983;Hylton et al., 1988), but they lack close C 4 relatives in their subtribe, Otachyrinae (Grass Phylogeny Working Group II, 2012). As a consequence of the discrepancy between C 3 -C 4 numbers in eudicots and monocots, our understanding of C 4 evolution is dominated by information from eudicot clades. If there is important variation in eudicot versus monocot patterns of C 4 evolution, as suggested by recent theoretical treatments (Williams et al., 2013), it could be missed because of low monocot representation in the C 3 -C 4 intermediate population.
The South American subtribe Arthropogoninae is a hotspot for C 4 evolution within the grasses, with four putative distinct C 4 origins, once in Mesosetum/Arthropogon, a second time in Oncorachis, and twice in Coleataenia (Grass Phylogeny Working Group II, 2012). As such, the Arthropogoninae is a strong candidate to contain numerous C 3 -C 4 species. This possibility is bolstered by an image from Homolepis aturensis in Supplementary fig. S1 of Christin et al. (2013) that illustrates enlarged bundle sheath (BS) cells with chloroplasts arranged around the periphery in addition to chloroplast clusters adjacent to the vascular tissue. Although this centripetal chloroplast arrangement led to tentative identification of H. aturensis as C 4 (Christin et al., 2013), the presence of centrifugal chloroplasts in the BS cells is a common feature in C 3 -C 4 intermediate species (Monson et al., 1984;Sage et al., 2014). Significantly, because the genus Homolepis is sister to a clade that contains only C 4 species, it has been identified as a genus that might exhibit precursor traits that could have enabled the evolution of the C 4 phenotype (Grass Phylogeny Working Group II, 2012). To evaluate this possibility, we have collected H. aturensis in Costa Rica for study.
A prominent physiological feature of C 3 -C 4 intermediates is the transport of photorespiratory glycine from mesophyll (M) to BS cells for decarboxylation by glycine decarboxylase (GDC), with the released CO 2 then being refixed by BS Rubisco (Monson and Rawsthorne, 2000). Photorespiratory glycine shuttling exhibited by C 3 -C 4 intermediates has also been termed C 2 photosynthesis in reference to the number of carbons shuttled from M to BS cells (Vogan et al., 2007;Bauwe, 2011). The BS mitochondria of most C 2 species examined to date contain the majority of the GDC within the leaf, with small amounts in M cells potentially for C1 metabolism (Hylton et al., 1988;Morgan et al., 1993;Rawsthorne et al., 1998;Voznesenskaya et al., 2001;Ueno et al., 2003;Marshall et al., 2007;Voznesenskaya et al., 2010;Muhaidat et al., 2011;. Decarboxylation of glycine in the BS cells establishes a glycine gradient between M and BS cells, and rapid movement to BS cells is facilitated by enhanced vein density in C 2 relative to C 3 species (Monson and Rawsthorne, 2000). Subsequent glycine decarboxylation within BS cells increases CO 2 around BS Rubisco ~3-fold (Keerberg et al., 2014), and the resulting increase in Rubisco efficiency reduces the CO 2 compensation point in C 2 species relative to C 3 by 10-40 μmol mol −1 (Holaday et al., 1984;Monson et al., 1984;Vogan et al., 2007;. The functional GDC holoenzyme consists of four subunits encoded by individual genes, GLDH, GLDL, GLDP, and GLDT (Bauwe 2011). Decarboxylase activity of the complex is located in the P-subunit encoded by the GDLP gene. In Flaveria, the C 4 photosynthetic mechanism was established through gradual pseudogenization of a ubiquitously expressed GLDP gene, and full activation of a second GLDP gene that shows BS-specific expression in C 3 Flaveria species . The ancestral duplication of the GLDP gene in Flaveria is considered a genetic enabler of C 4 evolution in the genus . BS cell-dominant expression of GDC has been reported in the C 2 grass Steinchisma hians (=Panicum milioides; Hylton et al., 1988). To date, the molecular evolution of this trait in S. hians or other C 2 and C 4 grasses has not been assessed. Schulze et al. (2013), highlighting the presence of two GLDP genes in rice [a member of the C 3 BEP (Bambusoideae, Ehrhartoideae, Pooideae) clade] and a single copy in maize, sorghum, and Setaria italica [of the C 4 PACMAD (Panicoideae, Arundinoideae, Chloridoideae, Micrairoideae, Aristoideae, Danthonioideae) clade], posited that the ubiquitously expressed GLDP gene(s) in C 4 grasses were pseudogenized as in Flaveria and subsequently lost from the genomes.
The purpose of this study was to determine whether H. aturensis exhibits C 2 photosynthesis or is a C 4 species as previously suggested (Christin et al., 2013). We compared anatomy, localization of GLDP, and photosynthetic physiology of H. aturensis with patterns previously identified in the C 2 grasses Steinchisma hians and Neurachne minor (Morgan and Brown, 1979;Hylton et al., 1988). Current phylogenies place the subtribe Otachyrinae, containing Steinchisma, as sister to the Arthropogoninae (Grass Phylogeny Working Group II, 2012). In addition, we examined S. laxum which has also been identified as a species that might provide information on the early stages of C 2 and C 4 evolution (Grass Phylogeny Working Group II, 2012; Sage et al., 2013). Finally, we use genome sequence data from publicly available databases (the Phytozome and NCBI), as well as assembled RNAsequencing (RNA-seq) from 16 additional grass species, to examine evolution of GLDP and genes encoding the other GDC subunits and provide a broader understanding of the evolution of BS-specific GDC expression in C 2 and C 4 grasses.

Plant material
Plants of Homolepis aturensis Chase., Steinchisma hians Raf., and S. laxum (Sw.) Zuloaga obtained from sources described in Supplementary Table S1 at JXB online were grown at the University of Toronto in a greenhouse in 10-20 liter pots of a sandy-loam soil and were watered daily to avoid water stress. Fertilizer was supplied weekly as a 50:50 mixture of Miracle-Grow 24-10-10 All Purpose Plant Food and Miracle Grow Evergreen Food (30-10-20) at the recommended dosage (22 ml of fertilizer salt per 6 liters; Scotts Miracle-Gro; www.scotts.ca). Plants of Neurachne minor from localities previously described (Christin et al., 2012) were grown in a naturally illuminated glasshouse with mean temperatures of 25 °C/13 °C (day/night) at the Plant Growth Facility (PGF) of the University of Western Australia, Perth, Western Australia (latitude 33°89'S). To provide C 3 and C 4 grass species for comparison, we also examined leaves of PACMAD species Dichanthelium oligosanthes (Schult.) Gould (C 3 ), Panicum bisulcatum Thunb. (C 3 ), and two C 4 species (Panicum virgatum L., NAD-ME subtype and Setaria viridis P. Beauv., NADP-ME subtype). Seed of these plants, obtained from sources described in Supplementary Table S1, were also grown at the University of Toronto.

Leaf anatomy, ultrastructure, and immunolocalizations
The internal anatomy of leaves was assessed on sections sampled from the middle of the most recent, fully expanded leaves (one leaf per plant; three plants per species). Plants were sampled from 09:00 h to 11:00 h between April and August when day length was >11.5 h and light intensity in the greenhouse regularly exceeded 1400 µmol photons m −2 s −1 . The youngest cohort of fully expanded leaves was sampled in full sun for all procedures. Samples were prepared for light and transmission electron microscopy (TEM) to assess anatomy as previously described (T.L. Sage et al., , 2013Stata et al., 2014). For immunolocalization, tissue from the same region of the leaf was fixed overnight in 1% (v/v) paraformaldehyde and 1% (v/v) glutaraldehyde in 0.05 M sodium cacodylate buffer. Tissue was then dehydrated and embedded in LR White (Voznesenskaya et al., 2013). Immunolocalization of GLDP was conducted as outlined by Khosravesh et al. (2012). Primary and secondary antibody (18 nm anti-rabbit IgG gold conjugate; Jackson Immunoresearch) dilutions were 1:50 and 1:20, respectively. Immunodetection of the Rubisco large subunit was modified from Ueno (1992). Sections were blocked in 0.5% BSA prior to incubation in primary antibody (1:100) for 3 h. Incubation in secondary antibody (1:40; 18 nm anti-rabbit IgG gold conjugate; Jackson Immunoresearch) was for 1 h. To quantify all BS and M cellular features, TEM images from BS and M cells of the same grids used for immunogold labeling were analyzed using Image J software (Schneider et al., 2012) as previously described Stata et al., 2014). The anti-GLDP antiserum was commercially produced (GL Biochem) against a 17 amino acid peptide showing high conservation in both monocots and dicots. Antisera recognizing the Rubisco large subunit (RBCL) were obtained from AgriSera. Three replicate immunolocalizations were conducted on different days with each replicate including sectioned tissue from all species.
Leaf gas exchange analysis Gas exchange of intact, attached leaves was determined using a LiCor 6400 gas exchange system as previously described . Measurement conditions were 31 ± 1 °C and a vapor pressure difference between leaf and air of 2 ± 0.2 kPa. For measurement of the response of net CO 2 assimilation rate (A) to intercellular CO 2 concentration (C i ) at light saturation, leaves were first equilibrated to 1200-1500 µmol photons m −1 s −1 at an ambient CO 2 concentration of 400 µmol m −2 s −1 . Ambient CO 2 levels were then reduced in steps to 30-50 µmol mol −1 (lower end of this range for C 2 and C 4 species, upper end of this range for C 3 species), with measurements at each step after rate equilibration. The ambient CO 2 was then returned to 400 µmol photons m −1 s −1 , and A re-measured. If A was within 10% of the original rate, the CO 2 concentration around the leaf was increased in steps to near 1600 µmol mol −1 , with measurements made at each step. The linear initial slope of the A/C i response was used as an estimate of carboxylation efficiency (CE).
For estimation of the apparent CO 2 compensation point in the absence of day respiration (C * ), the Laisk method was used as modified by Sage et al. (2013). Values of C * were not estimated in C 4 plants. Leaves were first equilibrated at 400 µmol mol −1 and a light intensity near saturation (900-1500 µmol photons m −2 s −1 ). The CO 2 was then reduced to provide a C i near 100 µmol mol −1 , and, after stabilization, the rate was reduced in a series of steps to near the CO 2 compensation point (Γ), with measurements at each step after signal stabilization. The C i was then returned to near 100 µmol mol −1 , and the procedure was repeated at a lower light intensity. This cycle was repeated such that when measurements were completed, there were 4-5 A versus C i response curves with over five measurements at C i values <100 µmol mol −1 . Each A versus C i response at a given light intensity was then fitted with a linear regression using the lowest 4-6 measurement points that fell on the regression. Points that fell below the regression line above ~80 µmol mol −1 were not included in the regression, as A/C i responses below light saturation may be prone to non-linearity above 60-100 µmol CO 2 mol −1 air. The estimate of C * was taken as the C i value where the 4-5 curves at different light intensities converged. Rarely, however, did all curves converge at the exact same C i ; instead, they intersected with each other over a 5-10 ppm range. In such cases, the middle of the intersects was taken as the C * estimate. However, if there was evidence of a shift to lower photosynthetic photon flux density (PPFD) of the high light response, which often occurs in C 2 species, the high light response was not included in the analysis .

Phylogenetic analysis of GDC subunit genes
Genome sequences for Zea mays, Sorghum bicolor, Panicum hallii, Se. viridis, S. italica, Oryza sativa, Brachypodium distachyon, and B. stacei were downloaded from the Phytozome v10.3 (https://phytozome.jgi.doe.gov/pz/portal.html). Polyploid species Triticum aestivum and P. virgatum were omitted. Amborella trichopoda was used as the outgroup, and we included the Brassicaceae species Arabidopsis thaliana, Capsella grandiflora, and Boechera stricta to show which duplications in the model plant A. thaliana are conserved across the land plants and which are lineage specific. RNA-seq data for 14 BEP clade species as well as the C 3 PACMAD Dicanthelium clandestinum and its close C 4 relative Megathyrsus maximus were downloaded from the NCBI short read archive (Supplementary Table  S2). Reads mapping to each GDC subunit gene were identified using BLASTN with orthologs in Z. mays and O. sativa as queries (S. bicolor was used in lieu of Z. mays for one of the two GLDL paralogs as Z. mays appears to have lost this copy). For each gene, alignments were generated based on the Phytozome genomes using Muscle (Edgar, 2004), a highly conserved region was selected, and a consensus sequence was generated. These consensus sequences were used as reference for assembling sequences based on the retrieved reads using Geneious 8 (http://www.geneious.com).
The assemblies and sequences from Phytozome genomes were aligned using Muscle, trimmed using Trimal (Capella-Gutierrez et al., 2009) with the 'strict' heuristic option, and used to generate Bayesian phylogenies using MrBayes (Ronquist and Huelsenbeck, 2003) as follows: four runs, four chains, GTR substitution model, 2 million generations for all trees except GLDT, which was run for 10 million generations to allow better convergence; ≥10 000 trees were sampled from portions of the end of each run where the average SD of split frequencies remained below 1%. Tree figures were generated using Fig Tree (http://tree.bio.ed.ac.uk/software/figtree).

Data analysis
Results were analyzed with Sigmaplot version 12.5 (Systat Software., San Jose, CA, USA) using one-way ANOVA followed by a Tukey's means comparison test. For characterization of leaf anatomy and ultrastructure, leaf samples were collected from three plants. For all traits measured, the values per plant were averaged to give one value for a plant. These individual plant values were the unit of replication for statistical analysis. For characterization of anatomical features, data from 3-5 sections per plant were averaged. For quantitative assessment of organelles in M and BS cells and GLDP immunodetection, the data from 10 imaged cells per cell type per plant were averaged. For leaf gas exchange, 4-15 measurements were conducted on 4-7 plants per species. In Neurachne species, vascular tissue is surrounded by two layers of cells, an outermost BS and innermost mestome sheath that functions in the C 4 species as the site of CO 2 refixation (Hattersley et al., 1986). Comparing data collected from the mestome sheath of N. minor with data from the BS of the other grasses makes statistical comparisons invalid except for comparisons between either the presence or absence of GDLP in M versus the BS or mestome sheath cells. Hence, N. minor was not included in the statistical tests involving these other grasses.

Homolepis aturensis possesses structural features common to C 2 species
Leaves of H. aturensis are anatomically similar to those of the C 2 species S. hians and S. laxum with respect to M cell structure (Fig. 1). One layer of M cells extends from BS cells to the adaxial and abaxial epidermis (Fig. 1). Approximately six M cells separate the BS of adjacent veins in S. laxum whereas four cells separate the BS of adjacent veins in H. aturensis and S. hians ( Fig. 1; Supplementary Table S3). The M:BS tissue ratio for S. laxum, H. aturensis, and S. hians is ~2 ( Fig. 1; Supplementary Table S3), which is similar to that of the C 4 species P. virgatum but >2.5 times less than that of the C 3 species D. oligosanthes and P. bisulcatum. A large M and small BS volume contributes to an M:BS of almost 5 in Se. viridis even though there are only two cells between each vein (Supplementary Fig. S1C; Supplementary Table S3). M cells in D. oligosanthes, P. bisulcatum, and Se. viridis are more loosely packed and elongate than those of P. virgatum ( Supplementary Fig. S1). The cellular features of N. minor are similar to those previously reported (Hattersley et al., 1986) Table  S3).

, with 2-3 M cells between veins (Supplementary
The BS cells of H. aturensis contain chloroplasts arranged around the periphery, with a significantly greater number clustered centripetally (Figs 1A Table  S4). BS mitochondria are commonly surrounded by chloroplasts in S. hians (Fig 3E), H. aturensis, and S. laxum in a pattern similar to previous reports for Steinchisma species (Brown et al., 1983). In contrast to centripetal chloroplasts displayed with their long axis parallel to the BS wall in S. laxum, the long axis of these chloroplasts in H. aturensis and S. hians are perpendicular to the BS wall (Figs 1, 2). BS chloroplasts are situated primarily in a centrifugal position in C 3 grasses D. oligosanthes and P. bisulcatum (Supplementary  Table S4; Supplementary Figs S1, S2), as noted in C 3 eudicots (Muhaidat et al., 2011;Sage et al., 2013). Approximately 65-69% of mitochondria and 19% (D. oligosanthes) and 41% (P. bisulctum) of peroxisomes, respectively, are located in the centripetal position of the C 3 grasses (Supplementary Table  S4).
Quantitative parameters of organelle traits of BS and M cells are summarized in Supplementary Table S5. BS organelle parameters of H. aturensis and S. hians are not statistically different from each other except for a significantly greater number of mitochondria per planar cell area in S. hians. Mitochondria planar area per planar BS cell area of H. aturensis and S. hians is significantly greater than for S. laxum. In addition, mitochondria number per planar BS cell area, chloroplast and peroxisome number, and planar area per planar BS cell area are significantly greater in S. hians than in S. laxum. Results from a one-way ANOVA on ranks comparing peroxisome planar area per planar BS cell area between S. hians, S. laxum, and H. aturensis only indicated that this trait was significantly higher in S. hians and H. aturensis than in S. laxum (P≤0.001). The C 3 grasses have the lowest mitochondria planar area per planar BS cell area; however, BS mitochondria parameters of C 4 grasses are not different from those of the C 2 grasses H. aturensis and S. hians. Mitochondria, peroxisomes, and chloroplasts are abundant in mestome sheath cells of N. minor (Hattersley et al., 1986;Sage et al., 2014). Unlike H. aturensis, S. hians, and S. laxum, organelles are not polarized in their distribution ( Supplementary Fig. S3), but are instead positioned around the mestome sheath cell periphery (Hattersley et al., 1986;Sage et al., 2014).
When considering M cell organelle features, significantly fewer mitochondria and peroxisomes are observed in C 4 species relative to the C 3 species and the C 2 species H. aturensis, S. hians, and S. laxum (Supplementary Table S5). The C 4 species as well as H. aturensis have significantly lower chloroplast planar area per M planar cell area relative to S. hians, S. laxum, and the C 3 grass species, and these changes result from either smaller (H. aturensis, P. virgatum) or fewer (Se. viridis) M cell chloroplasts (Supplementary Table S5). M cells of C 4like and C 4 Flaveria and other C 4 species have recently been reported to have significant reductions in chloroplast volume (Stata et al., 2014(Stata et al., , 2016.

Homolepis aturensis exhibits C 2 levels of GLDP in bundle sheath and mesophyll cells
Results from quantification of gold particles conjugated to secondary antibodies that bind to anti-GLDP are summarized in Supplementary Table S5. GLDP is almost exclusively located in BS cells of H. aturensis and S. hians, and mestome sheath cells of N. minor (Fig. 3A, B, E, F; Supplementary Fig. S3). Both BS and M cells contained high levels of GLDP labeling in S. laxum (Fig. 3C, D). In comparison with all other C 3 species examined, S. laxum had the highest GLDP labeling in BS mitochondria (Supplementary  Table S5). These patterns in GLDP distribution in M and BS mitochondria of S. laxum and S. hians are similar to earlier reports on these species (Hylton et al., 1988). The gold density in C 3 species D. oligosanthes and P. bisulcatum is higher in mitochondria of M than BS cells and is significantly greater on a planar M cell area basis than in those of H. aturensis, S. hians, and the C 4 species (Supplementary Table S5; Supplementary Fig. S4).

Homolepis aturensis exhibits C 2 levels of Rubisco in bundle sheath and mesophyll cells
Rubisco occurs in both M and BS cell chloroplasts of H. aturensis, and this pattern was similar to that observed in S. hians, S. laxum, and M and mestome sheath cells of N. minor (Supplementary Fig. S5). As expected, Rubisco labeling is present only in the BS chloroplasts of the C 4 species ( Supplementary Fig. S6). Although Rubisco is also present in the M and BS cells of the C 3 species D. oligosanthes and P. bisulcatum, there is qualitatively less labeling in the BS cells ( Supplementary Fig. S6).

Homolepis aturensis exhibits photosynthetic characteristics of a C 2 species
Values of A at 400 µmol CO 2 mol −1 air are statistically similar, being 21 ± 3 (mean±range) µmol m −2 s −1 for all species in the study except for the C 4 plant Se. viridis, which has higher A (Table 1). Thus, any differences in carboxylation efficiency (CE), Γ, or intrinsic water use efficiency (estimated as A/g s at 400 µmol CO 2 mol −1 air) should reflect photosynthetic pathway effects and not variation in photosynthetic capacity. Comparison of the A versus C i responses at saturating light intensities show three equivalent sets of curves that corresponded to the photosynthetic pathway (Fig. 4). C 3 and C 2 species have similar responses, with the exception that Γ was reduced ≥30 µmol mol −1 in the C 2 species; consequently, their A/C i responses were shifted to lower C i values. Homolepis aturensis has A/C i responses identical to those of S. hians and N. minor, leading us to classify H. aturensis as a C 2 species. Carboxylation efficiencies and A/g s of all the C 2 and C 3 species are statistically identical and less than those of the C 4 species (Table 1).
In the C 3 species D. oligosanthes and P. bisulcatum, we observed C * values near 50 µmol mol −1 (Table 1) which is typical for C 3 species at 31 °C . Steinchisma laxum had a similar C * value (53 µmol mol −1 ; Table 1; Fig. 5 A) indicating it is functionally C 3 despite the potential function of the BS organelles. The C * values of the C 2 species were 10.8−20 µmol mol −1 (Table 1; Fig. 5B-D); the 10.8 value is on the lower end of C * for species with this physiology (Edwards and Ku, 1987). In the grasses studied here, lower C * values correspond to higher values of BS mitochondria per planar cell area (Fig. 6A) and higher BS GLDP density per planar cell area, except for Se. viridis (Fig. 6C). C * was lower in species where BS chloroplast area per planar cell area was greater and M chloroplast area per planar cell area was lower (Fig. 6E, F).
There is little observed shift to lower C i in the high light response of A versus C i in H. aturensis, such that all A versus C i curves converge near a common intersection point (Fig. 5). In S. hians and N. minor there is a slight reduction by a C i of ~5-7 µmol mol −1 in the high light response. There is no change in Γ in the C 3 species with variation in light intensity (Fig. 7), while in each of the C 2 species, Γ increases at the lower light intensity (Fig. 7). Neurachne minor exhibits the greatest increase in Γ as light declines, while the light response of Γ is negligible in H. aturensis above 300 µmol m −2 s −1 . Notably, Γ of N. minor is twice that of H. aturensis across the range of light intensities.

GDC BS specificity in C 2 and C 4 grasses probably results from changes in expression of a single GLDP gene
Phylogenetic analyses reveal that within the Poaceae, O. sativa is the only species examined with two gene copies encoding GLDP, and these sequences are most closely related to each other, even with inclusion of 14 additional BEP clade species (Fig. 8). We find no evidence of broader GLDP gene duplication or loss of gene copies in C 4 grass species. Two paralogs encoding GLDH are present in all angiosperms, owing to an ancient duplication event (Supplementary Figs S7, S8). The two paralogs were treated independently as GLDH1 and GLDH2. Preliminary analysis indicates that there is a Poaceae-wide duplication only of GLDL, and targeted assemblies were made independently for each paralog, which we label GLDL1 and GLDL2 (Supplementary Fig.  S9). During each assembly, mapped reads were scrutinized manually for evidence of additional paralogs, and none was detected. No reads were found for GLDL2 in the BEP grass Dendrocalamus sinicus, suggesting that it may have been lost in this species. Most other grass species possess both GLDL gene copies, with the exception of Z. mays, which also lacks GLDL2; S. bicolor, which shares a common C 4 origin with Z. mays, has both paralogs. For all grass species, full GLDL1 and GLDL2 sequences from genomes and all assemblies which were completed to the start codon were strongly predicted to be mitochondrial-localized by TargetP (Emanuelsson et al., 2000;data not shown). This is consistent with targeting of the two copies in Arabidopsis (AT3G17240, AT1G48030; Fig. S9), which are both mitochondrial (Lutziger and Oliver, 2001;Rajinikanth et al., 2007). A plastidial dihydrolipoyl dehydrogenase exists in land plants as well, but is more distantly related and likely dates to a much earlier duplication (Lutziger and Oliver, 2000;Rajinikanth et al., 2007). Finally, GLDT is present as a single-copy gene in all species examined ( Supplementary Fig. S10). While we find evidence of local duplication (GLDP, GLDH2) and conserved paralogs (GLDL), we find no evidence of C 4 lineage loss of any GDC subunit genes.

Discussion
Photorespiratory glycine shuttling exhibited by C 2 species is considered to be the evolutionary bridge from C 3 photosynthesis to C 4 photosynthesis, based largely on studies from eudicot species, particularly Flaveria (Bauwe, 2011; Sage et al., Schulze et al., 2013;Mallmann et al., 2014). It has been posited that C 2 photosynthesis may serve a bridging role to C 4 photosynthesis in grasses as well , although the timing of trait acquisition, to include GDC BS cell specificity and abundance, has been proposed to differ between eudicots and monocots (Williams et al., 2013). Here, we evaluated photosynthetic pathway characteristics and cellular features of BS and M cells in H. aturensis, a candidate C 2 species which branches in a position sister to a C 4 clade in the subtribe Arthropogoninae. We used comparative studies with confirmed C 2 grasses, S. hians and N. minor, and the C 3 grass S. laxum to facilitate our classification of the photosynthetic type of H. aturensis. In addition, we conducted a phylogenetic study of genes for the GDC subunits to test the Schulze et al. (2013) inference that the evolution of BS-specific GDC expression in C 4 grasses was similar to that in Flaveria. From our physiological and structural results, we conclude that H. aturensis is indeed a C 2 species, supporting a hypothesis that the photorespiratory glycine shuttle is a bridge to C 4 photosynthesis in grasses in the subtribe Arthropogoninae. The characterization of S. laxum and S. hians also allows us to conclude that activation of the BS cells during transition from C 3 to C 2 in grasses in the subtribe sister to the Arthropogoninae is similar to what has been reported for eudicots with respect to BS organelle positioning and organelle and GDC enrichment (Muhaidat et al., 2011;Sage et al., 2013). Finally, our phylogenetic and immunohistochemical data are consistent with the notion that BS GDC in C 2 and C 4 grasses results from changes in expression levels of a single GLDP gene in M and BS cells.
Homolepis aturensis exhibits characteristic features common to species that concentrate photorespired CO 2 in BS cells using the C 2 metabolic cycle. One of the key proteins essential for GDC activity, GLDP, localizes almost exclusively in BS mitochondria in H. aturensis and S. hians. This pattern is ubiquitous in C 2 eudicots (Hylton et al., 1988;Voznesenskaya et al., 2001;Ueno and Sentoku, 2006;Voznesenskaya et al., 2010;Muhaidat et al., 2011;. A second characteristic of C 2 species is an abundance of Rubisco in M and BS cells (Monson et al., 1984). Rubisco is abundant in M and BS cells in H. aturensis and S. hians, in contrast to the typical C 3 pattern (high M Rubisco; low BS Rubisco) and C 4 pattern (Rubisco only in BS cells). Lastly, the number of Fig. 4. Response of net CO 2 assimilation rate to intercellular CO 2 concentration in Homolepis aturensis, two C 2 species (Neurachne minor and Steinchisma hians), a C 4 species (Setaria viridis), a C 3 species (Dicanthelium oligosanthes), and a C 3 species with proto-Kranz anatomy (Steinchisma laxum). Measurement conditions were 31 ± 1 °C and saturating light intensities (1200-1500 µmol m −2 s −1 ). The curves shown are representative of 2-3 individual A/C i responses per species. The linear regression for points below 100 µmol mol −1 is shown for H. aturensis (dashed line). Measurement temperature was 31 ± 1 °C. Sample sizes given are n = 3-5 for C * estimates, and n = 3-8 for all other data, with the exception of P. virgatum where n= 2). Letters indicate statistical groupings at P<0.5 via one-way ANOVA followed by a Student's-Neumann-Kuehls test. NA, not applicable. 400 refers to ambient CO 2 concentration in µmol mol −1 .

M cells between BS cells in H. aturensis and S. hians results
in a reduced M:BS ratio and increased vein density common to C 2 species (reviewed in Sage et al., 2012). These features promote rapid flux of photorespiratory metabolites between M and BS compartments (Monson and Rawsthorne, 2000), improve water relations under high photorespiratory conditions (Osborne and Sack, 2012), and facilitate an increase in the volume of leaf tissue where photorespired CO 2 is concentrated around Rubisco in C 2 species. As the refixed fraction of photorespiratory CO 2 increases, C * (and Γ) declines (von Caemmerer, 2000). The effectiveness of the C 2 process in H. aturensis as well as S. hians is reflected in the Γ and C * values that are at the lower range of values reported for C 2 species (Holaday et al., 1984;Monson et al., 1984;Ueno et al., 2003;Vogan et al., 2007;Voznesenskaya et al., 2010;. Values of Γ and C * below 15 ppm indicate either that a C 4 cycle is active to complement the C 2 cycle, or that the CO 2 trap in the inner BS is particularly effective at recapturing the photorespired CO 2 (von Caemmerer, 2000). Steinchisma hians has weak to negligible C 4 cycle activity (Edwards et al., 1982), and cellular characteristics of organelle orientation in BS cells may explain the low C * in H. aturensis and S. hians. The organelle-enriched BS cells of these two species exhibit polarity in organelle positioning such that almost all peroxisomes and GDCcontaining mitochondria, and over half of the chloroplasts are situated adjacent to the vascular tissue. This arrangement of BS organelles is posited to be particularly effective in enhancing recapture efficiency of photorespired CO 2 before it can escape the BS cell in C 2 species (Rawsthorne, 1992). One notable feature we observed is that the parallel orientation of the centripetal BS chloroplasts to the inner wall in S. laxum shifts to a perpendicular orientation in S. hians. A similar perpendicular chloroplast orientation is present in H. aturensis. This pattern allows for packaging of the more numerous, larger chloroplasts in the centripetal position, which could be important for increasing the surface area for refixing photorespired CO 2 from adjacent mitochondria. Moreover, BS mitochondria are physically surrounded by chloroplasts in S. hians and H. aturensis, and these close physical associations have been proposed to enhance refixation of photorespired CO 2 (Brown et al., 1983).
The fine structure of C 2 BS cells is defined as C 2 Kranz, reflecting a view that this photosynthetic carbon-concentrating mechanism is associated with its own enabling Kranzlike structure . Multiple convergence of C 2 Kranz in eudicots and grasses is strong evidence that this particular BS anatomy is specifically adapted for the C 2 pathway. The earliest recognizable subcellular events that 'increase the accessibility' (Grass Phylogeny Working Group II, 2012) of C 2 Kranz from C 3 in eudicots are an enhancement in numbers and size of mitochondria per BS cell, and positioning of

) and
Steinchisma laxum (C 3 proto-Kranz), (B) the C 2 species Steinchisma hians, (C) the C 2 species Homolepis aturensis, and (D) the C 2 species Neurachne minor. Measurement light intensities are indicated beside each curve in µmol photons m −1 s −1 . The C * estimate is indicated by arrows. Curves shown are representative of 3-6 measurement sets per species, except for P. virgatum where two sets of measurements were obtained. Measurement temperature was 31 ± 1 °C. these organelles from the centrifugal C 3 position to the centripetal BS pole. The BS chloroplast numbers also increase in tandem with alterations in mitochondria placement, along with a rearrangement of many, but not all, BS chloroplasts from the centrifugal to centripetal pole (Muhaidat et al., 2011;Sage et al., 2013;Voznesenskaya et al., 2013). The anatomy associated with these earliest subcellular events has been termed proto-Kranz (Muhaidat et al., 2011;Sage et al., 2013). The transition to full C 2 BS patterns from proto-Kranz in eudicots results from further amplification in centripetal mitochondria volume (size and numbers) and relocation of a greater fraction of enlarged chloroplasts to the centripetal pole (Muhaidat et al., 2011;Sage et al., 2013). Proto-Kranz and the shift to C 2 Kranz occurs with increasing vein density in Flaveria and Heliotropium (Muhaidat et al., 2011;Sage et al., 2013).
The subcellular framework of C 3 S. laxum BS cells and subsequent changes to that configuration from S. laxum to C 2 S. hians are similar to those observed in eudicots, supporting a hypothesis that proto-Kranz facilitates the C 3 to C 2 transition . In comparison with the C 3 species D. oligosanthes and P. bisulcatum, mitochondria and peroxisomes are situated along the centripetal poles of BS cells in S. laxum, classifying this species as proto-Kranz. Previous characterizations of proto-Kranz species have not presented data on peroxisomes, and this additional focus in the present study provides critical information on the positioning of the other organelle involved in C 2 photosynthesis in the BS. A 3-fold increase in mitochondrial volume and corresponding increase in GDC, as well as a 2.5-fold increase in peroxisome volume, and 2-fold increase in chloroplast volume accompany the transition to the C 2 BS pattern in S. hians. Also, as observed in eudicots, the increase in BS chloroplast volume in S. hians from proto-Kranz is associated with more of these organelles in the centripetal location. Notably, although the number and size of mitochondria per BS cell area are lower in S. laxum than in S. hians, the GLDP label intensity is similar per mitochondrion and significantly higher than that observed in the C 3 grasses. These results indicate that increased BS GDC per mitochondrion is also a functionally important development early in C 2 evolution in grasses. The enhanced BS GDC density per mitochondrion in S. laxum is present in tandem with C 3 -like values of vein density. The C 2 levels of BS GDC in S. hians are present in high vein density leaves. The high levels of BS GDC in S. laxum and S hians contrasts with the theoretical predictions of Williams et al. (2013) who modeled C 4 evolution in eudicots and monocots based on observed patterns of trait acquisition in C 3 -C 4 intermediates. For monocots, they predicted that an increase in vein density preceded enhanced GDC specificity and abundance in BS cells. However, their model relied on a relatively small data set with significant gaps. The results here indicate that C 2 evolution in grasses follows a pattern more typical of eudicots, which the model of Williams et al. (2013) may support when reparameterized with a richer data set.
In Neurachne, as in many other grasses, the mestome sheath cell is the site of the Calvin-Benson cycle in C 4 species (Hattersley and Browning, 1981;Hattersley et al., 1986;Dengler and Nelson, 1999;Edwards and Voznesenskaya, 2011). We demonstrate that the mestome sheath cells in the C 2 species N. minor are functionally similar to the C 2 BS cells of H. aturensis and S. hians because GDC is almost exclusively located in organelle-enriched mestome sheath cells in the high vein density leaves. However, unlike H. aturensis and S. hians, there is no polarized orientation of organelles in the GDC-enriched mestome sheath cells of N. minor (Hattersley et al., 1986;this study), indicating that N. minor utilizes a different strategy from H. aturensis and S. hians to trap photorespired CO 2 . In Neurachne, as in many other grasses, the thick mestome sheath cell wall with a suberized lamella becomes the Fig. 7. The response of the CO 2 compensation point of A (Γ) as a function of measurement light intensity in the C 2 species Homolepis aturensis, Neurachne minor, Steinchisma hians, the proto-Kranz species Steinchisma laxum, and the C 3 species Dicanthelium oligosanthes and Panicum bisulcatum. Values of Γ were determined from the A/C i curves used in the sequence of measurements to determine C * . Symbols are means ±SE (n=2-6), except where no error bar is shown, in which case n=1. trap (Hattersley and Browning, 1981;Hattersley et al., 1986;Dengler and Nelson, 1999;Edwards and Voznesenskaya, 2011). The C 2 species Alloteropsis semialata ssp semialata has a similar C 2 Kranz anatomy to N. minor; GDC levels are highest in mestome sheath cells with a suberized lamella and the abundant organelles are equally partitioned within those cells (Hattersley and Browning, 1981;Ueno and Sentoku, 2006). Intriguingly, although organelle orientation in mestome sheath cells is not important in the evolution of C 2 photosynthesis in N. minor, chloroplasts do have a centrifugal orientation in C 4 Neurachne species (Hattersley et al., 1986). Qualitative observations on C 3 Neurachne species indicate that some of the C 3 species have enhanced numbers of chloroplasts and mitochondria in mestome sheath cells (Hattersley et al., 1986), leading us to posit that organelle and GDC enrichment may have been important during the early stages of C 2 evolution in the genus.
The evolutionary transition from C 3 to C 2 has been proposed first to involve a change in cell type-specific expression of GDC from M to BS in tandem with a loss of M GDC (Bauwe, 2011). A comparison of the cellular site of GDC expression in proto-Kranz S. laxum with that of the sister species S. hians indicates that the severe reduction in M GDC in the C 2 species is preceded by increased expression in BS cells. This is consistent with patterns observed in the eudicots Flaveria and Heliotropium, and supports a model of gradual GDC loss in M cells following a physiological activation of the BS (Muhaidat et al., 2011;Sage et al., 2013;Schulze et al., 2013). C 3 species of Flaveria contain two copies of the gene encoding GLDP resulting from gene duplication . One of these is BS dominant in expression and the second is expressed ubiquitously throughout the leaf in C 3 Flaveria . During the evolution of C 4 photosynthesis, the loss of M GDC function in Flaveria resulted from pseudogenization of the gene coding for the ubiquitously expressed GLDP . Schulze et al. (2013) speculated that BS-dominant GLDP expression arose in a similar manner in C 4 grasses, because O. sativa (C 3 BEP clade) has two GLDP genes, but Z. mays and other C 4 species in the C 4 PACMAD clade have only one. To provide an understanding of the evolution of BS-specific GDC expression in C 2 and C 4 grasses, we conducted phylogenetic analyses of genes encoding GLDP using 17 BEP and seven PACMAD grass species, three Brassicaceae species, and Amborella as outgroup. Our analyses demonstrate that, with the exception of O. sativa, all grasses have one copy of GLDP. The two copies of the genes encoding GLDP in rice are more closely related to each other than to any other GLDP gene included in the analysis and therefore represent a local duplication. Combined, the phylogenetic and immunohistochemical observations on C 3 and C 4 PACMAD species are consistent with a hypothesis that BS-dominant quantities of GDC in C 2 and C 4 grasses resulted from modifications in regulatory mechanisms controlling the levels of expression of a single GLDP gene present in M and BS cells; C 3 species have high levels of M GDC and low levels of BS GDC, and the opposite pattern is present in the C 2 and C 4 species. Mesophyll tissue specificity of phosphoenolpyruvate carboxylase has evolved through modification of cis-regulatory elements in C 4 Flaveria (Gowik et al., 2004). Studies examining promoter regions of the GLDP subunits in closely related C 3 , C 2 , and C 4 species should provide insights into the evolution of the regulatory mechanisms that confer the requisite expression patterns for C 2 and subsequently C 4 photosynthesis in grasses.
Since it is conceivable that BS specificity of the GDC complex arose via duplication and pseudogenization of one of the other subunits, we also included analyses for GLDL, GLDH1 and GLDH2, and GLDT. In land plants, GLDH1 functions in photorespiration and GLDH2 is associated with C1 metabolism (Rajinikanth et al., 2007). We found no evidence of broad duplication for either of the ancient GLDH copies or GLDT in the grasses; however, GLDL is encoded by two conserved paralogs in most grass species, and may have arisen from the whole-genome duplication in the ancestor of the grass family. GLDL is the only GDC subunit gene for which we find evidence of subfunctionalized and conserved paralogs analogous to the two GLDP copies in Flaveria. There is, however, no evidence that either of these copies has been lost or pseudogenized (e.g. via nonsense or frameshift mutation) in any C 4 species except Z. mays, which has a local duplication of GLDL1 and lacks a gene encoding GLDL2. Yet it is unlikely that this loss was involved in the evolution of C 4 photosynthesis, as GLDL2 is present in S. bicolor, which shares a common C 4 origin with maize (Grass Phylogeny Working Group II, 2012). The two GLDL paralogs in grasses may be partially reduntant with one another (Rajinikanth et al., 2007). The nature of the subfunctionalization of GLDL which resulted in the evolutionary retention of two paralogs in Poaceae is not known, but our results indicate that these paralogs did not play a role in the evolution of C 4 photosynthesis analogous to the GLDP paralogs in Flaveria. It is similarly unlikely that the two ancestral copies of GLDH played a role in C 4 evolution analogous to that of GLDP in Flaveria as both are present in all grass species examined here, and only GLDH1 plays a role in photorespiration (Rajinikanth et al., 2007).

Conclusion
Evolution of C 2 photosynthesis in the grasses and eudicots is conferred by organelle enrichment, BS-or mestome sheathdominant GDC accumulation, and centripetal positioning of organelles when the BS is the carbon-concentrating tissue. In Steinchisma, the earliest recognized subcellular event that facilitates C 2 photosynthesis is the placement of BS mitochondria and peroxisomes exclusively to the centripetal pole. This feature, also present in eudicots, is posited to set in motion a feedforward facilitation cascade that leads to C 2 and subsequently C 4 photosynthesis . How and why changes in BS or mestome sheath GDC levels and organelle volume, and BS organelle positioning were initiated during the early stages of C 2 evolution in grasses remains a mystery. Identification of mechanisms controlling these processes should be a primary focus of research. The Arthropogoninae, the subtribe containing Steinchisma (Otachyrinae), and Neurachninae will be key to these future studies.

Supplementary data
Supplementary data are available at JXB online. Table S1. List of species studied and source of species. Table S2. Species for which RNA-seq data were downloaded and assembled for phylogenetic analysis of GDC subunits. Table S3. Anatomical parameters of C 3 , C 2 , and C 4 leaves. Table S4. Organelle distribution in bundle sheath cells of C 3 and C 2 species. Table S5. Quantification of organelle numbers, size, and density of gold labeling (GLDP) in mesophyll and bundle sheath cells of C 3 , C 2 , and C 4 leaves. Figure S1. Light micrographs of C 3 and C 4 species. Figure S2. Bundle sheath cell ultrastructure of C 3 and C 4 species. Figure S3. Leaf structure and anatomy and immunolocalization of GLDP in N. minor. Figure S4. Immunolocalization of GLDP in M and BS cells of C 3 and C 4 species. Figure S5. Immunolocalization of Rubisco large subunit in M and BS cells of C 3 and C 2 species. Figure S6. Immunolocalization of Rubisco large subunit in M and BS cells of C 3 and C 4 species. Figure S7. A Bayesian phylogenetic tree of GLDH1. Figure S8. A Bayesian phylogenetic tree of GLDH2. Figure S9. A Bayesian phylogenetic tree of GLDL. Figure S10. A Bayesian phylogenetic tree of GLDT.