Using breeding and quantitative genetics to understand the C4 pathway.

Reducing photorespiration in C3 crops could significantly increase rates of photosynthesis and yield. One method to achieve this would be to integrate C4 photosynthesis into C3 species. This objective is challenging as it involves engineering incompletely understood traits into C3 leaves, including complex changes to their biochemistry, cell biology, and anatomy. Quantitative genetics and selective breeding offer underexplored routes to identify regulators of these processes. We first review examples of natural intraspecific variation in C4 photosynthesis as well as the potential for hybridization between C3 and C4 species. We then discuss how quantitative genetic approaches including artificial selection and genome-wide association could be used to better understand the C4 syndrome and in so doing guide the engineering of the C4 pathway into C3 crops.

Introduction

Photosynthetic plants provide humanity’s food, many textiles, and building materials, and represent the source of numerous medicines and fuels. Understanding how improvements in photosynthesis could be achieved therefore has the potential to impact many aspects of human life. Photosynthesis requires the enzyme Rubisco to fix atmospheric carbon dioxide (CO2) into 3-phosphoglycerate (Calvin and Benson, 1948). Species that only use Rubisco for carbon fixation are known as ‘C3’ plants, as 3-phosphoglycerate contains three carbon atoms. Rubisco, however, is also able to react with oxygen in addition to CO2. This oxygenation reaction produces the toxic molecule 2-phosphoglycolate, which must be metabolized and recycled via the photorespiratory cycle. Photorespiration leads to loss of carbon fixed by Rubisco and release of ammonia from amino acids at the expense of both ATP and reducing power (Bowes ). Rates of photorespiration typically increase at higher temperatures because, under these conditions, the oxygenation reaction of Rubisco is favoured (Portis and Parry, 2007), but photorespiration can also increase during periods of drought when stomatal closure limits CO2 supply to the Rubisco active site. In extreme conditions, photorespiratory rates can use ~25% of photosynthetic outputs (Sharkey, 1988).

Land plants have evolved two carbon-concentrating mechanisms to reduce photorespiration. These are termed Crassulacean acid metabolism (CAM) and C4 photosynthesis. Whilst in both cases rates of photorespiration are reduced because compared with the C3 state, ~10-fold higher concentrations of CO2 are supplied to Rubisco, CAM and C4 species use temporal and spatial systems, respectively. It is estimated that the C4 pathway has evolved independently from C3 ancestors at least 60 times to yield numerous phenotypes that concentrate CO2 around Rubisco (Sage ). In all cases, in the C4 leaf Rubisco-dependent fixation of CO2 takes place in a specific compartment supplied with high concentrations of CO2 such that the oxygenase activity of Rubisco is almost completely abolished (Fig. 1A). In most C4 species, photosynthesis is compartmented between two cell types so that they are unified by a general pathway in which CO2 is converted to bicarbonate (HCO3–) by carbonic anhydrase (CA) in mesophyll cells, and then combined with the 3-carbon molecule phosphoenolpyruvate (PEP) by the enzyme phosphoenolpyruvate carboxylase (PEPC) into the 4-carbon molecule oxaloacetate (Fig. 1A). Oxaloacetate is then either reduced to malate or transaminated to aspartate. After diffusing to an adjacent cell layer such as the bundle or mestome sheath, malate or aspartate are decarboxylated such that high concentrations of CO2 accumulate around Rubisco and so allow high rates of carboxylation (Fig. 1A). Finally, in species that use NAD-dependent malic enzyme (NAD-ME) or NADP-ME to release CO2 around Rubisco, the 3-carbon molecule produced from decarboxylation is regenerated to PEP in mesophyll cells by pyruvate orthophosphate dikinase (PPDK) to continue the cycle (Fig. 1A).

Fig. 1.

Natural variation in C4 biochemistry and anatomy. (A) An overview of C4 biochemical subtypes. Although all forms of two-celled C4 photosynthesis involve initial CO2 fixation to generate four-carbon intermediates in mesophyll cells and diffusion to bundle sheath cells, the method of decarboxylation to create a high-CO2 environment around Rubisco varies between C4 species. Solid and dashed lines show enzymatic and diffusion steps of the C4 pathway, respectively. (B) Examples of leaf anatomies seen in C4 species. Exemplar species that use each anatomical variant are shown below each type. Many more anatomical types have been described, which suggests that multiple leaf morphologies can facilitate the C4 pathway. Abbreviations: M, mesophyll; B, bundle sheath; VB, vascular bundle; CCC, central cytoplasmic compartment; PC, peripheral chloroplast; WS, water storage cell; ch, chloroplast.

Traits underpinning C4 photosynthesis vary widely between species (Edwards and Voznesenskaya, 2011; Furbank, 2011; Sage and Stata, 2015; Sedelnikova ). This interspecific variation in C4 traits includes differences in leaf anatomy, cell biology, and biochemistry, as well as the patterns of gene expression that determine these characteristics. For example, the cell types and arrangement of veins used by C4 species vary between lineages that have independently evolved the pathway (Fig. 1B). At least nine anatomical types have been described in the grasses (Poaceae) (Edwards and Voznesenskaya, 2011). Examples of this variation include in the number of layers of mestome and/or bundle sheath cells, and whether Rubisco is compartmented into the bundle or the mestome sheath. Although much of this variation associated with C4 photosynthesis is found in lineages that are separated by deep evolutionary time, Kranz anatomy also differs in species within families including the Amaranthaceae (Kadereit ; Muhaidat ; Sage, 2016), Asteraceae (Peter and Katinas, 2003), Cleomaceae (Koteyeva ), Portulaceae (Voznesenskaya ), and Poaceae (Ohsugi and Murata, 1985; Edwards and Voznesenskaya, 2011). Of the ~8100 C4 species defined to date, six operate the C4 pathway in a single cell (Fig. 1B). In these single-celled C4 species, the pathway is distributed between separate populations of chloroplasts such that the cell biology of these species has been modified compared with the C3 state. However, modifications to the cell biology of C4 leaves is not restricted to these single-cell species. In C4 species that separate photosynthesis between two cell types, plasmodesmatal frequency is increased compared with the C3 state (Botha, 1992; Danila ). Some lineages contain suberin in the bundle sheath cell wall whilst others do not (Mertz and Brutnell, 2014), and whilst some C4 lineages arrange chloroplasts in bundle sheath cells centripetally, others do this centrifugally with respect to the veins (Edwards and Voznesenskaya, 2011).

Lastly, soon after the discovery of C4 photosynthesis, differences in the biochemistry of the pathway were discovered among C4 species (Hatch ). These different pathways were termed C4 ‘subtypes’ due to the fact that decarboxylation is associated with three separate C4 acid decarboxylases, NADP-ME, NAD-ME, and phosphoenolpyruvate carboxykinase (PEPCK). Although there is growing support for the notion that species can modify the extent to which each C4 acid decarboxylases is engaged (Omoto ; Sharwood ; Sales ), the differences in biochemistry associated with the subtypes exemplify the fact that the C4 pathway is a convergent phenomenon, and that its operation varies between species.

The differences in leaf anatomy, cell biology, and biochemistry between independent C4 lineages have frequently been summarized (Edwards and Voznesenskaya, 2011; Sage, 2016). In contrast, there have been fewer recent attempts to synthesize the literature relating to forced hybridizations between C3 and C4 species. Studies have included somatic hybridizations of phylogenetically distant C3 and C4 plants, as well as sexual hybridizations of congeneric species. Whilst these wide hybridizations have provided insight into the extent to which C4 traits can be maintained and inherited in C3 species, a growing body of evidence documents variation in C4 traits within a species. We summarize examples of this work and suggest that there are opportunities to use quantitative trait mapping to better understand the C4 pathway. Not only could these classical approaches provide insight into the evolution and genetic basis of C4 photosynthesis, they may also inform efforts to engineer more efficient C3 crops.

Somatic hybridization of C3 and C4 species

Approaches such as protoplast fusion allow somatic or asexual hybridization. Protoplasts from somatic cells from separate species are fused and regenerated into hybrid plants (Carlson ; Evans, 1983). In many cases, asexual hybridization can lead to fertile hybrids between species that are considered sexually incompatible. Attempts to form hybrids via somatic hybridization of C3 rice (Oryza sativa) and other C4 grasses have been moderately successful. Terada produced somatic hybrids between rice and C4Echinochloa oryzicola that were morphologically different from either parent. Some contained 60 chromosomes which corresponded to the full hybrid complement, but plants developed necrosis and died before forming roots. Moreover, rice and C4Panicum maximum (now Megathyrsus maximus) were successfully fused to form hybrids with abnormal floral structures with lowered fertility (Xin ). In all, 28 hybrids flowered but only five set fertile seed. To our knowledge, this work has never been repeated.

There have also been attempts to form hybrids between wheat and C4 grasses. A cell suspension of Trititrigia (a perennial hybrid of Triticum durum and Thinopyrum intermedium) was hybridized with maize (Wang ; Wang and Niizeki, 1994). Plants that regenerated were aneuploids carrying incomplete sets of chromosomes from both species. Although the progeny were not full hybrids, this study demonstrated that after asexual hybridization, maize and Triticum chromosomes were not eliminated during successive cell divisions despite the uniparental genome elimination that occurs when both species are hybridized sexually (Laurie and Bennett, 1986, 1989; Laurie ). Szarka fused a cell suspension of an albino maize mutant with wheat protoplasts. Plants that regenerated resembled maize but were green, indicating that photosynthesis from wheat rescued the albino phenotype in maize. Cytological observations showed the plants had all parental chromosomes, but no morphological traits associated with C4 photosynthesis were detected and, although the plants produced male and female flowers, all were sterile (Szarka ). Independently, Xu reported wheat–maize hybrids that contained nuclear and mitochondrial genomes of both species but plastid DNA only from wheat. These somatic hybrids resembled wheat and, although many flowered, they were all sterile. This may have been due, at least in part, to the fact that the wheat and maize cell suspension cultures had chromosomal aberrations prior to fusion. Thus, taken as a whole, work on asexual hybridization of C3 and C4 cereals indicates that chromosomes of both photosynthetic types are stable in fused cells. However, in reports such as those from Xu and Szarka , plants were not viable after transfer from tissue culture. In contrast, sexual hybridization of closely related C3 and C4 species has in some cases allowed production of fertile plants and their progeny assessed over multiple generations. We address this next.

Sexual hybridization of C3 and C4 species

A number of taxa containing either congeneric C3 and C4 species or C3, C3–C4 intermediates, and C4 species have been successfully hybridized (Fig. 2A, B). Although the outcome of these analyses varied, whilst wholesale transfer of C4 traits have not been reported in some instances, specific traits were introgressed into a C3 background. For example, crosses between C4Atriplex rosea and C3Atriplex prostrata (formerly A. patula ssp. hastata and A. triangularis, respectively), C3A. rosea and C3A. glabriuscula have been made (Björkman ; Nobs ). Populations derived from such crosses were progressed and C4-like characteristics assessed (Björkmann ). Among 200 F3 individuals screened for the CO2 compensation point, 178 individuals showed values similar to the C3 parent, 19 showed intermediate phenotypes, and three were similar to the C4 parent (Björkman ). Thus, in a small number of individuals, it appears that crossing was able to integrate loci associated with the compensation point. When F1 derived from a C4A. rosea×C3A. patula hybridization were backcrossed to C4A. rosea, these BC1 offspring segregated for either C4 or C3 photosynthesis, with only two individuals showing C4 photosynthesis (Rikiishi ), suggesting dominance towards a C3 state in this hybrid combination. In these reports above, no F1 individual, nor any within segregating F2 and F3 populations, showed a full transfer of C4 photosynthesis. More recently, F2 individuals derived from a resynthesized C4A. rosea×C3A. prostrata cross showed large variation in leaf anatomy and nearly intermediate CO2 compensation points, but individuals in the F3 generation seemed to revert to C3-like values (Oakley ). Hybrids have also been made between C3 and C4-like species of Flaveria (Apel ; Cameron ) and C3–C4 intermediate and C4Flaveria species (Brown et al., 1986, 1992). Significant F1 sterility was encountered (Brown and Bouton, 1993) but F2 were obtained and, although they possessed continuous variation with regard to C4 leaf anatomy and carbon isotope discrimination characteristics, it was skewed away from the mid-parental mean towards a C3 or C3–C4 phenotype. This would indicate dominance deviation towards a C3 phenotype despite the presence of genes that allow C4 photosynthesis. In F1 hybrids derived from a C3×C4-like Flaveria cross, enzyme activities of PEPC, PPDK, and NADP-ME were skewed towards those associated with C3 photosynthesis, but C4-like activities were reported for NADP-malate dehydrogenase (Holaday ), indicating that incomplete dominance for certain genes may exist while others show dominant activity patterns. In summary, although many C3×C4 hybrids in the dicotyledons showed reduced fertility and limited penetrance of C4 traits, these studies also indicate that aspects of C4 photosynthesis are heritable in a C3 background. As many other closely related C3 and C4 species exist (Fig. 2C), it is possible that additional stable hybrids could be generated that exhibit increased genomic stability and/or better trait segregation between the C3, C3–C4, and C4 types. Hybrids between different C4 decarboxylation subtypes may also be possible. Closely related species such as Blepharis cilaris and Blepharis attenuata that use NAD-ME and NADP-ME, respectively, have been described (Akhani ). To our knowledge, whilst no hybrids have been reported in Blepharis, natural hybrids between Cynodon dactylon (NAD-ME) and Chloris sp. (PEPCK) display intermediate activities of NAD-ME and PEPCK (Prendergast, 1987).

Fig. 2.

Examples of successful as well as potential hybridizations between C3 and C4 species. (A) Phylogenetic reconstruction of the orders constituting flowering plants according to The Angiosperm Phylogeny Group (2016). Orders containing C4 lineages are shown in bold. (B) Exemplar hybridization webs that have resulted in successful F1 hybrids between C3, C4, and C3–C4 intermediate photosynthetic types. (C) Taxa that contain closely related C3, C4, or C3–C4 intermediate species or accessions for which hybridization has not been reported, but may be possible. These groups are potential systems where C4 genes could be mapped. Arrows from the phylogenetic tree indicate from which order the plant species originate (B, C).

C3–C4 hybrids have been generated in the grasses by two broad approaches. First, as with dicotyledons, congeners using either C3 or C3–C4 photosynthesis have been crossed. Second, much wider crosses of distantly related species have been performed. Examples of crosses within a genus include C3 and C3–C4 intermediate Steinchisma (formally Panicum) species from the Poaceae (Bouton ; Brown ; Sternberg ). F2 and F5 individuals derived from hybridization of Steinchisma milioides (C3–C4) and Steinchisma laxum (C3), or S. spathellosum (C3–C4) and S. boliviense (C3) exhibited intermediate leaf morphologies, CO2 compensation points, and δ13C values. Also within the Poaceae, C3 and C4 accessions of Alloteropsis semialata have been hybridized, producing plants with intermediate anatomical traits as well as C4 gene expression (Bianconi , Preprint). Thus, in these hybridizations, some traits important for C4 photosynthesis could be introduced into an otherwise C3 leaf. A variety of attempts at wide hybridization have also been reported. For example, although maize pollen germinates and fertilizes the ovule of wheat to form zygotes containing a full haploid set of each parental genome (Laurie and Bennett, 1986), these hybrids were unstable and after three rounds of mitotic cell divisions during embryogenesis all maize chromosomes were lost (Laurie and Bennett, 1986, 1989). In contrast, after hybridization of oat and pearl millet (Pennisetum glaucum) (Gernand ; Ishii ), some oat embryos contained all pearl millet chromosomes, and embryo rescue allowed hybrids possessing the haploid genomes of both species to be obtained (Ishii ). It appears that the pearl millet chromosomes had incorporated centromeric oat histones (Ishii ), but these haploid oat–millet F1 hybrids developed necrosis and died. This may have been caused by incompatibility between the species or non-ideal tissue culture conditions. Crosses between wheat and grain pearl millet (Pennisetum americanum) or oat and maize both allowed individual chromosomes from one species to be incorporated into the other. In the case of wheat and grain pearl millet from 958 hybridizations, one wheat plant carrying an additional pearl millet chromosome was identified (Ahmad and Comeau, 1990). Although this chromosome was maintained until flowering, it was not detected in the next generation. Thus, wheat–pearl millet hybrids may be more stable than wheat–maize hybrids, but problems maintaining chromosomes from both parents still appear to exist. Unlike wheat–maize hybrids, maize chromosomes have successfully been integrated into oat. This allowed the synthesis of so-called oat–maize chromosome addition lines that stably inherit single chromosome pairs from maize (Kynast , 2004). As with the pearl millet–oat crosses (Ishii ), stability of the oat–maize addition lines appears to be mediated by incorporation of centromeric oat histones into the maize chromosomes such that proper chromosomal segregation can take place during mitosis (Jin ; Wang ). In some maize–oat lines, C4 characteristics such as abundant transcripts of PEPC or C4-like bundle sheath cell size and vein spacing were detected (Tolley ).

In summary, the findings based on wide hybridization of maize and oat indicate that breeding offers a possible route to incorporate some C4 traits into C3 crops without prior knowledge of the underlying genetics. Although additional parental combinations may exist that allow greater trait stability in progeny, this approach has not yet allowed loci controlling C4 traits to be identified. In contrast, quantitative variation in C4 characteristics within a C4 species would allow trait mapping, and there is increasing evidence that this could be informative.

Intraspecific variation in C4 photosynthesis

As PEPC discriminates less than Rubisco against the 13C isotope, a stronger C4 cycle leads to lower incorporation of 13C into tissue and so less negative δ13C values (Leary, 1988). Intraspecific variation in δ13C has been reported in maize and Gynandropsis gynandra (Voznesenskaya ; Kolbe and Cousins, 2018; Kolbe ; Reeves ; Twohey ). To our knowledge, the extent to which this variation in C4 efficiency is caused by differences in Kranz anatomy, cell biology, or C4 biochemistry has not been determined but, as summarized next, variation in some of these traits within a species has been reported. This includes variation in vein density in maize (Yabiku and Ueno, 2017; Kolbe and Cousins, 2018) as well as bundle sheath cell size in Alloteropsis semialata (Lundgren ) and G. gynandra (Reeves ). Thus, natural variation in Kranz anatomy is found within species of C4 monocotyledons and dicotyledons. Statistical modelling suggests that evolution of enlarged bundle sheath cells and vein density were among the first changes to occur during the transition from C3 to C4 photosynthesis (Williams ), and phylogenetic reconstructions reveal that these changes probably happened in response to reduced water availability (Edwards and Smith, 2010). As bundle sheath cell size and vein density were found to be correlated with water use efficiency in maize (Yabiku and Ueno, 2017) and G. gynandra (Reeves ), it is possible that analysis of C4 accessions adapted to different water availabilities will allow additional examples of intraspecific variation in Kranz anatomy to be identified.

While bundle sheath cells are always greener in C4 compared with C3 species, the proportion of leaf tissue allocated to bundle sheath cells compared with the mesophyll cells can be caused by either increased bundle sheath cell size or vein density (Sedelnikova ). Interestingly, within G. gynandra, these characteristics co-vary and correlate negatively with one another (Reeves ). In addition to variation in Kranz anatomy in a species, there is also evidence that the cell biology of C4 leaves can differ. For example, some accessions of Panicum coloratum possess a suberized bundle sheath whilst others do not (Ohsugi and Murata, 1985). There is also variation in chloroplast organization, with some accessions arranging chloroplasts centrifugally and others centripetally compared with veins (Ohsugi and Murata, 1985). Interestingly, Cynodon dactylon, an NAD-ME subtype with centripetal chloroplasts and a suberized bundle sheath, hybridizes naturally with Chloris that uses PEPCK as the primary C4 acid decarboxylase, has centrifugally arranged chloroplasts, and no suberization of the bundle sheath (Prendergast, 1987). F1s demonstrated intermediacy for these traits (Prendergast, 1987). Thus, these species offer an interesting system to study regulators of bundle sheath cell biology.

To our knowledge, there are no clear examples of quantitative variation in the extent to which accessions of an individual C4 species use the various C4 acid decarboxylases. However, there are two reasons to consider this likely. First, in 26 founder lines of a maize multiparent population, variation in the activities of C4 enzymes has been reported (McMullen ; Kolbe ). As the founders show differences in enzyme activity, it is likely that lines of the mapping population possess similar variation. Accessions of A. semialata (Dunning ) and G. gynandra (Reeves ) demonstrate differences in transcript abundance and so it appears likely that these species will also demonstrate variation in activity of C4 acid decarboxylases. Second, the extent to which the different C4 acid decarboxylases are engaged can vary with the environment. For example, in G. gynandra and maize, increased abundance of transcripts encoding C4 enzymes did not correlate with photosynthetic efficiency (Kolbe and Cousins, 2018; Reeves ) but in G. gynandra they were associated with increased water use efficiency. Additionally, the PEPCK subtype is considered more efficient under lower levels of light since it theoretically requires fewer quanta of light per CO2 molecule fixed (Furbank, 2011; Yin and Struik, 2020). Consistent with this, sugarcane (Saccharum offiniarum) and maize which predominantly use NADP-ME showed lower and higher activities of NADP-ME and PEPCK, respectively, after either shade or salt stress (Omoto ; Sharwood ; Sales ). Increased CO2 leakage from bundle sheath cells has also been reported, and it has been proposed that this is caused by increased use of cytosolic PEPCK compared with the chloroplastic NADP-ME (Sales ). If populations of these species have become reproductively isolated in habitats with distinct light supplies, differences in subtype preference may have evolved. Thus, C4 traits ranging from discrimination against δ13C, C4 leaf anatomy, bundle sheath cell biology, and C4 transcript abundance have been documented within a species. In each case, breeding and quantitative genetics offer an opportunity to identify loci controlling these traits. Within this context, we next assess opportunities associated with quantitative genetics to better understand C4 photosynthesis.

Quantitative genetics and C4 photosynthesis

Quantitative genetics allow traits exhibiting continuous variation to be linked to genomic regions termed quantitative trait loci (QTL). Advances in high-throughput phenotyping relevant to photosynthetic performance (reviewed by Choudhury ; van Bezouw ) mean that quantitative genetics now offers a path to dissect the genetics underlying photosynthesis.

Traditional QTL mapping requires a linkage map (or genetic map) to order loci. Using a population derived from two parents that differ in a trait of interest, associations between the trait and molecular markers can identify genes in close proximity to the trait (Mauricio, 2001). Advantages of QTL mapping are that limited knowledge of the genome is necessary and producing bi-parental populations is relatively rapid (Fig. 3A). Recombinant inbred lines (RILs) can be produced, for example, from a segregating F2 generation through rounds of self-fertilization and so generate an immortalized population that can be genotyped once but phenotyped repeatedly. This is especially useful for heritability estimates and mapping QTL in different environments or years (Broman, 2005). Due to considerable differences in the biochemistry and physiology of C3 and C4 plants, if mapping populations derived from C3 and C4 parents of Atriplex, Alloteropsis, or Flaveria were generated, QTL mapping could probably associate genes with a wide variety of C4 phenotypes. Alloteropsis semialata could be of particular interest here because of the presence of both C3 and C4 subspecies that hybridize to produce offspring with intermediate characteristics (Bianconi , Preprint). As self-fertilization is also possible, a population of RILs could be designed specifically for the investigation of C4 traits. High-throughput phenotyping combined with the convoluted neural network Mask R-CNN (He ) has been used for QTL mapping of C4-relevant traits in biparental populations. This allowed rapid assessment of thousands of images and identification of QTL for stomatal traits such as size and density (Xie ).

Fig. 3.

Quantitative genetics in the context of C4 photosynthesis. (A) A schematic for QTL mapping of leaf anatomical traits. Two homozygous parents, genotyped for four markers, A, B, C, and D, and differing in vein density are hybridized and advanced to form a bi-parental population that can be used to identify QTL associated with vein density (here located near markers C and D). Numbers show recombination fractions, which are used to position the QTL relative to flanking markers. (B) Population structure of a MAGIC pedigree followed by four generations of intercrossing and self-fertilization. Progeny contain more genetic variation than that derived from a bi-parental design. Hypothetical plot showing how QTL associated with individually mapped C4 phenotypes such as gene expression, bundle sheath cell size, or gas exchange parameters (e.g. stomatal conductance, CO2 assimilation, etc,) can be mapped with one population.

Although QTL mapping is used extensively, its power is limited if the trait is responsive to the environment and so has low heritability. The heritability of many C4 traits remains poorly understood, but there is growing evidence that variations in CO2 fixation processes and leaf anatomy exist (Table 1) and so estimates of heritability of such C4 traits should be possible. Given the complexity of photosynthesis, its ability to respond to the environment, and temporal variation in its efficiency, it is highly likely that low-heritability traits will be encountered (Flood ). Although traits with low heritability can be investigated using highly controlled environments, highly inbred populations in combination with high-density marker systems are necessary to capture the multiple small-effect QTL contributing to the low-heritability trait of interest. An alternative approach involves genome-wide association studies (GWAS) or linkage disequilibrium (LD) mapping, which identifies markers such as single nucleotide polymorphisms (SNPs) that are in LD with the phenotype of interest (Tam ). GWAS does not require a segregating population but rather uses many diverse accessions that represent thousands of years of recombination to capture multiple alleles, allowing marker groups (haplotypes) to be identified in close association with causal loci. Additionally, it has the advantage of being feasible for obligate outcrossers. In order to work successfully, GWAS requires many markers since it relies on LD decay (Mackay and Powell, 2007) and, as pedigrees are unknown, physical maps are also needed. Although population structure increases the number of false positives derived from GWAS (Korte and Farlow, 2013), this is increasingly being overcome by statistical modelling (Cortes ). GWAS has identified QTL associated with photosynthetic performance during chilling in maize (Strigens ) and sorghum (Ortiz ). More recently, a sorghum diversity panel of 756 African accessions was described (Faye ) and a diverse 869 line panel (Valluru ) was subjected to GWAS to identify genes controlling stomatal conductance and water use efficiency (Ferguson ; Pignon ). The latter two studies used transcriptome data to allow transcriptome-wide association as well as GWAS (reviewed by Wainberg ) to increase the likelihood of identifying candidate genes. Association mapping has also been used to study the light-dependent reactions of photosynthesis (van Bezouw ) but, to our knowledge, QTL determining differences in C4 carbon fixation or Kranz anatomy have not yet been identified. The sorghum and maize mapping panels present an avenue through which targeted phenotyping of C4-specific traits could be used to identify genes responsible for the C4 syndrome. For example, if a gene controlling bundle sheath cell size was identified through mapping in maize or sorghum, this could then be introduced in a C3 crop such as rice to determine whether this allowed engineering of this trait.

Table 1.

Summary of publications documenting intraspecific variation in traits relevant to C4 photosynthesis-associated traits

Species	Varying trait	Reference
Alloteropsis semialata (C4 accessions)	Abundance of PEPC and PEPCK transcripts	Dunning et al. (2017)
	PEPC contentCarbon isotope discriminationMesophyll cell sizeBundle sheath cell sizeLeaf physiology	Lundgren et al. (2016)
Gynandropsis gynandra	C4 transcript abundance, physiology, and leaf morphology	Reeves et al. (2018)
Panicum coloratum	Chloroplast locationBundle sheath suberization	Ohsugi and Murata (1985)
Setaria italica	Carbon isotopeDiffering intensities of green’	Lightfoot et al. (2016)
Sorghum bicolor	Net assimilation rate	Kataria and Guruprasad (2012)
Zea mays	CA transcript abundance	Zhang et al. (2015)
Zea mays	CA, PEPC, and Rubisco activityNet assimilation rateInterveinal distanceMesophyll thicknessMaximum assimilation rate	Kolbe and Cousins (2018)
	CA, PEPC, and Rubisco activityC4 transcript abundanceCarbon isotope	Kolbe et al. (2018)
	Vein densityGas exchange traitsPEPC, NADP-ME, PEPCK, and Rubisco activity	Yabiku and Ueno (2017)

CA, carbonic anhydrase, NADP-ME; NADP-dependent malic enzyme; PEPC, phosphoenolpyruvate carboxylase; PEPCK, phosphoenolpyruvate carboxykinase.

Association mapping can be combined with specific breeding pedigrees to capture multiple recombination events, account for population structure, and so allow higher resolution mapping. These include nested-association mapping (NAM) and multiparent advanced generation inter-crossing (MAGIC) population designs. Both address issues with GWAS and capture more allelic variation than bi-parental populations. Whilst allelic diversity is reduced in these multiparent designs compared with GWAS, linkage mapping as well as association mapping are possible, and this is particularly useful when a physical map is not available (Broman ). Thus, NAM and MAGIC are currently particularly relevant for C4 photosynthesis because although annotated genome sequences are being developed for, for example, Alloteropsis sp., Flaveria sp., and G. gynandra, complete and well-annotated genomes for many C4 model species have not yet been developed. The NAM design involves crossing one recurrent parent with many other accessions. Progeny from each cross are initially bulked and then self-fertilized for multiple generations, leading to multiple RIL families (one family per unique founder) that then constitute the final NAM population (Yu ; McMullen ). At least two NAM populations exist for maize (Yu ; Chen ) and, as mentioned above, significant variation for δ13C as well as CA, PEPC, and Rubisco activities has been reported in the founder lines (Zhang ; Kolbe ; Twohey ). Despite this, QTL for these traits have to our knowledge not yet been determined. A sorghum NAM population has been used in conjunction with an association panel to identify QTL for grain filling (Tao ). NAM populations offer the chance to study an extremely divergent line, such as a pre-domesticated species in the background of a stable population. This has been done with teosinte and maize as the recurrent parent (Chen ). Given the noted differences in maize and teosinte photosynthetic capacity (Yabiku and Ueno, 2017), this offers an interesting resource to map traits that differ between these species.

The MAGIC design also relies on homozygous founder lines that differ in traits of interest. Intercrossing for multiple generations allows segregating populations to be formed consisting of lines that capture the founder genomes in unique recombinants (Fig. 3B). Such segregating lines then undergo self-fertilization for several generations to generate RILs that capture multiple allele combinations from the various parents (Cavanagh ). With MAGIC, haplotype diversity is not limited by the use of a single recurrent parent (Ladejobi ) and, although the MAGIC design requires large amounts of hybridization and significant time to produce the final population (Huang ; Pascual ; Ongom and Ejeta, 2017; Mahan ), simplified strategies can be implemented (Stadlmeier ). In the context of C4 photosynthesis, MAGIC RILs are available for maize and sorghum (Dell’Acqua ; Ongom and Ejeta, 2017; Mahan ; Butrón ). Additionally, transcriptome data exist for the founders of one maize MAGIC population (Dell’Acqua ) and 94 of the MAGIC RILs (Baute ). Should these RILs possess variation in activity of C4 enzymes or components of Kranz anatomy, QTL could be identified. To our knowledge, there is currently no MAGIC population available for a C4 dicotyledon, nor a mapping panel designed explicitly to map C4 photosynthetic traits. As variation in C4 traits has been reported in A. semialata and G. gynandra (Lundgren ; Reeves ) and they can be crossed (Sogbohossou ; Bianconi ), mapping resources in these species would be useful.

Once a QTL is identified using any of the above population types, fine mapping enables causative genes to be identified (Hormozdiari ; Tam ). Parsing C4 photosynthesis into individual components, such genes controlling C4 enzyme activity or bundle sheath cell size (Dunning ) are identified by different phenotyping techniques which, combined with fine mapping, could identify additional genes required for C4 photosynthesis. Exploiting the high degree of natural variation among C3 and C4 species will enable genome-wide associations to help map critical photosynthesis regulators. Furthermore, inferences into the inheritance of C4 components such as cell-specific gene expression can be parsed even without proper segregation or recombination in C3 and C4 hybrids (Fig. 4). While such methods cannot identify QTL, they can at least establish broad modes of inheritance (Charlesworth and Willis, 2009). For example, sterile F1 populations derived from C3 and C4 parents that show altered transcript abundance or cellular localization of C4 enzymes can provide insight into whether genes are controlled in cis, trans, or a combination of both mechanisms, and whether these mechanisms are functioning in an activating or repressive manner (Fig. 4). This technique has been deployed in F1 hybrids derived from a cross between the C3–C4 intermediate Moricandia arvensis and the C3M. moricandiodes to show that cis-regulation dominates control of photosynthetic and anatomical phenotypes (Lin , Preprint). Information from such studies could inform mapping strategies and marker placement for associations.

Fig. 4.

Using breeding to understand the molecular basis of C4 gene regulation. Parental populations that differ in transcript abundance can be due to multiple genetic effects that can be parsed by quantitative genetics. A simplified two loci model where one locus is a cis-element and the other an activating trans-factor is presented to illustrate how the molecular basis underpinning variations in gene expression can be determined by inheritance of gene expression in F1 hybrids. If expression of a gene is controlled by changes in cis-regulation between parents, offspring exhibit additive expression patterns. If variation in expression is due to changes in trans between parents, then offspring exhibit dominance deviation towards one parent. Lastly, if differences in gene expression between parents is due to both cis and trans factors, offspring demonstrate heterosis or overdominance.

In summary, in order to modify C3 leaves to perform C4 photosynthesis, an improved understanding of C4 anatomy, cell biology, and biochemistry is needed. Wide hybridization by either sexual or asexual means to recombine interspecific variation found in C3 and C4 species or intraspecific photosynthetic variation in C4 species, combined with mapping populations and high-throughput phenotyping, should facilitate a better understanding of C4 photosynthesis. Quantitative genetics then offer robust methods to better understand the regulatory mechanisms behind these traits. Applying these techniques therefore promises to enhance photosynthetic efficiency of C3 and C4 crops and so contribute to a more robust world agriculture in the future.