Molecular phylogeny and forms of photosynthesis in tribe Salsoleae (Chenopodiaceae)

Evolution of C3–C4 intermediate and C4 lineages are resolved in Salsoleae (Chenopodiaceae), and a model for structural and biochemical changes for the evolution of the Salsoloid form of C4 is considered.


Introduction
Reconstructing the evolution of C 4 photosynthesis is challenging as it requires the complex coordination of anatomical, ultrastructural, biochemical, and gene regulatory changes from C 3 ancestors (Hibberd and Covshoff, 2010;Gowik and Westhoff, 2011;Sage et al., 2012Sage et al., , 2014Williams et al., 2012;Hancock and Edwards, 2014). In bringing together these aspects, a model of C 4 evolution where Kranz anatomy is formed around individual veins has been developed over the last 30 years, which includes potential evolutionary precursors and a number of transitional, evolutionary-stable states (Monson et al., 1984;Edwards and Ku, 1987;Sage, 2004;Gowik and Westhoff, 2011;Sage et al., 2012Sage et al., , 2014. These hypothetical states are based on distinct phenotypes observed in nature in close relatives of C 4 lineages and are characterized by a combination of C 3 and C 4 characteristics. From these, a stepwise progression from C 3 to proto-Kranz to photosynthetic intermediates, and finally to C 4 photosynthesis was proposed with a progressive reduction in photorespiration (Sage et al., 2014; hereafter named the 'Flaveria model' based on photosynthetic phenotypes studied in this genus).
In dicots, there are many anatomical forms of Kranz anatomy that differ in the arrangement of a dual layer of chlorenchyma cells performing the C 4 pathway. These includes forms where Kranz anatomy develops around individual veins; however, there are also nine forms where two concentric chlorenchyma layers surround all veins (Edwards and Voznesenskaya, 2011). According to Brown (1975), in C 4 plants we refer to cells of the inner chlorenchyma layer that become specialized for C 4 photosynthesis, irrespective of their position in the leaf, as Kranz cells (KC) and the outer layer as mesophyll (M) cells (Edwards and Voznesenskaya, 2011;Voznesenskaya et al., 2013). In C 3 -C 4 intermediate phenotypes the inner layer of chlorenchyma, which has become specialized to support the C 2 cycle, is referred to as Kranz-like cells (KLC; Voznesenskaya et al., 2013). In C 3 species vascular bundles (VB) are surrounded by non-specialized parenchymatic bundle sheath (BS) cells.
Proto-Kranz phenotypes, first described in Heliotropium and Flaveria, are suggested to represent the initial phase of C 4 evolution where overall vein density is increased and BS cells have an increased number of organelles, with enlarged mitochondria located internally to chloroplasts in a centripetal position towards the VB (Muhaidat et al., 2011;Sage et al., 2012Sage et al., , 2013Sage et al., , 2014. C 3 -C 4 intermediate phenotypes, which have been found in grasses and in a number of dicot families, have in common increased development of chloroplasts and mitochondria in the KLCs. Both M and KLC chloroplasts have Rubisco and the C 3 cycle. In the KLCs there is a distinctive layer of mitochondria that are located internally to the chloroplasts in a centripetal position. In C 3 -C 4 intermediate phenotypes glycine decarboxylase (GDC) is selectively localized in the KLC mitochondria, which support a C 2 cycle by establishing a photorespiratory CO 2 pump. In the C 2 cycle photorespiratory glycine produced in the M cells is shuttled for decarboxylation by GDC to the KLCs where photorespired CO 2 is concentrated, enhancing its capture by KLC Rubisco (see Edwards and Ku, 1987;Sage et al., 2012Sage et al., , 2014Voznesenskaya et al., 2013;Khoshravesh et al., 2016).
For the 'Flaveria model' C 3 -C 4 intermediate phenotypes have been classified into two general groups: Type I and Type II C 3 -C 4 species ; alternatively called Type 1 C 2 and Type 2 C 2 , Sage et al., 2014). Type I C 3 -C 4 species have developed little or no capacity for function of a C 4 cycle as activities/quantities of C 4 enzymes are low, similar to C 3 species. These intermediates mainly reduce losses of the CO 2 generated by photorespiration by its partial refixation in the KLCs. Type II intermediates have substantial expression of a C 4 cycle; e.g. the levels of the C 4 cycle enzymes phosphoenolpyruvate carboxylase (PEPC), pyruvate phosphate dikinase (PPDK), and NADP-malic enzyme (NADP-ME) are two-to five-fold higher in Type II C 3 -C 4 species than in C 3 species (Ku et al., 1983(Ku et al., , 1991Edwards and Ku, 1987;Moore et al., 1987;Muhaidat et al., 2011;Sage et al., 2012). The values of CO 2 compensation points (Γ) in C 3 -C 4 intermediate phenotypes are in between those of C 3 and C 4 species.
The fact that C 3 -C 4 intermediate phenotypes thrive, persist, and occasionally have been found in lineages without any C 4 relatives, suggests that they represent an evolutionary-stable condition in their own right (Monson et al., 1984;Edwards and Ku, 1987). Their predominant occurrence close to C 4 groups may be strongly biased by the more intensive screening in these lineages. Thus, C 3 -C 4 intermediate phenotypes do not necessarily represent transitional states that always lead to the establishment of C 4 photosynthesis. A C 2 cycle might already be favourable in conditions of high photorespiration, e.g. in hot, dry, and saline environments (Keerberg et al., 2014). The 'Flaveria model' is functionally plausible, and supported by phenotypes that actually exist in nature (Sage et al., 2014); however, phylogenetic evidence for the ancestry of the C 3 -C 4 intermediate condition is scarce, and is hampered by the generally low number of species with intermediate phenotypes.
C 3 -C 4 intermediate phenotypes have been recognized in 16 angiosperm genera (Sage et al., , 2014Khoshravesh et al., 2016). Often, the ancestry of the C 3 -C 4 intermediate condition is inferred from a sister-group relationship of a C 4 lineage and a C 3 -C 4 intermediate lineage  because the intermediate condition is a priori considered as less derived. However, in such cases it is impossible to distinguish between ancestry and a de novo evolution of the C 3 -C 4 intermediate condition (compare with Hancock and Edwards, 2014). If those cases in which C 3 -C 4 intermediate photosynthesis seems to precede C 4 photosynthesis, as suggested in Sage et al. (2011;, are critically tested for unequivocal phylogenetic evidence, only Flaveria (Asteraceae) studied by McKown et al. (2005) holds up. In this case, a stepwise acquisition of C 4 photosynthesis in one lineage of Flaveria was shown (McKown et al., 2005;Lyu et al., 2015).
There are four other promising groups that are rich in C 3 -C 4 intermediate phenotypes and therefore potentially informative lineages in terms of disentangling the steps of C 4 evolution and ancestral state reconstruction for C 3 -C 4 intermediacy: Blepharis (Fisher et al., 2015), Anticharis (Khoshravesh et al., 2012), Heliotropium , and Salsoleae sensu stricto (s.s.; Voznesenskaya et al., 2013). A better understanding of C 3 -C 4 intermediate phenotypes in Salsoleae is particulary important as these, in contrast to the other groups, seem to deviate from the 'Flaveria model' (see 'Salsoleae model', Voznesenskaya et al., 2013).
Salsoleae, especially the former section Coccosalsola, has long been known to contain C 3 and C 4 species (Winter, 1981;Akhani et al., 1997). In fact, Salsoleae has repeatedly been suspected to contain species that represent reversions from C 4 back to C 3 photosynthesis (Carolin et al., 1975;P'yankov et al., 1997;Kadereit et al., 2014); however, this has been questioned by Kadereit et al. (2003). According to a survey by Voznesenskaya et al. (2013, see table 5) there are at least 21 species with δ 13 C values within the typical range of C 3 species in Salsoleae. So far, four of these have been shown to possess either proto-Kranz (Salsola montana), or a C 3 -C 4 intermediate phenotype (S. arbusculiformis, S. divaricata, and S. laricifolia (Voznesenskaya et al., 2013 and references therein;Wen and Zhang, 2015). In the 'Flaveria model' Kranz anatomy is formed around individual veins, requiring a series of anatomical changes in progression from C 3 to C 4 . In Salsoleae, however, the photosynthetic tissue in leaves forms a continuous layer that surrounds all the vascular and water-storage tissue, i.e. in C 3 species by multiple layers of mesophyll tissue (Sympegmoid-type anatomy), and in C 4 species by a dual layer of chlorenchyma tissue forming a Kranz anatomy (Salsoloid-type anatomy). Voznesenskaya et al. (2013) proposed a model for transitions from C 3 to proto-Kranz to C 3 -C 4 intermediates to C 4 in Salsoleae, based on limited photosynthetic phenotypes, which would require very different changes in leaf anatomy and regulation of development of the dual layer of chlorenchyma cells compared to that in the 'Flaveria model' for development of Kranz anatomy around individual veins.
Here, we conduct a large-scale analysis of Salsoleae, including species with C 3 -type δ 13 C values. The results of a molecular phylogenetic study of 74 species and an anatomical survey of 77 species of Salsoleae s.s, and some outgroup species, are presented. Furthermore, in a search for additional C 3 -C 4 intermediates in the tribe, anatomical, ultrastructural, enzyme content, and gas exchange analyses were performed on a number of species that have C 3 -type δ 13 C values. Molecular clock and character optimization analyses were used to reconstruct the evolution of the C 4 pathway in Salsoleae. The following questions were addressed. (1) Is there evidence for additional C 3 -C 4 intermediates in Salsoleae? (2) What is the current model for evolution of C 4 in Salsoleae based on analyses of photosynthetic phenotypes? (3) In what ways does this model differ from the 'Flaveria model' proposed in Sage et al. (2012Sage et al. ( , 2014? (4) Where and when did C 4 photosynthesis originate in Salsoleae, and is there phylogenetic evidence for a reversion from C 4 back to C 3 ? (5) Does the C 3 -C 4 intermediate condition represent an ancestral state to C 4 in Salsoleae?

Plant material and sampling
Species and samples included in the analyses with their respective voucher information are listed in Supplementary Table S1 at JXB online. We used herbarium samples and plants grown in the greenhouse at the Botanical Gardens of the University of Mainz, Germany, and at Washington State University (WSU), Pullman, WA, USA, and leaves of specimens that were fixed during various expeditions, mainly by H. Freitag. A few samples were kindly provided by other institutions. Species of the Kali clade were mostly left out, because in the trees based on chloroplast sequence data they are separated from the Salsoleae s.s.
Our sampling of Salsoleae for the molecular phylogenetic analyses comprised 74 species representing all currently accepted genera (Supplementary Table S2) and included 15 species with C 3 -like carbon isotope (δ 13 C) values (compare with table 5 in Voznesenskaya et al., 2013). Furthermore, representatives of other primary clades of Salsoloideae (Supplementary Table S2) as well as representatives of Camphorosmoideae were included. For rooting and dating purposes Suaedoideae and Salicornioideae were sampled as outgroups (Supplementary Table S1). For light microscopy 77 species were examined, mostly from the same material (Supplementary Table S1), including 34 species studied for the first time. Data for eight species were taken from the literature. Six species having different anatomical types (most not previously known in that respect) were chosen for study by electron microscopy, and δ 13 C, in situ immunolocalization, western blot, and gas exchange analyses.
With respect to nomenclature, apart from a few exceptions we follow previous accounts of the different subfamilies, in particular Botschantzev (1989), , and Kadereit and Freitag (2011), although we are aware that more nomenclatural adjustments are required.

Sequencing and phylogenetic inference
Total DNA was extracted from dried or fresh leaf material using the DNeasy Plant Mini Kit (QIAGEN, Germany) or innuPREP Plant DNA Kit (Analytik Jena, Germany) following the manufacturers' protocols but increasing incubation times. PCRs for five markers (atpB-rbcL intergenic spacer, ndhF-rpL32 spacer, trnQ-rps16 spacer, rpl16 intron, ITS) were carried out in a T-Professional or T-Gradient Thermocycler (Biometra, Germany), or a PTC100 Thermocycler (MJ Research, USA). Primers sequences, PCR recipes, and cycler programs are documented in Supplementary Table  S3. PCR products were checked on 0.8% agarose gels and purified using the NucleoSpin® Gel and PCR clean-up-Kit (Macherey-Nagel, Germany) or ExoSAP (Affymetrix, USA) following the manufacturers' instructions. The Big Dye® Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems) combined with the primers mentioned above was used for the sequencing reactions, followed by a purification step using IllustraTM SephadexTM G-50 Fine DNA Grade (GE Healthcare, UK). Sequencing was performed following the Sanger method on a 3130xI Genetic Analyzer (Applied Biosystems Inc., USA). The raw forward and reverse sequences were checked and automatically aligned in Sequencher 4.1.4 (Gene Codes Corporation, USA). The refined alignment was performed in Mesquite 2.75 (http://mesquiteproject.org) and carefully checked visually. The program SequenceMatrix (v. 1.7.8; Vaidya et al., 2011) was used to combine the four chloroplast (cp) marker data sets.
Phylogenetic analyses under the settings outlined below were initially conducted individually for the five selected DNA regions. Results of the individual analyses of the four cp markers revealed no topological conflict [i.e. incongruence with ML Bootstrap ≥65% and Posterior Probability (PP) ≥0.90] among individual markers and combination of the cp markers distinctly increased the resolution and support values. For the cp data further analyses were performed using two large and different data sets: (1) a data set with 106 taxa (106t data set) including a broad outgroup sampling; and (2) a data set with 75 taxa (75t data set) in which only Salsola genistoides served as the outgroup. The 106t data set was used to reveal primary clades in Salsoloideae and for estimation of divergence time, whereas the 75t data set was used for character optimization (see below). Since the ITS tree and the cp tree showed supported conflict at basal branches, and the combination of cp data and ITS for the 75t data set led to a significant decrease of resolution and support values in some parts of the tree, we concentrated on the cp tree for further analyses and only include the ITS tree for comparison. First, the best-fitting substitution model for the combined cp data sets was inferred using jModel-Test (Posada, 2008). CIPRES (Cyberinfrastructure for Phylogenetic Research) Science Gateway V. 3.3. ML phylogenetic analyses were performed using RAxML (Stamatakis, 2006;Stamatakis et al., 2008), including bootstrapping that was halted automatically following the majority-rule 'autoMRE' criterion. Bayesian inference (BI) analysis was conducted using BEAST (Bayesian Evolutionary Analysis by Sampling Trees v.1.8.2; Drummond and Rambaut, 2007) with GTR+G (general time-reversible; best-fitting according to jModeltest under AIC criterion) with a gamma-distribution in four categories as the substitution model. A birth-and-death demographic model was used as the tree prior. Markov Chain Monte Carlo (MCMC) analysis was performed with the following settings: randomly generated starting tree, 20 000 000 (106t data set) or 10 000 000 iterations (75t data set), discarded burn-in of 10%, and sampling every 1000 steps (totalling 10 000). For the 106t data set a relaxed clock model was implemented in which rates for each branch are drawn independently from an exponential distribution (Drummond et al., 2006). The crown node of Salsoloideae and Camphoromoideae was set to 47.0-25.5 mya based on divergencetime estimates in the Chenopodiaceae/Amaranthaceae complex (Kadereit et al., 2012). We assumed a uniform distribution within the age bounds set. Other settings were left in default.

Light and electron microscopy
For light microscopy, after routine checks by manual sectioning, the middle parts of well-developed leaves were selected for transections by a rotary microtome (Leitz 1515). The semi-thin sections of material fixed in FAA [2% (v/v) formaldehyde, 0.5% (v/v) acetic acid, 70% (v/v) ethanol] were studied under a Dialux 20 (Leitz, Wetzlar). Some were first examined and photographed in water to get a better contrast between lignified (blue) and non-lignified (purple) cell walls before embedding into Depex (Serva) for documentation. For detailed study, middle parts of fully developed leaves were fixed and processed in a similar way to that described in Voznesenskaya et al. (2013).
For screening purposes, leaf samples from herbarium specimens were first boiled for about 1-3 minutes and hand-cut sections were preserved in glycerol-gelatin. Selected samples for microtome transections were soaked in a 10% solution of NH 3 for 10 d, dehydrated in ethanol, and embedded in Technovit 7100 (Heraeus Kulzer). The samples were sectioned at 5-20 µm using a rotary microtome. Sections were stained in a 6:6:5:6 mixture of Azur II, Eosin Y, methylene blue, and distilled water and mounted in Eukitt (O. Kindler) after drying. Images of the sections were taken using a Leitz Diaplan light microscope combined with Leica Application Suite 2.8.1.
For ultrastructural characterization, ultra-thin sections were taken from the same samples prepared for the light microscopy study and embedded in Spurr's resin as described in Voznesenskaya et al. (2013). The number and sizes of mitochondria in chlorenchyma cells were estimated per cell section (about 10-15 cell images from 2-3 separate leaf samples) using an image analysis program (ImageJ 1.37v, https://imagej.nih.gov/ij/index.html).
δ 13 C values Carbon isotope composition of plant samples was determined at Washington State University using a standard procedure relative to PDB (Pee Dee Belemnite) limestone as the carbon isotope standard (Bender et al., 1973). Leaf samples (from plants growing in the WSU growth chamber) were dried at 60 °C for 24 h, and then 1-2 mg were placed in a tin capsule and combusted in a Eurovector elemental analyser. The resulting N 2 and CO 2 gases were separated by gas chromatography and admitted into the inlet of a Micromass Isoprime isotope ratio mass spectrometer (IRMS) for determination of 13 C/ 12 C ratios (R). δ 13 C values were determined where δ = 1000 × (R sample /R standard ) − 1.
In situ immunolocalization Sample preparation and immunolocalization by transmission electron microscopy (TEM) were carried out according to Voznesenskaya et al. (2013). The antibody used (raised in rabbit) was against the P subunit of glycine decarboxylase (GDC) from Pisum sativum L. (courtesy of D. Oliver, Iowa State University). Pre-immune serum was used for controls. The density of labeling was determined by counting the gold particles on electron micrographs and calculating the number per unit area (µm 2 ) using ImageJ 1.37v. For each cell type, replicate measurements were made on parts of cell sections (n = 10-15 cell images). Immunolabeling procedures were performed separately for different species; the difference in the labeling intensity reflects the difference between cell types but not between species. The level of background labeling was low in all cases.

CO 2 compensation point
Measurements of CO 2 compensation points (Γ) were made on an individual lateral branch using a Li-Cor lighted chamber (LI-6400-22L; Li-Cor Biosciences, Lincoln, NE, USA) designed for terete or semiterete conifer leaves. For each species, a part of a branch of an intact plant was enclosed in the chamber and illuminated with a PPFD of 1000 μmol quanta m −2 s −1 under 400 μmol mol −1 CO 2 at 25 ºC until a steady-state rate of CO 2 fixation was obtained (generally 45-60 min). For varying CO 2 experiments, the CO 2 level was first decreased, and then increased up to 400 μmol mol −1 at 5 min intervals. Γ was determined by extrapolation of the initial slope of rate of CO 2 fixation (A) versus the intercellular CO 2 concentration in the leaf (C i ) through the x-axis where the net rate of CO 2 assimilation equals zero. The leaf area exposed to the incident light was calculated by taking a digital image of the part of the branch that was enclosed in the chamber, and then determining the exposed leaf area using ImageJ 1.37v.

Statistical analysis
Where indicated, standard errors were determined, and ANOVA was performed using Statistica 7.0 software (StatSoft, Inc.). Tukey's HSD (honest significant difference) test was used to analyze differences between amounts of gold particles in BS/KLC/KC versus M for each species, and δ 13 C and Г values in different species. All analyses were performed at the 95% significance level.

Character coding and analyses of character evolution
Analyses of character evolution were conducted for five traits: (1) type of photosynthesis; (2) KC/KLC function; (3) life form; (4) leaf sclerenchyma; and (5) leaf reduction (Table 1). Traits were optimized over 1000 trees of 74 Salsoleae and Salsola genistoides as the outgroup obtained in a Bayesian analysis (see above) using the ML criterion in Mesquite (http://mesquiteproject.org). The fit of singleversus two-rate models was tested for traits with two character states using a likelihood ratio test. Table 1 gives information about the coding of the character states of the five traits. Trait 1: type of photosynthesis according to carbon isotope value; coding trait 2: C 3 = 0, C 4 = 1. Trait 2: function of bundle sheath (BS), Kranz-like (KLC) and Kranz cells (KC); coding trait 2: C 3 type BS cells around peripheral VB (few or no organelles) = 0, KLC with increased number of organelles mostly in centripetal position, GDC only expressed in KLC cells, C 3 -C 4 species = 1, C 4 type Kranz cells = 2. Trait 3: Life form according to standard floras and our own observations in the field; coding trait 3: perennial l = 0, annual = 1. Trait 4: presence of sclerenchyma by replacement of major parts of the central water storage tissue; coding trait 4: no = 0, yes = 1. Trait 5: sites of major photosynthetic function; coding trait 5: predominantly in leaves and leaf-like structures (0), ± equally in leaves and stems (1), predominantly in stems due to reduction of leaves or their trans-formation into thorns (2). References for photosynthetic pathway and leaf anatomy see below table. Species are classified according to carbon isotope composition of leaf tissue as C 3 or C 4 . Species are classified as C 3 , proto-kranz, C 3 -C 4 intermediates, and C 4 based on analyses of leaf anatomy, Trait 2 and C isotope composition, (see text).   (2015). * In Voznesenskaya et al. (2013) this species was mentioned to have the C 4 pathway. However, samples for the carbon isotope value were taken from a wrongly identified specimen [D. Podlech 44954 (P)]. The correct identification for this specimen is Salsola oppositifolia, which indeed is a C 4 species. ** C 4 metabolism deduced from leaf anatomy, no carbon isotope values available. *** Classified as sympegmoid in Freitag and Duman (2000), Pyankov et al. (2001) and Khatib (1959).

Leaf anatomy
Light microscopy Figure 1A shows that S. abrotanoides (C 3 ) mostly had two layers of palisade M cells. The peripheral vascular bundles (VBs) were surrounded by a layer of bundle sheath (BS) cells, which looked similar to the adjacent cells of the water-storage (WS) tissue but were smaller. Four species (all C 3 -C 4 ), R. regelii, S. deschaseauxiana, S. gymnomaschala, and S. verticillata (Fig. 1B-E) were similar in having two layers of chlorenchyma cells underneath the epidermis. While the inner layer consisted of elongated palisade cells (M2), the cells of the outer layer (M1) were 2-5 times shorter with a reduction to varying degrees in the number of chloroplasts depending on species and growth conditions. In R. regelii the M1 cells were elongated (Figs 1B and 2), but in the other species the M1 cells appeared almost globular to polyhedral in shape; they were wider than the M2 palisade cells and more similar to the typical hypodermis of C 4 species (Fig. 1C-E). These species had a clearly defined continuous (or almost continuous in R. regelii) layer of chlorenchymatous Kranz-like cells (KLCs), which was situated above and between the peripheral VBs.
In S. oppositifolia (C 4 ) the palisade M2 cells were rather short and M1 cells were represented by a typical hypodermis consisting of large globular cells that were almost devoid of chloroplasts. The KCs had organelles in a centripetal position; they formed a continuous layer just beneath the palisade cells (Fig. 1F).
Among the Salsola species in this study, C 3 S. abrotanoides had the lowest volume of WS tissue. In R. regelii, the inner part of the WS tissue was replaced by massive sclerenchyma tissue, which accounted for half of the leaf diameter and for  the stiff appearance of the leaves (Fig. 2). Crystal-bearing idioblasts were preferentially located in the hypodermis or hypodermis-like layer, and in the Kranz-like layer between the peripheral bundles, but they could occur scattered elsewhere. In S. gymnomaschala and in S. verticillata the epidermis was partially doubled. In all Salsola species the main vein was located more or less in the center of the leaf and surrounded by 2-4 layers of WS tissue.
Transmission electron microscopy Figure 3 shows obvious differences in the quantity, position, size, and level of development of BS cell organelles in the C 3 species and in the corresponding KLCs in intermediates and the KCs of the C 4 species. C 3 Salsola abrotanoides (Fig. 3A, B) had the lowest number of organelles in BS cells; a few chloroplasts and mitochondria were distributed more or less evenly along the cell wall, with some mitochondria located in a centrifugal position. The structure of the thylakoid system was similar for BS and M chloroplasts (Fig. 3C, D).
In the KLCs of four species identified as C 3 -C 4 intermediates, R. regelii (Fig. 3E, F), S. deschaseauxiana (Fig. 3I, J), S. gymnomaschala, and S. verticillata (not shown), the chloroplasts were at least twice as numerous (per cell section) than those of the BS in S. abrotanoides; they were distributed along the cell wall but tended to be enriched in the centripetal position. The mitochondria were also twice as numerous (per cell section) and 1.5-2 times larger than in BS cells of S. abrotanoides; and most of them were located in the centripetal position, close to the inner periclinal or radial cell walls (Fig. 3F, J). KLC chloroplasts (Fig. 3G, K) and M chloroplasts (Fig. 3H, L) in R. regelii and S. deschaseauxiana (and the other two Salsola intermediates, not shown) had a similar structure with a well-developed system of medium-sized grana consisting of 7-11 thylakoids.
The KCs in C 4 S. oppositifolia contained numerous organelles in the centripetal position (Fig. 3M, N). The chloroplast structure differed remarkably among M cells and KCs: while the M chloroplasts had small to medium-sized grana of 2-5 thylakoids in stacks (Fig. 3P), the KC chloroplasts had numerous single thylakoids that interconnect small grana of paired thylakoids, or a few grana consisting mostly of 3-5 thylakoids (Fig. 3O).
Mitochondria in BS and M cells of S. abrotanoides had a similar size and structure (~0.4 µm), whereas in the KLCs of S. deschaseauxiana, S. gymnomaschala, S. verticillata, and R. regelii they were about 1.3-1.5 times larger compared to the M cells. In KCs and M cells of S. oppositifolia the mitochondria were almost identical in size (~0.5 µm).

Immunolocalization of GDC
In situ immunogold labeling for GDC using the antibody to the P protein was examined by electron microscopy, and a quantitative analysis was made based on the density of gold particles, in C 3 S. abrotanoides, C 3 -C 4 R. regelii, S. deschaseauxiana, S. gymnomaschala, and S. verticillata, along with the C 3 -C 4 species S. divaricata and C 4 S. oppositifolia. Analysis of the immunolabeling distribution showed that there was no significant difference in density of the gold particles between the mitochondria of M and BS cells in C 3 S. abrotanoides (Fig. 4, Supplementary Fig. S2). In contrast, in the C 4 species S. oppositifolia gold particles were selectively localized in KC mitochondria with low labeling in M mitochondria, with a 10-fold difference in their number. In the intermediates R. regelii, S. deschaseauxiana, S. gymnomaschala, and S. verticillata, as well as in S. divaricata, the number of gold particles was also ~5.8-10 times higher in KLCs compared to M mitochondria (Fig. 4).

Western blot analysis of key C 4 enzymes
Immunoblots for the key C 4 cycle enzymes PEPC, PPDK, NAD-ME, and NADP-ME from total soluble proteins extracted from leaves of the studied species are presented in Fig. 5. The C 4 species S. oppositifolia had very high labelling for the C 4 pathway enzymes, PEPC and PPDK, and the two decarboxylases, NADP-ME and NAD-ME. Compared to the C 4 species, the C 3 species S. abrotanoides and the C 3 -C 4 intermediates R. regelii, S. deschaseauxiana, S. gymnomaschala, S. verticillata, and S. divaricata had very low labelling for the C 4 cycle enzyme PPDK and, to varying degrees, less labelling for PEPC, NAD-ME, and NADP-ME.

Molecular phylogeny of Salsoleae and mapping of key traits
The molecular phylogenetic analysis of the chloroplast genome revealed two unambiguous C 4 lineages in Salsoleae s.s., (1) Halothamnus, and (2) Anabasis clade + Noaea clade + Haloxylon clade. Since the Anabasis clade, Noaea clade, and Haloxylon clade formed a polytomy with two C 3 -C 4 intermediate clades, a higher number of three or four C 4 origins is possible (Supplementary Fig. S3). When using a narrower outgroup (only Salsola genistoides) the two C 3 -C 4 intermediate clades merged into a monophyletic group that still formed a polytomy with the Anabasis clade, Noaea clade, and Haloxylon clade (Fig. 6). From the crown group age of Halothamnus (7.9-1.3 mya), Anabasis (10.2-2.6 mya), and Noaea (10.3-2.8 mya) it can be assumed that in these lineages C 4 photosynthesis has been present since the Late Miocene/ Early Pliocene. Only for the Haloxylon clade can a distinctly older minimum age for the origin of C 4 photosynthesis of 16.9-6.6 mya be inferred from the molecular dating. In the case of common ancestry of the Anabasis clade + Noaea clade + Haloxylon clade, C 4 photosynthesis might date back to 19.2-7.6 mya ( Supplementary Fig. S3). The ML character optimization inferred a perennial life form and fully developed leaves without massive central sclerenchyma as the ancestral condition in Salsoleae s.s. (Fig. 6). An annual life form evolved at least six times independently in the tribe. A massive central sclerenchyma also evolved repeatedly (Fig. 6). In some cases, this feature was characteristic at the generic level, as in Girgensohnia, Horaninovia, Cornulaca, Raphidophyton, and Noaea. The reduction or complete loss of a true leaf lamina and a shift of photosynthetic function to the young stems was a common feature in Salsoleae and evolved multiple times in C 4 lineages of Salsoleae s.s., but also in S. genistoides. The occurrence of leafless species was clustered in certain genera, such as Anabasis, Haloxylon, and Hammada, the latter of which seemed to be highly polyphyletic.
Furthermore, the ML character optimization inferred a C 3 metabolism and C 3 -type BS cells as ancestral in Salsoleae s.s. According to the ancestral character state reconstruction, a switch towards C 4 seems to have already occurred along the branch leading to the large sister group of the C 3 species Salsola pachyphylla, which contains three C 4 subclades but also one clade of C 3 and C 3 -C 4 intermediates (highlighted green in Fig. 6). This clade of C 3 and C 3 -C 4 intermediate species did not contain any C 4 species, and the clade was part of a polytomy of C 4 clades; thus, there is no indication in the cp tree that the C 3 -C 4 intermediates represent ancestral states leading towards full C 4 photosynthesis.
Resolution in the ITS tree was weak in many parts of the tree (Supplementary Fig. S4). Combining cp and ITS data resulted in very low resolution (tree not shown) due to conflicting topologies. Branches that were in conflict between the two data sets (with bootstrap >75) are marked on the ITS tree ( Supplementary Fig. S4).

Evidence for newly identified C 3 -C 4 species in Salsoleae
Results from gas exchange (Γ), compartmentation of GDC between M cells and KLCs, analyses of carbon isotope composition, and analyses of levels of C 4 enzymes, along with the structure of the respective cells, indicated that four species, S. deschaseauxiana, S. gymnomaschala, S. verticillata, and R. regelii, are C 3 -C 4 intermediates, while  S. abrotanoides operates C 3 photosynthesis. Analyses of the carbon isotope composition of these four intermediates as well as the C 3 -C 4 intermediate S. divaricata showed they all have values in the range of those of C 3 species compared to the C 4 -type value in S. oppositifolia (Table 2). However, values for plants grown in growth chambers are more negative (i.e. up to 4-7‰) than samples from natural habitats . The more positive δ 13 C values in the natural habitat may be due to growth under arid conditions limiting CO 2 diffusion into leaves (Cerling, 1999), or to induction of a partially functional C 4 cycle. According to our study, among Salsoleae there are at least 19 species with C 3 isotope values, seven of which are C 3 -C 4 species (Table 1). Fig. 6. Molecular phylogenetic tree of Salsoleae s.s. (Chenopodiaceae) based on four cp markers (atpB-rbcL spacer, ndhF-rpl32 spacer, trnQ-rps16 spacer, rpl16 intron) and 74 representative species. The tree was calculated using the program package BEAST (posterior probabilities are shown above branches) and was rooted with Salsola genistoides. Character optimization was conducted using Mesquite and ancestral conditions are indicated for selected nodes (pL = proportional likelihood).
Structural, biochemical, and functional analyses are needed in order to determine whether species having C 3 -type δ 13 C values are C 3 , proto-Kranz, or C 3 -C 4 . An important test is measurement of Γ, since values are lower in C 3 -C 4 than in C 3 plants, which is indicative of a reduction in photorespiration . Gas exchange analyses of S. deschaseauxiana, S. gymnomaschala, S. verticillata, and R. regelii showed that all these species have Γ values that are intermediate between C 4 S. oppositifolia and C 3 S. abrotanoides (Table 2). Additionally, C 3 -C 4 intermediates, like C 4 species, have selective compartmentation of GDC in KLC mitochondria (Rawsthorne et al., 1988;Voznesenskaya et al., 2001Voznesenskaya et al., , 2013Sage et al., 2012Sage et al., , 2014, supporting refixation of photorespired CO 2 . Analysis of GDC levels by immunolocalization in these four intermediates indicated selective localization in mitochondria of Kranz-like cells (KLCs), while in the C 3 species S. abrotanoides the density of immunolabeling for GDC was similar in M and BS mitochondria.
Western blot analysis of C 4 enzymes showed that levels in the C 3 species S. abrotanoides and the C 3 -C 4 intermediates S. deschaseauxiana, S. gymnomaschala, S. verticillata, and R. regelii were very low compared to the C 4 species S. oppositifolia. The levels of PEPC in the four C 3 -C 4 intermediate species were higher than in the C 3 species S. abrotanoides. However, except for R. regelii, levels of PPDK were low and barely detectable in both the C 3 and intermediate species. R. regelii had higher levels of PPDK, but low levels of C 4 decarboxylases similar to the C 3 species.
Currrently, the results suggest that all seven known C 3 -C 4 species of Salsoleae, R. regelii, S. arbusculiformis, S. deschaseauxiana, S. divaricata, S. gymnomaschala, S. laricifolia, and S. verticillata (Voznesenskaya et al., 2001Wen and Zhang, 2015; this study, Table 1), are Type I, where the reduction of Г comes from refixation of photorespired CO 2 in KLCs with little or no function of a C 4 cycle . Whether there is a contribution from a limited C 4 cycle to photosynthesis in these intermediates could be more directly analyzed by the method of Alonso-Cantabrana and von Caemmerer (2016) via online measurements of photosynthesis and carbon isotope discrimination.

A model for evolution of C 4 photosynthesis in Salsoleae based on identified photosynthetic phenotypes
Of the 77 species of Salsoleae analyzed, without those of the Kali clade (Table 1), 24 species were studied for the first time. Our sampling was comprehensive and surpasses Carolin et al. (1975) with 43 species and Pyankov et al. (2001) with 38 species. Of the 77 species, 19 had C 3 -type carbon isotope composition (consisting of seven C 3 species, five proposed proto-Kranz species, and seven C 3 -C 4 intermediates) while 58 were C 4 species with C 4 -type carbon isotope composition and Salsoloid-type leaf anatomy (Table 1). In Fig. 7 five photosynthetic phenotypes in Salsoleae are described based on the anatomical, ultrastructural, and biochemical analyses of species in the current study together with a few species described by Voznesenskaya et al. (2013). In this model, C 4 is proposed to have evolved structurally and functionally from C 3 Sympegmoid to Proto-Kranz Sympegmoid to C 3 -C 4 Kranz-like Sympegmoid to C 3 -C 4 Kranz-like Salsoloid to C 4 Salsoloid-type anatomy. There are two subtypes of Kranzlike Salsoloid C 3 -C 4 intermediates (with or without sclerenchyma) and five anatomical subtypes with Salsoloid type.

Non-Kranz anatomy, functionally C 3
The Sympegmoid leaf type is anatomically and functionally C 3 . It is characterized by usually two well-developed layers of palisade M cells (M1 and M2) and indistinct C 3 -type BS cells around peripheral VBs containing only a few organelles. Species of this type have C 3 δ 13 C values, C 3 -type Γ values, and structural features of M and BS cells characteristic of C 3 plants (including the occurrence of GDC in both M and BS mitochondria). It is found in S. abrotanoides (this study, Figs. 1A, 3, 4), S. genistoides, S. oreophila, S. pachyphylla, S. webbii (Carolin et al., 1975;Voznesenskaya, 1976;P'yankov et al., 1997;Pyankov et al., 2001;Voznesenskaya et al., 2013), and Sympegma regelii (Wen and Zhang, 2011). Based on anatomical evidence alone, we conclude that Salsola tianschanica. belongs to this group, which would then comprise seven species in total (Table 1). An additional trait observed in this group is the comparatively low volume of water storage (WS) tissue and the position of peripheral VBs embedded in the WS tissue rather than at its periphery (Fig. 1A). From known data and the taxonomic literature, in particular the pertinent revisions of section Coccosalsola by Botschantzev (1976Botschantzev ( , 1989, the occurrence of this leaf type in other species is unlikely.

Proto-Kranz anatomy, functionally C 3
Proto-Kranz species have anatomical changes in BS cells that may be the earliest phase of C 4 evolution, preceding development of the C 2 cycle (Sage et al., 2014). In Salsoleae, the Proto-Kranz Sympegmoid type only differs from the Sympegmoid type by having distinct cells with chloroplasts and mitochondria arranged preferentially along the inner and the radial walls between peripheral VBs and the chlorenchyma (Fig. 7). Currently this type is only documented in S. montana . It has C 3 -like δ 13 C and Γ values, and immunolabeling for GDC is similar for M and BS mitochondria. However, based on analysis of leaf anatomy (by light microscopy of fresh leaf or herbarium samples fixed in FAA) there are additional probable candidates for proto-Kranz anatomy among the Central Asian Salsola species that have C 3 δ 13 C values, namely S. botschantzevii, S. drobovii, S. junatovii, and S. lipschitzii (Table 1).

Kranz-like anatomy, functionally C 3 -C 4 intermediate
Anatomically there are two types of intermediates in Salsoleae, the Kranz-like Sympegmoid type, and the Kranzlike Salsoloid type (Fig. 7). They resemble C 4 Salsola species in having KLCs with numerous organelles in the centripetal position. Both have C 3 -type δ 13 C values, selective localization of GDC in KLC mitochondria, and intermediate Γ values indicating functionally C 2 -type species.
The Kranz-like Sympegmoid type is very similar to the aforementioned Sympegmoid forms; but the outer M cells (M1) are distinctly shorter and smaller than the inner M cells (M2) and the KLCs are restricted to the peripheral VBs. This type of structure has so far only been found in S. arbusculiformis (Voznesenskaya et al., 2001), and it is suggested to represent the first functional step towards C 4 -type anatomy.
In the Kranz-like Salsoloid type of intermediacy, the M1 appear more like the hypodermal cells in C 4 species, while still containing more or less numerous chloroplasts. M1 cell size and M/KLC ratio is reduced in comparison with the Sympegmoid types, and the KLCs form a more or less continuous layer (interrupted by crystal-containing idioblasts) around the leaf, as in C 4 Salsola species. The KLCs contain chloroplasts and numerous large mitochondria positioned towards the inner cell wall, characteristic of other C 3 -C 4 intermediate species Rawsthorne and Bauwe, 1998;Voznesenskaya et al., 2007Voznesenskaya et al., , 2010Muhaidat et al., 2011;Sage et al., 2012Sage et al., , 2014. This type is currently found in six species (Table 1): S. laricifolia Zhang, 2011, 2015), S. divaricata (Voznesenskaya et al., 2013), and S. deschaseauxiana, S. gymnomaschala, S. verticillata, and R. regelii (this study , Figs 1 and 7). However, R. regelii represents a different subtype by its very strong central sclerenchyma (Fig.  2), which, according to our knowledge, is unique among the C 3 -C 4 intermediates identified in Salsoloideae.
Of note in the Kranz-like Salsoloid type, considerable variation occurs, mainly in the size, shape and the number of organelles in the M1 cell layer and the arrangement of the KLCs. Sometimes multiple sections within species revealed a certain degree of variation, showing phenotypes more similar to the Kranz-like Sympegmoid type or phenotypes approaching C 4 plants with typical Salsoloid leaf anatomy. Therefore, more detailed studies are needed to assess the phenotypic plasticity of the functionally intermediate types.

Kranz-type anatomy, functionally C 4
The Salsoloid leaf anatomy in C 4 lineages of Salsoleae differs substantially from C 4 eudicots having Atriplicoid-type leaf anatomy with Kranz anatomy around individual veins in flat leaves. Species with Salsoloid-type anatomy are functionally C 4 , with a continuous layer of Kranz cells (KC) around WS tissue and VBs. If the M1 layer of cells is present it occurs as a hypodermis with few or no organelles. There is a further reduction in the M/KC ratio, with organelles in the KCs in a centripetal (or, rarely, in centrifugal) position. In other lineages, as in Halothamnus, Noaea, Kali, Nanophyton, and Climacoptera, the hypodermis is lacking (Kadereit et al., 2003;Wen and Zhang, 2011). Our data on S. oppositifolia, and on several other C 4 species that had not previously been studied, do not add substantially to the well-known Salsoloid-type anatomy. Together with 40 other species, S. oppositifolia displayed the most common Salsoloid type that has a hypodermis and lacks central sclerenchyma; on the other hand only 12 species account for the variant with central sclerenchyma. The Salsoloid type without a hypodermis and without central sclerenchyma was represented in our sampling by four species of Halothamnus only. This form is present in many other species of that genus (Kothe-Heinrich, 1993), and in almost all species of Kali (Rilke, 1999), while the variant with a central sclerenchyma was seen only in the two Noaea species.

Comparison with the 'Flaveria model' for C 4 evolution
The various photosynthetic phenotypes in Salsoleae fit the general model of evolution from C 3 to proto-Kranz, to intermediates, to Kranz anatomy with a progressive reduction in functional losses due to photorespiration. However, in the 'Flaveria model' where Kranz anatomy forms around individual veins Sage et al., 2012) the structural modifications are very different from the modifications from a Sympegmoid type to a Salsoloid Kranz type where the Kranz anatomy is formed around all veins and WS tissue. In evolution from C 3 to C 4 with Kranz anatomy around individual veins, there is increased vein density and size of BS cells around veins as they develop KC features. In contrast, in the Salsoleae the vein density in C 4 species does not appear to be higher than in the C 3 species. In addition, the size of the KCs is not significantly increased in the C 4 species in comparison with their forerunners in C 3 species . Furthermore, in the Salsoleae model (see fig. 9 in Voznesenskaya et al., 2013) a decrease in the M/KC ratio might also be a precondition, as in grasses (Christin et al., 2013) and dicots ; however, it happens by development of a continuous layer of KCs and reduction in the M1 layer rather than by an increase of veins and BS size.
Where and when did C 4 photosynthesis originate in Salsoleae?
The chloroplast gene tree of Salsoloideae resolves five primary clades in the subfamily, namely the Nanophyton clade, the Caroxyloneae clade, the Salsola genistoides clade, the Kali clade, and the Salsoleae s.s. clade (Supplementary Fig.  S3; Kadereit and Freitag, 2011), with the first four forming the sister group to Salsoleae s.s. (Supplementary Fig.  S3). Molecular trees based on the nrDNA marker ITS (Supplementary Fig. S4) also reveal these clades but they are contradictory in their positions (for a short discussion on this matter see Supplemetary Fig. S3). According to our results in both data sets only two clearly independent C 4 lineages in Salsoleae seem likely. The first lineage is Halothamnus, with a crown age of 7.9-1.3 mya, and probably plus the Kali clade in the ITS data set. Most species of both clades have Salsoloid leaf anatomy and lack a hypodermis (e.g. as in Traganum). In the cp tree the monospecific C 3 genus Sympegma is sister to Halothamnus while in the ITS tree it is sister to all C 3 -C 4 intermediates and C 4 clades in the Salsoleae sensu lato (s.l., Supplementary Fig. S4). The Halothamnus and Kali clades seem to have no close relatives with a C 3 -C 4 intermediate phenotype. The second C 4 lineage in Salsoleae s.s. probably consists of all other C 4 species. Since the C 4 clades of Anabasis, Noaea, and Haloxylon form a polytomy with the C 3 -C 4 intermediates clade in the cp tree (Fig. 6), one large monophyletic C 4 clade and a sister-group relationship to the C 3 -C 4 intermediates remains possible. Unfortunately, the weak resolution of this particular part of the ITS tree within Salsoleae s.s. does not support this ( Supplementary Fig. S4). The overall similar Salsoloid leaf anatomy with hypodermis (except for the two species of Noaea) and NADP-ME biochemistry is in favor of common ancestry of the C 4 syndrome in this lineage. The predicted age of this large C 4 lineage in Salsoleae s.s. is 19.2-7.6 mya ( Supplementary Fig. S3), which is in accordance with the origin of many other C 4 lineages during the Middle to Late Miocene .

Does the C 3 -C 4 intermediate condition represent an ancestral state to C 4 in Salsoleae? Is there phylogenetic evidence for a reversion from C 4 back to C 3 ?
The ML optimization suggests that early Salsoleae were shrubs or subshrubs that performed C 3 photosynthesis in well-developed leaves with a Sympegmoid leaf type (Fig. 6). Along with one C 3 species (S. oreophila), 10 species (consisting of proto-Kranz and C 3 -C 4 intermediates) in Salsoleae form one (Fig. 6) or two clades ( Supplementary Fig. S3) in the cp trees. Lack of resolution in the phylogenetic trees just at the node where these phenotypes and their closest C 4 relatives arise hampers a reconstruction of these proto-Kranz and C 3 -C 4 intermediates as ancestral (see also Supplementary Fig.  S4). The ML optimization even suggests that the node from which these phenotypes arise was most likely C 4 (pL=0.98, Fig. 6), which would imply the origin of these C 3 -C 4 species from C 4 . However, in general a reversion from C 4 back to C 3 or intermediate phenotypes seems to be exceedingly rare, if not improbable. Although a few cases have been reported in which C 3 species or C 3 -C 4 intermediates are nested in a C 4 clade, there has also been plausible evidence for a scenario of multiple C 4 origins (e.g. Ocampo et al., 2013;Bohley et al. 2015). In the case of Salsoleae s.s., we assume that a convincing reconstruction is severely hampered by the low number of C 3 -C 4 intermediate lineages (not by the low number of intermediate species) since all C 3 -C 4 intermediate species seem to belong to just one lineage.
This C 3 -C 4 intermediate-rich lineage might, however, be of major interest for future studies. Studying the ecology, physiology, and biochemistry of closely related proto-Kranz and C 3 -C 4 intermediate species with and without selective localization of GDC might provide further insights into the selective advantage of proto-Kranz anatomy and the C 2 pathway. So far the selective advantage of displacement of BS organelles towards the centripetal position in proto-Kranz compared to C 3 species is not clear. Possibly the proto-Kranz state leads to a slight increase in refixation of the CO 2 generated by GDC in BS cells; however, this could be difficult to detect experimentally (e.g. by measurements of Γ). A C 2 cycle is indeed able to generate distinctly higher CO 2 levels in leaves (Keerberg et al., 2014) and therefore has an ecophysiological advantage. Since a reversion from intermediate to C 3 still seems possible in this clade (S. oreophila), there is a need for further sampling and deeper resolution within this lineage.
Reduction of leaf lamina combined with photosynthesis being taken over by the green stem cortex evolved multiple times in Salsoleae; however, except for S. genistoides, this is only observed in the C 4 clades. We hypothesize that the higher productivity of the C 4 cycle in Salsoleae allows for a reduction in the surface area and the amount of photosynthetic tissue, reduction of transpiration, and an increase in water use efficiency. This is obviously advantageous in the extremely dry habitats of the Eurasian deserts and semi-deserts that the Salsoleae have successfully colonized.

Conclusions
In the model for evolution of C 4 in Salsoleae, putative C 3 ancestors have M tissue surrounding the entity of veins with a limited volume of WS tissue; differentiation occurs with development of KLCs next to minor veins, reduction in size of M cells, and ultimately development of an internal layer of KLCs that surrounds all the vascular and WS tissue. Compared to the 'Flaveria model' for C 4 development around individual veins, in Salsoleae the proposed biochemical and functional transitions suggest convergence; however, there is obvious divergence in how structural changes were made in C 3 ancestors to develop Kranz anatomy. In Salsoleae, a number of structural changes that are important in the evolution of C 4 flat-leaved species are missing: individual KC size often does not increase, but KC volume increases due to the formation of the continuous layer, a decrease of the M/KC ratio occurs mainly due to the reduction of the M1 layer, and the density of venation does not change. These differences might be related to the succulent nature of Salsoleae, with an increase of the volume of WS tissue during the transition from C 3 to C 3 -C 4 intermediates and to C 4 species.
The Salsoleae phylogenies unambiguously reveal all C 3 species, except for S. oreophila, in basal positions, and both the C 4 species and the C 3 -C 4 intermediates as derived, but occurring in different clades. Intermediates, proto-Kranz, and one C 3 species are clustered in one (cp tree) or four (ITS tree) monophyletic groups that might either be sister to a large C 4 clade or nested within it. In the absence of closely related C 4 species, their intermediacy cannot be determined as ancestral; although they display logical stepwise, 'modelconforming' phenotypes from C 3 to C 4 photosynthesis. From a phylogenetic point of view, they may represent an evolutionarily independent solution, enabling the respective species to survive in harsh environments, even in competition with distantly related species of the same tribe possessing full C 4 photosynthesis.