Two ecotypes of Phragmites communis, i.e., swamp reed and desert-dune reed, were obtained in southern margin of the Badan jilin desert in Northwest China. Swamp reed (SR) naturally grows in a nameless rivulet; desert-dune reed (DR) grows on natural sand dunes (Wang et al., 1995; Cui et al., 2009). They are typical sympatric populations distributing a narrow area with about 6.5 km² (39°31′–58′N, 100°4′–36′E; elevation 1300 m). The mean annual precipitation of the sampling site is 118 mm, with the annual potential evaporation of 2392 mm and daily fluctuation of air temperature (data from local Meteorological Bureau). Every year, the leaf tissues of the two ecotypes were collected at the beginning of June. For avoiding single clone, nine plots with 50 m in distance were selected for three biological replicates of each ecotype (Supplementary Figure S1). After harvesting, the samples were rapidly frozen and stored at -80°C for protein extraction, or immediately used for physiological analysis.

To evaluate the water status in leaf tissues, three physiological indicators, i.e., water content, levels of free and bound water, water potential, were determined in the field. The water content was calculated as the difference between the fresh and dry masses and expressed on a fresh weight basis. The levels of both free and bound water were determined according to the method described (Zhou, 2000). Water potentials were directly monitored using a dew-point water potential meter (WP4C, Decagon, USA). For every assay, three replicates were performed.

Protein fractionation was performed according to a previous protocol (Cui et al., 2005, 2009) with modifications. Leaf tissues (1.0 g) from SR and DR were ground, respectively, to a fine powder in liquid nitrogen. For the soluble fraction of the proteins, extraction buffer (50 mM Tris-HCl, pH 7.8, 10% glycerol, 2% β-mercaptoethanol, 1 mM PMSF, 1 mM EDTA) was added and homogenized. Ultrasonic crushing was carried out under the condition of 10 cycles of 10 s/10 s to promote protein extraction. After centrifugation at 43,600 × g for 30 min, the supernatant containing the soluble proteins was collected into a new test tube. The pellet was frozen again in liquid nitrogen and ground in the same buffer. After ultrasonication and centrifugation, the resulting supernatant was pooled into the test tube. Following the removal of the soluble proteins, the remaining pellet was extracted for insoluble proteins. Specifically, the pellet was suspended in 100 mM phosphate buffer (pH 7.1) containing the detergent CHAPS (4% w/v), 0.2 M KCl, 2 mM MgSO₄, 10% glycerol, 2% β-mercaptoethanol, 1 mM PMSF and 1 mM EDTA. After ultrasonication and incubation at 4°C for 30 min, 7 M urea, 2 M thiourea and 30 mM DTT were added and incubated for an additional 30 min at room temperature. After centrifugation at 18°C, the supernatant was collected as the first part of the insoluble fraction. Further, the remaining pellet was re-suspended in a solution containing ASB-14, TBP, and ampholyte (3/10) for obtaining more insoluble proteins. After centrifugation, the second supernatant was pooled together with the above protein solution as an intact insoluble fraction. For purifying protein samples, phenol extraction coupled with saturated ammonium acetate precipitation was applied as described previously (Cui et al., 2012). The protein concentration in the fractions was determined using our established method (Cui et al., 2009). Unless otherwise stated, all extraction steps were carried out at 4°C.

For labeling with DIGE dyes, lyophilized protein samples were re-suspended in lysis buffer, and then labeled reciprocally with Cy3 and Cy5 according to the manufacturer’s instructions (GE Healthcare). An internal standard generated by pooling equal amounts of proteins from each sample were labeled with DIGE-specific Cy2 (Supplementary Table S1). Three biological replicates (including 12 samples) were carried out in the experimental set.

Pairs of Cy3- and Cy5-labeled protein samples and a Cy2-labeled internal standard were mixed together and subjected to IEF (24-cm IPG strips at pH 4-7, GE Healthcare) and SDS-PAGE, as described previously (Cui et al., 2009, 2012). After scanning using the Typhoon 9400 Imager (GE Healthcare), the DIGE images were analyzed using DeCyder v6.5 software. Spot intensities were normalized based on the internal standard labeled by Cy2. The ratios of soluble and insoluble proteins were obtained for each reed ecotype using the DeCyder software. Based on the ratios (Student’s t-test, P < 0.05), the proportion of each protein spot in the soluble and insoluble fraction was calculated. The rate of protein translocation, i.e., the difference of insoluble fraction between SR and DR, was obtained. The protein spots with more than 25% in translocation rate were selected and subjected to MS identification.

For protein identification, traditional 2-DE gels stained with Coomassie brilliant blue (CBB) were prepared using the same parameters (see Supporting Information). The spots of interest were picked and submitted to in-gel digestion with trypsin. After washing and dehydrating, the spots were covered with trypsin solution (12.5 ng/μl in 50 mM NH₄HCO₃) for 45 min in ice. Trypsin digestion was performed overnight at 37°C and stopped by adding 5% formic acid. Extracted peptides were analyzed by MALDI TOF/TOF MS (Ultraflex III, Bruker Daltonics, Germany) as previously described (Cui et al., 2012). In brief, samples were spotted onto a MALDI target plate using cyano-4-hydroxycinnamic acid matrix. Mass data acquisitions were piloted automatically by an auto Xecute method within the Flex Control software v3.0. In MS mode, spectra were acquired in a mass range m/z 800–4500 by summing up 2000 laser shots with an acceleration of 23 kV. The MS spectra were externally calibrated using Peptide Calib Standard II (Bruker Daltonics) (1046.542, 1296.685, 1347.735, 1619.822, 2093.086, 2465.198, and 3147.471), resulting in mass errors of <50 ppm. The MS peaks were detected with a minimum signal/noise (S/N) ratio > 20 and cluster area S/N threshold > 25 without smoothening and raw spectrum filtering. Peptide precursor ions corresponding to contaminants including keratin and the trypsin autolytic products were excluded in a mass tolerance of 0.2 Da. For acquiring MS/MS spectra, the filtered precursor ions with a user-defined threshold (S/N ratio > 50) were selected. Fragmentation of precursor ions was performed using the LIFT positive mode. MS/MS spectra were accumulated from 4000 laser shots. The MS/MS peaks were detected on a minimum S/N ratio > 3 and a cluster area S/N threshold > 15 with smoothing. Mass spectra were piloted using Flex Analysis software.

The obtained MS and MS/MS spectra per spot were combined, and submitted to MASCOT search engine by Biotools 3.1 (Bruker Daltonics). Parameters selected included: MS/MS Ion Search, the NCBInr database of green plants created on 31 July 2016 (91579380 sequences; 33741759339 residues), trypsin of the digestion enzyme, up to one missed cleavage site, parent ion mass tolerance at 100 ppm, MS/MS mass tolerance of 0.5 Da, carbamidomethylation of cysteine (global modification), and methionine oxidation (variable modification). The probability score (95% confidence level) calculated by the software, and a matching of at least 2 peptides was used as a criterion for correct identification. The MS/MS data set for all proteins are provided in Supplementary Table S2.

To obtain the full-length cDNA of the rbcL and rbcS genes, DOP-PCR (degenerated oligonucleotideo-primed-PCR), 3′-RACE (rapid-amplification of cDNA ends) and 5′ TAIL-PCR (Thermal asymmetric interlaced polymerase chain reaction) were performed by using the double-stranded cDNA of SR and DR as a template, respectively. All primers, conditions and cycle settings for PCR were provided in Supplementary Tables S3, S4. The resulting PCR products were ligated into vector TA, cloned, and sequenced. Every sequence was confirmed by examining both strands; in addition, at least two bacterial clones obtained by TA cloning were examined. By assembling the sequences of 3′, 5′ and the core partial sequences on Contig Express (Vector NTI Advance11.5.1), the full-length cDNA sequence of rbcL and rbcS gene was deduced. The nucleotide sequence data for rbcL and rbcS, from SR and DR, have been deposited in the GenBank nucleotide sequence databases under accession no. KF697233, KF697234, KF697235, and KF697236, respectively. Finally, the sequence alignments between SR and DR were determined using AlignX (Vector NTI Advance11.5.1) (see Supplementary Figure S2).

For determining both the amount and activity of Rubisco, leaf soluble proteins were obtained from SR and DR using extraction buffer containing 40 mM Tris-HCl (pH 7.6), 10 mM MgCl₂, 0.25 mM EDTA, 5 mM glutathione, 2% (v/v) β-mercaptoethanol and a small amount of PVP-40. The extraction process was repeated once, and the resulting supernatants were pooled. In the process, no supersonic step was conducted to avoid loss of the enzyme activity. For quantifying the Rubisco content specifically, proteins from the supernatant were precipitated with methanol/chloroform according to the method of Wessel and Flügge (1984). IEF and SDS-PAGE for 2-DE were carried out as described previously (Cui et al., 2009). Specifically, only 200-μg protein samples were loaded onto 24-cm IPG strips (pH 4-7) in order to alleviate the saturation effect of the highly abundant protein. The resulting 2D gels were stained using CBB and digitized with a calibrated densitometer (GS-800, Bio-Rad). The images were analyzed using PDQuest software v7.1.1 (Bio-Rad). The intensity level of the Rubisco protein (including the large and small subunits) was obtained by expressing the intensity of each spot in a gel as a proportion of the total protein intensity detected using the entire gel. The Rubisco contents were calculated based on the loading amount.

Rubisco activity was measured by rate of NADH oxidation at 340 nm according to a method described (Reid et al., 1997) with minor modifications. Briefly, the initial Rubisco activity was measured at 340 nm by adding 50 μL of the supernatant to 950 μL of assay buffer containing 40 mM Tris-HCl (pH 7.6), 0.25 mM EDTA-Na₂, 10 mM MgCl₂, 5 mM glutathione, 0.2 mM NaHCO₃, 5 mM ATP, 0.5 mM NADH, 5 mM creatine phosphate, 0.5 mM RuBP, 10 units of phosphocreatine kinase, 10 units of glyceraldehydes-3-phosphate dehydrogenase and 10 units of phosphoglycerate kinase. All regents listed above were obtained from Sigma (USA). Subsequently, the mixture was measured at 340 nm every 30 s for 3 min. Rubisco activity was expressed in CO₂ nmol min^-1 mg^-1Fresh weight/Soluble proteins/Rubisco.

According to the method above, soluble proteins were extracted in the presence or absence of the reducing agent β-mercaptoethanol (ME) buffer. The amount and activity of Rubisco were determined. Three biological repeats were carried out for evaluating the effect of the reductant on the different distribution of Rubisco in SR and DR.

The two ecotypes of Phragmites, SR and DR, are typical sympatric populations with contrasting habitats of water availability (Cui et al., 2002, 2009). Table 1 lists five morphometric parameters measured at the mature plant stage to evaluate the effect of environmental stresses on growth and development. Significant differences between SR and DR were evident for all five traits (P < 0.01). Undoubtedly, harsh desert habitats lead to growth retardation of DR, resulting in a 3.2 and 10.6-fold reduction in height and biomass, respectively, in comparison to SR with a good water supply (Figure 1 and Table 1).

Considering the huge difference in the water supply of the habitats, the leaf water content of both SR and DR was carefully measured. Unexpectedly, the water contents were highly similar, i.e., 61.2% (SR) and 60.8% (DR) (Figure 2A). There were no significant differences in any of the leaves collected from any leaf position. The similarity did not match the moisture state in their habitats. To explore the inherent reason for this result, we aimed to test the water activity through two independent methods. First, water components (free water and bound water) were measured (Figure 2B). As expected, the free water was remarkably decreased in DR, with a concomitant increase in the bound water. The level of bound water in DR (31.9%) was more than twofold higher than that in SR (15.1%). Second, we monitored the leaf water potential in a real time in the field using a portable water potential meter (WP4C). Compared with the potential of -2.07 MPa in SR, the DR value was more negative (-3.8 MPa) (Figure 2C), implying that the water molecules were bound to other biomolecules in DR. Therefore, we conclude that the water activity in DR was down-regulated for coping with the water deficit in its habitat. The underlying causes of the strategy need to be investigated.

In our previous proteomic research, a step-by step workflow extraction was established for obtaining entire proteins from the leaves of Phragmites (Cui et al., 2009). We noticed that the protein contents were similar between SR and DR, but the percentages of the soluble and insoluble proteins were dramatically altered in the two ecotypes. Compared with SR, DR exhibited a very low level of soluble proteins, and a high level of insoluble proteins (Cui et al., 2009). The alteration in the protein components showed a similar trend as that observed for the water components.

To confirm the finding and decipher the underlying meaning of the information, we collected reed leaf samples from their natural habitats in 2009, 2012, and 2014 for the current protein fractionation assay. The fractional protocol was optimized for maintaining the integrity of each protein fraction (see Materials and Methods; Supplementary Figure S3). Figure 3 shows the results that were generated from three sampling years in 2009, 2012, and 2014. Based on the optimized fractionation method and dozens of experiments, we confirmed that the soluble proteins in DR accounted for about 28.7% (on average) of total proteins, a lower level compared with 55.6% for SR. On the other hand, the insoluble proteins in DR appeared to exhibit an increasing trend, i.e., 71.4% on average (Figure 3). We assume that regulation of the protein proportion between the soluble and insoluble fractions plays a role in DR resistance to environmental stresses.

For analyzing the protein proportion alteration, high-resolution 2D gels were generated within a pH range of 4–7 based on 24-cm IPG strips. Figure 4A shows the four protein profiles corresponding to each fraction of DR and SR. We observed that numerous, highly abundant proteins, which were identified in our previous paper (Cui et al., 2009), were shared in two fractions, but with varying proportions (Figure 4A). For instance, two subunits (C60α and C60β) of the chloroplast chaperonin 60 kDa were mainly distributed in the soluble fraction of SR, but emerged dominantly in the insoluble fraction of DR (Figure 4Ba); the large subunit of Rubisco (RL) and the alpha subunit of ATP synthase (CF1α) were located mainly in the soluble fraction of SR, but were highlighted in the insoluble fraction of DR (Figure 4Bb), as was 2-cystenin peroxiredoxin (2-Cys Prx) (Figure 4Bc).

For obtaining accurate quantitative information on the alteration in protein distribution, we further designed and performed a comprehensive proteomic analysis using the DIGE technique. Pairs of Cy3- and Cy5-labeled proteins from the soluble and insoluble fractions, as well as a Cy2-labeled internal standard were mixed together and then subjected to IEF and SDS-PAGE. Triplicates with independent biological samples were carried out in the experimental set. Two representatives of the DIGE maps, corresponding to SR and DR, are shown in Figure 5. Using the DeCyder v6.5 BVA (Biological Variation Analysis) module, approximately 1730 protein spots were detected on the DIGE gels. Based on the soluble/insoluble protein ratios in each reed ecotype, the protein translocation rate, i.e., the value of the translocating into the insoluble fraction of DR in comparison to SR, was calculated and listed in Table 2 and Supplementary Table S5. A user-defined threshold (translocation ≥ 25%) was employed to identify the translocation proteins.

In comparison to SR, 93 protein spots shifted ≥ 25% into the insoluble fraction of DR (Figure 5). The majority of the protein spots (about 80%) were identified by combined analysis of MALDI TOF/TOF MS and databases using the previous method (Cui et al., 2009). Supplementary Table S2 describes the detailed information pertaining to the protein identification. Based on the established features of metabolism (Buchanan et al., 2015) and the functional classification of the Gene Ontology¹, the proteins were classified into five functional categories, i.e., photosynthesis, transport ATPases, protein synthesis, protein folding and stability, as well as disease/defense (Table 2). Nineteen proteins with an unknown function could not be identified using current databases (see Supplementary Table S5).

Six proteins are involved in the dark reaction of photosynthesis (Table 2). The most conspicuous change occurred in Rubisco, the chloroplast protein with the greatest relative abundance (six isoforms in spot 156 and spots 210, 123, 120, 70). We found that the majority (66.9 and 65.7%) of the large and small subunits of Rubisco was distributed in the soluble fraction of SR; the remainder (33.1 and 34.3%) was in the insoluble fraction of SR. The two-phase distribution could be associated with the stroma and thylakoid membrane of the chloroplast (Jin et al., 2006). That is, most of the Rubisco in SR was located and functioned in the stroma of the chloroplast, whereas only a small amount was located in the thylakoid membrane. On the contrary, only 14.1 and 13.4% of the large and small subunits of Rubisco were distributed in the soluble fraction of DR compared with 85.9 and 86.6%, respectively, in the insoluble fraction of DR. By contrast, 52.8 and 52.3% of large and small subunits of Rubisco shifted, respectively, to the insoluble fraction of DR. The distribution result indicates that the majority of Rubisco was sequestered in the thylakoid membrane of DR. Interestingly, the protein translocation mainly occurred in the main product with an observed molecular weight of 54 KDa (spots 156). Another protein Rubisco activase (RCA), an activator of Rubisco, had similar behavior (Figures 5, 6A). At least 4 isoforms of RCA (10 spots) with different molecular weights and isoelectronic points altered their binding status in the chloroplasts (Table 2). Soluble RCA in SR accounted for about 58.8% (on average), whereas the proportion decreased significantly to 14.0% in DR. The result also implied that the majority of RCA was attached to the thylakoid membrane of DR. As early as 2001, a similar translocation response was reported in spinach leaves during heat treatment (Rokka et al., 2001). In addition, a proportional change in RCA in the stroma and thylakoids has been observed in antisense rca rice (Jin et al., 2006). We speculate that the distribution of RCA in the chloroplasts is dynamically regulated based on alterations in environmental conditions. Other enzymes functioning in the dark reaction of photosynthesis, including glyceraldehyde-3-phosphate dehydrogenase (spot 318, Figure 6A), fructose-1,6-bisphosphatase (spot 5), and phosphoribulokinase (spots 169, 420), exhibited similar dynamic changes in DR (Table 2). In addition, three subunits of ATP synthase, i.e., α (spots 149, 146), β (spot 166), and 𝜀 subunit (spot 596), also exhibited the dynamic changes. Because the three subunits are components of the soluble catalytic core F₁, we suspect that the change in the fractions reflects a model of activity regulation of the ATPase complex.

Twelve proteins identified are involved in protein synthesis, folding, and stability. In the category of protein synthesis, four proteins (spots 567, 592, 600, and 571) belong to ribosomal protein, and one protein is the translational elongation factor Tu (spots 206, 573). Compared with SR, these proteins apparently translocated (>25%) into the insoluble fraction of DR (Table 2). Notably, eight proteins functioning in protein folding and stability also exhibited altered distribution proportions in the soluble and insoluble fractions (Table 2). They are α and β subunit 60-kDa chaperonin (Cpn60, 14 spots), two HSPs with 70 kDa (spots 131, 108, Figure 6B) and a 90-kDa HSP (spot 560), as well as peptidyl-prolyl cis–trans isomerase (PPI, spot 580, Figure 6C), lumenal binding protein (spot 436) and the ATP-binding subunit of ATP-dependent Clp protease (spots 214, 239). With respect to Cpn60, we identified at least 3 isoforms of Cpn60-α and 4 isoforms of Cpn60-β according to differences in molecular weight (Table 2). Compared with SR, the two subunits of Cpn60 in DR exhibited a significantly altered proportion between the soluble and insoluble fractions. About 45.7% of the proteins were transferred to the insoluble fraction in DR. Although there is evidence that Hsp60 in animal cells undergoes changes in subcellular compartments under stress conditions (Itoh et al., 2002), the inherent physiological function is unknown in plant cells.

In the categories of disease/defense, four chloroplast proteins were found to translocate evidently into the insoluble fraction in DR. They are 2-Cystenin peroxiredoxin (2-Cys Prx, four spots), peroxiredoxin-2E-2 (spot 602), thioredoxin M (spot 597) and plastidic glutamine synthetase (GS2, spots 132, 574). Plant 2-Cys Prx is a key participant in ROS responses, reduce alkyl hydroperoxidase, hydrogen peroxide and even complex lipid peroxides (König et al., 2003). The chloroplast protein usually presented in the form of 3–4 isoforms on a 2D map (Figures 4Bc, 6D), revealed a 35:65% distribution between the soluble and insoluble fractions in SR, and a 10:90% distribution in DR. Compared with SR, the protein translocated about 35% (mean of four isoforms) to the insoluble fraction of DR (Table 2). Two other proteins, peroxiredoxin-2E-2 and thioredoxin M, are also two components linked to ROS responses. They are also dominantly distributed in the insoluble fraction of DR. Considering the oxidizing environment caused by environmental stresses in DR cells, the hydrophobization of proteins associated with ROS metabolism might be positive for protection itself.

In addition, a 14-3-3 protein (spot 185) and an adenosine kinase-like protein (spot 579) as well as two RNA binding proteins (spots 7 and 598) presented a significant translocation into the insoluble fraction of DR. Nineteen protein with an unknown function could not be identified using current databased (Supplementary Table S5). Their roles in the translocation sub-proteome need to be characterized in future.

To provide an explanation for the protein translocation, Rubisco was selected for the following experiments because of its abundance and high translocation rate (Figure 5 and Table 2). By gene cloning and sequencing, full-length cDNAs of the rbcL and rbcS genes were obtained separately from SR and DR, and then deposited in the GenBank. There were no differences between the nucleotide sequences of Rubisco from SR and DR, for either the large (1431 bps) or small (516 bps) subunits (see Supplementary Figure S2). The results suggested that the protein translocation did not result from the amino difference in the sequence and probably depended on a conformational change in the Rubisco itself.

Given the inevitable oxidative stress in a desert environment, we analyzed both the amount and activity of Rubisco in the presence and absence of reductant. The amount of Rubisco was calculated based on the intensity of each protein spot. We found that the extracting buffer with the reductant ME could significantly increase the extracted amounts of soluble proteins in DR, but hardly impacted the proteins in SR (Figure 8). Notably, the effect of the reductant on Rubisco was more efficient. In the presence of ME, the amount of soluble proteins and Rubisco increased by 20.8 and 128.7%, respectively, in DR. The results suggested that some soluble proteins, including Rubisco in DR, were released from the insoluble fraction. The activity assay of Rubisco supported this observation. The addition of the reductant remarkably increased Rubisco activity in DR, expressed especially by the amount of soluble proteins (Figure 9B). By contrast, the effect of the reductant on Rubisco was limited in SR. The difference in the effect of the reductant between SR and DR was reflected in the differences in the state of Rubisco. Zhao and Zhang (1993) reported that Rubisco from DR contained about 50% of titratable SH groups compared with that from SR, suggesting that the harsh desert environment induced a conformational change in Rubisco. We speculate that the conformational change allowed the enzyme to attach to the thylakoid membrane, resulting in high levels of insoluble proteins in DR. Interestingly, if expressed as the amount of Rubisco, the enzyme had higher activity in DR than in SR (Figure 9C). Taken together, the Rubisco distribution between the soluble and insoluble fractions was dynamic and partially regulated by cellular redox status. We propose that the translocation proteome, including Rubisco, may be an unknown plant response to stressful environments.