The Phytozome 10 dataset (Goodstein et al., 2012) of annotated protein sequences and the Pfam dehydrin sequence dataset (Pfam00257) were used as the sources for all dehydrin sequences. Pfam00257 entries were first filtered to remove duplicate sequences and sequences marked as fragments. Selection of the dehydrin sequence was performed as outlined in the flowchart (Figure 1A). By definition, a dehydrin must contain at least one K-segment. The initial K-segment pattern was based on a compilation of K-segment definitions from the literature (Supplementary Table S1). All Phytozome annotated protein sequences were filtered using a regular expression (search pattern) created during the literature search K-segment (Figure 1B), which represents a broad definition of the K-segment. To narrow the definition of dehydrins, the software program MEME (version 4.9.1) was run on the list of dehydrin sequences to create an unbiased narrower search expression for the K-segment (Supplementary Figure S1A) with the software parameters shown in Supplementary Table S4. This search expression was used to create a narrow dehydrin sequence dataset from the Phytozome sequences (Supplementary Table S5 and File S1) and Pfam00257 that was used for all subsequent analyses.

The MEME program was used again as an unbiased approach to determine the Y- and S-segment sequences, and to potentially determine new motifs in the dehydrin family. The dehydrin sequence dataset was used in all searches except for species specific searches where only Phytozome 10 sequences were used. For all motif searches, the ‘any number of repeats’ mode was chosen. Parameters for the searches performed using MEME are shown in Supplementary Table S4, with all other parameters left at their default values. Results of the motif analyses were visualized using the LOGO format (Schneider and Stephens, 1990) and as position-weighted matrices. The statistical significance of discovered motifs is usually examined by calculating the E-value using MEME (McGrath et al., 2007). The E-value represents the expectation value of finding a motif with an equally conserved pattern in random sequences (Bailey et al., 2006). Because dehydrins, like many IDPs, have low sequence complexity regions, we repeated the MEME search using randomly shuffled dehydrin sequences using the program ‘shuffleseq’ (Rice et al., 2000), and calculated the E-value again.

For the GT-motif search, the motif width was varied from 5 to 20 in each individual MEME run. In addition to calculating the E-value, we determined whether the detection of the GT-motif occurred by chance by randomly shuffled the K_n dehydrin sequences enriched in the GT-motif before repeating the MEME search. The shuffling and search were performed five times.

The Shapiro–Wilk test for normality (Shapiro and Wilk, 1965) was calculated using ‘R’ (R Core Team, 2016) to determine if the ϕ-segment sequences are random. Amino acids were converted into the numbers 1–20 for the calculation. Sequences shorter than 30 residues were discarded from the test since they may result in false positives (Razali and Wah, 2011). The Shapiro–Wilk test was performed with the null hypothesis that the ϕ-segment followed a normal distribution. A normal quantile–quantile (Q–Q) plot was used to visualize the results.

We separately searched for dehydrins in the genomes of a lycophyte (Selaginella moellendorffii), non-vascular plants (Marchantia polymorpha, Physcomitrella patens, Sphagnum fallax), and green algae (Chlamydomonas reinhardtii, Dunaliella salina, Coccomyxa subellipsoidea C-169, Micromonas pusilla CCMP1545, Micromonas sp. RCC299, Ostreococcus lucimarinus, Volvox carteri) using BLASTP (Altschul et al., 1997) against known P. patens dehydrins (Ruibal et al., 2012), and searching with MAST (Bailey and Gribskov, 1998) using K-segment motifs found in all vascular plant dehydrins and discovered non-vascular plant dehydrins.

Dehydrin sequences are modular in nature and as such the conserved Y-, S-, and K-segments could occur in any order and with different frequencies. To examine segment order in dehydrin architectures, a script was written to search for the various motifs and report the distances between them. To ensure an as broad as possible number of relevant hits, the two of three rule, as explained in Section “Results,” was used to search for K-segments (Supplementary Figure S1A), Y-segments (Supplementary Figure S1B) and S-segments (Supplementary Figure S1C). Dehydrin sequences were assigned to one of the five common architectures (K_n, Y_nSK_n, Y_nK_n, SK_n, or K_nS). Novel and rare motifs were discovered using MEME (Bailey and Elkan, 1994) and GLAM2 (Frith et al., 2008).

Dehydrin sequences were submitted to the EXPASY server’s isoelectric point (pI) and molecular mass (M_r) calculator to compute the theoretical pI and M_r (Gasteiger et al., 2005). GRAVY scores were calculated using the Sequence Manipulation Suite (Stothard, 2000). The molecular masses, pI values, GRAVY scores, and motif compositions were visualized in R using the bean plot package (Kampstra, 2008).

Patterns of dehydrin architecture were examined using microarray data from Genevestigator (Laule et al., 2008). Gene expression perturbations were examined during different abiotic stresses (drought, cold, and salinity), by plant anatomy and by developmental stage. The plant species examined were Arabidopsis thaliana, Zea mays, Solanum lycopersicum, Oryza sativa, Medicago truncatula, Triticum aestivum, and Hordeum vulgare. Affymetrix genome arrays were used for all species except for Zea mays, where the mRNA-seq Gene Level Zea mays platform was used. For species found in Phytozome 10, dehydrin genes were manually selected. For the remaining genomes the search term dehydrin was used to select genes, which were verified to match with our definition of a K-segment.

For gene upregulation during the abiotic stresses, the experimental data were obtained using the perturbation tool in Genevestigator. The data were then organized to exclude experiments that were performed in seeds since we are interested in protection in adult plants. The log2 fold change for all experiments were recorded at a significance level of α = 0.05, with log2 values of one or lower considered to be insignificant (i.e., assigned a value of zero). The data were arranged into groups based on the three perturbations being examined (i.e., drought, cold, and salinity). No down regulation was observed for any of the genes. The dehydrins from each organism were then organized based on their architecture, being grouped into K_n, SK_n, Y_nSK_n, or K_nS. Insufficient data were available to examine the Y_nK_n architecture. For localization of the different architectures, the gene expression of dehydrins in various anatomical regions of plants (seedling, inflorescence, shoots, roots, and, where available, callus) was examined. The log2 fold change for each experiment were recorded at a significance level of α = 0.05. For the developmental stages experiments, the specific developmental stages were grouped into ‘Early,’ ‘Middle,’ and ‘Late’ developmental stages as shown in Supplementary Table S5.

CATPCA was performed using the SPSS Statistics (version 22) software package (2013). To prevent the averaging out of the biochemical properties, the variables pI, M_r, and GRAVY score of each dehydrin were binned into low, medium and high ranges as defined in Supplementary Table S2. Architecture classification was based on the five major architectures found in the Phytozome dehydrins (K_n, Y_nSK_n, Y_nK_n, and SK_n). The presence of the SK-segment was treated as a separate variable. These properties were then tallied for each dehydrin in each species (Supplementary Table S3). After the CATPCA, three components were selected for further analysis on the basis that (1) the elbow in the Scree plot occurred at the third component, (2) the Eigenvalues were much greater than one (the fourth principal component had an Eigenvalue of 1.06 so it was cut) and (3) 68.3% of the variance is explained by the three components. Rotation was performed using the ‘varimax’ protocol (Kaiser, 1958). Coefficients with absolute values less than 0.3 were considered insignificant and therefore ignored in the subsequent analysis.

Thirty-five annotated vascular genomes were acquired from the Phytozome, version 10 (Goodstein et al., 2012). While this work focuses on dehydrins from higher (i.e., vascular) plants, an analysis of non-vascular plants is also performed. The Phytozome 10 file containing all protein-coding sequences, with alternative splice variants, was used as the source for the genome sequences, since a study on Vitis riparia dehydrins showed that alternative splicing can partially account for the sequence variation (Xiao and Nassuth, 2006). The Pfam dehydrin dataset (PF00257) is a collection of protein sequences that have been annotated or detected to be dehydrins (Finn et al., 2014). Duplicate sequences and fragment entries in the Pfam dataset were removed before further use.

Our first goal was to obtain a broad sample of K-segments in order to perform an unbiased motif search for sequence conservation of this motif. However, in order to obtain a pool of dehydrin sequences, we had to first find protein sequences that contain at least one K-segment. To circumvent this circular problem, the dehydrin sequences were collected and filtered over multiple steps (Figure 1A). In the first step, we obtained the definition of K-segments from a large number of papers to develop a very broad, very comprehensive regular expression pattern (Figures 1B,C and Supplementary Table S1). Note that this search expression does not give any weight to any of the positions, but merely notes the presence of a particular amino acid. This filter results in a very large list of putative dehydrins, including some matches that were not representative of the dehydrin K-segment. This broad set of dehydrin sequences was used as input for the Multiple Expectation maximization for Motif Elicitation (MEME) program to find a consensus sequence of the K-segment (Bailey and Elkan, 1994). This consensus sequence was used to develop a matching rule where a sequence segment must match two of the three rules shown in Supplementary Figure S1A in order to be called a K-segment. This type of search allows for substitutions, deletions, and insertions that a normal regular expression does not allow, but at the same time prevents a large number of false positives from occurring. This technique works under the assumption that the K-segment sequences being searched for are biologically functional and that the conserved core sections are responsible for this functionality. If we assume that the functionality is important in vivo, then proteins missing large parts of this motif would be considered questionable dehydrins and may need to be classified as dehydrin-like proteins instead. The result of this filtering step, 643 sequences, contains all proteins that we propose to be dehydrins. This narrow list was used for all subsequent analyses, including the motif searches.

The narrow list of dehydrin sequences was used as input for the MEME program to allow for an unbiased search for the Y-, S-, and K-segments, and for the potential discovery of new motifs (Bailey and Elkan, 1994). A LOGO representation (Schneider and Stephens, 1990) of the MEME K-segment is shown in Figure 2A, while a position-weighted matrix (PWM) is shown in Figure 2B. As the prefix ‘K-’ suggests, this motif is rich in Lys amino acids, with the amino acid being highly conserved (>90%) at positions 2, 8, 10, and 12. Other highly conserved positions include a Gly at position 4 (93%) and a Pro-Gly at the end of the motif (89 and 95%). Several other positions are conserved in terms of the chemical properties of the amino acid: position 5, 9, and 13 are hydrophobic (95%), while positions 7 and 11 have the acidic residues Asp and/or Glu (>90%). The remaining positions are generally variable in terms of both amino acid type and property, although the N-terminal residues 1 and 3 are generally polar or charged, while position 6 is mostly hydrophobic.

The Y-segment is named after the presence of a Tyr amino acid in the middle of this motif. While tyrosine was a frequently found amino acid during the MEME search, alternate residues were also commonly found at this position. Tyr accounts for 74% of the amino acid present, while the next two most conserved amino acids are His (12%) and Phe (8%) (Figure 3). The presence of these three amino acids suggests that it is the aromatic character that is important at this position, although interestingly Trp was never detected. The most conserved group of residues is at positions 4–6, which consist of a Gly at position 4 (99%), an Asn at position 5 (96%) and a Pro at position 6 (94%). The first residue in the Y-segment is often Asp (94%), and is often followed by another acidic amino acid (Glu, 73%). In the literature, the Y-segment is often defined as being six residues long, however, as Figure 3 shows, this motif ends with a hydrophobic amino acid at the 7th position 96% of the time.

An analysis of the S-segment motif required a slightly different approach since the length of the Ser-tract is variable (though generally agreed to be 4–8 Ser residues long), and another work has reported the conservation of residues N-terminal to the Ser-tract (Svensson et al., 2002). To examine these two aspects, we performed the MEME searches with widths of 5–20 residues, and found that 16 residues was the optimum value (i.e., most conserved length) for the longer S-segment. As can be seen in Figure 4, the highest probability length is six serines. At the C-terminus of this tract a pair of Asp and/or Glu residues often follows the Ser-tract, and at the N-terminus there is often a Ser-Gly pair, which can be seen in the LOGO representation at positions 4–5 and 6–7. Even further toward the N-terminus the motif Leu-His-Arg (90, 74, and 94%, respectively), followed by Ser or Thr (82 or 15%) is found. These N-terminal residues show that Leu needs to be followed by a positively charged amino acid (His and Arg), and then a hydroxyl group containing amino acid (Ser or Thr). The fifth position has a preference for Gly (49%), but could be any non-hydrophobic amino acid.

We examined the E-values of the Y-, S-, and K-segments discovered by MEME, and compared them to E-values calculated from randomly shuffled dehydrin sequences using the same MEME run parameters (Table 1). As can be seen, the segment E-values are many fold higher, strongly showing the discovered motifs are statistically significant.

To test whether the ϕ-segment is a random sequence, we took all sequences located between the Y-, S-, and K-segments to perform the Shapiro–Wilk test for normality (Shapiro and Wilk, 1965). A Q–Q plot can be used to visualize the data; deviation from a diagonal line would suggest that the data are not normally distributed (Wilk and Gnanadesikan, 1968). The Q–Q plot for the ϕ-segments is presented in Figure 5A, and shows that the sequence composition of the ϕ-segments is not random. We also looked at the distribution of the lengths of the ϕ-segment. Figure 5B shows a highly left-skewed distribution. This is partly caused by the large number of ϕ-segments that are located at the end of a protein after a K-segment, which tend to be very short (1–10 residues). The majority of the ϕ-segments (95%) are <100 residues in length, but can possibly extend out to several hundred residues for the larger dehydrins.

The ϕ-segments are often said to be rich in Gly, Ala, Ser, and Thr amino acids, but an examination of a large number of ϕ-segments shows that this is not completely correct in a larger dataset (Figure 5C). The top three amino acids in all ϕ-segment sequences are Gly (17.5%), Thr (11%), and Glu (10.7%). Other amino acids which occur with >5% frequency are Ala (7.0%), His (7.6%), and Gln (6%). The bottom three amino acids are Trp (0.05%), Cys (0.4%), and Phe (1.1%). All other amino acids occurred with frequencies between 1.6 and 4.6%. We also examined the ϕ-segment amino acid composition in the five different dehydrin architectures. For most amino acids, there is only a moderate difference between the architectures (Figure 5C). Exceptions are the more abundant amino acids such as Gly, Glu, and Lys. For the most part these can be differentiated by the presence of the Y-segment, such that the Y_nSK_n and Y_nK_n dehydrins contain higher amounts of Gly (24% versus 17% on average) but lower amounts of Glu (4% versus 15% on average) and Lys (2.5% versus 10% on average). The K_nS dehydrins tend to be higher in His (14.7% versus 7.6% on average), Gln (11.6% versus 6.1% on average), and Lys (13.1% versus 7.6% on average).

We also performed an extensive motif discovery search on the ϕ-segment sequences to see if any recurrent motifs can be found that were previously missed. Individual MEME runs were performed on the full dehydrin sequences with the widths varied between 5 and 20 residues in each search, and a single run with the width allowed to vary between 8 and 50 residues. Only one novel motif was consistently observed, which is a GT-rich motif with an optimal width of 17 residues (Figure 5D). This sequence is very rich in Gly and Thr (dominating in 41 and 29% of the residue positions), and may contain two Gln residues in the middle.

The E-value of unshuffled and shuffled GT-motif searches is reported in Table 1. Once again, the unshuffled search has a much lower statistical probability of having occurred by chance. Nevertheless, the top hit in all of the shuffled sequence searches returned Gly-rich motifs (data not shown). To examine whether this sequence has occurred simply by chance due to the high Gly and Thr content in ϕ-segments, we took 236 K_n dehydrin sequences that are rich in GT-motifs and repeated the MEME search using shuffled sequences. The search was repeated using five different sets of shuffled sequences. No GT-motifs were detected, suggesting that this motif did not occur by chance.

We also examined between which Y-, S-, and K-segments the GT-motif is found. Figure 5E shows that GT-motifs are especially prevalent between Y- and S-segments and between K- and K-segments. They are infrequently found between Y- and K-segments and rarely found between S- and K-segments and were never found between K- and S-segments. With regards to dehydrin architecture, the GT-motif was found in all Y_nSK_n architectures and in 82% of all Y_nK_n architectures, suggesting that the presence of the Y-segment is a strong indicator for the presence of this motif. It is also found in 50% of the K_n proteins, nearly 40% of the K_nS proteins, and approximately 25% of all of the SK_n dehydrins (Table 2). To determine whether these GT-motifs have a specific function or are merely a remnant of dehydrin evolution will require further evaluation.

There have been reports on the presence of Lys-rich (Mouillon et al., 2006; Eriksson and Harryson, 2011) and His-rich (Hara et al., 2005) motifs in dehydrins. For our purposes, we define the motif as being rich in that particular amino acid if it contains four or more contiguous His or Lys residues. A search of the dehydrin sequences revealed very few His-rich sequences (Table 2). The Lys-rich sequences are rare in K_n, Y_nSK_n, and Y_nK_n architectures (less than 15%), but occur in all K_nS dehydrins and are present 71% of the time in SK_n dehydrins (Table 2). In the literature the Lys-rich segments have sometimes been labeled as K-segments, but the lack of hydrophobic residues in alternating positions with Lys would suggest that they are different form the K-segment, and most likely have a different function in dehydrins.

The conserved Y-, S-, and K-segments in dehydrins are modular in nature and can be found with different frequencies (including zero) (Close, 1996). The five commonly listed architectures are K_n, Y_nSK_n, SK_n, Y_nK_n, and K_nS. We counted the number of these different architectures in different species in Phytozome 10 and in Pfam 00257 (Figure 6A). Searches for the motifs were performed using the segment definitions found in Supplementary Figure S1. The last column in the figure lists the total number of dehydrins discovered in each species and in Pfam. An examination of the distribution by species reveals different preferences for some of the architectures between three groups (grasses, non-rosid dicots, and rosid dicots). Among the grasses, the S-segment appears quite prevalent since only Y_nSK_n, SK_n, and K_nS architectures are found. Among the non-rosid dicots, the K_n and Y_nK_n architectures are somewhat rare but less so than in the grasses, while for rosids all of the four major architectures are seen. The SK_n, K_nS, and Y_nSK_n architectures show no preference among the different species.

The order of the three major dehydrin motifs and the distribution of the distances between them were examined next. Even in this large dehydrin dataset, the order of the three different segments is highly conserved with only very few exceptions (see Figure 7). That is, the Y-segment is the most N-terminal motif, the S-segment is next, and the K-segment(s) occur at the C-terminal end of the protein. With regards to distances (Figure 6B), the median distances between the Y- and S-segments and between two K-segments is similar (∼30–40 residues), while the median distance between Y- and K-segments is longer at 70 residues and, with two exceptions, never shorter than 30 residues. We observed that the distances between S- and K-segments are unusually left-skewed to short distances in comparison to the others, with 3–6 residues occurring 56% of the time. To investigate this more closely, dehydrin sequences with short distances were extracted and examined with the gapped motif finder GLAM2 (Frith et al., 2008). The LOGO representation of the highest scoring sequence is shown in Figure 6C. This motif, which we have named as the “SK-segment,” is 40–42 residues long, and is a combination of the S-segment and K-segment with a short linker. The S-segment portion runs from position 1 to approximately 18, while the K-segment runs from approximately positions 27–42. Variability in these lengths is due to the variability in the gap, that is, in determining the end of the S-segment and the beginning of the K-segment. The gap, essentially the linker between the S- and K-segments, has some conservation of its own. The first part consists of Gly-Xaa-Gly-Gly-Arg (where Xaa is any amino acid), while the second part consists of Arg-[Arg, Glu, Lys]-[Lys, Arg]-Arg-Lys, where the square brackets denote alternative choices at that position. A search for the SK-segment among the different species found in Phytozome 10 did not show any distinct distribution pattern (Supplementary Figure S2). The SK-segment (Table 2) is found predominantly in the Y_nSK_n architecture (85%) and sometimes in the SK_n architecture (17%).

We subsequently searched in our dehydrin sequence dataset for architectures that did not follow the typical order of Y-, S-, and K-segments. Only two dehydrins were found that could not be classified under one of the five common architectures. This includes a SKKS dehydrin from Stellaria longipes (Figure 7A) (Zhang et al., 1993) and a SKKYKYK dehydrin from Cerastium arcticum (Kim et al., 2013) (Figure 7B). The SKKS architecture consists of SK-K-S motifs, where the second S-segment is of the short form. In the SKKYKY dehydrin, the single S-segment is of the long form. The two Y-segments both contain Tyr, but the seventh residue is not hydrophobic (Thr in both cases), while several of the K-segments are not well-conserved because of a lack of Lys residues in the N-terminal part of this motif.

We subsequently examined how some biochemical properties are affected by the dehydrin architecture. These biochemical properties include their molecular mass (M_r, a measure of their size), isoelectric point (pI, a measure of their net charge), GRAVY score (a measure of their net hydrophobicity), and FoldIndex (propensity of a protein to fold) (Prilusky et al., 2005) which are shown in Figure 8 as bean plots (Kampstra, 2008). The distribution of the pI scores is bimodal for four architectures (K_n, Y_nSK_n, YnK_n, and SK_n), where all of them show a basic pI value centered around pH 9 (Figure 8A). The exception to this, the KnS architecture, consisted mainly of dehydrins with a near neutral pI and a few with a basic pI. The bimodal properties are interesting; unimodality would suggest a random sequence selection of that property while bimodality suggests that the proteins specifically evolved these properties, possibly in reference to their specific function. In the acidic range there is dissimilarity among the pI values of the different architectures; the K_n and Y_nSK_n architectures show a wide range of values (between pH 5.0–7.0 and 6.0–7.0, respectively). Two other architectures show a narrower acidic pI range, with Y_nK_n dehydrins having a center near pH 6 while the SK_n dehydrins have a center near pH 5, while the K_nS dehydrins have an average pI near pH 7. For the GRAVY scores (Figure 8B), the K_n architecture shows a bimodal distribution, with the centers near -1.6 and -1.3. The four other architectures showed unimodal distributions. In the case of Y_nSK_n there is a preponderance of dehydrins with GRAVY scores around -1. For Y_nK_n and SK_n, there is a large spread of GRAVY scores, centered at -1.2 for Y_nK_n, -1.4 for SK_n, and -1.9 for K_nS. The molecular mass plot (Figure 8C) did not show any bimodal distribution, but instead had a region where a majority of the M_rs were found. For K_n and K_nS this is around 12 kDa, while for the other three architectures it is around 20 kDa. Even with these average values it should be noted that there is a very large range of molecular masses for the dehydrins. With FoldIndex, negative scores represent proteins that are predicted to be unlikely to fold, and hence be intrinsically disordered. For the most part, the distribution of the FoldIndex scores (Figure 8D) follow that of the GRAVY scores (Figure 8C), suggesting that the disorder in dehydrins is driven by hydrophobic residues rather than the balance of charged residues. One exception to this pattern is K_nS; the FoldIndex scores are spread over a large range of values, suggesting that in this architecture differences in the number of charged residues has an effect on the potential amount of disorder.

We examined the change in expression of the different architectures under different stress conditions, in different tissues, and at different life stages (Figure 9). Details on how the data points were collected is described in Section “Materials and Methods.” Expression levels showing log2 values <1 are shown as being zero-fold change. An examination of the expression of the various dehydrins sorted by drought, cold, and salinity stress suggests that there is some preference for architecture by stress (Figure 9A). The K_n dehydrins are mostly upregulated during cold and drought. Under all stress conditions examined, the SK_n dehydrins showed on average very little change in expression. The Y_nSK_n dehydrins are most upregulated during drought, but were only weakly upregulated during cold and salinity stresses. The K_nS dehydrin shows the most upregulation during cold stress, minimally with drought stress, but not with salinity.

Examination of the relationship between architecture and anatomy (Figure 9B) or developmental stage (Figure 9C) did not show any substantial change for any of the architectures. In all cases, the log2 changes in expression levels were similar across all conditions.

The plotting of architectures versus these properties is one way to look for relationships among these proteins, but many other relationships are possible (in fact between each variable) and pairwise plots cannot show multiple relationships at the same time. We evaluated such potential relationships using a principal component analysis (Hotelling, 1933), specifically categorical principal component analysis (CATPCA) (Linting et al., 2007). The advantages of CATPCA are that it does not require assumptions of normality or linearity in the variables, and can be used in the analysis of categorical data such as dehydrin architecture. CATPCA also allows us to see what associations may exist between the five variables (pI, M_r, GRAVY score, and architecture, to which we added the SK-segment motif). To prevent an averaging out of the values, properties are classified according to the ranges shown in Supplementary Table S2 and as described in Section “Materials and Methods.” The resulting CATPCA variable inputs are shown in Supplementary Table S3.

The variable coefficients after CATPCA are shown in Table 3 where they are broken down by their contribution to the first three principal components (PC1-3). PC1 is dominated by the SK_n and K_n architectures (i.e., in the absence of a Y-segment), and is highly linked with acidic pI and medium M_r proteins. The GRAVY scores in PC1 are low, showing that these dehydrins are highly hydrophilic even in this family of disordered proteins. PC2 is dominated by the Y_nSK_n architecture, and is linked with dehydrins that have high M_r, higher GRAVY scores and basic pI. PC3 is associated with the SK_n architecture, and has a strong linkage with dehydrins with neutral pI. The K_nS architecture has no notable association with PC1 or PC2, but has a strong negative association with PC3, showing that this architecture is not associated with a neutral pI.

In non-vascular plants, the K-segment has been suggested to have weaker similarity to the K-segment found in vascular plants (Velten and Oliver, 2001; Saavedra et al., 2006). We therefore used a combined approach of four different methods to search for potential dehydrins in non-vascular embryophytes: (i) BLASTP search (Altschul et al., 1997) using previously identified P. patens dehydrins (Ruibal et al., 2012); (ii) BLASTP search using dehydrins identified in (i); (iii) MAST search (Bailey and Gribskov, 1998) using the vascular K-segment; (iv) MAST search using the non-vascular K-segment identified in (i). The resulting motif is shown in LOGO form in Figure 10A. Interestingly, the most prevalent amino acids appear to be similar to the vascular K-segment in Figure 2A, and different from the K-segments identified in P. patens (Saavedra et al., 2006) and a dehydrin-like protein in Tortula ruralis (Velten and Oliver, 2001).

At this point we made a visual inspection of the non-vascular dehydrins, and noted that there were two kinds – one that resembled the vascular K-segment, and another that contained a less conserved Lys and Glu/Asp residues (Figure 10B). We also realized that an analysis of the lycophyte Selaginella moellendorffii genome did not detect any dehydrins. We repeated the search using the K-segment found in Figure 10B and interestingly found one sequence, shown in Figure 10C, that matched the K-segment motif from the non-vascular plants.

Using the K-segment motif in Figure 10B we did not discover any new dehydrin sequences in the other vascular plant genomes. Note also that no matches were found in any of the seven green algae genomes that were analyzed (data not shown).