(C) PLOS One
This story was originally published by PLOS One and is unaltered.
. . . . . . . . . .



Evolutionary rate covariation is pervasive between glycosylation pathways and points to potential disease modifiers [1]

['Holly J. Thorpe', 'Department Of Human Genetics', 'University Of Utah School Of Medicine', 'Salt Lake City', 'Utah', 'United States Of America', 'Raghavendran Partha', 'Department Of Biological Sciences', 'University Of Pittsburgh', 'Pittsburgh']

Date: 2024-12

Mutations in glycosylation pathways, such as N-linked glycosylation, O-linked glycosylation, and GPI anchor synthesis, lead to Congenital Disorders of Glycosylation (CDG). CDG typically present with seizures, hypotonia, and developmental delay but display large clinical variability with symptoms affecting every system in the body. This variability suggests modifier genes might influence the phenotypes. Because of the similar physiology and clinical symptoms, there are likely common genetic modifiers between CDG. Here, we use evolution as a tool to identify common modifiers between CDG and glycosylation genes. Protein glycosylation is evolutionarily conserved from yeast to mammals. Evolutionary rate covariation (ERC) identifies proteins with similar evolutionary rates that indicate shared biological functions and pathways. Using ERC, we identified strong evolutionary rate signatures between proteins in the same and different glycosylation pathways. Genome-wide analysis of proteins showing significant ERC with GPI anchor synthesis proteins revealed strong signatures with ncRNA modification proteins and DNA repair proteins. We also identified strong patterns of ERC based on cellular sub-localization of the GPI anchor synthesis enzymes. Functional testing of the highest scoring candidates validated genetic interactions and identified novel genetic modifiers of CDG genes. ERC analysis of disease genes and biological pathways allows for rapid prioritization of potential genetic modifiers, which can provide a better understanding of disease pathophysiology and novel therapeutic targets.

Congenital Disorders of Glycosylation (CDG) are a group of rare disorders resulting from impaired protein glycosylation. Glycosylation is the addition of sugar chains onto proteins and is required for proper protein function. CDG patients typically present with seizures and hypotonia. However, they can have a large amount of clinical variability, which is likely influenced by modifier genes. Modifier genes are genes that affect a phenotype without causing the disease. Using an evolutionary method that examines proteins that evolve at similar rates, we identified proteins within glycosylation pathways and among other unexpected pathways, such as ncRNA modification and DNA repair, that could be potential genetic modifiers of CDG genes. We also tested top protein pairs using the Drosophila eye as a model and identified novel genetic modifiers of CDG genes. Broadening our understanding of CDG modifiers can help us to better understand why loss of glycosylation results in specific patient symptoms and could provide new treatment targets.

Funding: This work was supported by NIGMS R35 GM124780 to CYC and NHGRI R01 HG009299 to NLC. HJT was supported by NIDDK T32 DK1109660. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability: The authors confirm that all data underlying the findings are fully available without restriction. All relevant data are within the paper and its Supporting Information files, and at doi.org/10.6084/m9.figshare.25734708 .

Copyright: © 2024 Thorpe et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Applying a computational evolutionary method allows for more rapid and broad detection of disease modifiers. Identifying genetic modifiers of CDG will lead to a better understanding of the physiology underlying variable symptoms in patients and could point to new therapeutic targets. Using ERC, we performed an analysis to discover proteins that have correlated evolutionary rates with glycosylation proteins to discover genetic modifiers of CDG. We found strong intra-pathway ERC signatures among glycosylation proteins in each of the three glycosylation pathways, with the highest average score in GPI anchor synthesis. There were also high inter-pathway ERC values between proteins in different glycosylation pathways. Strikingly, GPI anchor synthesis proteins had many high ERC values with N-linked glycosylation proteins. Protein complexes and proteins with similar functions in glycosylation were also enriched for strong ERC. When examining the genome for non-glycosylation modifiers of GPI anchor synthesis proteins, we found high ERC scores with proteins involved in multiple unrelated pathways, including ncRNA and DNA repair. We found that the strength of these ERC scores is driven by the GPI anchor synthesis enzyme’s location in the cell. In vivo testing of top scoring protein pairs, including GPAA1 with ALG1 and PIGA with RBSN, showed a genetic interaction in a Drosophila eye model. Using computational evolutionary methods, we identified patterns of ERC between CDG proteins and novel potential modifier proteins, many of which validated in vivo.

A number of recent studies suggest that other glycosylation genes modify CDG genes. In 2002, a study examined a small group of PMM2-CDG patients, the most common CDG, and identified a variant in ALG6, an N-linked glycosylation gene, segregating with more severe outcomes [ 20 ]. A modifier screen and ERC analysis of NGLY1, a deglycosylating enzyme associated with a CDG, identified multiple glycosylation genes showing high ERC with NGLY1, implicating them as potential modifiers [ 15 ]. A CRISPR screen in cells with reduced DPAGT1 function, the first step in N-glycan synthesis, identified multiple glycosylation genes able to rescue glycosylation defects caused by inhibition of DPAGT1 [ 14 ]. In yet another study, a yeast model of PMM2-CDG was evolved over 1000 generations to identify genetic changes that increased the viability of the PMM2 mutant yeast strains [ 16 ]. Several top genes that accumulated compensatory mutations were also involved in glycosylation. These studies indicate that genetic variation in glycosylation genes might potentially modify the outcomes of CDG.

Standard methods for identifying genetic modifiers typically involve utilizing natural variation in the population or performing large mutagenesis screens [ 1 ]. Several recent studies reported modifier screens in animal models of specific CDG [ 14 – 16 ]. However, with over 150 CDG, performing similar studies for each CDG is impractical. PMM2-CDG, the most common CDG, has just over 1000 diagnosed patients worldwide [ 17 ]. Most CDG, however, have fewer than 100 patients, making it challenging to have enough statistical power to distinguish impactful variation within the population [ 8 , 18 ]. It is likely CDG share common genetic modifiers because different forms can have similar effects on glycosylation and similar clinical presentations. Instead of examining each CDG independently, we sought to identify common modifier genes for all or a subset of CDG. To do this, we used evolutionary rate covariation (ERC), a computational evolutionary method developed to identify proteins with evolutionary rates that covary across species [ 19 ]. Protein pairs with high ERC values have similar changes in evolutionary rate and are thought to function together, either in a complex or in a pathway, and may genetically modify each other. ERC allows us to identify proteins with evolutionary rates that covary with CDG proteins, potentially identifying genetic modifiers.

GPI anchors are glycolipids that tether proteins to the cell membrane [ 8 , 13 ]. GPI anchor synthesis begins by adding an N-acetylglucosamine (GlcNAc) onto a phosphatidylinositol on the cytoplasmic side of the ER membrane. The acyl group is then removed before the glycan is flipped onto the lumen side of the ER membrane. In the ER lumen, three mannoses (four in some species/tissues) and three phosphoethanolamine groups are added before a protein is attached to the GPI anchor. The GPI anchor then undergoes fatty acid remodeling before it is sent to the Golgi for further remodeling. Then, the GPI-anchored glycoprotein is trafficked to the cell surface.

Protein glycosylation is essential for proper protein folding, stability, and localization [ 4 , 5 , 11 ]. There are three main types of glycosylation: N-linked glycosylation, O-linked glycosylation, and glycosylphosphatidylinositol (GPI) anchor biosynthesis. N-linked glycosylation is the addition of a glycan onto an asparagine [ 4 , 5 ]. The glycan is synthesized on the endoplasmic reticulum (ER) membrane, with the stepwise addition of sugars onto a dolichol-phosphate. The glycan is then transferred from the dolichol-phosphate onto a protein and undergoes further glycan additions and trimming in the ER and Golgi. O-linked glycosylation is the addition of a glycan to the hydroxyl group of a serine, threonine, or hydroxylysine [ 4 ]. O-linked glycosylation encompasses the most extensive variety of glycans and can occur in the ER, Golgi, or cytoplasm [ 12 ]. O-linked glycans are classified by the first sugar attached to the protein. The rest of the glycan is built directly onto that sugar.

Congenital disorders of glycosylation (CDG) are a group of ultra-rare multisystemic disorders typically characterized by seizures, hypotonia, and neurodevelopmental delays [ 4 , 5 ]. However, because of the ubiquity of glycosylation, symptoms of CDG can affect most organ systems. Loss-of-function mutations in 189 glycosylation genes lead to CDG [ 6 ]. Patients with the same CDG and even the same disease-causing variant can show phenotypic variability, indicating that genetic background and environment can influence clinical presentation [ 7 , 8 ]. Current treatment options for CDG are limited, primarily focusing on symptom management [ 9 , 10 ]. Identifying genetic modifiers of glycosylation genes could lead to new therapeutic targets for CDG patients.

Phenotypic variation is common in Mendelian diseases, even in patients with the same disease-causing variant [ 1 – 3 ]. Understanding the factors underlying variable disease presentation can be difficult despite often knowing the causal disease gene. Genetic modifiers can enhance or suppress the phenotype of a primary disease-causing mutation, but they may not independently lead to disease phenotypes.

Results

Strong evolutionary covariation between proteins in different glycosylation pathways Glycosylation is a broad term encompassing multiple pathways with hundreds of proteins in different organelles. While some proteins associated with CDG impact different types of glycans, nearly all enzymatic proteins are specific to one of the three main glycosylation pathways. We asked if the high mean ERC score observed across all glycosylation proteins (mean ERC = 0.41, Significance tested against 100,000 permutations p < 1x10-5) was driven primarily by high scores between proteins in the same pathway or if high ERC scores between proteins in different pathways were also observed. Proteins in all three glycosylation pathways showed strong ERC with proteins in different pathways (Table 2, and, S5 and S6 Figs). As within the pathway, GPI anchor synthesis had the highest scores with other glycosylation pathways. The average ERC score between proteins in GPI anchor synthesis and proteins in N-linked glycosylation is 0.81 (p < 1x10-5), O-linked glycosylation is 0.46 (p < 0.014), and Other is 0.61 (p < 1.30−4). The average scores between GPI anchor synthesis and either N-linked, O-linked glycosylation, or Other are higher than the average of all glycosylation proteins (mean ERC = 0.41). This strong signal between glycosylation proteins in different pathways, especially those in GPI anchor synthesis and N-linked glycosylation, could indicate unexpected interactions between pathways. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Table 2. Average ERC values between glycosylation pathways. https://doi.org/10.1371/journal.pgen.1011406.t002

Not all glycosylation complexes have high correlated evolutionary rates We next examined whether glycosylation proteins that form physical complexes have high ERC values. The GlcNAc transferase complex (GlcNAc-T) performs the first step in GPI anchor synthesis, attaching a GlcNAc to the phosphatidylinositol base [13]. GlcNAc-T consists of seven proteins: PIGA, PIGC, PIGH, PIGP, PIGQ, PIGY, and DPM2. The proteins in the GlcNAc-T complex do not have enriched rate covariation with each other (mean ERC = 0.18, p = 0.41) (Figs 2A and S7A). Only one pair of proteins shows significant ERC, PIGQ and DPM2 (ERC = 3.58). The low ERC scores between proteins in this complex suggest that physical interaction does not drive evolutionary rate changes in these proteins. The other complex in GPI anchor synthesis is the GPI transamidase complex which is comprised of five proteins: PIGS, PIGT, PIGU, PIGK, and GPAA1 [13]. The GPI transamidase complex attaches the protein to the synthesized GPI anchor. The mean ERC score among proteins in the GPI transamidase complex is 5.86 (p < 1x10-5), which is more than 11 standard deviations higher than the average expected based on permutation testing (permutations mean = 0.12) (Figs 2B and S7B). All pairwise ERC scores between GPI transamidase proteins show significant rate correlation (ERC ≥ 3), except between PIGU and PIGK (ERC = 0.80). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 2. Heatmaps of significant pairwise ERC values for proteins in complexes involved glycosylation. The GlcNAc-T complex (A) does not show significant ERC (p = 0.41). The GPI transamidase complex (B), the OST complex (C), and the COG complex (D) all have significant ERC (p < 1x10-4). https://doi.org/10.1371/journal.pgen.1011406.g002 In N-linked glycosylation, the oligosaccharyltransferase (OST) complex attaches the synthesized glycan onto a protein [11]. This complex is made up of 13 proteins. The average pairwise ERC score of the OST complex is 1.26 and is higher than expected based on permutation testing (p = 4.0x10-5) (Figs 2C and S7C). The highest value is between MLEC and RPN2 (ERC = 7.02). Both MAGT1 and OST4 have no significant ERC scores with any other proteins in the complex. MAGT1 acts as a scaffold protein for STT3B in the OST complex, but it also functions as a magnesium transporter [21]. Because MAGT1 has functions outside of this complex, it might have multiple evolutionary pressures that reduce its covariation rate with OST complex proteins. OST4 plays a critical role in the OST complex, binding the catalytic subunits STT3A or STT3B to the rest of the complex [22]. Despite this, it does not show strong ERC with OST complex components suggesting that it might have another unknown, unrelated function. The COG complex is a group of eight proteins (COG1-8) that acts as a tethering factor in intra-Golgi trafficking [23]. Pairwise ERC scores between the proteins in the COG complex show significant ERC (mean ERC = 3.00, p < 1x10-5) (Figs 2D and S7D). Only half of the protein pairs show a significant correlation (ERC ≥ 3). The COG complex is organized into two lobes, with COG1-4 making up one lobe and COG5-8 making up the second lobe. Within the first lobe, COG1 has significant ERC with both COG2 and COG3. Strikingly, COG4 does not have any significant ERC scores with the other proteins in the same lobe; however, COG4 has the highest ERC value of the whole complex with COG6 (ERC = 8.46), a protein in the opposite lobe. The second lobe has slightly more protein pairs with significant ERC. COG6, COG7, and COG8 all have significant ERC with each other. COG5 has no significant ERC with other proteins in the same lobe. A bond between COG1 and COG8 connects the two lobes. COG1 and COG8 have the second highest ERC score of the complex (ERC = 7.16). The average ERC score between proteins in the first lobe is 2.70, and the average ERC score between proteins in the second lobe is 2.51. Unexpectedly, the average ERC score among inter-lobe protein pairs is 3.29. Strong ERC scores in this complex are not solely driven by direct physical contact between the proteins and are, on average, higher between proteins in different lobes.

Glycosylation enzymes with similar functions have increased ERC Next, we examined whether glycosylation proteins with similar enzymatic functions, but from different pathways display high ERC. There are twelve mannosyltransferases across the three glycosylation pathways. These mannosyltransferases are not known to physically interact with each other, and each one functions in only one specific glycosylation pathway. Pairwise ERC scores between these twelve mannosyltransferases show significant ERC by permutation testing (mean ERC = 3.07, p < 1x10-5) (Figs 3 and S7E). Of these 55 ERC values, 27 are above the significance cut-off (≥3). The top score is between ALG12 and ALG2 (ERC = 8.02), which are both in the N-linked glycosylation pathway. The second highest score is between POMT1, involved in O-mannosylation, and ALG2, involved in N-linked glycosylation (ERC = 7.97). The six mannosyltransferases involved in N-linked glycosylation have an average ERC of 4.40. The four GPI anchor synthesis specific mannosyltransferases have an average of 2.32. O-linked glycosylation has only two mannosyltransferases (POMT1 and POMT2), and their average ERC score is 6.86. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 3. Heatmap showing significant ERC values between mannosyltransferases. N-linked glycosylation (yellow), GPI-anchor synthesis (red), O-linked glycosylation (blue). Mannosyltransferases have significant ERC with each other across pathways (p < 1x10-5). https://doi.org/10.1371/journal.pgen.1011406.g003

Glycosylation protein pairs with strongest ERC Some of the top ten highest-scoring glycosylation protein pairs show overlapping functions, which could explain the high ERC scores (Table 3). For example, COG3 and TRIP11 (ERC = 9.62) are proteins bound to the outer membrane of the Golgi involved in vesicle tethering and assembly and maintenance of the Golgi. PIGZ and PIGW also have a high ERC score (ERC = 9.10). PIGZ and PIGW are both involved in GPI anchor synthesis in the ER, and they share a common GPI recognition domain in their transmembrane domains [24]. While these two pairs have high ERC scores supported by current knowledge, the top scores also include pairs with high scores that are not easily explained based on our current knowledge of their function. FKRP, an enzyme that adds ribitol to O-linked alpha-dystroglycans in the Golgi [25], and MOGS, an enzyme that cleaves glucose from the synthesized N-glycan in the ER [26], have an ERC score of 10.63. EXTL3, a GlcNAc transferase for O-linked glycosaminoglycan synthesis located in the Golgi [27], and GMPPA, a proposed inhibitor of GDP-mannose synthesis localized to the cytoplasm [28], have an ERC score of 9.62. The remaining six pairs in the top 10 show similar patterns of high ERC with no obvious overlap in function, localization, or glycosylation pathway. These unrelated pairs of proteins with high ERC values could signal novel interactions between glycosylation proteins. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Table 3. Top 10 pairwise ERC scores among glycosylation proteins. https://doi.org/10.1371/journal.pgen.1011406.t003

Genome-wide patterns of evolutionary rate correlation for GPI Anchor Synthesis Proteins We next examined whether proteins with unrelated, non-glycosylation functions might also have high ERC values with glycosylation proteins. To do this, we extended our analyses to a full genome-wide comparison. Because the GPI anchor synthesis pathway had the highest average ERC score of all glycosylation pathways (GPI mean ERC = 1.61), we focused the genome-wide analyses on the GPI anchor biosynthesis pathway. We included the 27 proteins specific to the GPI anchor synthesis pathway and DPM1, DPM2, and DPM3. DPM2 is a component of the GlcNAc-transferase complex in GPI anchor synthesis and the dolichol-phosphate-mannose (DPM) synthase complex with DPM1 and DPM3. DPM synthase is required to add mannose to the GPI anchor [29]. We calculated pairwise ERC values for each of the 30 proteins involved in GPI anchor biosynthesis with all 19,149 proteins in our dataset (Fig 4 and S1 Data). This resulted in 568,438 pairwise values ranging from -8.97 to 11.28, with an average of 0.32. 31,483 protein pairs had significant ERC scores (ERC ≥ 3). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 4. Histogram of genome-wide ERC values for with GPI anchor synthesis proteins. Vertical red line indicates top 1% of values. https://doi.org/10.1371/journal.pgen.1011406.g004

Top scoring pairs across the genome Many top-scoring pairs from the genome-wide analysis fall under the categories enriched in the GO analyses. Four of the top 30 scoring pairs were between known glycosylation proteins (Table 4). PIGN, a phosphoethanolamine transferase [13], had high ERC with MAN2B2, a protein involved in glycan recycling (ERC = 10.01) [10]. PIGG, a phosphoethanolamine transferase [13], had strong ERC with DOLK, a kinase that creates dolichol-phosphate upon which the N-glycan is built (ERC = 9.59) [30], and TRIP11, a protein involved in Golgi vesicle tethering (ERC = 9.50) [31]. GPAA1, a member of the GPI transamidase complex [13], had high ERC with ALG1, a mannosyltransferase in N-linked glycosylation (ERC = 9.29) [11]. All seven glycosylation proteins in these top-scoring pairs are associated with their own CDG, further suggesting that known CDG genes are likely candidates for genetic modifiers of other CDG [14–16,20]. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Table 4. Top scoring protein pairs from genome-wide analysis of GPI anchor synthesis proteins. https://doi.org/10.1371/journal.pgen.1011406.t004 Top scores in our dataset revealed high ERC between PIGB and UTP20 (ERC = 10.33) and PIGN and WDR36 (ERC = 9.74). UTP20 and WDR36 are both involved in rRNA processing [32,33]. Top scores also included PIGG and PRKDC (ERC = 10.01), a DNA damage sensing protein [34], ERCC6 (ERC = 9.76), a protein important for transcription-coupled excision repair [35], and AP5Z1 (ERC = 9.26), a protein involved in homologous recombination DNA repair [36]. GPAA1 had high ERC with MCM7 (ERC = 9.97), a protein that signals DNA damage [37]. These top pairs further support the hypothesis of a relationship between GPI anchor synthesis genes and the GO terms identified and offer targeted protein pairs to begin testing to explore these pathways.

Cytoplasmic GPI anchor synthesis proteins have lower ERC Among the top 1% of genome-wide GPI anchor scores, nearly 10% are with PIGG (552/5685). To determine if this enrichment was significant, we compared the actual number of ERC values each GPI anchor synthesis protein had in the top 1% to an expected number of values, assuming an equal distribution of all the GPI anchor synthesis proteins (expected ~189 occurrences per protein) (Fig 7A and S1 Table). PIGG (552 occurrences, 2.91 fold change, p < 2.2x10-16) was the most overrepresented, with nearly three times as many ERC values in the top 1% as expected, followed by PIGW (526 occurrences, 2.78 fold change, p < 2.2x10-16). On the other hand, PIGY (8 occurrences, -23.68 fold change, p < 2.2x10-16), PIGH (7 occurrences, -27.06 fold change, p < 2.2x10-16), and DPM2 (6 occurrences, -31.57 fold change, p < 2.2x10-16) all have more than 20 fold less than the expected frequency. We tested whether this enrichment/depletion followed the pathway order (Fig 7B and 7C). Analyses of the GPI anchor synthesis proteins by pathway order demonstrated that 8/10 proteins on the cytoplasmic side are underrepresented, and none are overrepresented. In the ER lumen, only 4/17 proteins are underrepresented, and 9/17 of the proteins are overrepresented. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 7. Proportion of expected occurrences of all GPI anchor synthesis proteins in the top 1% of genome-wide scores. Overrepresented proteins are shown in red, underrepresented in blue, and proteins not significantly different from expected are in grey. A) Proteins sorted from most to least represented. B) Proteins sorted in pathway order with lines dividing genes active on the cytoplasmic side of the ER, the lumen side of the ER, and in the Golgi. C) GPI anchor synthesis pathway diagram. Synthesis begins on the cytoplasmic side of the ER. The anchor is flipped to the lumen side. After the protein is attached, the glycan is sent to the Golgi for lipid remodeling. https://doi.org/10.1371/journal.pgen.1011406.g007 Proteins acting on the cytoplasmic side of the ER were significantly underrepresented in the top 1% of ERC values, and proteins acting on the lumen side were highly overrepresented. We hypothesized that high ERC with other ER resident proteins may be driving the overrepresentation of lumen proteins. However, GO analysis for cellular compartment of the top 1% of proteins with high ERC with GPI anchor synthesis proteins revealed the top enriched compartments are associated with the exosome, ribosome, and mitochondria (S8 Fig and S2 Data). Some ER-related terms were enriched, but over 50 other terms were more enriched. This indicates that while ER resident proteins may contribute to some of the high scores observed among lumen GPI anchor synthesis proteins, it is unlikely that high scores with ER resident proteins are solely responsible for driving the overrepresentation of lumen GPI anchor synthesis proteins in the top 1% of ERC scores. We examined the N-linked glycosylation pathway to determine if lower ERC scores for cytoplasmic enzymes were unique to GPI anchor synthesis or common in glycosylation pathways. The N-linked glycosylation pathway is similar in that the first few steps occur on the cytoplasmic side, and the glycan is completed and attached to a protein on the ER lumen side. We calculated ERC scores for proteins in the N-linked glycosylation pathway across the genome and ran the same analysis as we did for GPI anchor synthesis. We took the top 1% of values and determined the number of scores for each N-linked glycosylation protein and compared this to the expected distribution, as we did above (S9 Fig). Among the eight cytoplasmic proteins, three were under-represented, and four were over-represented. Among the ER lumen proteins, eight of the 23 proteins were under-represented, and ten of the 23 lumen proteins were over-represented. The N-linked glycosylation enzymes acting on the cytoplasmic did not have an enrichment of lower scores as observed for GPI anchor synthesis. The ERC pattern for cytoplasmic vs. ER lumen proteins appears specific to the GPI anchor synthesis pathway.

[END]
---
[1] Url: https://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1011406

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/

You are viewing proxied material from magical.fish. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.