Journal of Biomolecular Structure and Dynamics

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tbsd20

De novo structure prediction of meteorin and

meteorin-like protein for identification of
domains, functional receptor binding regions, and
their high-risk missense variants

S. Shiva Shankar, Reema Banarjee, Swaraj M. Jathar, S. Rajesh, Sureshkumar

Ramasamy & Mahesh J. Kulkarni

To cite this article: S. Shiva Shankar, Reema Banarjee, Swaraj M. Jathar, S. Rajesh,
Sureshkumar Ramasamy & Mahesh J. Kulkarni (2023): De novo structure prediction of meteorin
and meteorin-like protein for identification of domains, functional receptor binding regions,
and their high-risk missense variants, Journal of Biomolecular Structure and Dynamics, DOI:
10.1080/07391102.2023.2220804

To link to this article: https://doi.org/10.1080/07391102.2023.2220804

View supplementary material

Published online: 08 Jun 2023.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tbsd20
JOURNAL OF BIOMOLECULAR STRUCTURE AND DYNAMICS
https://doi.org/10.1080/07391102.2023.2220804

De novo structure prediction of meteorin and meteorin-like protein for

identification of domains, functional receptor binding regions, and their
high-risk missense variants
S. Shiva Shankara,b, Reema Banarjeea, Swaraj M. Jathara,b, S. Rajesha, Sureshkumar Ramasamya and
Mahesh J. Kulkarnia,b
a
Proteomics Facility, Division of Biochemical Sciences, CSIR-National Chemical Laboratory, Pune, India; bAcademy of Scientific and
Innovative Research (AcSIR), Ghaziabad, India
Communicated by Ramaswamy H. Sarma

ABSTRACT ARTICLE HISTORY

Meteorin (Metrn) and Meteorin-like (Metrnl) are homologous secreted proteins involved in neural Received 7 March 2023
development and metabolic regulation. In this study, we have performed de novo structure prediction Accepted 29 May 2023
and analysis of both Metrn and Metrnl using Alphafold2 (AF2) and RoseTTAfold (RF). Based on the
KEYWORDS
domain and structural homology analysis of the predicted structures, we have identified that these
Meteorin; meteorin-like;
proteins are composed of two functional domains, a CUB domain and an NTR domain, connected by structure prediction;
a hinge/loop region. We have identified the receptor binding regions of Metrn and Metrnl using the missense variants; domains;
machine-learning tools ScanNet and Masif. These were further validated by docking Metrnl with its protein–protein interactions
reported KIT receptor, thus establishing the role of each domain in the receptor interaction. Also, we
have studied the effect of non-synonymous SNPs on the structure and function of these proteins using
an array of bioinformatics tools and selected 16 missense variants in Metrn and 10 in Metrnl that can
affect the protein stability. This is the first study to comprehensively characterize the functional
domains of Metrn and Metrnl at their structural level and identify the functional domains, and protein
binding regions. This study also highlights the interaction mechanism of the KIT receptor and Metrnl.
The predicted deleterious SNPs will allow further understanding of the role of these variants in modu-
lating the plasma levels of these proteins in disease conditions such as diabetes.

1. Introduction Additionally, it regulates angiogenesis at the gliovascular

interface to promote brain vascular maturation during devel-
Meteorin (Metrn) and Meteorin-like (Metrnl) proteins are
opment (Jeong et al., 2008) and in the placenta, serving as a
novel homologous secreted proteins of around 30 kDa, with
biomarker in preeclampsia (Garc
es et al., 2015). It is also
40% identity in amino acid sequence. Metrn is a neuro-
secreted by the glial cells of the retina into the optic macula
trophic factor located on chromosome 16p13.3 in humans
and has protective action in age-related macular degener-
and 17A3.3 in mice and is expressed in neuronal cells as a ation via its anti-angiogenic activity (Delaunay et al., 2021).
potent neurotrophic growth factor (Nishino et al., 2004). On Besides its role in neurogenesis and angiogenesis, the knock-
the other hand, Metrnl was first reported as a neurotrophic out of the Meteorin gene in mice is lethal in the embryonic
factor-like Metrn. However, it functions as an adipokine and stages, indicating its involvement in early embryogenesis
cytokine as well and is termed Cometin or Subfatin based on (Kim et al., 2014). It has also been recently shown to be
its different roles (Jørgensen, Fransson, et al., 2012; Li et al., secreted by macrophages during hypoxic conditions and
2014). It is located on chromosome 17q25.3 in humans and reduce mobilization of hematopoietic stem cells into circula-
11qE2 in mice (Ushach et al., 2015) and is expressed in adi- tion (Dai et al., 2022).
pose tissue, skeletal muscle, and intestinal epithelial cells Metrnl is secreted by macrophages and monocytes during
(Zheng et al., 2016). inflammation (Ushach et al., 2015) and promotes metabolic
Both proteins regulate a wide range of physiological func- adaptation and tissue protection under stressful conditions
tions in the body. Metrn regulates glial cell differentiation (Zheng et al., 2016). Its expression increases in adipocytes
from neuronal stem cells and axonal extension via activation and muscle cells after aerobic exercise (Bae, 2018) and dur-
of the Jak-STAT signalling pathway (Lee et al., 2010). Due to ing acute exposure to cold, wherein it is involved in the cold
its neuroprotective action, it has been reported as a potential adaptation by induction of thermogenesis-associated genes
drug for neuropathic pain (Jørgensen et al., 2012). and adipose browning by alternate macrophage activation

CONTACT Mahesh J. Kulkarni [email protected] Proteomics Facility Biochemical Sciences Division CSIR-National Chemical Laboratory, Pune 411008,
India
Supplemental data for this article can be accessed online at https://doi.org/10.1080/07391102.2023.2220804.
ß 2023 Informa UK Limited, trading as Taylor & Francis Group
2 S. S. SHANKAR ET AL.

(Rao et al., 2014). It also acts as an adipokine to activate the only structural study reported till now (Wen et al., 2017).
PPAR-c in adipocytes and increase insulin sensitivity via AKT Therefore, in our present study, we have attempted to per-
phosphorylation, and its deficiency reportedly leads to adipo- form a detailed structural analysis of these proteins and their
cyte insulin resistance (Jung et al., 2018). Metrnl expression functional regions and further characterize the effect of
levels in the heart have been shown to increase after myo- nsSNPs on the functional regions by using in silico
cardial infarction in both mouse models and human tissue approaches.
samples (Rup
erez et al., 2021). Recently, it was discovered to
be the ligand for the stem cell factor receptor KIT (KIT recep-
tor tyrosine kinase), wherein it promoted infarct repair in 2. Materials and methods
mouse models of myocardial infarction by promoting the 2.1. Structure prediction of human Metrn and Metrnl
proliferation of KIT-expressing endothelial cell population at
the infarct border zone (Reboll et al., 2022). Metrn and Metrnl protein sequences of humans were
Many clinical studies have been done for Metrn and retrieved in Fasta format from the UniProt database (UniProt
Metrnl in different disease conditions; however, there are cer- id: Q9UJH8 and Q641Q3) and used for structure prediction
tain caveats concerning their role in the diseases. For using the ColabFold online platform, which is an Alphafold2
instance, there have been numerous conflicting reports (AF2) pipeline with MMseq2 in Google Colab (Mirdita et al.,
regarding the plasma levels of Metrnl as a biomarker in type 2022) (Jumper et al., 2021). The quality of the AF2-predicted
2 diabetes (Chung et al., 2018; Lee et al., 2018; Wang et al., models was assessed using the pLDDT confidence scores and
2019; Alkhairi et al., 2019; El-Ashmawy et al., 2019; Onalan predicted aligned error (PAE). The RoseTTAfold (RF) models
et al., 2020). A recent meta-analysis using nine previous clin- were prepared using the Robetta web server (Kim et al.,
ical studies, however, indicated no correlation between the 2004; Baek et al., 2021), and the models were validated
serum levels of Metrnl and the severity of type 2 diabetes based on the confidence score. In both the tools, the model
(Wu et al., 2020). Similar inconsistent correlations of Metrnl was developed using the ’without template’ method, and
with BMI have also been reported in obesity (Chung et al., the best model was selected and validated further using the
2018; Du et al., 2020; Qi et al., 2020; Schmid et al., 2021). PROCHECK and ERRAT tools on the online SAVES server
Various factors regulate the expression of proteins, one (Colovos & Yeates, 1993; Roman Laskowski et al., 1983). The
such factor that can modulate protein expression levels is structural analysis, visualization and alignment were done
Single Nucleotide Polymorphisms (SNPs) especially. Those using PyMOL (Schro €dinger, LLC, 2015). The TM-score was
SNPs that cause amino acid residue substitution in the pro- analyzed using the TM-align tool (Zhang & Skolnick, 2005),
tein product are also referred to as missense or non-syn- and RMSD was calculated using the CE-align command in
onymous SNPs (nsSNPs). nsSNPs can alter protein structure PyMOL.
and function, leading to altered expression and disease
development (Islam et al., 2019; Choudhury et al., 2021). In
fact, many experimental studies have reported that nearly 2.1. Domain boundary prediction
one-third of the nsSNPs were found to be deleterious
The domain boundaries of human Metrn and Metrnl AF2
(Tokuriki & Tawfik, 2009). Particularly in the case of secretory
model structures were predicted using the default parame-
proteins, missense mutations have been reported to affect
ters of the TopDomain tool (Mulnaes et al., 2021) and the
protein localization and expression by causing intracellular
Sword2 tool (Cretin et al., 2022). TopDomain is a meta-pre-
retention of the mutant proteins (Boulling et al., 2007;
dictor that combines various predictors’ sequence homology
Gregor et al., 2022). Both Metrn and Metrnl are secreted pro-
and accurately predicts the domain parsing. Sword2 web ser-
teins, and there is a lack of information regarding how their
ver provides three levels of hierarchical decomposition of
genetic variations affect stability, function, and expression.
protein structures, namely secondary structure, protein units,
We hypothesize that missense mutations in these proteins
affect their stability and secretion, thus leading to the wide and domains. This tool combines protein peeling and sword
variation in the observed plasma levels in different patients algorithm, giving protein units in contact probability and
under disease conditions. Several novel computational meth- ambiguity index.
ods now allow rapid and robust screening of functional SNPs
and detection of the deleterious effect of nsSNPs on protein 2.2. Structure similarity analysis
structure and function (De Alencar & Lopes, 2010). Therefore,
to understand the variability of Metrn and Metrnl levels in The structure similarity analysis was performed using DALI
diseases, we have performed a comprehensive in silico ana- and Foldseek search tools. DALI is a web server used to com-
lysis of SNPs from these genes. pare protein structures using the distance matrix alignment
Although there have been numerous studies on the func- method (Holm, 2022), and Foldseek is a fast and accurate
tional implications of these proteins in clinical, cell culture structural search tool that uses the 3Di alphabet-based
and animal models, the crystal structures for Metrn and method (van Kempen et al., 2022). For the structure similarity
Metrnl have not been studied. A crosslinking mass spectrom- analysis, we separated the domains of AF2 models based on
etry-based study that identified the presence of 5 disulfide our domain prediction and used them individually for homo-
bonds between the ten conserved cysteines in mice Metrn is logs search. From the resultant structures, the top 5 models
 JOURNAL OF BIOMOLECULAR STRUCTURE AND DYNAMICS 3

were selected based on the Z-score and TM-score from DALI 2.6. Protein–protein docking
and Foldseek, respectively.
Docking of the Metrnl and KIT receptor ectodomain was per-
formed using ClusPro and HDock. The KIT receptor structure
2.3. Conservation analysis was retrieved from RCSB PDB (PDB ID: 2E9W), and the AF2
predicted Metrnl was used as a ligand. Default parameters
The Consurf web server was used for the conservation ana- were used for docking analysis in both software. The ClusPro
lysis of Metrn and Metrnl using the sequences and predicted tool performs rigid docking using PIPER, followed by RMSD-
models (Ashkenazy et al., 2016). Using Bayesian inference, based clustering of the 1000 lowest-energy structures. These
the server predicts the structurally and functionally important 1000 structures are then refined using CHARMM energy mini-
residues based on sequence homology and phylogenetic mization to generate 30 models (Desta et al., 2020; Kozakov
relations. The server scores each amino acid of a given et al., 2013, 2017; Vajda et al., 2017). The HDock server per-
sequence from the range 1 to 9 based on its conservation forms global docking for the input PDB structures through
throughout evolution. The number 9 represents the highest, an FFT-based search method (Yan et al., 2020).
and the number 1 represents the lowest level of conserva- The top 10 docked structures in both tools were used for
tion. We used the Uniref90 database for sequence search further analysis. The docked structures were selected based
using the MMseq2 algorithm with an E-value cut-off of on their binding position, similar to SCF, which is the native
0.0001 to find 500 homologous sequences. Default parame- ligand for a KIT receptor, and the interaction energy was
ters were used for the CD-HIT cut-off for homolog selection, analyzed using their docking score and PRODIGY values (Xue
multiple sequence alignment, and phylogenetic reconstruc- et al., 2016). The docked complex structures were visualized
tion for the calculation of conservation scores. and analyzed using ChimeraX.

2.4. Identification of motifs and saturated mutagenesis 2.7. Identification of deleterious/disease SNP variants

The unique sequences of all species used for Consurf analysis The complete list of SNPs includes synonymous mutations,
were taken for motif analysis. The motif identification was intronic regions SNPs and UTR region SNPs of human Metrn
performed using the MEME suite server (Bailey et al., 2015) and Metrnl were obtained from the Ensembl database (Howe
with a given length of 8–20 of at least 10 motifs. The motifs et al., 2021; Hunt et al., 2018) and curated by selecting only
were analyzed based on the occurrence and their E-value. the coding region missense SNPs of full-length protein
The saturated mutagenesis was performed using the SNAP2 sequences from the entire list. We used seven different tools
server (Bromberg & Rost, 2007; Hecht et al., 2013, 2015) for viz., SIFT (Ng & Henikoff, 2003), Provean (Choi & Chan, 2015),
the given human Metrn and Metrnl sequences. MetaSNP (Capriotti et al., 2013), Polyphen-2 (Adzhubei et al.,
2010), pMut (Lo
pez-Ferrando et al., 2017), PhD-SNP (Capriotti
& Fariselli, 2017), and SNAP (S. Li et al., 2007) to identify the
2.5. Prediction of possible binding or functional deleterious and disease variants which can alter the structure
interface regions and function of Metrn and Metrnl protein based on the
sequence alteration. SIFT predicts protein function based on
The protein-protein interaction regions were predicted using
the sequence homology and the physical properties of
two tools ScanNet and MaSIF. ScanNet (Spatio chemical
amino acids in the mutated structure. SNPs with scores less
arrangement of ’neighbours’ Neural network) is a geometric than 0.05 were considered deleterious, while those with
deep learning method that builds representations of atoms scores of 0.05–1.0 were considered neutral or benign. The
and amino acids based on the spatio chemical arrangement PROVEAN uses an alignment-based scoring approach by
of neighbours (Tubiana et al., 2022). Its predictions are local, measuring the changes in sequence similarity with wild type
invariant on Euclidean transformations, and integrate infor- before and after the substitution of amino acids to deter-
mation from atoms, amino acids, structure, and MSA mine the deleterious effects of the substitution. If the score
(Multiple Sequence Alignment) in a synergistic method. The was below or equal to the predefined threshold, the SNP
tool is accessible using a web server. MaSIF (Molecular sur- was considered to have a deleterious effect on protein func-
face interaction fingerprinting) is a tool accessible on Google tion. Meta-SNP uses a random forest-based binary classifier
Colab dMaSIF that is based on the geometric deep learning to discriminate between disease-related and polymorphic
method to identify the regions involved in biomolecular non-synonymous substitutions. It is an integration of the
interactions (Gainza et al., 2020). We used the AF2 models of four existing methods: PANTHER, PhD-SNP, SIFT and SNAP. If
Metrn and Metrnl and calculated the protein–protein inter- the score is more than 0.5, it indicates that the mutant will
action regions using default parameters on both these tools. have a disease-causing or deleterious effect. Polyphen-2 pre-
The electrostatic potential was analyzed using APBS dicts the impact of amino acid substitution based on a clus-
(Adaptive Poisson-Boltzmann solver) with AMBER force field tering algorithm that uses eight sequence-based and three
parameters (Baker et al., 2001). The resultant structures were structure-based predictive methods. The functional signifi-
visualized and analyzed using ChimeraX (Pettersen et al., cance of the substitution was predicted by a naive Bayes
2021). classifier. The scoring function predicts the range from 0.0 to
4 S. S. SHANKAR ET AL.

0.15 as benign and 0.15 to 1.0 as damaging. pMut uses a the pLDDT score (Metrn-81.4, Metrnl-82.5) and its predicted
neural network-based classifier based on a Random Forest aligned error (PAE) (Supplementary Figure 1A–H and 2A–H).
classifier trained with SwissProt datasets for understanding The RoseTTAfold-based structure prediction using Robetta
the effect of amino acid substitution using sequence conser- online web server yielded a predicted structure confidence
vation and predicted physio-chemical properties. It provides score of 0.63 and 0.66 for Metrn and Metrnl, respectively
scores between 0 and 1, wherein those from 0 to 0.5 are (Supplementary Figures 4 and 5). The de novo models of
considered neutral, and from 0.5 to 1 are considered dele- both Metrn and Metrnl are beta-sheet-rich proteins (Figure
terious mutations. SNAP (Screening for non-acceptable poly- 1A, B) containing 15 beta-sheets, two alpha-helical regions,
morphisms) is another neural network-based prediction tool and 15 loop regions each. We also found that the AF2 and
that uses evolutionary information and predicted aspects of RF predicted all the conserved five disulfide bonds (Figure
protein structure to predict the functional effects of amino 1C, E), which had been previously identified (Wen et al.,
acid substitutions. PhD-SNP is a binary classifier based on a 2017). This indicates the accuracy of both prediction tools
Gradient Boosting algorithm and trained using
36,000 and that the predicted models would be highly similar to
benign and pathogenic variants extracted from the Clinvar the native structure of the protein.
dataset. The scoring function provides a score between 0 We also validated the predicted models using the
and 1, where an SNP with a score of more than 0.5 was pre- Ramachandran plot and the ERRAT plot using the SAVES ser-
dicted as a deleterious mutation. The variants which were ver. The Ramachandran plot analysis of Metrn and Metrnl
predicted to show deleterious and disease effects in all seven AF2 models (Supplementary Figure 3A–D) showed residues
tools were selected and further analyzed for effect on pro- in the favoured regions of 92.1%, and 92% and the ERRAT
tein stability. score quality factor of around 86.1 and 93.9. The RF Metrn
models showed the favoured areas of about 86% and add-
itional allowed regions of 12%. In the case of Metrnl, RF
2.8. Prediction of variant effect on protein stability models showed favoured areas of 81.7%, allowed regions of
To understand the effect of the selected SNP variants on pro- 13.8% and disallowed regions of around 3.1%, and the
tein dynamics and stability, we used the following tools: ERRAT scores of Metrn and Metrnl were 83.4 and 95.7%
mCSM, INPS-3D, CUPSAT, DeepDDG, DynaMut and FoldX. (Supplementary Figure 6A–D). Based on the stereo-chemical
mCSM predicts the impact of missense mutations using validation, AF2 models were selected for further structural
graph-based signatures (Pires et al., 2014). The INPS tool is and docking analysis. The PAE plot of AF2 models showed
trained to predict the thermodynamic free energy change two domain regions (Supplementary Figures 1F and 2F) for
using support vector machine regression (Savojardo et al., both Metrn and Metrnl. Thus, we deduced that both are
2016). CUPSAT web tool uses torsion angle potential and multi-domain proteins connected by a flexible loop/hinge
structure-specific atom potentials to predict the free energy region. The N-terminal domain contains an anti-parallel beta-
values (Parthiban et al., 2006). The DeepDDG tool is based sheet, and the C-terminal domain contains a beta-barrel-like
on a neural network-based method to understand the point structure and the AF2 model alignment of Metrn and Metrnl
mutations induced changes in the stability of the proteins showed RMSD of 3.76 Å and TM-score-0.74 (Figure 1D). We
(Cao et al., 2019). DynaMut calculates the vibrational entropy further performed structural alignment of the AF2 and RF
changes upon mutations (Rodrigues et al., 2018). FoldX cal- models and analyzed the RMSD and TM-score. We observed
culates the free energy changes between the native and that only one domain aligned between the models; in the
mutant structure of a protein using an empirical force field Metrn model, the N-terminal domain was aligned and
approach (Buß et al., 2018). The SNPs that were predicted to showed an RMSD of 2.65 Å (128 residues). In the case of
be structurally destabilizing by all six tools were further ana- Metrnl, the C-terminal domain was aligned and showed an
lyzed. The RMSF analysis of selected Metrn and Metrnl RMSD of 3.66 Å (136 residues) as shown in Supplementary
mutants was performed using CABS-flex 2.0 web server to Figure-7.
understand the fluctuations in protein stability caused by the We further investigated the domain boundary regions of
mutation (Kuriata et al., 2018). Metrn and Metrnl AF2 models using Sword2 and TopDomain
web servers. As shown in Figure 2A, B, both proteins contain
two domains connected by a loop/hinge (L11-Metrn, a3-
3. Results and discussion Metrnl) region. Both the tools predicted the same residue
number for the domain boundary, viz., Gly-154 in the case of
3.1. Structure prediction and domain identification of
Metrn, and Arg-175 in Metrnl (Supplementary Figure 10).
human Metrn and Metrnl
Based on the domain boundary and structural analysis, we
Alphafold2 (AF2) and RoseTTAfold (RF) were used to gener- named the N-terminal residues from 24 to 152 in Metrn and
ate the de novo structural models since they are the best 46 to 175 in Metrnl as Domain-1 (D1) and the C-terminal res-
structure prediction tools in CASP-14 (Pereira et al., 2021). idues from 170 to 293 in Metrn and 189 to 311 in Metrnl as
Using the ColabFold pipeline for AF2 model prediction with Domain-2 (D2). Also, the residues from 153 to 170 in Metrn
default settings, a total of 5 models were predicted. Of these, and 176 to 188 in Metrnl form the hinge region responsible
model No. 3 was ranked as the topmost model for both for the conformational shift between the domains, as shown
Metrn and Metrnl (Figure 1A, B) based on the validation of in Supplementary Figure-7. The D1 contains two disulfide
 JOURNAL OF BIOMOLECULAR STRUCTURE AND DYNAMICS 5

Figure 1. Cartoon representation of Alphafold2 predicted structures of Meteorin (A) and Meteorin-like (B), the five disulfide linkages of both proteins are shown in
ribbon structure (C-E), and the structural alignment of Metrn and Metrnl and its RMSD values and TM-score (D).

bonds, and D2 contains three disulfide bonds. Further struc- and Foldseek search using the full-length protein structure,
tural alignment of the domains between AF2 and RF models but it showed an alignment of structures to D2 predomin-
(Supplementary Figure-7) showed RMSD of 2.5 Å (D1-Metrn) antly (data not shown). In both tools, the sequence similarity
and 2.06 Å (D1-Metrnl). The alignment of D2 regions showed of the homologous structures was found to be < 10%, the
an RMSD of 2.08 Å (D2-Metrn) and 1.46 Å (D2-Metrnl). There Z-score of DALI was a maximum of 8.1 and the Foldseek TM-
were structural as well sequence differences between Metrn score was between 0.5 and 0.6. We selected the top five
and Metrnl in loop regions L2, L8, L9 and L13. The connect- models for structural homologs assessment in both tools.
ing loop/hinge region in Metrn was found to be larger Based on the structural similarity analysis of each domain,
than Metrnl. These regions also showed low confidence we found that D1 was similar to the CUB domain and D2
pLDDT scores in AF2 models which are highlighted in was similar to the NTR domain with an oligonucleotide-fold
Supplementary Figure 8A, B and 9. Further validation of (OB-fold) structure (Figure 3).
these regions may help to understand their role in domain The CUB domain comprised ten b-strands with the jelly
conformational changes and overall protein dynamics. roll fold topology and was so named since it was first identi-
Overall, this is the first study to assess Metrn and Metrnl fied in Complement subunits C1r/C1s, sea Urchin epidermal
structure and their domain regions using advanced predic- growth factor, and Bone morphogenetic protein 1. The bio-
tion tools, which will help in better understanding of these logical functions of CUB domain-containing proteins are
proteins in various functional characterization. mainly complement activation, calcium binding, axon guid-
ance, synaptic functions, tissue regeneration, and fertilization.
From the Foldseek and DALI search, the D1 of Metrn and
3.2. Structural similarity analysis
Metrnl were like CUB domain-containing proteins (Figure 3,
Based on our predicted models, we tried to understand the Supplementary Figure-11) Spermadhesin (1SPP), collagen
functional regions of both proteins using the structure simi- binding protein (5IKU, 3JQX), complement protein C1s, C1r
larity approach to identify the structural homologs of Metrn (4LMF, 6F1D), Wnt modulator Kremen (5FWT, 5FWV),
and Metrnl. For this, we searched the D1 (N-terminal) and D2 Neuropilin domains (2QQI, 4GZ9), Mannan binding serine
(C-terminal) domains of both proteins separately in DALI and peptidase-1 (5CKM), periplasmic lysozyme inhibitor (4DZG)
Foldseek servers against known structures in the PDB data- and cubilin (3KQ4). In all these proteins, the functional
base. The resulting structures were assessed based on their regions are mainly present in the putative surface-exposed
Z-score and TM-score, respectively. We also performed DALI loops and used for receptor signalling. Similarly, both Metrn
6 S. S. SHANKAR ET AL.

Figure 2. Contact map representation of domain boundary analysis using Sword2 webserver Metrn (A) and Metrnl (B).

and Metrnl N-terminal domains contained the loop regions and we identified 6 unique conserved motifs in Metrn and
L2, L4, and L10 in the structure (Figure 3, Supplementary fig- Metrnl, as shown in Figure 4C. The Motifs 1–4 were predom-
ure-11), which indicates that this loop region of D1 could be inantly present in most of the given sequences. These motifs
a possible receptor binding region of Metrn and Metrnl. were unique and present only in Metrn and Metrnl proteins.
The D2 of both proteins was similar to the NTR (Netrin- The CXXC active site was found in motif-2 of Metrn and
like) domain identified in the C-terminal part of netrins that Metrnl, which is known to be important for the activity of
encompasses 130 residues forming a beta-barrel structure thioredoxin-related proteins (Quan et al., 2007) and is pre-
with two terminal alpha-helices. As shown in Figure 3 and sent in zinc-finger domains (Frauer et al., 2011). But so far,
Supplementary Figure-11, the Metrn and Metrnl D2 were there are no reports of Metrn or Metrnl having any redox
similar to N-terminal part of TIMP1 (tissue inhibitor of matrix function and zinc-binding or DNA-binding activity. Also, sat-
metalloprotein, 1UEA, 6N9D, 2J0T), human PCOLCE-1 (Type-1 uration mutagenesis using the SNAP2 tool indicated that
procollagen C-proteinase enhancer protein, 1UAP), comple- mutations in these identified motif regions showed a nega-
ment proteins C3, C4, C5 (4FXG, 5JPN, 4E0S, 5I5K, 3PVM) and tive effect (Figure 4D, E) indicating that they play a major
excretory-secretion protein (3NSW). The NTR domains of role in the protein structure stability and function. This is the
these proteins are involved in various functions like inhibi- first study to identify the conserved motif regions of Metrn
tory activity of metalloproteinase and complement activation. and Metrnl, and it will help in identifying the functional
The functional regions of the NTR domain reside around properties of these motifs in future studies.
the disulfide-linked beta sheets (Bodden et al., 1994). As mentioned earlier, it was recently identified that
Metrnl activates the endothelial c-KIT receptor protein and
Interestingly, Metrn and Metrnl also contained the three
promotes heart repair. However, the functional regions of
disulfide bonds in D2. Thus, the beta-sheets b-13, 14, 15 and
Metrnl involved in activation have not been evaluated. So,
L15 regions in D2 could play a role in receptor activation.
we further investigated the protein–protein interaction
Another protein that contains both CUB and NTR domains is
regions of our models to identify the functional regions
PCOLCE-1 (CUB1-CUB2-NTR), wherein one CUB domain car-
using ScanNet and Masif algorithms, which use deep learn-
ries a protease function (Blanc et al., 2007), whereas the NTR
ing and machine learning approaches to find the possible
domain is involved in controlling the metalloproteinase activ-
protein binding regions, respectively. As shown in Figure 5A,
ity (Liepinsh et al., 2003). Since Metrn and Metrnl also show
B, both the tools predicted the surface exposed loops L2, L4,
both CUB-NTR domains, it raises the question about each
L5, L10 and surface residues of beta-sheets b-1, 2, 4, 9 and 6
domain’s individual function and the role of the hinge in the CUB domain of both Metrn and Metrnl as receptor-
region. interacting regions. In the case of the NTR domain, the beta-
sheets b-13 and loops L14 and 15 were identified by both
3.3. Prediction of motifs and protein–protein interaction the tools as protein-protein interaction regions. The residues
regions of the hinge region and the two helical regions in the NTR
domain were identified as binding regions only in the
We also identified the conserved motif regions of Metrn and Metrnl model. We further validated the predicted regions
Metrnl proteins using Consurf and MEME-suite server. Based using electrostatic potential and the analysis of conserved
on the Consurf analysis, the CUB and NTR domain regions of residues. We found that the identified surface residues of
both Metrn and Metrnl were found to have more evolution- loop and beta-sheets of protein-protein interaction regions
arily conserved residues, and the hinge regions contained contain negatively charged residues and are highly con-
highly variable residues (Figure 4A, B). The sequences ana- served (Figure 5C, D). The structural difference of the hinge
lyzed for Consurf analysis were used for motif prediction, region between Metrn and Metrnl also indicates that it is
 JOURNAL OF BIOMOLECULAR STRUCTURE AND DYNAMICS 7

Figure 3. Structural alignment of similar fold proteins identified using foldseek search for each domain of Metrn (A) and Metrnl (B). The PDB-id and TM-score are
mentioned for each structure.

not only used for the conformational change of domains,

3.4. Protein-protein docking of Metrnl and KIT receptor
but it may also play an important role in receptor binding
or activation in the case of Metrnl. These results increased To understand further about the protein interacting regions,
the confidence of our structure similarity studies and con- we performed the protein-protein docking of the KIT-ectodo-
firmed the functional regions of Metrn and Metrnl. This is main receptor and Metrnl using ClusPro and HDock server.
the first study to identify the receptor binding regions of The KIT receptor contains five immunoglobulin-like domains
Metrn and Metrnl which will help in the functional charac- (Ig-like) domains, and it dimerizes upon binding of its ligand
terization of each domain. SCF. SCF binds mainly to the concave surface of domains 1,
8 S. S. SHANKAR ET AL.

Figure 4. The sequence conservation analysis using Consurf (A-Metrn, B-Metrnl) and the identified motifs of both proteins using MEME is shown as a Stacked
sequence logo (C). Heatmap showing the effect of saturated mutagenesis for both Metrn (D) and Metrnl (E).

2 and 3 of the KIT receptors and induces dimerization with chain A-site 2 and 3 of the receptors. In the case of RF-
between domains 4 and 5 (Yuzawa et al., 2007). The ligand Metrnl, the CUB domain interacts with the chain A-site 1, 2
binding residues of each domain of the KIT receptor are and 3 predominantly in ClusPro and with the chain B-site 1,
named site1, site2 and site3, respectively, and are mainly 2 and 3 in HDock (Figure 7). We further analyzed the inter-
composed of charged residues. SCF’s interaction is mainly acting profiles of the Metrnl-KIT docked complexes and, the
based on complementary electrostatic interaction, salt number of calculated salt bridges, hydrogen bonds and non-
bridges, and a few hydrogen bonds (Liu et al., 2007). We bonded contacts are shown in Supplementary table-1. From
used both the AF2 and RF-predicted models of Metrnl (AF2- domain-2 of the KIT receptor, residues Lys247, and Asn293
Metrnl and RF-Metrnl) to understand the different conforma- showed hydrogen bond interactions with Asp160 and
tions of receptor binding. Based on the docking score and Gln159 of Metrnl, respectively. The positively charged amino
Prodigy binding energy (DG) values we analyzed the interact- acids Lys247 and His263 of the KIT receptor form a salt
ing regions of complexes using PDBsum. The ClusPro com- bridge with negatively charged Asp160 and Glu259 of
plex of AF2-Metrnl and the KIT receptor showed a docking Metrnl, respectively. Further, the amino acid residues of
score of 1217.6 and a Prodigy binding energy of domain 2-KIT receptor, Tyr125, Lys127, Lys203 and Phe208
20.6 kcal/mol. In the case of the RF-Metrnl-KIT complex, the showed non-bonded electrostatic interactions with Glu122,
docking score was 1063.6, and the binding energy was Lys66, Glu67, and Gly125 of Metrnl. Similarly, in domain 3
19.5 kcal/mol. In both models, the top score docked com- the amino acids Ser240, Ser241, Val242, Tyr243, Lys247,
plexes were mainly from electrostatic-favoured clusters. The Tyr259, Asn260, Trp262, His263, His264, Asp266, Glu270,
docking score and binding energy of the HDock docked Tyr291, Asn293, Thr295 of the KIT receptor showed electro-
complexes were 323.36, 14.1 kcal/mol and 307.19, static interaction with Asn118, Glu154, Thr156, Asp160,
17.1 kcal/mol in AF2 and RF-Metrnl models, respectively. As Gly134, Trp250, Leu258, Glu259, Gly261, Arg263, Pro190,
shown in Figure 6A–D, we identified that both models of Gln159, Arg65 of Metrnl. The analysis noted that charged
Metrnl prefer the same ligand binding site of the KIT recep- and acidic residues participate predominantly in the total sta-
tor, similar to SCF in both docking methods. In ClusPro, the bilization energies. As shown in our electrostatic potential
CUB domain AF2-Metrnl bound to chain A-site 2 and the analysis, the AF2-Metrnl CUB domain loops L5 and L10, and
NTR domain bound to chain B-site 2 of the KIT receptor the NTR region a-6 and b-13 contain electronegative resi-
using the electrostatics-favoured method. In HDock, both dues, and these regions are involved in the binding of the
domains of AF2-Metrnl interaction were mainly observed KIT receptor (Supplementary Figure-12). As mentioned above,
 JOURNAL OF BIOMOLECULAR STRUCTURE AND DYNAMICS 9

Figure 5. Surface representation of protein–protein interaction regions of Metrn and Metrnl using ScanNet (A) and MASIF (B). The electrostatic potential (C) and
evolutionary conservation analysis (D) of Metrn and Metrnl.

Figure 6. Crystal structure of KIT-SCF complex (A) Pdb-id-2E9W. The ClusPro docked complex of AF2-Metrnl-KIT (B) and RF-Metrnl-KIT(C). The HDock docked com-
plex of AF2-Metrnl-KIT (D) and RF-Metrnl-KIT (E).

SCF forms a homodimer to dimerize the KIT receptor, and in dimerization of the receptor and activates the KIT signalling
the case of Metrnl, we hypothesize that each domain of cascade. Based on our docking experiments, the interface
Metrnl binds to the individual KIT receptors and, by possible regions of these proteins involved in KIT binding were also
conformational changes of the hinge region, initiates the identified in our predicted motifs.
10 S. S. SHANKAR ET AL.

Figure 7. Surface representation of KIT receptor binding regions of Metrnl AF2 (A, B), RF (C, D).

Figure 8. Number of nsSNPs having a deleterious/disease effect on the biological function of Metrn (A) and Metrnl (B) based on seven different tools. Cartoon rep-
resentation of nsSNPs which can cause structural instability on Metrn (C) and Metrnl (D) based on the instability prediction tools.

3.5. Identification of high-risk SNP variants based on

sequence and structure analysis the effect of a mutation in the functional regions, we selected
and investigated the non-synonymous (nsSNPs)/missense var-
Previous studies have shown altered plasma protein levels of iants of Metrn and Metrnl using the sequence variant effect
Metrnl in disease conditions like diabetes and obesity, how- and structural effect prediction tools. A total of 268 missense
ever, the results are inconclusive. We hypothesize that the SNPs of Metrn and 240 of Metrnl were selected for analysis
plasma protein levels of Metrnl are affected by the presence using seven different tools (Figure 8A, B). In the case
of missense SNPs that can affect protein stability. To identify of Metrn, we found 124 mutations in SIFT, 86 in Polyphen-2,
 JOURNAL OF BIOMOLECULAR STRUCTURE AND DYNAMICS 11

Figure 9. Cabs-flex RMSF analysis of the resultant mutant structure and regions fluctuating by the mutation were highlighted for Metrn (A) and Metrnl (B).

104 in Provean, 75 in PhD-SNP, 136 in SNAP, 65 in Meta-SNP, to the above criteria, we selected the variants which had a
and 65 in pMut that had a deleterious or disease-causing destabilizing effect in all six tools. Out of 36 missense SNPs in
effect. In the case of Metrnl, 118 mutations were found to be Meteorin, 16 were found to be structurally destabilizing,
causing disease or deleterious effects in SIFT, 112 in whereas, in the case of Metrnl, 10 missense SNPs (Figure 8C, D)
PolyPhen-2, 112 in Provean, 65 in PhD-SNP, 129 in SNAP, 61 were predicted to be destabilizing from the selected 32 SNPs.
in Meta-SNP, and 57 in pMut. Since each tool has a different The selected SNPs were investigated for their evolutionary
approach for predicting the effect of the SNPs, we selected conservation using the Consurf tool. Based on the Consurf
only the mutations that showed a deleterious effect in all the analysis, out of the 16 SNPs of Metrn, 12 were present in the
tools to obtain the consensus variants with high confidence in highly conserved regions with high-ranking scores of 8 and 9
the prediction. As a result, 36 and 32 mutations in Metrn and (L49P, G63E, G192R, P246L, C288S, E149G, T185P, R217G/C,
Metrnl respectively were found to have deleterious effects G260R, F273L/S, W58L, F259S) and only one residue G249S
and these variants were analyzed further for structural stability was found in the variable region with an average score of 6
(Supplementary File-1). (Figure 4A, B). In the case of Metrnl, out of 10 SNPs, nine
The stability of the protein determines its function, and a were present in the score range of 8 and 9 (G78A, G87S,
decrease in the stability leads to aggregation or misfolding. R163W, L172V, E195A, C201S, D204N, A285G, T202P) and
This will affect the protein levels as well as its function. Here, residue R293G was present in the variable region with a
the protein stability of the selected variants was analyzed based score of 4. This indicates that these amino acids/positions
on the free energy DDG values from six different tools. Similar are essential in structural stability and function.
12 S. S. SHANKAR ET AL.

We have also investigated wild-type and mutants’ flexibil- analysis can be used to design therapeutic strategies for
ity and dynamic protein fluctuations using the RMSF (Root such diseases. In addition, we have also identified the pos-
Mean Square Fluctuation) values from the CABS-flex 2.0 web sible high-risk variants which can influence the protein struc-
server. Compared to the wild type, the Meteorin mutants ture and expression levels of these proteins. Based on our
W58L, T185P, G260R, and G63E showed the highest fluctu- study, further genome-wide association studies of these pro-
ation in loop regions L2, L5, L10, and L15, respectively teins SNPs can be designed to correlate the presence of
(Figure 9A and Supplementary Figure-13A). Moreover, G26V SNPs with plasma protein levels to elucidate the basis of
and C288S showed the highest fluctuation in the loop their altered expression levels in various disease conditions.
regions L11. On the other hand, the ten Metrnl mutants
T202P, E195A, A285G, T202P, G87S, G78A, C201S, and E195A
showed the highest fluctuation in loop and helix regions L2, Disclosure statement
L5, L8, L10, a3, L14, L15 respectively (Figure 9B and All the authors declare no conflict of interest.
Supplementary Figure-13B).
Interestingly, the predicted receptor-binding regions L2,
L5, L8, and L10 of the CUB domain, the L14 and L15 of the
Funding
NTR domain and the hinge regions of both proteins were The project work is supported by a research grant from the Council of
affected by the SNP variations. This showed that the muta- Scientific and Industrial Research (CSIR), New Delhi, India, and CSIR-
tions predominantly affect the stability of the functional National Chemical Laboratory-Pune.
regions and may also affect the stability at the protein level.
Thus, the variation in plasma Metrnl level observed in differ- References
ent disease conditions could also result from the presence of
Adzhubei, I. A., Schmidt, S., Peshkin, L., Ramensky, V. E., Gerasimova, A.,
SNPs in different patients. Further studies of SNPs of Metrnl
Bork, P., Kondrashov, A. S., & Sunyaev, S. R. (2010). A method and ser-
at the genome level in disease conditions will be helpful in ver for predicting damaging missense mutations. Nature Methods,
giving a clear indication of the factors affecting these protein 7(4), 248–249. https://doi.org/10.1038/nmeth0410-248
levels. Alkhairi, I., Cherian, P., Abu-Farha, M., Al Madhoun, A., Nizam, R.,
Melhem, M., Jamal, M., Al-Sabah, S., Ali, H., Tuomilehto, J., Al-Mulla, F.,
& Abubaker, J. (2019). Increased expression of meteorin-like hormone
4. Conclusion in type 2 diabetes and obesity and its association with irisin. Cells,
8(10), 1283. https://doi.org/10.3390/cells8101283
Metrn and Metrnl are secretory proteins involved in various Ashkenazy, H., Abadi, S., Martz, E., Chay, O., Mayrose, I., Pupko, T., & Ben-
functions related to neural development, metabolic diseases, Tal, N. (2016). ConSurf 2016: An improved methodology to estimate
and immune regulation. However, both proteins’ structures and visualize evolutionary conservation in macromolecules. Nucleic
and functional regions have not yet been reported. In the Acids Research, 44(W1), W344–W350. https://doi.org/10.1093/NAR/
GKW408
current study, we have analyzed two different structural Bae, J. Y. (2018). Aerobic exercise increases meteorin-like protein in
aspects of these proteins. First, we have analyzed the muscle and adipose tissue of chronic high-fat diet-induced obese
Alphafold2 and RoseTTAfold predicted models of these pro- mice. BioMed Research International, 2018, 6283932. https://doi.org/10.
teins to understand their unique structural features and 1155/2018/6283932
receptor binding regions. Secondly, we have studied the Baek, M., DiMaio, F., Anishchenko, I., Dauparas, J., Ovchinnikov, S., Lee,
G. R., Wang, J., Cong, Q., Kinch, L. N., Schaeffer, R. D., Mill
an, C., Park,
effect of Single Nucleotide Polymorphisms in these proteins
H., Adams, C., Glassman, C. R., DeGiovanni, A., Pereira, J. H., Rodrigues,
on their structure and function. This is the first study to A. V., van Dijk, A. A., Ebrecht, A. C., … Baker, D. (2021). Accurate pre-
report that both proteins contain two domains and are con- diction of protein structures and interactions using a three-track
nected by a linker loop region. We have identified that the neural network. Science (New York, N.Y.), 373(6557), 871–876. https://
D1 belongs to the CUB domain family and the D2 belongs doi.org/10.1126/science.abj8754
to the NTR domain family and have determined the regions Bailey, T. L., Johnson, J., Grant, C. E., & Noble, W. S. (2015). The MEME
Suite. Nucleic Acids Research, 43(W1), W39–W49. https://doi.org/10.
involved in their function by analyzing the protein-protein
1093/nar/gkv416
interacting regions and protein-receptor docking. The Baker, N. A., Sept, D., Holst, M. J., & McCammon, J. A. (2001). The adap-
domain analysis study will be helpful in understanding the tive multilevel finite element solution of the Poisson-Boltzmann equa-
function of these proteins, and the role of these domains tion on massively parallel computers. IBM Journal of Research and
can be further determined using in vitro studies. We report Development, 45(3.4), 427–438. https://doi.org/10.1147/rd.453.0427
Blanc, G., Font, B., Eichenberger, D., Moreau, C., Ricard-Blum, S., Hulmes,
that the two domains of Metrnl mimic the role of SCF homo-
D. J. S., & Moali, C. (2007). Insights into How CUB domains can exert
dimer in the binding and activation of the KIT receptor. specific functions while sharing a common fold: Conserved and spe-
Further studies using Molecular dynamics simulation of cific features of the CUB1 domain contribute to the molecular basis
the receptor with Metrnl are required to give a better under- of procollagen C-proteinase enhancer-1 activity. The Journal of
standing of their assembly and the conformational changes. Biological Chemistry, 282(23), 16924–16933. https://doi.org/10.1074/
Previous studies have shown that Metrn is involved in neur- JBC.M701610200
Bodden, M. K., Harber, G. J., Birkedal-Hansen, B., Windsor, L. J., Caterina,
onal repair and development (Jeong et al., 2008), and while
N. C., Engler, J. A., & Birkedal-Hansen, H. (1994). Functional domains
Metrnl is found to play a role in heart repair (Reboll et al., of human TIMP-1 (tissue inhibitor of metalloproteinases). Journal of
2022), increasing insulin sensitivity (Jung et al., 2018), and Biological Chemistry, 269(29), 18943–18952. https://doi.org/10.1016/
adipose browning (Rao et al., 2014). Therefore, our structural S0021-9258(17)32258-5
 JOURNAL OF BIOMOLECULAR STRUCTURE AND DYNAMICS 13

Boulling, A., Le Mar
echal, C., Trouve
, P., Ragu
en

es, O., Chen, J. M., & atherosclerosis. Diabetes Research and Clinical Practice, 150, 57–63.
F
erec, C. (2007). Functional analysis of pancreatitis-associated mis- https://doi.org/10.1016/j.diabres.2019.02.026
sense mutations in the pancreatic secretory trypsin inhibitor (SPINK1) Frauer, C., Rottach, A., Meilinger, D., Bultmann, S., Fellinger, K.,
gene. European Journal of Human Genetics : EJHG, 15(9), 936–942. Haseno €der, S., Wang, M., Qin, W., So €ding, J., Spada, F., & Leonhardt, H.
https://doi.org/10.1038/sj.ejhg.5201873 (2011). Different binding properties and function of CXXC zinc finger
Bromberg, Y., & Rost, B. (2007). SNAP: Predict effect of non-synonymous domains in Dnmt1 and Tet1. PLoS One, 6(2), e16627. https://doi.org/
polymorphisms on function. Nucleic Acids Research, 35(11), 3823–3835. 10.1371/journal.pone.0016627
https://doi.org/10.1093/nar/gkm238 Gainza, P., Sverrisson, F., Monti, F., Rodola, E., Boscaini, D., Bronstein,
Buß, O., Rudat, J., & Ochsenreither, K. (2018). FoldX as Protein M. M., & Correia, B. E. (2020). Deciphering interaction fingerprints
Engineering Tool: Better than random based approaches? from protein molecular surfaces using geometric deep learning.
Computational and Structural Biotechnology Journal, 16, 25–33. https:// Nature Methods, 17(2), 184–192. https://doi.org/10.1038/s41592-019-
doi.org/10.1016/j.csbj.2018.01.002 0666-6
Cao, H., Wang, J., He, L., Qi, Y., & Zhang, J. Z. (2019). DeepDDG: Garc
es, M. F., Sanchez, E., Cardona, L. F., Simanca, E. L., Gonz
alez, I., Leal,
Predicting the stability change of protein point mutations using L. G., Mora, J. A., Bedoya, A., Alzate, J. P., S
anchez, A.
Y., Eslava-
neural networks. Journal of Chemical Information and Modeling, 59(4), Schmalbach, J. H., Franco-Vega, R., Parra, M. O., Ru
ız-Parra, A. I.,
1508–1514. https://doi.org/10.1021/acs.jcim.8b00697 Di
eguez, C., Nogueiras, R., & Caminos, J. E. (2015). Maternal serum
Capriotti, E., Altman, R. B., & Bromberg, Y. (2013). Collective judgment meteorin levels and the risk of preeclampsia. Plos One, 10(6),
predicts disease-associated single nucleotide variants. BMC Genomics. e0131013. https://doi.org/10.1371/journal.pone.0131013
14(Suppl 3), S2. https://doi.org/10.1186/1471-2164-14-S3-S2 Gregor, A., Meerbrei, T., Gerstner, T., Toutain, A., Lynch, S. A., Stals, K.,
Capriotti, E., & Fariselli, P. (2017). PhD-SNP g: A web server and light- Maxton, C., Lemke, J. R., Bernat, J. A., Bombei, H. M., Foulds, N., Hunt,
weight tool for scoring single nucleotide variants. Nucleic Acids D., Kuechler, A., Beygo, J., Sto €be, P., Bouman, A., Palomares-Bralo, M.,
Research, 45(W1), W247–W252. https://doi.org/10.1093/nar/gkx369 Santos-Simarro, F., Garcia-Minaur, S., … Zweier, C. (2022). De novo
Choi, Y., & Chan, A. P. (2015). PROVEAN web server: A tool to predict the missense variants in FBXO11 alter its protein expression and subcellu-
functional effect of amino acid substitutions and indels. Bioinformatics lar localization. Human Molecular Genetics, 31(3), 440–454. https://doi.
(Oxford, England), 31(16), 2745–2747. https://doi.org/10.1093/bioinfor- org/10.1093/hmg/ddab265
matics/btv195 Hecht, M., Bromberg, Y., & Rost, B. (2013). News from the protein mut-
Choudhury, A., Mohammad, T., Samarth, N., Hussain, A., Rehman, M. T.,
ability landscape. Journal of Molecular Biology, 425(21), 3937–3948.
Islam, A., Alajmi, M. F., Singh, S., & Hassan, M. I. (2021). Structural gen-
https://doi.org/10.1016/j.jmb.2013.07.028
omics approach to investigate deleterious impact of nsSNPs in con-
Hecht, M., Bromberg, Y., & Rost, B. (2015). Better prediction of functional
served telomere maintenance component 1. Scientific Reports,
effects for sequence variants. BMC Genomics, 16(S8), S1. https://doi.
11(10202) https://doi.org/10.1038/s41598-021-89450-7
org/10.1186/1471-2164-16-S8-S1
Chung, H. S., Hwang, S. Y., Choi, J. H., Lee, H. J., Kim, N. H. N. H., Yoo,
Holm, L. (2022). Dali server: Structural unification of protein families.
H. J., Seo, J. A., Kim, S. G., Kim, N. H. N. H., Baik, S. H., & Choi, K. M.
Nucleic Acids Research, 50(W1), W210–W215. https://doi.org/10.1093/
(2018). Implications of circulating Meteorin-like (Metrnl) level in
nar/gkac387
human subjects with type 2 diabetes. Diabetes Research and Clinical
Howe, K. L., Achuthan, P., Allen, J., Allen, J., Alvarez-Jarreta, J., Amode,
Practice, 136, 100–107. https://doi.org/10.1016/j.diabres.2017.11.031
M. R., Armean, I. M., Azov, A. G., Bennett, R., Bhai, J., Billis, K., Boddu,
Colovos, C., & Yeates, T. O. (1993). Verification of protein structures:
S., Charkhchi, M., Cummins, C., Da Rin Fioretto, L., Davidson, C.,
Patterns of non-bonded atomic interactions. Protein Science : A
Dodiya, K., El Houdaigui, B., Fatima, R., … Flicek, P. (2021). Ensembl
Publication of the Protein Society, 2(9), 1511–1519. https://doi.org/10.
2021. Nucleic Acids Research, 49(D1), D884–D891. https://doi.org/10.
1002/pro.5560020916
1093/NAR/GKAA942
Cretin, G., Galochkina, T., Vander Meersche, Y., De Brevern, A. G., Postic,
Hunt, S. E., McLaren, W., Gil, L., Thormann, A., Schuilenburg, H.,
G., & Gelly, J. C. (2022). SWORD2: Hierarchical analysis of protein 3D
Sheppard, D., Parton, A., Armean, I. M., Trevanion, S. J., Flicek, P., &
structures. Nucleic Acids Research, 50(W1), W732–W738. https://doi.
org/10.1093/nar/gkac370 Cunningham, F. (2018). Ensembl variation resources. Database, 2018,
Dai, Y., Ma, J., Jiang, R., Lu, J., Chen, J., Lu, X., Dai, Y., Ma, J., Jiang, R., bay119. https://doi.org/10.1093/database/bay119
Zhan, X., Chen, S., Feng, L., & Zhang, Q. (2022). Article Meteorin links Islam, M. J., Khan, A. M., Parves, M. R., Hossain, M. N., & Halim, M. A.
the bone marrow hypoxic state to hematopoietic stem/progenitor (2019). Prediction of deleterious non-synonymous SNPs of human
cell mobilization ll ll Meteorin links the bone marrow hypoxic state to STK11 gene by combining algorithms, molecular docking, and
hematopoietic stem/progenitor cell mobilization. Cell Reports, 40(12), molecular dynamics simulation. Scientific Reports, 9(16426). https://doi.
111361. https://doi.org/10.1016/j.celrep.2022.111361 org/10.1038/s41598-019-52308-0
De Alencar, S. A., & Lopes, J. C. D. (2010). A Comprehensive in silico ana- Jeong, A. P., Hye, S. L., Keum, J. K., Soo, Y. P., Jeong, H. K., Choe, G.,
lysis of the functional and structural impact of SNPs in the IGF1R Kweon, H. S., Hyun, S. S., Ahn, J. C., Young, S. Y., & Kim, K. W. (2008).
gene. Journal of Biomedicine & Biotechnology, 2010, 715139. https:// Meteorin regulates angiogenesis at the gliovascular interface. Glia,
doi.org/10.1155/2010/715139 56(3), 247–258. https://doi.org/10.1002/glia.20600
Delaunay, K., Sellam, A., Dinet, V., Moulin, A., Zhao, M., Geliz
e, E., Jørgensen, J. R., Fransson, A., Fjord-Larsen, L., Thompson, L. H., Houchins,
Canonica, J., Naud, M. C., Crisanti-Lassiaz, P., & Behar-Cohen, F. (2021). J. P., Andrade, N., Torp, M., Kalkkinen, N., Andersson, E., Lindvall, O.,
Meteorin is a novel therapeutic target for wet age-related macular Ulfendahl, M., Brunak, S., Johansen, T. E., & Wahlberg, L. U. (2012).
degeneration. Journal of Clinical Medicine, 10(13), 2973. https://doi. Cometin is a novel neurotrophic factor that promotes neurite out-
org/10.3390/jcm10132973 growth and neuroblast migration in vitro and supports survival of spi-
Desta, I. T., Porter, K. A., Xia, B., Kozakov, D., & Vajda, S. (2020). ral ganglion neurons in vivo. Experimental Neurology, 233(1), 172–181.
Performance and Its Limits in Rigid Body Protein-Protein Docking. https://doi.org/10.1016/j.expneurol.2011.09.027
Structure (London, England : 1993), 28(9), 1071–1081.e3. https://doi. Jørgensen, J. R., Xu, X. J., Arnold, H. M., Munro, G., Hao, J. X., Pepinsky,
org/10.1016/j.str.2020.06.006 B., Huang, C., Gong, B. J., Wiesenfeld-Hallin, Z., Wahlberg, L. U., &
Du, Y., Ye, X., Lu, A., Zhao, D., Liu, J., Cheng, J., & Yang, T. (2020). Inverse Johansen, T. E. (2012). Meteorin reverses hypersensitivity in rat mod-
relationship between serum Metrnl levels and visceral fat obesity els of neuropathic pain. Experimental Neurology, 237(2), 260–266.
(VFO) in patients with type 2 diabetes. Diabetes Research and Clinical https://doi.org/10.1016/j.expneurol.2012.06.027
Practice, 161, 108068. https://doi.org/10.1016/j.diabres.2020.108068 Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O.,
El-Ashmawy, H. M., Selim, F. O., Hosny, T. A. M., & Almassry, H. N. (2019). Tunyasuvunakool, K., Bates, R., Z
ıdek, A., Potapenko, A., Bridgland, A.,
Association of low serum Meteorin like (Metrnl) concentrations with Meyer, C., Kohl, S. A. A., Ballard, A. J., Cowie, A., Romera-Paredes, B.,
worsening of glucose tolerance, impaired endothelial function and Nikolov, S., Jain, R., Adler, J., … Hassabis, D. (2021). Highly accurate
14 S. S. SHANKAR ET AL.

protein structure prediction with AlphaFold. Nature, 596(7873), 583– cell differentiation and promotes axonal extension. The EMBO Journal,
589. https://doi.org/10.1038/s41586-021-03819-2 23(9), 1998–2008. https://doi.org/10.1038/sj.emboj.7600202
Jung, T. W., Lee, S. H., Kim, H. C., Bang, J. S., Abd El-Aty, A. M., Onalan, E., Cavlı, C., Dogan, Y., Onalan, E., Gozel, N., Buran, _I., Yakar, B., &
Hacım€ uft€  lu, A., Shin, Y. K., & Jeong, J. H. (2018). METRNL attenu-
uog Donder, E. (2020). Low serum levels of meteorin-like/subfatin: An indi-
ates lipid-induced inflammation and insulin resistance via AMPK or cator of diabetes mellitus and insulin resistance? Endokrynologia
PPAR$d$-dependent pathways in skeletal muscle of mice. Polska, 71(5), 397–403. https://doi.org/10.5603/EP.A2020.0038
Experimental & Molecular Medicine, 50(9), 1–11. https://doi.org/10. Parthiban, V., Gromiha, M. M., & Schomburg, D. (2006). CUPSAT:
1038/s12276-018-0147-5 Prediction of protein stability upon point mutations. Nucleic Acids
Kim, D. E., Chivian, D., & Baker, D. (2004). Protein structure prediction Research, 34(Web Server), W239–W242. https://doi.org/10.1093/nar/
and analysis using the Robetta server. Nucleic Acids Research, 32(Web gkl190
Server issue), W526–W531. https://doi.org/10.1093/NAR/GKH468 Pereira, J., Simpkin, A. J., Hartmann, M. D., Rigden, D. J., Keegan, R. M., &
Kim, Y. Y., Moon, J. S., Kwon, M. C., Shin, J., Im, S. K., Kim, H. A., Han, Lupas, A. N. (2021). High-accuracy protein structure prediction in
J. K., & Kong, Y. Y. (2014). Meteorin regulates mesendoderm develop- CASP14. Proteins, 89(12), 1687–1699. https://doi.org/10.1002/prot.
ment by enhancing Nodal expression. PLoS One, 9(2), e88811. https:// 26171
doi.org/10.1371/journal.pone.0088811 Pettersen, E. F., Goddard, T. D., Huang, C. C., Meng, E. C., Couch, G. S.,
Kozakov, D., Beglov, D., Bohnuud, T., Mottarella, S. E., Xia, B., Hall, D. R.,
Croll, T. I., Morris, J. H., & Ferrin, T. E. (2021). UCSF ChimeraX:
& Vajda, S. (2013). How good is automated protein docking? Proteins,
Structure visualization for researchers, educators, and developers.
81(12), 2159–2166. https://doi.org/10.1002/prot.24403
Protein Science : A Publication of the Protein Society, 30(1), 70–82.
Kozakov, D., Hall, D. R., Xia, B., Porter, K. A., Padhorny, D., Yueh, C.,
https://doi.org/10.1002/pro.3943
Beglov, D., & Vajda, S. (2017). The ClusPro web server for protein-pro-
Pires, D. E. V., Ascher, D. B., & Blundell, T. L. (2014). MCSM: Predicting the
tein docking. Nature Protocols, 12(2), 255–278. https://doi.org/10.1038/
effects of mutations in proteins using graph-based signatures.
nprot.2016.169
Kuriata, A., Gierut, A. M., Oleniecki, T., Ciemny, M. P., Kolinski, A., Bioinformatics (Oxford, England), 30(3), 335–342. https://doi.org/10.
Kurcinski, M., & Kmiecik, S. (2018). CABS-flex 2.0: A web server for fast 1093/bioinformatics/btt691
simulations of flexibility of protein structures. Nucleic Acids Research, Qi, Q., Jun Hu, W., Li Zheng, S., Long Zhang, S., Ying Le, Y., Yong Li, Z.,
46(W1), W338–W343. https://doi.org/10.1093/nar/gky356 Yu Miao, C., Hu, W., Jun, Zheng, S., Li, Zhang, S., Long, Le, Y., Ying, Li,
Lee, H. S., Han, J., Lee, S. H., Park, J. A., & Kim, K. W. (2010). Meteorin pro- Z.-Y., & Miao, C. y (2020). Metrnl deficiency decreases blood HDL chol-
motes the formation of GFAP-positive glia via activation of the Jak- esterol and increases blood triglyceride. Acta Pharmacologica Sinica,
STAT3 pathway. Journal of Cell Science, 123(Pt 11), 1959–1968. https:// 41(12), 1568–1575. https://doi.org/10.1038/s41401-020-0368-8
doi.org/10.1242/jcs.063784 Quan, S., Schneider, I., Pan, J., Von Hacht, A., & Bardwell, J. C. A. (2007).
Lee, J. H., Kang, Y. E., Kim, J. M., Choung, S., Joung, K. H., Kim, H. J., & The CXXC motif is more than a redox rheostat. The Journal of
Ku, B. J. (2018). Serum Meteorin-like protein levels decreased in Biological Chemistry, 282(39), 28823–28833. https://doi.org/10.1074/
patients newly diagnosed with type 2 diabetes. Diabetes Research and jbc.M705291200
Clinical Practice, 135, 7–10. https://doi.org/10.1016/j.diabres.2017.10. Rao, R. R., Long, J. Z., White, J. P., Svensson, K. J., Lou, J., Lokurkar, I.,
005 Jedrychowski, M. P., Ruas, J. L., Wrann, C. D., Lo, J. C., Camera, D. M.,
Liepinsh, E., B
anyai, L., Pintacuda, G., Trexler, M., Patthy, L., & Otting, G. Lachey, J., Gygi, S., Seehra, J., Hawley, J. A., & Spiegelman, B. M.
(2003). NMR structure of the netrin-like domain (NTR) of human type (2014). Meteorin-like is a hormone that regulates immune-adipose
I procollagen C-proteinase enhancer defines structural consensus of interactions to increase beige fat thermogenesis. Cell, 157(6), 1279–
NTR domains and assesses potential proteinase inhibitory activity and 1291. https://doi.org/10.1016/j.cell.2014.03.065
ligand binding. The Journal of Biological Chemistry, 278(28), 25982– Reboll, M. R., Klede, S., Taft, M. H., Cai, C.-L., Field, L. J., Lavine, K. J.,
25989. https://doi.org/10.1074/JBC.M302734200 Koenig, A. L., Fleischauer, J., Meyer, J., Schambach, A., Niessen, H. W.,
Li, S., Ma, L., Li, H., Vang, S., Hu, Y., Bolund, L., & Wang, J. (2007). Snap: Kosanke, M., van den Heuvel, J., Pich, A., Bauersachs, J., Wu, X.,
An integrated SNP annotation platform. Nucleic Acids Research, Zheng, L., Wang, Y., Korf-Klingebiel, M., Polten, F., & Wollert, K. C.
35(suppl_1), D707–D710. https://doi.org/10.1093/nar/gkl969 (2022). Meteorin-like promotes heart repair through endothelial KIT
Liu, H., Chen, X., Focia, P. J., & He, X. (2007). Structural basis for stem cell receptor tyrosine kinase. Science, 376(6599), 1343–1347. https://doi.
factor-KIT signaling and activation of class III receptor tyrosine kin- org/10.1126/science.abn3027
ases. The EMBO Journal, 26(3), 891–901. https://doi.org/10.1038/sj. Rodrigues, C. H. M., Pires, D. E. V., & Ascher, D. B. (2018). DynaMut:
emboj.7601545 Predicting the impact of mutations on protein conformation, flexibil-
Li, Z. Y., Zheng, S. L., Wang, P., Xu, T. Y., Guan, Y. F., Zhang, Y. J., & Miao,
ity, and stability. Nucleic Acids Research, 46(W1), W350–W355. https://
C. Y. (2014). Subfatin is a novel adipokine and unlike meteorin in adi-
doi.org/10.1093/nar/gky300
pose and brain expression. CNS Neuroscience & Therapeutics, 20(4),
Roman Laskowski, B. A., Macarthur, M. W., & Thornton, J. M. (1983).
344–354. https://doi.org/10.1111/cns.12219
PROCHECK: a program to check the stereochemicai quality of protein

pez-Ferrando, V., Gazzo, A., De La Cruz, X., Orozco, M., & Gelp
ı, J. L.
Lo
structures. J. Appl. Cryst., 26, 283–291.
(2017). PMut: A web-based tool for the annotation of pathological
Rupe
rez, C., Ferrer-Curriu, G., Cervera-Barea, A., Florit, L., Guitart-Mampel,
variants on proteins, 2017 update. Nucleic Acids Research, 45(W1),
M., Garrabou, G., Zamora, M., Crispi, F., Fernandez-Sola, J., Lupo
n, J.,
W222–W228. https://doi.org/10.1093/nar/gkx313
Mirdita, M., Sch€ utze, K., Moriwaki, Y., Heo, L., Ovchinnikov, S., & Bayes-Genis, A., Villarroya, F., & Planavila, A. (2021). Meteorin-
Steinegger, M. (2022). ColabFold: Making protein folding accessible to like/Meteorin-b protects heart against cardiac dysfunction. Journal of
all. Nature Methods, 19(6), 679–682. https://doi.org/10.1038/s41592- Experimental Medicine, 218(5), e20201206. https://doi.org/10.1084/jem.
022-01488-1 20201206
Mulnaes, D., Golchin, P., Koenig, F., & Gohlke, H. (2021). TopDomain: Savojardo, C., Fariselli, P., Martelli, P. L., & Casadio, R. (2016). INPS-MD: A
Exhaustive protein domain boundary metaprediction combining mul- web server to predict stability of protein variants from sequence and
tisource information and deep learning. Journal of Chemical Theory structure. Bioinformatics (Oxford, England), 32(16), 2542–2544. https://
and Computation, 17(7), 4599–4613. https://doi.org/10.1021/acs.jctc. doi.org/10.1093/bioinformatics/btw192
1c00129 Schmid, A., Karrasch, T., & Sch€affler, A. (2021). Meteorin-like protein
Ng, P. C., & Henikoff, S. (2003). SIFT: Predicting amino acid changes that (Metrnl) in obesity, during weight loss and in adipocyte differenti-
affect protein function. Nucleic Acids Research, 31(13), 3812–3814. ation. Journal of Clinical Medicine, 10(19), 4338. https://doi.org/10.
https://doi.org/10.1093/nar/gkg509 3390/jcm10194338
Nishino, J., Yamashita, K., Hashiguchi, H., Fujii, H., Shimazaki, T., & Schro€dinger, LLC. (2015). The fPyMOLg Molecular Graphics System,
Hamada, H. (2004). Meteorin: A secreted protein that regulates glial Version
1.8.
 JOURNAL OF BIOMOLECULAR STRUCTURE AND DYNAMICS 15

Tokuriki, N., & Tawfik, D. S. (2009). Stability effects of mutations and pro- Wen, D., Xiao, Y., Vecchi, M. M., Gong, B. J., Dolnikova, J., & Pepinsky,
tein evolvability. In Current Opinion in Structural Biology, 19(5), 596– R. B. (2017). Determination of the disulfide structure of murine mete-
604. https://doi.org/10.1016/j.sbi.2009.08.003 orin, a neurotrophic factor, by LC-MS and electron transfer dissoci-
Tubiana, J., Schneidman-Duhovny, D., & Wolfson, H. J. (2022). ScanNet: ation-high-energy collisional dissociation analysis of proteolytic
An interpretable geometric deep learning model for structure-based fragments. Analytical Chemistry, 89(7), 4021–4030. https://doi.org/10.
protein binding site prediction. Nature Methods, 19(6), 730–739. 1021/acs.analchem.6b04600
https://doi.org/10.1038/s41592-022-01490-7 Wu, Q., Dan, Y.-L., He, Y.-S., Xiang, K., Hu, Y.-Q., Zhao, C.-N., Zhong, X.,
Ushach, I., Burkhardt, A. M., Martinez, C., Hevezi, P. A., Gerber, P. A., Wang, D.-G., & Pan, H.-F. (2020). Circulating meteorin-like levels in
Buhren, B. A., Schrumpf, H., Valle-Rios, R., Vazquez, M. I., Homey, B., & patients with type 2 diabetes mellitus: A meta-analysis. Current
Zlotnik, A. (2015). METEORIN-LIKE is a cytokine associated with barrier Pharmaceutical Design, 26(44), 5732–5738. https://doi.org/10.2174/
tissues and alternatively activated macrophages. Clinical Immunology 1381612826666201007163930
(Orlando, Fla.), 156(2), 119–127. https://doi.org/10.1016/j.clim.2014.11. Xue, L. C., Rodrigues, J. P., Kastritis, P. L., Bonvin, A. M., & Vangone, A.
006 (2016). PRODIGY: A web server for predicting the binding affinity of
Vajda, S., Yueh, C., Beglov, D., Bohnuud, T., Mottarella, S. E., Xia, B., Hall, protein-protein complexes. Bioinformatics (Oxford, England), 32(23),
D. R., & Kozakov, D. (2017). New additions to the ClusPro server moti- 3676–3678. https://doi.org/10.1093/bioinformatics/btw514
vated by CAPRI. Proteins, 85(3), 435–444. https://doi.org/10.1002/prot. Yan, Y., Tao, H., He, J., & Huang, S. (2020). The HDOCK server for inte-
25219 grated protein – protein docking. Nature Protocols, 15(5), 1829–1852.
van Kempen, M., Kim, S. S., Tumescheit, C., Mirdita, M., So €ding, J., https://doi.org/10.1038/s41596-020-0312-x
Steinegger, M., Kempen, M., van, Kim, S. S., Tumescheit, C., Mirdita, Yuzawa, S., Opatowsky, Y., Zhang, Z., Mandiyan, V., Lax, I., & Schlessinger,
M., Gilchrist, C. L. M., So€ding, J., & Steinegger, M. (2022). Foldseek: J. (2007). Structural basis for activation of the receptor tyrosine kinase
Fast and accurate protein structure search. BioRxiv. 2022.02.07.479398 KIT by stem cell factor. Cell, 130(2), 323–334. https://doi.org/10.1016/j.
https://doi.org/10.1101/2022.02.07.479398 cell.2007.05.055
Wang, K., Li, F., Wang, C., Deng, Y., Cao, Z., Cui, Y., Xu, K., Ln, P., & Sun, Zhang, Y., & Skolnick, J. (2005). TM-align: A protein structure alignment
Y. (2019). Serum levels of meteorin-like (Metrnl) are increased in algorithm based on the TM-score. Nucleic Acids Research, 33(7), 2302–
patients with newly diagnosed type 2 diabetes mellitus and are asso- 2309. https://doi.org/10.1093/nar/gki524
ciated with insulin resistance. Medical Science Monitor : International Zheng, S. L., Li, Z. Y., Song, J., Liu, J. M., & Miao, C. Y. (2016). Metrnl: A
Medical Journal of Experimental and Clinical Research, 25, 2337–2343. secreted protein with new emerging functions. Acta Pharmacologica
https://doi.org/10.12659/MSM.915331 Sinica, 37(5), 571–579. https://doi.org/10.1038/aps.2016.9