4 receptor (MC4R), one of 5 melanocortin receptors (MCRs) in the G-protein coupled receptor (GPCR) superfamily, is expressed primarily in the hypothalamus, particularly in the paraventricular nucleus and the lateral hypothalamus, and at high levels in the brain stem, notably the nucleus of the solitary tract (1)
. Endogenous melanocortin ligands (e.g.
, α-MSH) activate the MCRs by stimulating the formation of cAMP and the subsequent activation of protein kinase A. Additional cAMP-dependent and -independent signaling pathways involving MAPK (2)
, and protein kinase C (4)
have also been reported. In vitro
, the MC4R has been shown to be both constitutively active (5)
and physically recycled (6)
; the receptor is internalized from the plasma membrane into an intracellular compartment and subsequently transported back to the cell surface, where the molecule is reused for signaling.
The MC4R, and two other members of the melanocortin family (the MC1R and MC3R), are unique in that they are the only members of the GPCR family known to have endogenous antagonists (“inverse agonists”; see refs. 7, 8
): Agouti signaling protein (ASIP) and Agouti-related protein (AgRP). ASIP, a peptide normally expressed in skin cells, antagonizes the melanocortin 1 receptor (MC1R) in the hair follicle, causing pigment-type switching from eumelanin (black) to pheomelanin (yellow) (9)
. When overexpressed ectopically, ASIP blocks MC4R and MC1R signaling in the brain and the hair follicle (10)
, respectively. AgRP, released by neurons in the arcuate nucleus of the hypothalamus, is a natural antagonist of the MC3R and MC4R (11)
. The molecular mechanisms by which ASIP and AgRP act as inverse agonists remain unclear (8)
MC4R is involved in weight regulation: mice homozygous for a null allele of the Mc4r
are obese because of hyperphagia and decreased energy expenditure (12)
. In humans, MC4R
mutations constitute the most commonly identified cause of monogenic obesity (~2.5–3% of all cases of severe human obesity, BMI>40; see refs. 13, 14
). Over 80 nonsynonymous mutations in the MC4R
coding sequence have been reported, yet analyses of their consequences for protein structure and function remain limited. Dominantly inherited nonsense and missense mutations that completely disable MC4R represent the most common apparent molecular mechanism for obesity; however, recessive modes of inheritance are also encountered (15)
Experimental studies have elucidated some of the structure-function aspects of MC4R and found that all of its conserved residues are involved in agonist binding and signaling (16)
. In most cases, however, the exact residues related to any specific function remain unknown. Single amino acid (AA) mutations associated with obesity most often result in intracellular retention of the mutant receptor, followed by receptors with impaired ligand-binding capacity and receptors with defective signal transduction (17)
. In addition, obese humans segregating for MC4R
alleles with nonsynonymous AA variants that appear to function normally in vitro
have been reported (17)
A comprehensive mutagenesis survey that substitutes all MC4R residues by each of the 19 nonnative AAs and tests the functional consequences is not experimentally feasible. In silico
modeling may provide insights into the relationships among sequence, structure, and function. The most reliable technique for such modeling and structure prediction is comparative modeling (18)
. This technique finds a protein with a structure (template) that is similar to the protein to be modeled (target) and then copies the coordinates of the backbone of the template onto the target (19, 20)
; unaligned residues typically remain without a model. The two proteins with known structure that are most sequence-similar to MC4R are rhodopsin (RHO; 24% sequence ID in 143 aligned residues; ref. 21
) and the β2-andrenegic receptor (B2AR; 24% sequence ID in 307 aligned residues; ref. 22
). Structural models for MC4R (and other GPCRs) have been primarily based on the structure of RHO. However, because of the substantial sequence differences between the two receptors, RHO-based MC4R models are not sufficiently accurate to provide reliable per-residue data regarding function. Models based on B2AR might be more useful, because this receptor is more similar in sequence to MC4R. However, there are no comprehensive and convincing B2AR-based structural models for MC4R available.
To reduce the level of detail, rather than mapping residues onto 3-dimensional (3-D) coordinates, we limited our analyses to a 1-dimensional perspective including information regarding the location of putative transmembrane helices (). Even at this reduced resolution, however, minor differences remain in the details of protein structure: for example, the locations of the membrane helices are not exact but are resolved to a few residues. This minor uncertainty also affects the accuracy of our predictions of the regions in the extracellular and cytoplasmic regions.
Figure 1. Important hMC4R residues. One-dimensional structure of the hMC4R receptor (adapted from Govaerts et al.; ref. 42). At present, no reliable 3-D structure at higher level of detail is available. Circles represent constituent AAs (1-letter alphabet). Shading (more ...)
For each human MC4R (hMC4R) AA residue, we used SNAP (Supplemental and ref. 23
) to predict the functional consequences of exchanging the native AA with each of the 19 other naturally occurring AAs. SNAP is a neural network-based method that uses evolutionary conservation and structure/function relationships to predict the functional consequences of AA substitutions. SNAP was optimized on the Protein Mutant Database (PMD) (24)
. Cross-validation tests indicate that SNAP identifies >80% of the nonneutral variants with 77% accuracy and >76% of the neutral mutations with 80% accuracy at its default threshold (hMC4R was not used in the development of SNAP). SNAP performed only slightly worse for the subset of Swiss-Prot (25)
annotated transmembrane proteins in the test set, identifying 74% of the nonneutral variants at 73% accuracy and 72% of the neutral mutations with 74% accuracy. The advantage of SNAP is that it can be applied to serial analyses of the effects of substitution at a specific position of all 19 nonnative AAs. This paper reports the first attempt to perform such an analysis of a membrane protein.
Our comprehensive in silico mutagenesis predicted functionally essential residues in hMC4R, mouse MC4R (mMC4R), and human MC1R (hMC1R) without relying on experimental information, and without knowledge of the specific function in which a given residue is involved. To assess the sensitivity and specificity of our computational approach we compared experimentally determined functional consequences of mutations to our predictions. We also examined the specific prediction differences and similarities between the receptors providing some insight into their molecular physiology.