What makes antibodies specific

Phagocytic enhancement by antibodies is called opsonization. In another process, complement fixation, IgM and IgG in serum bind to antigens, providing docking sites onto which sequential complement proteins can bind.

The combination of antibodies and complement enhances opsonization even further, promoting rapid clearing of pathogens. Not all antibodies bind with the same strength, specificity, and stability. In fact, antibodies exhibit different affinities attraction depending on the molecular complementarity between antigen and antibody molecules.

An antibody with a higher affinity for a particular antigen would bind more strongly and stably. It would be expected to present a more challenging defense against the pathogen corresponding to the specific antigen. Antibody affinity, avidity, and cross reactivity : a Affinity refers to the strength of single interactions between antigen and antibody, while avidity refers to the strength of all interactions combined.

The term avidity describes binding by antibody classes that are secreted as joined, multivalent structures such as IgM and IgA. Although avidity measures the strength of binding, just as affinity does, the avidity is not simply the sum of the affinities of the antibodies in a multimeric structure.

The avidity depends on the number of identical binding sites on the antigen being detected, as well as other physical and chemical factors. Typically, multimeric antibodies, such as pentameric IgM, are classified as having lower affinity than monomeric antibodies, but high avidity.

Antibodies secreted after binding to one epitope on an antigen may exhibit cross reactivity for the same or similar epitopes on different antigens. Cross reactivity occurs when an antibody binds not to the antigen that elicited its synthesis and secretion, but to a different antigen. Because an epitope corresponds to such a small region the surface area of about four to six amino acids , it is possible for different macromolecules to exhibit the same molecular identities and orientations over short regions.

Cross reactivity can be beneficial if an individual develops immunity to several related pathogens despite having been exposed to or vaccinated against only one of them. For instance, antibody cross reactivity may occur against the similar surface structures of various Gram-negative bacteria. Conversely, antibodies raised against pathogenic molecular components that resemble self molecules may incorrectly mark host cells for destruction, causing autoimmune damage.

These antibodies may have been initially raised against the nucleic acid of microorganisms, but later cross-reacted with self-antigens. This phenomenon is also called molecular mimicry. Privacy Policy. Skip to main content. The Immune System. Search for:. Antibody Structure Variations in antibody structure allow great diversity of antigen recognition among different antibodies.

Learning Objectives Differentiate among the classes of antibodies. Each antibody has a unique variable region, which is responsible for antigen detection and specificity.

IgAs, secreted in the milk, tears and mucous, are the most numerous antibodies produced; inside of the body, circulating IgGs are the most abundant.

Key Terms immunoglobulin : any of the glycoproteins in blood serum that respond to invasion by foreign antigens and that protect the host by removing pathogens; also known as an antibody antigen : a substance that binds to a specific antibody; may cause an immune response B cell : a lymphocyte, developed in the bursa of birds and the bone marrow of other animals, that produces antibodies and is responsible for the immune system epitope : that part of a biomolecule such as a protein that is the target of an immune response; the part of the antigen recognized by the immune system.

Antibody Functions Antibodies, part of the humoral immune response, are involved in pathogen detection and neutralization. These antibodies then bind specifically with the foreign molecule and allow the immune system to eliminate the molecule from the system.

In some cases, these antibodies can disable pathogens such as viruses directly due to the binding action.

After the foreign molecule has been eliminated, B cells remain in the bloodstream ready to produce antibodies if the antigen is encountered again. From the perspective of developing a custom antibody against a protein antigen, the immune system captures the protein, breaks it down into individual epitopes and presents these epitopes to the B cells so that development of antibodies specific to those epitopes can begin.

The Zernike Moments, being translation, scale and rotation invariant, provide a detailed yet robust representation of a surface. In short, the Zernike moments give a compact description of an image by deconvoluting it into a set of primitive functions centered in the middle of the image, each describing a different type of shape. The Zernike Moments were calculated using a modified version of the python package by Scott Grandison et al. Examples of the shape description of each moment are shown in Figure 1D.

To optimize the computational time and focus on a coarse description of the patch, less dependent on the correct side chain positioning in the structure, only Zernike Moments of order 3 and 4 mainly describing the vertical and horizontal tilts were included.

Figure 1. A Conjoint Triads amino acid classes and representation of method on a sequence level. B Structural representation of Conjoint Triads classes mapped to an epitope patch. C The three principal components illustrated on an epitope patch. D Illustration of 4th order of Zernike Moments' descriptive shape excluding order 0 and 1. The amino acid composition and the conjoint triads were used to describe the patch composition statistics.

The amino acid composition was calculated as the frequency of each amino acid type in the patch. The conjoint triads are based on structural neighbors illustrated in Figures 1A,B Amino acids were assigned to one out of 7 classes displayed in Figure 1A. Finally, the frequencies of the total non-directional triad types were computed. Finally, a few features were included describing the physicochemical characteristics of the patch.

The exposed acceptor and donor atoms were calculated as the number of exposed h-bond acceptors and donor atoms, respectively, regardless of their actual involvement in any h-bond in the original structure. Additionally, the number of positively and negatively charged residues and number of aromatic residues were included. Surface patches were generated using a Monte Carlo MC approach The moves are then accepted or rejected using the Metropolis criterion.

Here, the set of features F includes the three ratios between the three principal components, PC1, PC2, and PC3 of the patch and the paratope, the ratio of the size of the patch to the size of the paratope, the ratio between the summed residue surface area of the patch and the surface of the paratope, and the ratio between paratope and epitope patch density.

The mean and standard deviation values of each feature were determined from the actual epitope-paratope pairs in a cross-validated manner, so that the patches generated for any antigen in a given partition are constructed from values obtained from the remaining 4 partitions.

Using this MC approach with a total of MC moves per simulation, patches MC patches were generated per antigen. Target values were assigned to MC generated epitope-paratope pairs based on their overlap with the real pairs as the product of the precision proportion of residues in the patch that are part of the actual epitope and recall proportion of epitope residues included in the patch. This target value is hence 1 if the patch overlaps perfectly with the actual epitope, and zero if no overlap is present.

To evaluate how well a model predicts patches overlapping to the real epitope, we defined patches with a target value above 0. Moreover, for each complex, we included 10 mis-paired paratope-epitope patches, obtained by pairing the real epitope patch with the paratope of an antibody from a different antibody cluster.

Given the very high specificity of antibodies, we assumed that they do not bind a random antigen, and therefore assigned a target score of 0 to the mis-paired patches. Three models were made Full, Minimal and Antigen model using different features to encode the patches.

Table 1 shows a summary of which features were used in the different models. The Full model included all calculated features, i. The Minimal model did not include the last three feature sets resulting in 62 features, 31 for each antibody and antigen patch.

The Antigen model was similar to the Full model, however, only including the antigen features. Feed forward neural networks were trained and their performance were evaluated using a nested 5 partition fold cross validation: one of the 5 partitions was in turn left out from the model training, and then the remaining 4 partitions were next split into 10 random sub-partitions maintaining the original clustering, and models were trained using fold cross-validation with early stopping.

Finally, the ensemble of these 10 models was used to predict the left-out partition in the outer 5-fold cross validation. As an initial analysis, we investigated correlations between structural and physicochemical properties of actual paratope and epitope patches. We compared the correlations of various structural features PC, size, and surface measured on both the paratope and the epitope patch, as shown in Figure 2.

Figure 2. Correlation matrix of structural and physicochemical features of the true paired paratope and epitope patches. As expected, these analyses demonstrated a high correlation between corresponding structural properties of the paratope and epitope; i. The same holds for the epitope shape PCs and surface. Similar results were obtained for physicochemical features: hydrophobicity, h-bond acceptors and donors.

Unsurprisingly, the number of available acceptors in a patch correlates with the number of available donors in the corresponding partner patch. These observations overall suggest that information contained within the antibody is of potentially use to gain insight into the shape and physicochemical properties of the cognate epitope. In order to prove that the observed correlations could be used to generate an improved prediction, we tested different prediction models each trained and evaluated using nested cross-validation on the training data for detail on model architecture, training hyper-parameters and model features see methods.

We trained 3 such methods using different subsets of the features from each patch. The first model was trained on the Antigen features only i. Here, no information on the paratope patch was included. The second model was a Minimal model , which was trained using a minimal set of features of antigen and paratope patches.

The third was the Full model expanding the Minimal model to include the additional features of structural conjoint triads, Zernike Moments, and maximal, minimal and average relative surface exposure of the paratope and epitope patches for details see materials and methods. We evaluated the performance of each of the three models by scoring, for each structure in the independent test partition, the actual epitope and MC patches.

The patches were sorted according to their score, and the performance was reported as the relative number of MC patches with a score higher than the epitope F rank.

In this way, a perfect prediction where the epitope is ranked at the top of the list would get a F rank score of 0, and a F rank score of 0. In Figure 3 , we show the results of this analysis for the 3 models in terms of a boxplots of the F rank values for each of the three prediction.

These plots confirm the superior performance of the Full model compared to the other two models with a median rank performance of 7. The results demonstrate that incorporating information of the antibody in the prediction model results in a high gain in predictive power. Figure 3. Box plot showing the distribution of the real epitope ranks within each Antibody-Antigen structure for the three prediction models; Antigen, Minimal, and Full.

Figure 4 provides another way to illustrate the predictive performance of the different models. Here, we show how many of the structures have a highly overlapping patch within the top 1, 5, 10, 15, 20 up to top MC patches for each of the 3 models. Additionally, we compared the performance of each models to DiscoTope Here, the patch score was calculated as the sum of DiscoTope predictions over all residues in a given patch. This analysis again clearly demonstrates that the Full model has the highest performance of all models confirming that the predictive power is increased by integrating information from the cognate antibody.

It is also interesting to observe that even the Antigen model achieves results that are slightly improved compared to Discotope Figure 4. X-axis indicating number of top predicted patches included and Y-axis showing the percentage of structures having at least one high overlapping within the selected pool.

One of the main goals of an antibody-guided epitope prediction tool would be to identify the cognate antigen target of a given antibody from a pool of potential antigens. This benchmark resulted in average and median ranks of Further details on the different performance measures used are given in Table 2. Next, for each antibody, we similarly score the paratope against the epitopes from all structures within the given data partition, and identified the rank of the cognate epitope patch within this sorted list Antigen Rank , resulting in an average rank of These analyses demonstrate that the model is capable of differentiating between real and mis-paired epitope-paratope pairs.

Table 2. Description of the ranking measurements used to describe performance of the models. As the PCs are highly correlated between paratope and epitope patches, one could speculate that the above performance values were driven by these structural similarities.

To investigate this, we repeated the experiment only including paratopes of similar shape. We performed a K-means clustering with 5 clusters within each test partition, based on a vector of the three PCs of the paratope. We then recalculated the ranking of each paratope only against paratopes in the same cluster, hence with similar PCs Structurally Similar Antibody Rank. This resulted in an average Structurally Similar Antibody Rank of This result indicates that the geometric differences i.

In the benchmark calculations conducted so far, we have focused on identifying surface patches corresponding to or overlapping the cognate epitope of a given antibody.