Large-scale analysis of human heavy chain V(D)J recombination patterns

Abstract

Background: The processes involved in the somatic assembly of antigen receptor genes are unique to the immune system and are driven largely by random events. Subtle biases, however, may exist and provide clues to the molecular mechanisms involved in their assembly and selection. Large-scale efforts to provide baseline data about the genetic characteristics of immunoglobulin (Ig) genes and the mechanisms involved in their assembly have recently become possible due to the rapid growth of genetic databases.

Results: We gathered and analyzed nearly 6,500 productive human Ig heavy chain genes and compared them with 325 non-productive Ig genes that were originally rearranged out of frame and therefore incapable of being biased by selection. We found evidence for differences in n-nucleotide tract length distributions which have interesting interpretations for the mechanisms involved in n-nucleotide polymerization. Additionally, we found striking statistical evidence for pairing preferences among D and J segments. We present a statistical model to support our hypothesis that these pairing biases are due to multiple sequential D-to-J rearrangements.

Conclusion: We present here the most precise estimates of gene segment usage frequencies currently available along with analyses regarding n-nucleotide distributions and D-J segment pair preferences. Additionally, we provide the first statistical evidence that sequential D-J recombinations occur at the human heavy chain locus during B-cell ontogeny with an approximate frequency of 20%.

Figure 1

Adjusted Residuals for D-J Segment Pairings. A heat map showing adjusted residual values for D-J segment pairings based on contingency table analysis of the P sequence data. Adjusted residuals are approximately independent and distributed as standard normals. Values greater than 1.96 (white) or less than -1.96 (dark gray) represent a significant departure from the expected value at a 95% confidence level.

Figure 2

Estimation Algorithm Depiction. Flow diagram depicting the steps for estimating the multiple recombination parameters in our statistical model. Rho (ρ) is the parameter for multiple recombination; it represents the probability of a subsequent recombination occurring given that one just occurred and that segments are available for another recombination. Changes to the parameters are accepted stochastically according to the Metropolis-Hastings criterion: with probability 1 if the new chi-square (${\chi }_{new}^2$) value is lower than the old value (${\chi }_{old}^2$), or with probability exp(0.5(${\chi }_{old}^2-{\chi }_{new}^2$)) (37).

Figure 3

Fitting Plots of n-Nucleotide Data. Plots of the observed n-nucleotide data for both the P and NP genes in both the VD and DJ junctions fit to a zero-inflated negative binomial distribution.

Figure 4

Confidence Regions for n-Nucleotide Data Fits. We fit our observed n-nucleotide addition data to the negative binomial distribution, and calculated both the maximum likelihood estimators plotted at (r, p) and the corresponding confidence regions.

Figure 5

J Gene Segment Usage Frequencies. Observed relative frequencies of JH gene segment usage in the P and NP gene sets.

Figure 6

D Gene Segment Usage Frequencies. Relative observed frequencies of DH gene segment usage by family in the P and NP gene sets, and comparison to germline complexity of each gene segment family. The germline complexity refers to the number of segments within the locus assigned to each family.

Figure 7

V Gene Segment Usage Frequencies. Observed relative frequencies of VH gene segment usage by family in P and NP sequences, and comparison to germline complexity of each gene segment family.