We have analyzed a large collection of coding junctions generated in h
uman cells. From this analysis, we infer the following about nucleotid
e processing at coding joints in human cells, First, the pattern of nu
cleotide loss from coding ends is influenced by the base composition o
f the coding end sequences. AT-rich sequences suffer greater loss than
do GC-rich sequences, Second, inverted repents can occur at ends that
have undergone nucleolytic processing, Previously, inverted repents (
P nucleotides) have been noted only at coding ends that have not under
gone nucleolytic processing, this observation being the basis for a mo
del in which a hairpin intermediate is formed at the coding ends early
in the reaction, Here, inverted repeats at processed coding ends were
present at approximately twice the number of junctions as P nucleotid
e additions, Terminal deoxynucleotidyl transferase (TdT) is required f
or the appearance of the inverted repeats at processed ends (but not f
ull-length coding ends), get statistical analysis shows that it is vir
tually impossible for the inverted repeats to be polymerized by TdT, T
hird, TdT additions are not random, It has long been noted that TdT ha
s a G utilization preference. In addition to the G preference, we find
that TdT adds strings of purines or strings of pyrimidines at a highl
y significant frequency. This tendency suggests that nucleotide-stacki
ng interactions affect TdT polymerization. All three of these features
place constraints on the extent of junctional diversity in human V(D)
J recombination.