Biology Reference
In-Depth Information
algorithm that could build a model of the sickle cell fi ber from individual hemoglobin
molecules (Levinthal et al.
1975
). Shoshana Wodak, one of Levinthal's co-authors,
joined me in Pr. Georges Cohen's laboratory at the Pasteur Institute in Paris, and we
decided together to investigate what computer simulations could tell us about pro-
tein-protein recognition. For that purpose, we designed a procedure that generated
all the orientations of one protein relative to another, and brought the two surfaces
into contact by translation. To gain computer time and memory space, we borrowed
from Michael Levitt a simplified protein model that represented each amino acid
residue by a sphere of appropriate radius (Levitt
1976
). We allowed a degree of
penetration between the spheres, and estimated the quality of the fit by the number
of intersubunit residue-residue contacts. Our test system was the same trypsin/BPTI
complex as in Blow et al. (
1972
), but by then, Huber's lab had determined a X-ray
structure of the complex (Huber et al.
1974
), and issued coordinates that could serve
to assess the accuracy of the docking models. In the summer of 1976, we were given
access to a state-of-the-art computer in Orsay, France - one that was only about
10,000 times slower than a laptop today - during a workshop of the Centre Européen
de Calcul Atomique et Moléculaire (CECAM). In about an hour of cpu time, our
software (named DOCK like several others after it) generated models of the inhibitor
filling the active site of the protease in 2,300 different orientations. To our satisfac-
tion, an orientation close to Huber's X-ray structure showed a good fit, but there
were several other that achieved a similar score. In other terms, the procedure had
produced a native-like model of the assembly, plus some false positives. We attrib-
uted the false positives to the coarse nature of our score, which took into account the
geometric complementarity of the two molecular surfaces, but ignored their chemical
nature and the physics of their interaction (Wodak and Janin
1978
) .
Computational biology had no established status in the mid-1970s, and we had a
difficult time convincing journal editors that protein-protein docking was more than
a futile game. Yet, Levinthal had addressed a related question, protein folding, several
years before, and ambitious attempts were already being made to solve it in the
computer (Levitt and Lifson
1969
; Levitt
1976
; Némethy and Scheraga
1977
) . With
rigid molecules, docking is a much simpler problem than protein folding. Whereas
folding has thousands of degrees of freedom, docking has only six, and by restrict-
ing the search to the active site of trypsin, we had reduced that number to four,
which had made the calculation feasible.
The next application of our software was to simulate the allosteric transition of
hemoglobin. Hemoglobin is an order of magnitude larger than BPTI, but its twofold
symmetry also reduces the search to four degrees of freedom, and the computation
was within the reach of extant computers. It was done at the Free University,
Brussels, in the summer of 1981, also during a CECAM workshop (Janin and Wodak
1981
). We used a much improved version of DOCK to build hemoglobin tetramers
from alpha-beta dimers in a range of orientations that covered the T and R quater-
nary structures described by Perutz. The results showed that the allosteric transition
from R to T could not proceed along a linear pathway, due to steric hindrance at the
dimer-dimer interface, and it drew an alternative pathway in excellent agreement
with the classical description of Baldwin and Chothia (
1979
) .
Search WWH ::
Custom Search