HHrep: de novo protein repeat detection and the origin of TIM barrels

Abstract

HHrep is a web server for the de novo identification of repeats in protein sequences, which is based on the pairwise comparison of profile hidden Markov models (HMMs). Its main strength is its sensitivity, allowing it to detect highly divergent repeat units in protein sequences whose repeats could as yet only be detected from their structures. Examples include sequences with beta-propellor fold, ferredoxin-like fold, double psi barrels or (betaalpha)8 (TIM) barrels. We illustrate this with proteins from four superfamilies of TIM barrels by revealing a clear 4- and 8-fold symmetry, which we detect solely from their sequences. This symmetry might be the trace of an ancient origin through duplication of a betaalphabetaalpha or betaalpha unit. HHrep can be accessed at http://hhrep.tuebingen.mpg.de.

Figure 1

Start page (upper), sample help page (middle), repeat alignment displayed with JalView (28) (lower), and main results page with a profile–profile dot plot, summary hit list and pairwise self-alignments found by HMM–HMM comparison (right). The self-alignments, which are also shown as blue traces in the upper right half of the dot plot, can be selected and deselected by clicking on them.

Figure 2

ROC plot comparing HHrep with and without secondary structure scoring to the method TRUST (12). Filled diamonds and triangles mark a P-value of 10⁻³ and 10⁻², respectively, the open diamond and triangle indicate a probability of 50% and 10%, respectively.

Figure 3

The 2-, 4- and 8-fold repeat pattern in TIM barrel sequences. The figure shows profile–profile dot plots of HisF (1thf_D) (upper row) and KDPG aldolase (1fq0_A) (lower row), before (left) and after (middle) incorporation of transitivity information through alignment merging. The dot plots on the right were obtained from the middle ones by lowering the score threshold from the default value (0.4 bits) to 0.1 bits.