Bridging topological and functional information in protein interaction networks by short loops profiling

Abstract

Protein-protein interaction networks (PPINs) have been employed to identify potential novel interconnections between proteins as well as crucial cellular functions. In this study we identify fundamental principles of PPIN topologies by analysing network motifs of short loops, which are small cyclic interactions of between 3 and 6 proteins. We compared 30 PPINs with corresponding randomised null models and examined the occurrence of common biological functions in loops extracted from a cross-validated high-confidence dataset of 622 human protein complexes. We demonstrate that loops are an intrinsic feature of PPINs and that specific cell functions are predominantly performed by loops of different lengths. Topologically, we find that loops are strongly related to the accuracy of PPINs and define a core of interactions with high resilience. The identification of this core and the analysis of loop composition are promising tools to assess PPIN quality and to uncover possible biases from experimental detection methods. More than 96% of loops share at least one biological function, with enrichment of cellular functions related to mRNA metabolic processing and the cell cycle. Our analyses suggest that these motifs can be used in the design of targeted experiments for functional phenotype detection.

Figure 1

(a–d) Number of loops of length 3–6 in the H. sapiens V (BP-MS) network during randomisation. The number of loops per edge during Markov Chain Graph Dynamics (MCGD) is reported for loops of length 3 (a), 4 (b), 5 (c) and 6 (d) in the H. sapiens V (BP-MS) network. Simulations were performed under two set of constraints: 1) degree distribution {k} (red line) and 2) both degree distribution {k} and degree-degree correlation {k,k′} (blue). (e–g) Number of loops of length 3 in four different H. sapiens networks during randomisation. The number of loops per edge during MCGD is reported for loops of length 3 in four human PPINs from different experimental techniques and research groups: H. sapiens I (e), II (f), III (g) and VI (h). Colour coding as in Figure 2a. Details on network properties, experimental techniques and related literature references are reported in Table 1.

Figure 2

(a) Representative of four recurrent trends in the change of loop number during randomisation. Four distinct and recurring trends were identified in the change of loop number during MCGD: (Purple frame; top left) increase under both constraint conditions (e.g. M. loti network); (Pink frame; top right) increase under constraint on {k} and decrease under constraint on both {k} and {k,k′} (e.g. H.sapiens VIII); (Cyan frame) decrease under both constraints with a steeper reduction under constraint on {k} and {k,k′} (e.g. C. elegans II); and (Green frame; bottom right) decrease under both constraints with a steeper reduction under constraint on {k} (e.g. H. sapiens VI). Colour coding as in Figure 1. (b) Network classification by Principal Component Analysis. Biplot of the first two principal components (PC1-2) of the measured network topological properties. The 30 PPINs are reported as circles numbered according to Table 1 and colored according to the trend in change of loop numbers (Figure 2a). Vectors representing the original variables included in the PC analysis are projected into the PC1/PC2 plane and reported as oranges arrows. Details on network properties are reported in Table 1.

Figure 3. Sub-network of resilient loops preserved after randomization of H. sapiens V network.

The network of proteins and interactions included in loops of length 3–4 preserved after MCGD in the H. sapiens V network. Only loops consistently preserved in five independent simulations are reported. A core set of ribosomal proteins was detected and is reported in grey in the figure.

Figure 4. Example of functional consensus in a loop of length 3.

Functional consensus was defined as the percentage of GO terms shared by all the proteins in a loop. An example is reported for the loop of length 3 including ACTR1A, DCTN2 and MCM7. Common terms are reported in the central circle. The functional consensus is calculated as the percentage of common GO terms (see the main text for details).

Figure 5. Functional consensus in loops of length 3–5 in the human PPINs.

The barplots in panel (a–d) report the fraction of loops of length 3, 4, and 5 by percentage of functional consensus binned at intervals of 25%: (a) H. sapiens V (BP-MS) network; (b) H. sapiens V (BP-MS) without the largest complex of ribosomal proteins; (c) integrated human PPIN obtained by combining H. sapiens IV, H. sapiens V (BP-MS), H. sapiens VII and H. sapiens VIII; d) integrated human including only H. sapiens IV, H. sapiens VII and H. sapiens VIII. The density plots in panel (e–g) report the comparison for the distribution of functional consensus in loops of length 3 (e), 4 (f) and 5 (g) with corresponding randomised samples (Methods for details).

Figure 6. Frequency of GO terms in loops of length 3–5 and in the network of H. sapiens V (BP-MS).

The plots report the relative frequency (in %) of GO terms in the network and in loops of length 3, 4 and 5: (a) H. sapiens V (BP-MS) and (b) H. sapiens V (BP-MS) without the largest ribosomal complex. The values for each GO term are coloured according to their trend: higher frequencies in the loops than in the network (Trend 1 in red and Trend 4 in purple); higher frequency in the network (Trend 2 in blue and Trend 5 in orange) and lower frequencies in the loops with a consistent value independent by the loop length (Trend 3 in green). The relationship between terms before and after the removal of ribosomal proteins is summarised in the diagram in panel (c).

Figure 7. Example of annotation enrichment of the cell cycle pathway by inclusion of loops of length 3.

The diagram reports the KEGG cell cycle pathway (yellow background) annotated with the proteins and interactions from loops of length 3 that have a “cell cycle” GO term. Loop proteins mapping directly onto the KEGG pathway are represented in red boxes. Large functional complexes and the cell cycle stage-specific complexes are highlighted by green backgrounds.