IMProv: A Resource for Cross-link-Driven Structure Modeling that Accommodates Protein Dynamics

Abstract

Proteomics methodology has expanded to include protein structural analysis, primarily through cross-linking mass spectrometry (XL-MS) and hydrogen-deuterium exchange mass spectrometry (HX-MS). However, while the structural proteomics community has effective tools for primary data analysis, there is a need for structure modeling pipelines that are accessible to the proteomics specialist. Integrative structural biology requires the aggregation of multiple distinct types of data to generate models that satisfy all inputs. Here, we describe IMProv, an app in the Mass Spec Studio that combines XL-MS data with other structural data, such as cryo-EM densities and crystallographic structures, for integrative structure modeling on high-performance computing platforms. The resource provides an easily deployed bundle that includes the open-source Integrative Modeling Platform program (IMP) and its dependencies. IMProv also provides functionality to adjust cross-link distance restraints according to the underlying dynamics of cross-linked sites, as characterized by HX-MS. A dynamics-driven conditioning of restraint values can improve structure modeling precision, as illustrated by an integrative structure of the five-membered Polycomb Repressive Complex 2. IMProv is extensible to additional types of data.

Keywords: Polycomb Repressive Complex 2; crosslinking; cryo-electron microscopy; hydrogen-deuterium exchange; integrative modeling; structural mass spectrometry.

Graphical abstract

Fig. 1

IMProv, a workflow for integrative structural modeling in the Mass Spec Studio. A graphical user interface enables the configuration of integrative modeling routines based on IMP (26). IMProv works in concert with MS data analysis apps in the Studio to receive XL and HX datasets that are properly configured for use within IMP, with an option for footprinting data as well. These datasets are combined with structures from the Protein DataBank (PDB), available cryo-EM density maps (EMD), and sequence information as needed. A deployment bundle including IMP and its dependencies is exported, which can be executed on a high-performance computing cluster for integrative modeling of the protein system.

Fig. 2

Mapping XL-MS and HX-MS data onto PRC2.A, circos plot showing the aggregate of unique DSS and BS³ cross-linking sites, with stability zones marked on the inner ring according to the HX-MS data. B, unique cross-links displayed on the hybrid high-resolution PRC2 structure; green linkages are ≤35 Å, orange are ≥35 Å. Only cross-links that correspond to resolved structure are shown (approximately one-third of total XLs). C, box plot of cross-link distances grouped by stability class, as shown in (A). Bars indicate quartiles, whiskers indicate 1.5 × IQR (interquartile range).

Fig. 3

HX-MS heat maps. Data mapped onto the sequences of all five core PRC2 proteins, displaying 90% sequence coverage.

Fig. 4

Concept of conditioning of XL-MS distance restraints with HX-MS structural stability measurements.A, a protein complex is analyzed by both HX-MS and XL-MS methods. Both datasets are imported into the IMProv module of the Mass Spec Studio where cross-links are partitioned according to structural dynamics. A three-tiered example is shown. Red-violet-blue color scheme indicates stable, semistable, or unstable structural classes, respectively. Each class informs a scoring function based on a modified distance restraint, in this example presented as the ambiguity of a cross-linked site in space. B, altered distance restraint function based on the slope (left) and offset (right) methods. The slope method is related to the HX stability class

Fig. 5

Classification of cross-links with HX protection factors. Screenshot of IMProv where the user selects protection factor cutoff values for downstream stability classification of cross-linked sites, guided by a graphical display of the composite protection factor values spanning the given sites.

Fig. 6

Comparison of IMP models with benchmark structure. Distribution of ensemble structure RMSDs for the main (i.e., largest) cluster compared with the high-resolution benchmark structure of PRC2 (∼2/3rds of sequence). Black bars indicate medians.

Fig. 7

Integrative modeling of PRC2.A, bead and density model of full-length PRC2, from the O-4 modelling trial. Subunits coloring: EZH2 in green, EED in red, RBAP48 in violet, SUZ12 in orange, AEBP2 in dark cyan. B, density of AEBP2_209–503 from the O4 modeling trial overlaid on the high-resolution structure of PRC2 (PDB 6c23), colored as above, except with AEBP2_440–503 in blue. Domain labels for EZH2 provided.