Software maintenance and evolution are among the most challenging and cost-intensive activities in software engineering. This is not different for software product lines due to their complexity and long life-span. New customer requirements, technology changes and internal enhancements lead to the continuous evolution of a product line's reusable assets. Due to the size of product lines, single stakeholders or teams can only maintain a small part of a system which poses additional challenges for evolution. This paper presents an approach supporting product line evolution by organizing variability models of large-scale product lines as a set of interrelated model fragments defining the variability of particular parts of the system. The approach allows semi-automatic merging of fragments into complete variability models. We also provide tool support to automatically detect changes that would make models and the architecture inconsistent. Furthermore, our approach supports the co-evolution of variability models and their respective meta-models. We illustrate the approach with examples from an ongoing industry collaboration.