The increasing prevalence of software vulnerabilities continues to pose serious threats to system security, underscoring the need for accurate and scalable techniques for vulnerability detection and classification. While Pre-trained Language Models (PLMs) have shown strong potential in vulnerability analysis, most existing methods provide no explicit guidance on which parts of the input code are more likely to be vulnerable. As a result, the model must infer token-level relevance without any indication of which parts are important, making it harder to learn the characteristics of vulnerable code during training. To address this limitation, we propose LOSVER (Line-level mOdifiability Signal-guided VulnERability analyzer), a novel two-stage framework that enhances PLM-based vulnerability analysis by incorporating line-level modifiability signals. In the first stage, LOSVER localizes modifiable lines. These are code segments likely to be changed in the future due to instability or complexity, which are often associated with vulnerabilities. In the second stage, the model assigns greater importance to the predicted modifiable lines, allowing the PLM to focus on potentially vulnerable regions during both training and inference. We evaluated LOSVER with two widely used benchmark datasets: Devign, for function-level vulnerability detection, and Big-Vul, for function-level vulnerability classification with Common Weakness Enumeration (CWE) ID labels. Experimental results show that LOSVER improves detection accuracy on Devign by approximately 4 percentage points and increases the weighted F1-score for CWE ID classification on Big-Vul by over 2 points, when applied on top of the UniXcoder baseline. We also conducted experiments on the PrimeVul dataset, which focuses on vulnerability–patch pairs, and observed meaningful improvements in pair-wise detection. These results demonstrate that integrating line-level modifiability signals significantly enhances the effectiveness of PLM-based software vulnerability analysis across both detection and classification tasks.