The reason for the reduced efficacy of lung cancer therapy is the existence of lung cancer stem cells (CSCs). Targeting CSCs results in evolved phenotypes with increased malignancy, leading to therapy failure. Here, we propose a new therapeutic strategy: investigating the "transitional" cells that represent the stage between normal lung stem cells and lung CSCs. Identifying and targeting the key molecule that drives carcinogenesis to inhibit or reverse this process would thus provide new perspectives for early diagnosis and intervention in lung cancer. We used Gprc5a-knockout (KO) mice, the first animal model of spontaneous lung adenocarcinoma established by the deletion of a single lung tumor suppressor gene. We investigated the interaction of lung progenitor cells AT2 with Lgr5 cells in the generation of CSCs and related signaling mechanism. In the present study, using Gprc5a-KO mice, we found the initiator Sca-1+Abcg1+ subset with a CSC-like phenotype within the lung progenitor AT2 cell population in mice that had not yet developed tumors. We confirmed the self-renewal and tumor initiation capacities of this subset in vitro, in vivo, and clinical samples. Mechanistically, we found that the generation of Sca-1+Abcg1+ cells was associated with an interaction between AT2 and Lgr5 cells and the subsequent activation of the ECM1-α6β4-ABCG1 axis. Importantly, Sca-1+Abcg1+ and SPA+ABCG1+ cells specifically existed in the small bronchioles of Gprc5a-KO mice and patients with pneumonia, respectively. Thus, the present study unveiled a new kind of lung cancer-initiating cells (LCICs) and provided potential markers for the early diagnosis of lung cancer.