Alzheimer's disease can be classified based on amyloid, tau and neurodegeneration status. The Default Mode Network is notably vulnerable to these processes, making early structural alterations in this network of particular interest for identifying prodromal biomarkers. In this longitudinal cross-sectional study, we analysed data from 279 participants in the Alzheimer's Disease Neuroimaging Initiative (mean age = 73.7 ± 9 years, 53.2% males). Structural measures-sulcal depth, gyrification and cortical thickness-were extracted for all Default Mode Network regions. Their ability to predict memory performance (encoding, retrieval and recall) was tested at baseline and 2-year follow-up by means of multiple linear regression models, which were all corrected for the risk of multiple comparisons. Covariates included Mini Mental State Examination scores, amyloid status and regional tau burden, to examine interactions with structural changes. Our results showed distinct Default Mode Network alteration patterns based on tau burden and amyloid status, highlighting patterns of morphological features with different susceptibility to proteinopathy. In individuals with concordant (both positive or both negative) amyloid and tau status, preserved structural integrity and complexity were linked to better cognitive performance and appeared protective against decline. However, mainly negative associations were instead observed in individuals with discordant amyloid or tau status (i.e. positive for only either amyloid or tau accumulation). We discuss these findings as a possible reflection of a mismatch between abnormal protein accumulation and structural damage in these populations. The multimodal nature of this study helps clarifying the heterogeneous findings reported in existing literature regarding structural integrity and cognitive outcomes in Alzheimer's disease.