Bike enthusiasts argue that bikesharing programs can be an important element of sustainable mobility planning in the urban cores of large metropolitan areas. However, the objective long-term impact of bikesharing on reducing auto-dependence is not well-examined, as prior studies have tended to rely on self-reported subjective mode substitution effects. We use a unique longitudinal dataset containing millions of geo-referenced vehicle registrations and odometer readings in Massachusetts over a six-year period - the Massachusetts Vehicle Census - to examine the causal impact of bikesharing on various metrics of auto-dependence in the inner core of Metro Boston. The difference-in-differences (DiD) framework is extended to accommodate spatial spillover effects with the inclusion of a spatial autoregressive lag leading to the spatial DiD (SpDiD) model. We also account for seasonal variation in bikeshare operations, where several stations are shut down for the winter months, by setting up a dynamic treatment definition. We find that a new bikeshare station reduces vehicle ownership per household by 2.2%, vehicle miles traveled per person by 3.3%, and per-capita vehicular GHG emissions by 2.9%. We also find strong evidence to support the use of bikesharing as a first/last-mile connector to mass transit. Auto-dependence reductions are around 10% (more than thrice as high as average) where bikeshare connections to transit stations are less than one kilometer long. Finally, we find that vehicle ownership reductions are almost immediate and last up to a year, while vehicle use and emission reductions are lagged over 1.5 years. These sizeable and measurable auto-substitution effects do support some of the claims of bikesharing advocates. These findings are especially important in the post-COVID-19 era, as cities strive to counter the pandemic-inspired safety skepticism about non-car travel.