This is completely false, and is a poor extrapolation from a faulty initial data point. There have now been several publications -- including the J&J clinical trial data itself  -- which show that immune responses induced by non-variant virus (or vaccine) are protective against the Brazilian variant.
There is some evidence that a particular mutation in the spike protein of the Brazilian variant can partially evade neutralization by particular antibodies, but this is far from "totally avoiding natural immunity".
> The proof is from the city of Manaus, which was over 75% infected in October 2020.
This number is based on a study that was written in summer 2020 , and used a number of questionable "adjustments" to the raw seroprevalence data to arrive at their conclusion. Raw seroprevalence data (Figure 2) was closer to 30%, not 75%.
The parsimonious conclusion is that the paper citing 75% seroprevalence in Manaus was wrong. Science really should retract this paper, or publish an addendum. It's an example where low-quality science is leading to a lot of unnecessary panic and speculation.
Basically, folks are going back to this paper (or more realistically: not reading it at all, and simply parroting headlines), ignoring the intense massaging of the data, and claiming that reality disagrees with the predictions from the paper, therefore reality must be wrong.