This is just marginal variance tho. You could also account for compositional nature by forming TPM per infRep, and then forming SD-based or quantile-based intervals across the infReps (quantile requires a lot more boots, eg 200, for reasonable 95% CI)
2/2