This is kind of a note-to-self, but it seems the issue is this code

n_row_emeta <- if (is.null(results$e_meta)) {
  # Use either the original or isoform e_meta depending on the input.
  if (is.null(isoformRes)) {
    nrow(unique(pepData$e_meta))
  } else {
    nrow(e_meta_iso)
  }
} else {
  nrow(unique(results$e_meta))
}

is using the original (non rolled up) e_meta, which means the n_row_emeta is not equal to n_row_edata, so the column gets removed. As a test I tried using length(unique(pepData$e_meta[[emeta_cname]])) in place of nrow(unique(pepData$e_meta)), but it ended up with problems further down the line (e_meta was getting duplicate rows, meaning as.proData was throwing an error).

As it turns out, the above example is a more complex version of the problem, where sometimes the extra columns have multiple possible entries when rolled up. Appending the following lines to the above example should illustrate the simpler case:

mypepData$e_meta$Org <- round(runif(nrow(mypepData$e_meta)), 1)*10
results <- protein_quant(pepData = mypepData, method = 'rollup', combine_fn = 'median', emeta_cols = "Org")

Were gonna now keep those columns and paste() together their elements