Single-end or Paired-end?

I want to run Kneaddata on bacterial shotgun sequences and then Humann 3 on filtered sequences. I have paired-end FASTQ files.

I understand that Humann will not pair the reads and analyze paired-end reads as single-end reads (I plan to concatenate two files before running Humann).

My question is, should I run Kneaddata in single-end or paired-end mode? Since, Humann will consider only one read at a time, it does not matter if paired mate is present or not. Correct?