DNA extraction and amplicon sequencing
For the amplicon sequencing, we randomly selected 4 out of the 10 replicates in the experiment. The preserved filters, through which the suspension of epiphytes was filtered, were cut into fragments using sterilized scissors. Subsequently, DNA was extracted using the ZYMO Fecal/soil microbe kit (D6102; ZYMO Research, Irvine, CA, USA). Based on the two-step PCR protocol from Gohl et al. (Gohl et al., 2016), we prepared 16S-V4 amplicon libraries as described in Bonthond et al. (2020), using the KAPA HIFI HotStart polymerase (Roche, Basel, Switzerland) and the 16S-V4 target and indexing primers (515F & 806R, Klindworth et al., 2013). All amplicons were pooled into two libraries, such that populations and treatments were balanced. We included four negative DNA extraction controls and four negative and positive PCR controls (mock communities; ZYMO D6311) in each library. After the final purification step, in which the pooled amplicons were run and re-extracted from agarose gels, libraries were quantified using qPCR and sequenced at the Max-Plank-Institute for Evolutionary Biology (Plön, Germany) on the Illumina MiSeq platform as paired-end 300bp reads. As most of the samples from the final timepoint (t12) yielded only a low number of reads, these samples were re-amplified with 5 extra cycles in the first PCR and re-sequenced from a new amplicon library that was prepared following the same methods. Together with the original fastq files from the field samples that had been included in the study of Bonthond et al. (2020), files were de-multiplexed allowing zero barcoding mismatches, assembled, quality filtered and classified using the software Mothur v1.43.0 (Schloss et al., 2009) and the SILVA reference alignment v132 (Quast et al., 2013),. After removing mitochondrial, chloroplast, eukaryotic and unclassified sequences the remaining sequences were clustered open-reference to the OTUs from the field study in Bonthond et al. (2020), with the opticlust algorithm (Westcott & Schloss, 2017). To curate the final community matrix, we removed OTUs that were singleton in the full dataset, excluded samples with <1000 read counts and deleted OTUs that had zero reads as a result of the preceding step. The de-multiplexed sequences are deposited in the SRA database (accession: PRJNA564581, PRJNA612003 and PRJNA842363).