Sample collection Bacteria and phages were isolated from whole leaves as previously described in Koskella (2014) and Koskella & Parr (2015). Briefly, 2 leaves from the same branch were collected over 4 years from the same set of 8 individual horse chestnut trees (Aesculus hippocastanum; N = 64) located in an urban park in Oxfordshire, UK. Sampling occurred once at the end of the growing season on 09/15/2011, 08/19/2012, 09/16/2013, and 09/03/2014. Leaves were immediately processed in the lab for long-term freezer storage in order to preserve the leaf and its microbes. Individual leaves were washed in a surface sterilizing solution of 10% bleach and a 0.01% Tween detergent solution, then rinsed 3 times in sterile water and rolled inside a 15ml falcon tube. Tubes were filled with 0.1 M potassium phosphate and 20% glycerol buffer and stored at -20°C. After all sampling was completed, random leaves were rapidly thawed at 37°C for 14 minutes, then homogenized using a fast prep 24 (MP Biomedicals) at 4 rotations/sec for 1 minute, with 4 sterile ceramic beads to break up tissue. 100µL of the homogenate was diluted to 1:10, 1:100 and 1:1000 in KB and 200µL of each dilution was plated onto KB 1.2% agar plates using sterile glass beads. Plates were then incubated at 28°C upside down for 48 hours. 48 colonies per plate were picked, transferred to a 96-deep well plate with 700µL KB and incubated overnight at 28°C, then stored with 300 µL of 50% KB glycerol at -80°C. This was repeated for both leaves of each tree, resulting in 96 hosts per tree in a given year. To extract phage, the remaining leaf homogenate was centrifuged for 20 minutes at 550g, then passed through a 0.45µm filter to remove any bacteria and stored in 2mL aliquots at 4°C in the dark. Leaf homogenate from the 2 leaves were bulked together to give 1 phage inoculum per tree per year, as phage infectivity is similar across leaves from the same tree in this system (Koskella et al. 2011). Time shift experiments Each of the 96 bacterial isolates per tree per year were crossed with all 4 phage inocula from the same tree (sympatric). To increase throughput we initially crossed individual bacterial hosts against a pooled inoculum of the sympatric phages from all four years. Each bacterial isolate that was sensitive to this bulk inoculum was then crossed separately with each of the individual year phage inocula. All crosses used soft agar overlays, as described previously in Koskella et al. (2011), that were modified for a 24-well plate. Briefly, each bacterial isolate was grown from freezer stock overnight at 28°C in 200µL KB. 40µL of the culture was added to warm soft agar in 2mL tubes, then vortexed and pipetted on top of a hard agar base in each well of a 24-well plate. Once cooled, 10µL of phage filtrate (bulked or individual year) was spotted into each well, and the plates were incubated upside down overnight at 28°C. Two control wells were included for each bacterial isolate by spotting sterile water rather than phage inocula. A bacterial host was considered susceptible if a clearance (plaque) in the lawn were visible, and the absence of bacterial growth overlapped with where the phage inoculum had been spotted. Phage clones Individual phage were isolated from bacterial hosts that formed plaques in response to individual year phage inoculum. An agar plug of the plaque was added to 900µL KB, along with 3 plugs of the surrounding bacterial lawn. The host and phage were co-cultured overnight at 28°C to amplify the phage. 100µL chloroform was added to the co-culture and vortexed for 30 seconds to remove bacteria. Phage were removed from the chloroform by centrifuging at 13,000 rpm for 3 minutes, then the supernatant removed and stored at 4°C in the dark. To ensure the capture of a single phage, phages were single-plaque purified on the originally sensitive host, co-cultured for 24 hours and then filter-purified to generate a high titer inoculum. For the 20 phages that were successfully isolated and amplified, we then tested their infectivity against all sympatric host isolates from either past, contemporary, or future time points. Data Cleaning/Preparation Data are in 2 .csv files: one that contains the data for the main suite of crosses between isolated bacterial hosts and bulked (per tree per year) phage sample inocula [Data for submission_Time Shift_All phage All bacteria.csv]; the other contains the data for the suite of crosses between isolated phage clones and their sympatric bacterial hosts [Data for submission_Time Shift_Phage clones.csv]. Data were cleaned using a combination of R (version 4.0.0) and Microsoft Excel (version 16.44). For both data sheets, column headings were changed for ease of use (standardized capitalization, shortened for legibility), typos were fixed in the columns for Genus and Species, and the column "Unique.Host" was added to identify each individual isolated bacterial host such that it can be matched to sequencing data. Additional edits were not made to "Data for submission_Time Shift_All phage All bacteria.csv", but were made to "Data for submission_Time Shift_Phage clones.csv", as follows. Typos in the names of the individual phage clones were corrected. The original data sheet for individual isolated phage clones included both sympatric and allopatric pairs of bacterial hosts and phage clones, but with no outcome recorded (as we tested only sympatric crosses) - these rows were removed from the submitted data set. In addition, the data from crosses with the phage "Tree#1 2013 D4" were accidentally duplicated, and the copied rows were removed. The phage "Tree#6 2014 H2" also had the outcomes ("Infected" column) for crosses with sympatric hosts from 2011-2013 (288 rows) accidentally copied over the blank outcome entries for the rows containing allopatric hosts from Tree 7. The values in the "Infected" column for the Tree 7 hosts were side-by-side verified to be identical in pattern to those from Tree 6 hosts crossed with "Tree#6 2014 H2" for all 288 rows, making it extremely unlikely that this was anything but an accidental duplication during data entry. These Tree 7 host rows were thus removed. No other changes were made to "Data for submission_Time Shift_Phage clones.csv".

Note that when grouping was chosen based on Geneious grades and ability to call to a given taxonomic resoluation, the class Gammaproteobacteria was recorded as Genus = "Gamma" and Species = "Gamma proteobacterium". This grouping was not removed from the data, but it is important to be aware that this is not a genus- nor species-level identification, and is actually a class of bacteria that contains many genera, including Erwinia, Pantoea, and Pseudomonas. Metadata for datasheets: Column Description Tree Tree from which phage and bacteria samples were collected Host.year Year in which bacteria sample was collected Host Randomized name given to individual hosts in each set of 96 hosts from a given tree in a given year (A1-H12) Genus Genus assigned to successfully to individual hosts from NCBI blast Species Species assigned to successfully sequenced individual hosts from NCBI blast E.value The Expect value from NCBI blast Geneious.Grade Geneious grade given to individual hosts when identifying via sequencing Phage.inoculum (All phage/bacteria dataset only) Year in which phage sample was collected Phage.clone (Phage clones dataset only) Identifier of the isolated phage clone used in the second set of time shift crosses. Identifiers assigned based on collection data: [Tree# (1-8)] [Year (2011-2014)] [Initial susceptible host (A1-H12)]. Infected 0 = Not infected; 1 = Successful infection Unique.Host Identifier of individual bacterial isolates used in both sets of time shift crosses, to allow for cross-reference to sequencing data. Identifiers assigned based on collection data: [Host (A1-H12)]_[Tree# (1-8)]_[Year (2011-2014)].

Coevolutionary dynamics shape diversity within and among populations but are difficult to study directly. Time shift experiments, where populations of species A from one point in time are experimentally challenged against populations of species B from past, contemporary, and/or future time points, and vice versa, are a particularly powerful tool to measure coevolution. This approach has been primarily applied to study host-parasite interactions and proven useful in directly measuring coevolutionary change and distinguishing among coevolutionary models. However, these data are only as informative as the time window over which they were collected, and data from shorter coevolutionary windows might conflict with data collected over longer time periods. Previous work on natural microbial communities from horse chestnut tree leaves used time shifts to uncover an apparent asymmetry, whereby hosts tended to be resistant to phages from all earlier points in the growing season while phages tended to be most infective on hosts from the recent past. Here we extend the time window over which these infectivity and resistance ranges are observed from within a growing season to across years and confirm that the previously observed asymmetry holds over longer timescales.