(a) LTEE design. Three chemostat vessels w ere founded with a clonal ancestral population and maintained under constant dilution with defined media containing 0.4 mM (NH 4) 2 SO 4 , (0.8mM nitrogen). Chemostats were maintained in continuous mode for 250 generations (~ 3 months) and samples obtained every 10-20 generations for archiving. (b) Dynamics of population fitness. The rate of fitness improvement decelerates over time for three independently replicated LTEE (R1, R2 and R3). Error bar are 95% CI of linear regression analysis of competition assays, which comprised six time points each. (c) Allele dynamics in parallel LTEE. Whole population while genome sequencing was performed on samples from the three LTEEs at 50, 100 and 250 generations to identify high frequency mutations (> ~10%) using an Illumina HiSeq2500 in 2 x 50 bp paired end mode with an average read depth of ~50X. The frequency of the MEP2 amplification was defined as the proportion of clones bearing more than 2 copies of MEP2 among 96 randomly selected clones. GAT1 (red lines) variants are a primary target in the earliest generations of selection but, in some LTEE, are ultimately replaced by other alleles including CNVs that encompass MEP2 (blue lines) and mutations in genes that control cell cycle and growth. x- and y-axes represent time in generation and AF (allele frequency) in %, respectively. (d) Dynamics of minor frequency mutations. 12 genes that have previously been identified as adaptive targets of selection in different nitrogen-limited environments (see method) were subjected to targeted deep sequencing. Multiple missense mutations in GAT1 are present simultaneously in each population and compete with each other during the earliest generations. Discrepancies in the estimated AFs of the same mutations identified in Figure 1c are likely due to noise in the sequencing assay. x- and y-axes represent time in generation and AF (allele frequency) in %, respectively (e) Mutational landscape of adaptive GAT1 mutations. All mutations in GAT1 with population AF greater than 1% are shown. The DNA binding domain of GAT1 is a mutational hotspot in all ammonium-limited LTEE. Multiple missense (protein coding alteration) mutations were observed but no nonsense or frame-shift mutations were detected. (f), The DNA binding domain of GAT1 is under purifying selection in the wild. dN and dS values (see method for their definition) for GAT1 at each amino acid position were calculated using SNAP v2.1.1 (http://www.hiv.lanl.gov/content/sequence/SNAP/SNAP.html) Korber, 2000 using sequences from 42 different wild yeast strains. Unlike LTEE in ammonium-limited chemostats, the GAT1 DNA binding domain is under purifying selection implying that non-synonymous mutations are likely to be detrimental in dynamic environments.