Relationship between network connectivity and adaptive evolution
Across both gardens, QST was positively associated with module connectivity (kWithin; Cold garden: r = 0.36, p < 0.0001; Warm garden: r = 0.46,p < 0.0001) and with total network connectivity (kTotal; Cold garden r = 0.53, p < 0.0001; Warm garden: r = 0.56, p < 0.0001). Additionally, transcripts in the Warm-condA and Cold-condA categories were significantly over-represented at the core of their respective modules (Warm garden enrichment = 1.7, p < 0.0001; Cold garden enrichment = 2.5, p < 0.0001), as well as at the core of the garden-specific network (Warm garden enrichment = 2.0, p< 0.0001; Cold garden enrichment = 3.9, p < 0.0001) (Fig. 4; Fig. S3). This pattern persisted even after accounting for differences in expression levels between the core and the periphery transcripts. For most modules exhibiting strong population differentiation, the transcript with highest QSTalso had the highest module membership and hence were located at the module core (Fig. S3). The average QST for core transcripts, however, was higher than that of the periphery for only 38% of the strong differentiated modules at the cold garden and for only 20% of the differentiated modules at the warm garden (Table S3). Overall, results from our module and network level relative representation analyses do not support H4.
Only five modules out of the 31 identified at the cold garden and four out of the 26 modules identified at the warm garden were strongly preserved across gardens using our joint threshold (Table S3; Fig. 4). Generally, modules enriched for QST categories were weakly preserved across gardens, while QSTdepleted modules had consistently strong preservation across gardens (Fig. 4a,c) providing further support for H2 at the module level.