Comparison of the significant trends (p ≤ 0.05) when analyzed for 6 different factors.

Trends with two categories were tested using either the t-test (normal data) or the Mann-Whitney test (non-normal data): Ice Extent (<2007 vs. >2007), Geography (Franklin Bay [FB] vs. open-ocean stations), Season (summer vs. fall) and searching for a difference within vs. between groups for the # shared OTUs. Trends with three categories were tested using the Kruskal-Wallis test (non-parametric version of ANOVA): Photosynthetically-Active Radiation (PAR; low vs. medium vs. high) and Chl a Distribution (picoplankton-dominated vs. shared vs. nanoplankton-dominated). The temporal trends (over 8 years, 2003–2010) were tested using linear regression (normal data) or the Spearman Rank correlation (non-normal data). Only those trends showing statistical significance out of the parameters tested are shown. Differences are indicated as either: overall percent increases (+X%) or decreases (−X%) for the two Ice Extent categories; FB more or less abundant than open-ocean for Geography; increasing or decreasing linear trends (r value) over the 8 Years; or “ns” for non-significant. For the # of shared OTUs it is a little different: using the example of Ice Extent, each sample from <2007 is compared to the other samples from <2007 (# shared OTUs within the group) and then to the samples from >2007 (between the groups). If the community structure is the same, they should share the same # of OTUs within vs. between groups. However, if there is a significant difference (≠), then there has been a change in structure. For the linear decreasing trends, this implies that the more years separating samples, the less OTUs they share – indicating change in community structure over time. See Supplementary Table S5 for the assignments of the 11 samples into each of the respective categories.