Proc freq

The frequency of each agr type and MSCRAMM genes was calculated as the proportion of positive isolates for each gene divided by the total number of isolates in the clinical and subclinical categories. Chi-square or Fisher Exact tests (PROC FREQ; SAS Institute, Cary, NC, USA) were used, with a significance level of 0.05.
Adhesion and invasion cell counts were log-transformed for analysis. Subsequently, the ratio of adhesion to invasion was calculated by dividing the log of adhesion counts by the log of invasion counts. The normality of the response variables was assessed using both statistical tests and graphical analysis. Differences in the medians between clinical severity groups were conducted using the Kruskal–Wallis test followed by Dunn’s post-hoc analysis. Statistical significance was considered when p < 0.05. A correlation of adhesion and invasion genes was obtained using the Mann–Whitney test. Statistical analysis and graphical representation were generated using GraphPad Prism version 8.0.1.

We used three different analysis approaches to test if the probability of flowering differed between plants according to their collection position along the transect. The three approaches capture the UHIE in different ways (Amsterdam urban–rural transect) to reveal at which scale adaptation is occurring. First, we used logistic regressions using distance from the start of the transect as a continuous variable to test if the flowering response showed a gradual change along the transect. Logistic regressions were performed using PROC GENMOD (SAS OnDemand for Academics, SAS Institute Inc., Cary, NC, USA) using Wald chi-square tests. Second, we used Fisher’s exact tests (PROC FREQ, SAS OnDemand for Academics) to test if the proportion of plants that flowered in our vernalization experiments differed between the three main transect districts: urban, suburban, and rural. Third, we used Fisher’s exact tests to test if the proportion of plants that flowered in our vernalization experiments differed between the five subhabitats: street, urban roadside verge, park, rural roadside verge, and dairy farm grassland. For the second experiment involving the short vernalization and control treatments, we fitted separate statistical models to data for each of the two treatment levels, because the vernalization treatment not only induces flowering capacity but also causes a delay in plant development, which makes it difficult to compare flowering dates between plants from the different treatments. However, we also analyzed a single model containing both short and no vernalization treatment levels, which is useful for evaluating the interaction effect between vernalization and different levels of urbanization (PROC GENMOD, SAS OnDemand for Academics, Supplementary Material Part 3).
For the heat experiment, we used linear mixed models to test the effects of temperature, collection position, and their interaction, on total plant biomass. In all models, we included replicate block and temperature treatment as categorical fixed factors and third leaf length at the start of the temperature treatment as a continuous fixed cofactor. In all models, we included “collection location of the accession” and its interaction with temperature treatment as fixed factors, where three separate models used different definitions of “collection location” to reveal the scale at which adaptation occurs: (1) distance from the start position of the transect (continuous cofactor); (2) main transect district (urban, suburban or rural; categorical factor); and (3) subhabitat type (street, urban roadside verge, park, rural roadside verge and dairy farm grassland; categorical factor). Linear models were performed using PROC GLM (SAS OnDemand for Academics). After plotting temperature treatment results as a function of distance to the start of the transect, we visualized trends in growth response at different temperatures along the urban–rural transect by fitting the regression line for each temperature treatment as estimated from a linear model as described above, when fitted to data from each temperature treatment separately. All p-values reported in this manuscript correspond to two-sided statistical testing.

Data on risk factors likely to impact perinatal mortality were collected from a pre-designed farmer questionnaire (Table 1). A written herd-level questionnaire to determine farm management practices plausibly linked to perinatal bovine mortality was drafted and piloted with two farm managers at Moorepark Research Centre to assess understanding of the questions. The redesigned questionnaire was emailed to each farmer in January 2010 and in January 2012. The questionnaire contained 17 questions grouped into four sections. Three sections dealt with pre-calving factors; breeding (heifer and cow breeds in the herd, breeds of service sires used on heifers and on cows), pre-calving diet and body condition score (forage, concentrate ration and macro and micronutrient supplementation of heifers and of cows pre-calving and body condition score of heifers and of cows pre-calving), endemic infectious diseases (recent clinical history of endemic infectious diseases, number of dogs and presence of foxes on the farm and vaccines used in heifers and in cows) and one dealt with calving management (timing of transfer of pregnant heifers and of cows to the calving facility, type of calving facility, frequency of observations for imminent calving, duration of natural calving allowed before intervention, type of calving aid used, number of personnel managing calvings, methods used to prevent milk fever and techniques used to resuscitate weak newborn calves). The topics were chosen based on management factors known to impact bovine perinatal mortality [1 (link)]. The questionnaire returns were examined prior to data entry and obvious errors indicating misunderstanding of a question were clarified. Data analysis consisted of tests of association between risk group (High and Low) and the per question responses. To facilitate comparison of preferable and non-preferable variable levels, some outcome variable levels were collapsed. Contingency tables were constructed for each year of observation and for the combined results and tested using Chi Square tests or Fisher’s Exact Test. The latter was included to cover those tables where one or more cell frequencies were ≤5. The LOGISTIC procedure was used to generate odds ratios (OR) and their 95% confidence intervals for these tests. The final model fit was evaluated using the Hosmer and Lemeshow goodness-of-fit test. Differences were considered significant if P ≤ 0.05. All data editing and statistical analyses were carried out using appropriate procedures (Proc FREQ and LOGISTIC in SAS (SAS Institute Inc., Cary, NC, USA)).