Supplementary Material

Supplementary Table S1: Average (± SE) value of EFN traits, foliar herbivory and plant performance in each population, and ANOVA statistics between the two A.album populations.

Plant variables / Cristália / Grão Mogol / Statistics / p value
EFN density base (clustered)1 / 39.0 ± 2.3 / 43.9 ± 2.2 / F(1,223) = 5.08 / 0.025
EFN size / 0.168 ± 0.002 / 0.174 ± 0.002 / F(1,192) = 4.01 / 0.046
EFN density abaxial (scattered)1 / 0.40 ± 0.1 / 0.62 ± 0.1 / F(1,223) = 4.05 / 0.045
EFN density adaxial (scattered)1 / 1.08 ± 0.1 / 1.18 ± 0.1 / F(1,223) = 0.05 / 0.816
Herbivory (%)2 (range) / 13.1 ± 2.3 (0-100 %) / 10.4 ± 1.6 (0-85 %) / F(1,223) = 1.03 / 0.312
Number of flowers3 (proportion of reproductive plants) / 18.53 ± 4.89 (45%) / 59.43 ± 12.68 (54%) / F(1,207) = 9.11 / <0.001

1 Log transformation; 2 Arcsine transformation; 3 Negative binomial distribution used.

Supplementary Table S2: Modeling the relationship among variables in order to test the full causal hypothesis related to the pattern of phenotypic selection on EFNs mediated by the herbivory process. Here, we performed a simultaneous test of all N independence claims described such an inferential test for a particular ‘‘basis set’’ (BU) as detailed by Shipley (2009). We detailed below the complete BU basis set of d-separation claims (d-sep; directional separation) that are implied by the correct causal structure expected in the case of plant defense mediated by extrafloral nectaries.

Population / D-set claim of independence / General or generalized regression model / Variable whose partial regression slope should be zero / Null probability (distribution)
Cr / (EFN_ab, EFN_size)|{Ø} / EFN_ab ~ EFN_size / - / 0.547 (normal)
Cr / (Success, EFN_ab)|{Herb.} / Success ~ Herb. + EFN_ab / EFN_ab / 0.084 (binomial)
Cr / (Success, EFN_size)|{Herb.} / Success ~Herb. + EFN_size / EFN_size / 0.206 (binomial)
Gm / (EFN_ab, EFN_size)|{Ø} / EFN_size ~ EFN_ab / - / < 0.001 (normal)
Gm / (EFN_ab, Sucess)|{Herb.} / Sucess ~ Herb. + EFN_ab / EFN_ab / 0.154 (binomial)
Gm / (EFN_size, Sucess)|{Herb.} / Sucess ~Herb. + EFN_size / EFN_size / 0.526 (binomial)

Key to variables: EFN_ab: Abundance of extrafloral nectaries on the base of leaflets (abaxial surface); EFN_size: Size of the biggest extrafloral nectary on the base of leaflets (abaxial surface); Herb.: Foliar herbivory (%); Success: Positive production of flowers/fruits (1) or zero production (0) during the rainy season.

Supplementary Table S3: Standardized path coefficients obtained by fitting models following the causal structure detailed in the Figure 6 (main text). Coefficients here were obtained only regressing variables on each of its direct causes as indicated by Shipley (2009), and in the absence of normality we performed regressions with randomization procedures.

Population / Models (distribution) / Path coefficient estimated / Statistics / p value / Graphs
Cr / Herb. ~ EFN_ab (binomial) / - 0.15 ± 0.09 / F(1,112) = 2.77 / 0.07 / Appendix E1
Cr / Herb. ~ EFN_size (binomial) / - 0.21 ± 0.10 / F(1,95) = 4.26 / 0.04 / Appendix E2
Cr / Success ~ Herb. (binomial) / -0.03 ± 0.01 / F(1,113) = 11.14 / 0.01 / Appendix E3
Gm / Herb. ~ EFN_ab (binomial) / - 0.06 ± 0.09 / F(1,115) = 0.41 / 0.52 / Appendix E4
Gm / Herb. ~ EFN_size (binomial) / 0.10 ± 0.10 / F(1,94) = 0.88 / 0.35 / Appendix E5
Gm / Success ~ Herb. (binomial) / 0.01 ± 0.01 / F(1,108) = 0.77 / 0.33 / Appendix E6

Key to variables: EFN_ab: Abundance of extrafloral nectaries on the base of leaflets (abaxial surface); EFN_size: Size of the biggest extrafloral nectary on the base of leaflets (abaxial surface); Herb.: Foliar herbivory (%); Success: Positive production of flowers/fruits (1) or zero production (0) during the rainy season.

Supplementary Figure S1: Parent-offspring regression between parental midpoint from field and offspring mean from common garden. A: Positive relationship between parents and offspring in the Cristália population, including midparent-offspring regression (F(1,5) = 6.6; p = 0.049) and sire-offspring regression (F(1,5) = 14.3; p = 0.012). B: Coefficient of regression not different from zero in the Grão Mogol population (midparent-offspring and sire-offspring models). Statistics are available in Table 1 of the main results.

Supplementary Figure S2: Each graph illustrates the relationship among causal variables based on the estimates of each path coefficient in the Appendix D. Cristália population: E1, E2 and E3. Grão Mogol population: E4, E5 and E6. Solid red line (p < 0.05), and dashed red line (p < 0.08).