Supporting Information Methods S1, Tables S1–S6, Figs S1 & S2 Methods S1 Phylogenetic regression and phylogenetic signal statistics with branchlength transforms and measurement error Here we present briefly the statistical model that we used to perform phylogenetic regressions with the inclusion of within-species variation (measurement error). The approach is a merger of phylogenetic regression with branch-length transforms as implemented by REGRESSIONv2.m (Lavin et al. 2008) and phylogenetic regression with measurement error that assumes a Brownian motion model of evolution (Ives et al. 2007). These references provide further background and place the analyses into the context of the broader literature. Here we focus on model structure. The Matlab code, MERegPHYSIGv2.m, is available from Ted Garland (theodore.garland@ucr.edu) upon request. The statistical model for dependent variable y measured with error, independent variable x measured with error, and independent variable u measured without error is X* = ax + x Y* = a0 + a1U + bX* + y (S1) X = X* + x Y = Y* + y where X* and Y* are N N species, and X and Y are the values observed with measurement errors x and y. Values of trait u measured without error are contained in the N U. Because values of the independent trait x will likely be phylogenetically correlated, we assume x has covariance matrix Vx, and because residual variance in Y* given by y is also likely to be phylogenetically correlated, we assume y has covariance matrix Vy() where is a parameter that changes the 1 phylogenetic correlation matrix Vy according to a specific branch-length transformation (see next paragraph). Because values of U are fixed, their distribution does not affect the analysis. Following standard regression, we assume that variation in x and y are independent. We also assume that measurement error may be correlated for traits x and y within species. If there is no measurement error (Mx = My = 0), then this problem reduces to phylogenetic regression under a branch-length transform (as implemented in REGRESSIONv2.M program; Lavin et al., 2008). If the parameter is not estimated but instead defined by the user, then this model reduces to a phylogenetic measurement error model (as implemented in MERegPHYSIG.m; Ives et al., 2007). Although we present this model for single independent variables with and without measurement error (trait x and trait u, respectively), the model expands naturally to include p independent variables measured with error and q independent variables measured without error. In model S1, we implement two different branch-length transforms to the residual covariance matrix Vy(): the Ornstein-Uhlenbeck (OU) model of evolution and Pagel's . The OU model of evolution (Felsenstein 1988, Martins and Hansen 1997) assumes that there is stabilizing selection around a trait optimum, with the strength of stabilizing selection given by the parameter d (Blomberg et al. 2003). Our particular model of OU evolution assumes that the basal node of the phylogenetic tree is at this optimum, in contrast to other variants of the OU model (Hansen 1997). This assumption means that the case of Brownian motion (BM) evolution is recovered when d = 1. When there is stabilizing selection, "phylogenetic memory" is lost, so the strength of phylogenetic signal decreases. At the limit of very strong selection, all taxa (tips of the phylogenetic tree) are independent, which corresponds to d = 0. Disruptive selection can lead to stronger-than-BM signal in which recently separated taxa have stronger trait correlations than predicted by the BM model; these cases are given by d > 1. 2 Pagel's changes the phylogenetic correlations among taxa by adding branch length to the tips of the phylogenetic tree. In our implementation, when = 1 the added branch lengths are zero, returning the original "starter" phylogenetic tree. As approaches zero, the proportional length of the tip branches increases to one, thereby implying no phylogenetic signal. By adding uncorrelated variation to the tip taxa of a phylogenetic tree, Pagel's is structurally similar to measurement error in the residual variation (or dependent variables) of a regression, since measurement errors will also give uncorrelated variation in the residuals. Thus, Pagel's in some contexts is assumed to contain measurement error (e.g., Housworth et al. 2004). The important distinction here, however, is that we assume measurement error is measured from multiple samples from the same taxa. This means that there is no confounding between the lengthening of tip branches due to measurement error (which is estimated separately) and lengthening using Pagel's to account for less-than-BM phylogenetic signal. Finally, unlike the OU transform, greater-than-BM phylogenetic signal cannot be modeled using Pagel's . Although the independent variables may contain phylogenetic signal, it is more appropriate to estimate this signal independently. Thus, for example, model S1 could be applied by treating the independent variable as the dependent variable (and including no independent variable) to estimate Vx(), with this estimate then used in S1 to regress y on x. Specifying the branch-length transforms mathematical, for the OU transformation with d > 0, the elements of Vy(d) are d vij = (cii +c jj -2cij ) (1- d ) 2cij (S2) (1- d ) 2 where cij are the shared branch lengths between taxa i and j obtained from the "starter" phylogeny that is derived under the assumption of Brownian motion evolution (Blomberg et al., 3 2003). In the absence of phylogenetic signal (d = 0), the matrix Vy(0) has all zeros on the offdiagonals. For Pagel's vij = cij (S3) for off-diagonal elements, while the diagonal elements remain unchanged. The regression model can be written W=A++ (S4) where W is the 2N X and Y, and A is the 2N first N elements are ax and second N elements are given by a0 + a1U. The resulting covariance matrix E{(W–A)(W–A)’} = 2 is æ 2 s x2 Vx + s mx Mx bs x2 Vx + rms mxs my M xy ç =ç 2 ç bs x2 Vx + rms mxs my M xy ' b2s x2 Vx + s y2 Vy (q ) + s my My è 2 ö ÷ ÷÷ . ø (S5) The terms involving measurement error are derived from the covariance matrix given by æ 2 s mx Mx rms mxs my M xy ç E{'} = ç 2 rs s M ' s my My è m mx my xy ö ÷ ÷ ø (S6) x on top of y, mx2Mx and my2My are where is the 2N matrices containing the measurement error variances, and rmmxmyMxy is the matrix containing covariances in measurement errors between traits for each species. If measurement errors for each trait are independent among species, then Mx, My, and Mxy will be diagonal matrices (i.e., all off-diagonal elements will be zero). If measurement errors for the two traits within species are correlated (e.g., the measurements of traits x and y for a given species tend to err either high or low in unison), then this correlation is given by rmMxy. These terms are all estimated independently from data using multiple measurements on the same taxa; these within-species 4 variances include both true measurement error and variation among individuals or populations within taxa. We estimated parameter values using Restricted Maximum Likelihood (REML) estimation. The log-likelihood function from which variance parameters are estimated is (Harville 1974): (S7) For regression with p independent variables measured with error and q independent variables measured without error, â GLS is the p + q + 1) vector containing the "mean" components of the model consisting of ax (p values, one for each trait x measured with error) and ai (i = 0, …, q). The p + 1)N p + q + 1) matrix Z contains ones in the first p columns identifying rows corresponding to the p traits x, and (q + 1) columns containing in the last N rows the values of trait u corresponding to values of trait y, with the first of these columns containing ones. To speed the numerical maximization, the likelihood function is concentrated by using the GLS estimates â GLS and 2 conditioned on the remaining parameters. In addition to standard asymptotic approaches for computing standard errors and confidence intervals, MERegPHYSIGv2.m also performs parametric bootstrapping. This is the only reliable way to obtain confidence in estimates of d and . Furthermore, parametric bootstrapping is useful in the estimation of the other parameters because it identifies bias in the estimates. Parametric bootstrapping for a given model is performed by first estimating parameters and then using the model fitted with these estimates to simulate a large number of data sets. For each of the simulated data sets, the parameters of the model are re-estimated. The 5 resulting sets of parameter estimates approximate the distribution of the parameter estimators under the assumption that the model (with its fitted parameter values) is correct. Thus, the 95% inclusion intervals for the estimates give the approximate 95% confidence intervals for the parameter estimates. Bias in the estimates is revealed if the mean of the simulated parameter estimates differs substantially from the parameter values estimated from the data and used to construct the simulated data sets. References Blomberg, S. P., T. Garland, Jr., and A. R. Ives. 2003. Testing for phylogenetic signal in comparative data: behavioral traits are more labile. Evolution 57:717-745. Felsenstein, J. 1988. Phylogenies and quantitative characters. Annual Review of Ecology and Systematics 19:445-471. Hansen, T. F. 1997. Stabilizing selection and the comparative analysis of adaptation. Evolution 51:1341–1351. Harville, D. A. 1974. Bayesian inference for variance components using only error contrasts. Biometrika 61:383-385. Housworth, E. A., E. P. Martins, and M. Lynch. 2004. The phylogenetic mixed model. American Naturalist 163:84-96. Ives, A. R., P. E. Midford, and T. Garland, Jr. 2007. Within-species variation and measurement error in phylogenetic comparative methods. Systematic Biology 56:252270. Lavin, S. R., W. H. Karasov, A. R. Ives, K. M. Middleton, and T. Garland, Jr. 2008. Morphometrics of the avian small intestine, compared with non-flying mammals: a phylogenetic approach. Physiological and Biochemical Zoology 81:526-550. Martins, E. P. and T. F. Hansen. 1997. Phylogenies and the comparative method: A general approach to incorporating phylogenetic information into the analysis of interspecific data. American Naturalist 149:646–667. Erratum 153:448. 6 Supporting Information Tables S1–S6 Table S1 Summary of the common phenolics identified from leaf samples by UPLC-MS across 26 Oenothera species. The table gives peak names, retention times (Rt), molecular weights (Mw) and assignments of peaks measured during this study. There were many less abundant and/or rare phenolics that contributed to our quantification of phenolic diversity and total concentrations of the three main phenolic types which are not included here. Peak Name Rt (min) Peak ID Mw (g/mol) ET1 2.90 oenothein B 1568 ET2 3.23 oenothein A 2352 ET3 2.69 oxidized oenothein A 2366 2.48 caffeoyl tartaric acid 312 3.23 feruloyl tartaric acid 326 3.81 3.90 3.99 4.08 4.21 4.21 4.25 4.27 4.28 4.28 4.35 4.40 4.40 4.51 4.61 4.62 4.62 4.66 4.76 4.78 quercetin glycoside quercetin glycoside flavonoid glycoside quercetin glycoside kaempferol glycoside kaempferol glycoside kaempferol glycoside quercetin glycoside flavonoid glycoside quercetin glycoside kaempferol glycoside kaempferol glycoside quercetin glycoside flavonoid glycoside flavonoid glycoside flavonoid glycoside myricetin glycoside kaempferol glycoside quercetin glycoside kaempferol glycoside 616 610 464 478 448 594 600 630 550 434 448 462 448 492 614 432 506 534 564 432 Ellagitannins (ET) Caffeic acid derivatives (CA) CA1 CA2 Flavonoid glycosides (FG) FG1 FG2 FG3 FG4 FG5 FG6 FG7 FG8 FG9 FG10 FG11 FG12 FG13 FG14 FG15 FG16 FG17 FG18 FG19 FG20 7 Table S2 Eigenvalues (A) and eigenvectors (B) from the principal components analysis of phenolic chemistry variation among species. Values in bold show the traits with the two highest and two lowest loadings on PC 1 and PC 2. A) Eigenvalues Eigenvalue Proportion of total variance Cumulative variance explained B) Eigenvectors peaks 280 nm peaks 349 nm oenothein B oenothein A ox-oenothein A total ET total CA FG4 - quercetin total kaempferol total quercetin other flavonoids total FG PC1 3.08 0.26 PC2 2.50 0.21 PC3 1.32 0.11 PC4 1.17 0.10 PC5 1.00 0.08 0.26 PC1 -0.36 -0.35 0.32 -0.02 0.34 0.43 0.22 0.09 -0.07 0.05 -0.29 -0.44 0.47 PC2 -0.09 0.09 -0.15 0.11 0.25 0.21 -0.35 -0.53 0.32 -0.57 0.02 -0.13 0.58 0.67 0.76 8 Table S3 Eigenvalues (A) and eigenvectors (B) from the principal components analysis of all traits combined among species. Values in bold show the traits with the two highest and two lowest loadings on PC 1 and PC 2. A) Eigenvalues Eigenvalue Proportion of total variance Cumulative variance explained B) Eigenvectors peaks 280 nm peaks 349 nm oenothein B oenothein A ox-oenothein A total ET total CA FG4 - quercetin total kaempferol total quercetin other flavonoids total FG toughness % water Specific leaf area tannins trichomes PC1 2.08 0.25 0.25 PC1 -0.22 -0.05 0.03 0.23 0.32 0.34 -0.10 -0.21 0.12 -0.27 -0.07 -0.25 0.32 -0.34 -0.31 0.39 -0.04 PC2 1.83 0.20 0.45 PC2 0.24 0.33 -0.34 0.11 -0.14 -0.27 -0.34 -0.28 0.21 -0.28 0.24 0.30 0.12 -0.12 -0.26 -0.19 -0.16 9 PC3 1.33 0.10 0.56 PC4 1.20 0.08 0.64 PC5 1.13 0.08 0.71 Table S4 Species-level correlations between traits. Correlations were performed while taking intraspecific variation into account (see Methods), and 95% CI were calculated using parametric bootstrapping with 1000 iterations. We show correlation coefficients in bold when their 95% CI did not overlap 0. peaks 280 peaks 349 oeno B oeno A oxoeno A total ET total CA FG 4querc total querc total kaemp other FG total FG 0.39 peaks 349 -0.19 -0.08 -0.56 -0.54 -0.17 0.04 -0.07 -0.01 0.24 0.18 -0.22 -0.25 -0.43 -0.14 -0.12 -0.02 0.01 0.54 0.25 0.3 0.16 0.17 -0.11 0.05 0.53 total ET 0.04 -0.03 total CA -0.21 oeno B -0.13 oeno A 0.06 oxoeno A toughness % H20 SLA PPC trichomes 0.31 0.03 0.45 0.19 -0.45 -0.27 0.36 0.6 -0.06 -0.17 -0.18 -0.17 -0.16 -0.05 -0.29 -0.34 -0.17 0.08 0.15 0.29 0.23 -0.13 0.12 0.08 -0.17 0.59 -0.62 -0.63 0.58 -0.06 -0.21 -0.16 0.05 -0.2 -0.5 0.45 -0.41 -0.24 0.53 -0.09 -0.03 -0.23 0.02 -0.4 -0.62 0.29 -0.29 -0.18 0.88 0.05 0.26 FG 4querc 0.42 -0.3 -0.08 -0.25 -0.38 -0.04 0.5 0.12 0.27 0.86 total querc -0.33 -0.19 0.09 -0.19 0.43 0.22 -0.06 0.16 -0.39 total kaemp -0.13 0.28 -0.33 0.39 0.35 -0.23 0.42 -0.24 other FG 0.22 0.3 -0.19 -0.38 -0.11 -0.01 0.34 total FG -0.11 -0.3 -0.11 -0.17 -0.05 -0.31 0.07 -0.08 -0.65 0.2 -0.32 -0.66 0.44 -0.18 % H20 0.63 -0.6 -0.05 SLA -0.36 -0.1 PPC 0.03 toughness 10 Table S5 Associations between herbivore performance and variation in plant traits. The measures of herbivore performance are shown as subheaders in the left-most column, and bivariate regressions with each of the 17 traits follows. The regression statistics shown correspond to the phylogenetic generalized least-squares (PGLS) OrnsteinUhlenbeck model of stabilizing selection. When the best fitting model had a slope different from 0 at P<0.1, we show the slope, t-value and P-value in bold. We chose P<0.1 here not as a hard line of statistical significance testing, but to indicate traits of potential interest because the non-measurement error PGLS models show lower statistical power. We also performed these analyses while incorporating measurement error, and although there were many more significant relationships, a large fraction of the models failed to run and so for consistency we only show the non-measurement error models here. For completeness, we underline those traits that were also significant (P<0.05) according to non-phylogenetic ordinary least-squares regression. Pvalues are not corrected for multiple comparisons. Response/Predictor Caterpillar herbivory No. peaks at 280nm No. peaks at 345nm Oenothein B Oenothein A Ox-oenothein A Total ET Total CA FG 4-quercetin Total kaempferol Total quercetin Other flavonoids Total FG Toughness % water SLA PPC Trichomes Caterpillar mass No. peaks at 280nm No. peaks at 345nm Oenothein B Oenothein A Ox-oenothein A Total ET Total CA FG 4-quercetin Intercept Slope t P R2 d -0.1476 0.0131 0.0544 0.0549 0.0483 0.0849 0.0410 0.0431 0.0464 0.0517 0.0409 0.0240 0.0667 0.0036 -0.0068 0.0870 0.0621 0.0024 0.0025 -0.0002 -0.0009 0.0000 -0.0005 0.0013 0.0052 0.0006 -0.0016 0.0043 0.0022 -0.0002 0.0006 0.0003 -0.0190 -0.0002 1.19 0.98 0.62 1.47 0.11 1.93 0.71 1.51 0.14 0.50 0.98 0.77 0.98 0.28 1.22 1.53 1.75 0.245 0.339 0.543 0.154 0.912 0.066 0.484 0.144 0.893 0.622 0.338 0.447 0.335 0.785 0.236 0.139 0.093 0.06 0.04 0.02 0.09 0.00 0.14 0.02 0.09 0.00 0.01 0.04 0.03 0.04 0.00 0.06 0.09 0.12 0.11 0.03 0.07 0.08 0.18 0.02 0.21 0.48 0.19 0.12 0.09 0.11 0.09 0.22 0.18 0.11 0.12 -0.5840 0.0809 0.2830 0.2680 0.2620 0.4270 0.2210 0.2230 0.0100 0.0116 -0.0012 -0.0033 -0.0010 -0.0025 0.0044 0.0148 1.13 0.93 0.85 1.10 0.53 2.01 0.51 0.88 0.270 0.363 0.402 0.282 0.601 0.056 0.612 0.388 0.05 0.03 0.03 0.05 0.01 0.14 0.01 0.03 0.35 0.34 0.36 0.36 0.40 0.34 0.41 0.50 11 Total kaempferol Total quercetin Other flavonoids Total FG Toughness % water SLA PPC Trichomes Caterpillar survival No. peaks at 280nm No. peaks at 345nm Oenothein B Oenothein A Ox-oenothein A Total ET Total CA FG 4-quercetin Total kaempferol Total quercetin Other flavonoids Total FG Toughness % water SLA PPC Trichomes Mite eggs No. peaks at 280nm No. peaks at 345nm Oenothein B Oenothein A Ox-oenothein A Total ET Total CA FG 4-quercetin Total kaempferol Total quercetin Other flavonoids Total FG Toughness % water SLA 0.2360 0.2380 0.2600 0.0551 0.3148 -0.7945 -0.0196 0.4651 0.2839 0.0029 0.0014 -0.0102 0.0181 -0.0006 0.0134 0.0013 -0.1096 -0.0005 0.12 0.09 0.43 1.35 0.83 1.27 1.16 1.82 1.26 0.904 0.928 0.668 0.190 0.412 0.216 0.256 0.081 0.220 0.00 0.00 0.01 0.07 0.03 0.06 0.05 0.12 0.06 0.40 0.40 0.43 0.35 0.37 0.48 0.41 0.35 0.37 -1.2900 0.3610 0.4210 0.3930 0.3970 0.5660 0.3660 0.3390 0.3230 0.4080 0.2910 0.1030 0.4630 -0.5775 0.2292 0.6084 0.4383 0.0200 -0.0005 -0.0020 -0.0046 -0.0018 -0.0028 -0.0023 0.0324 0.0203 -0.0176 0.0440 0.0243 -0.0010 0.0123 0.0006 -0.1222 -0.0008 1.99 0.04 1.13 1.38 0.82 2.18 0.24 1.74 0.81 1.05 2.00 1.61 1.00 1.05 0.50 1.99 1.73 0.058 0.972 0.270 0.180 0.419 0.039 0.813 0.095 0.427 0.304 0.057 0.120 0.327 0.303 0.625 0.058 0.097 0.14 0.00 0.05 0.07 0.03 0.17 0.00 0.11 0.03 0.04 0.14 0.10 0.04 0.04 0.01 0.14 0.11 0.20 0.14 0.00 0.05 0.20 0.00 0.13 0.50 0.16 0.03 0.02 0.01 0.00 0.31 0.12 0.00 0.18 -0.5050 0.6960 0.5240 0.4580 0.4670 0.5680 0.4140 0.4860 0.2970 0.4990 0.3580 0.4820 0.3830 2.7932 0.6997 0.0118 -0.0173 -0.0015 0.0017 0.0002 -0.0013 0.0122 -0.0085 0.0789 -0.0116 0.0681 -0.0011 0.0008 -0.0299 -0.0012 0.93 0.93 0.69 0.36 0.08 0.62 0.97 0.32 2.14 0.50 1.94 0.05 0.68 1.86 0.64 0.361 0.360 0.496 0.724 0.934 0.540 0.343 0.749 0.043 0.624 0.064 0.959 0.503 0.075 0.529 0.03 0.04 0.02 0.01 0.00 0.02 0.04 0.00 0.16 0.01 0.14 0.00 0.02 0.13 0.02 0.56 0.70 0.56 0.59 0.57 0.55 0.63 0.56 0.82 0.58 0.67 0.58 0.60 0.61 0.58 12 PPC Trichomes Mite survival No. peaks at 280nm No. peaks at 345nm Oenothein B Oenothein A Ox-oenothein A Total ET Total CA FG 4-quercetin Total kaempferol Total quercetin Other flavonoids Total FG Toughness % water SLA PPC Trichomes Beetle herbivory No. peaks at 280nm No. peaks at 345nm Oenothein B Oenothein A Ox-oenothein A Total ET Total CA FG 4-quercetin Total kaempferol Total quercetin Other flavonoids Total FG Toughness % water SLA PPC Trichomes Field herbivory No. peaks at 280nm No. peaks at 345nm Oenothein B Oenothein A 0.4684 0.4800 0.0016 -0.0001 0.02 0.16 0.987 0.872 0.00 0.00 0.58 0.56 -0.1340 0.3080 0.6500 0.6330 0.6390 0.6230 0.6370 0.6110 0.6720 0.6370 0.6260 0.5020 0.6956 0.6827 0.6098 0.5891 0.6309 0.0094 0.0243 -0.0002 0.0007 0.0000 0.0002 0.0005 0.0121 -0.0186 0.0009 0.0098 0.0139 -0.0005 -0.0006 0.0001 0.0224 0.0001 1.35 3.06 0.20 0.29 0.02 0.18 0.07 0.99 1.14 0.07 0.60 1.36 0.74 0.08 0.17 0.48 0.30 0.190 0.005 0.843 0.778 0.985 0.858 0.941 0.332 0.266 0.941 0.554 0.186 0.467 0.940 0.867 0.637 0.764 0.07 0.28 0.00 0.00 0.00 0.00 0.00 0.04 0.05 0.00 0.01 0.07 0.02 0.00 0.00 0.01 0.00 0.15 0.35 0.18 0.21 0.22 0.26 0.22 0.27 0.19 0.22 0.19 0.14 0.16 0.24 0.24 0.28 0.19 0.2247 0.1653 0.0845 0.0931 0.1006 0.0752 0.0816 0.1002 0.0816 0.1016 0.1041 0.1431 0.0918 0.6766 0.1900 0.0496 0.0595 -0.0015 -0.0053 0.0004 0.0013 0.0000 0.0003 0.0029 0.0002 0.0209 -0.0004 -0.0019 -0.0048 0.0001 -0.0075 -0.0004 0.0231 0.0006 0.59 1.77 1.19 1.79 0.01 0.87 1.22 0.04 2.40 0.09 0.30 1.48 0.37 3.39 1.31 1.36 2.63 0.560 0.094 0.250 0.091 0.996 0.399 0.238 0.971 0.028 0.927 0.765 0.156 0.713 0.004 0.206 0.192 0.018 0.02 0.16 0.08 0.16 0.00 0.04 0.08 0.00 0.25 0.00 0.01 0.11 0.01 0.40 0.09 0.10 0.29 0.55 0.83 0.57 0.47 0.58 0.56 0.65 0.58 0.52 0.59 0.59 0.71 0.55 0.61 0.47 0.54 0.54 0.6689 0.2164 0.1577 0.1249 -0.0062 -0.0044 0.0000 0.0031 0.93 0.56 0.00 2.00 0.366 0.586 0.998 0.061 0.05 0.02 0.00 0.19 0.01 0.06 0.04 0.01 13 Ox-oenothein A Total ET Total CA FG 4-quercetin Total kaempferol Total quercetin Other flavonoids Total FG Toughness % water SLA PPC Trichomes 0.1183 0.1118 0.0809 0.1429 0.1062 0.1435 0.1781 0.2127 0.1153 0.9871 0.3093 0.0064 0.1087 0.0017 0.0006 0.0112 0.0058 0.0429 0.0044 -0.0131 -0.0055 0.0004 -0.0108 -0.0008 0.0657 0.0006 14 1.76 0.72 2.18 0.54 1.80 0.39 1.09 0.64 0.63 2.09 1.21 1.90 1.06 0.097 0.483 0.044 0.599 0.089 0.701 0.292 0.531 0.536 0.052 0.245 0.074 0.303 0.15 0.03 0.22 0.02 0.16 0.01 0.07 0.02 0.02 0.20 0.08 0.18 0.06 0.00 0.02 0.30 0.08 0.00 0.06 0.03 0.03 0.01 0.02 0.01 0.00 0.05 Table S6 Results from phylogenetic regressions of herbivore performance on principal component summaries of variation in chemical traits (PCchemistry) and chemical and nonchemical traits (PCall traits). When the best fitting model had a slope different from 0 at P<0.1, we show the slope, t-value and P-value in bold. We chose P<0.1 here not as a hard line of statistical significance testing, but to indicate traits of potential interest because the non-measurement error PGLS models show lower statistical power. For completeness, we underline those traits that were also significant (P<0.05) according to non-phylogenetic ordinary least-squares regression. P-values are not corrected for multiple comparisons. Response/Predictor Caterpillar herbivory PC 1 - chemistry PC 2 - chemistry PC 1 - all traits PC 2 - all traits Caterpillar mass PC 1 - chemistry PC 2 - chemistry PC 1 - all traits PC 2 - all traits Caterpillar survival PC 1 - chemistry PC 2 - chemistry PC 1 - all traits PC 2 - all traits Mite eggs PC 1 - chemistry PC 2 - chemistry PC 1 - all traits PC 2 - all traits Mite survival PC 1 - chemistry PC 2 - chemistry PC 1 - all traits PC 2 - all traits Beetle herbivory PC 1 - chemistry PC 2 - chemistry PC 1 - all traits PC 2 - all traits Field herbivory PC 1 - chemistry Intercept Slope t P R2 d 0.0458 0.0499 0.0476 0.0460 -0.0089 -0.0042 -0.0059 0.0061 1.53 0.69 1.28 1.13 0.140 0.499 0.215 0.271 0.09 0.02 0.07 0.05 0.04 0.29 0.21 0.04 0.2190 0.2554 0.2418 0.2246 -0.0522 -0.0269 -0.0382 0.0279 1.76 0.90 1.75 0.99 0.091 0.379 0.093 0.330 0.11 0.03 0.11 0.04 0.29 0.47 0.42 0.32 0.3406 0.3564 0.3562 0.3425 -0.0635 -0.0030 -0.0340 0.0509 2.23 0.09 1.34 1.87 0.036 0.929 0.192 0.073 0.17 0.00 0.07 0.13 0.01 0.16 0.19 0.01 0.4614 0.4654 0.4715 0.4524 -0.0375 0.0153 0.0133 0.0441 0.78 0.33 0.38 1.01 0.443 0.747 0.706 0.322 0.02 0.00 0.01 0.04 0.53 0.57 0.59 0.53 0.6380 0.6412 0.6396 0.6374 -0.0319 -0.0128 -0.0107 0.0118 1.52 0.58 0.63 0.58 0.143 0.566 0.535 0.570 0.09 0.01 0.02 0.01 0.09 0.20 0.19 0.19 0.1000 0.1004 0.1026 0.0992 0.0126 0.0018 0.0084 -0.0094 1.61 0.22 1.49 1.09 0.126 0.830 0.155 0.290 0.13 0.00 0.12 0.07 0.64 0.58 0.52 0.65 0.1597 0.0216 1.33 0.201 0.09 0.02 15 PC 2 - chemistry PC 1 - all traits PC 2 - all traits 0.1577 0.1528 0.1568 0.0007 0.0229 -0.0171 16 0.04 1.87 0.91 0.972 0.079 0.374 0.00 0.17 0.05 0.04 0.00 0.10 Supporting Information Figs S1 & S2 Fig. S1 Representative UV chromatograms and spectra showing quantitative and qualitative variation in polyphenolic profiles among Oenothera species. A1-A6 show chromatograms of six species at 280 nm (highlighting ellagitannin diversity and abundance) and 349 nm (highlighting caffeic acid derivative and flavonoid diversity and abundance). Species were selected to illustrate differences in polyphenolic profiles identified in principle component and hierarchical cluster analyses. A1-2: Plants with high levels of total ellagitannins and high levels of the oxidized oenothein A (ET3). A3-4: Plants with high levels of oenothein B (ET1), caffeic acid derivatives (CA1 and CA2), and quercetin glycosides (particularly peak FL4, a quercetin glycoside). A5-6: Plants with high levels of total flavonoids and high diversity of polyphenolics (number of peaks at 280 and 349 nm). Figure inset shows the chromatogram between 0.95 and 2 minutes to illustrate smaller peaks observed at 280 nm. B: representative UV spectra of different classes of polyphenolics measured in this study, including: ellagitannins, caffeic acid derivatives, and flavonoid glycosides. 17 A A1 1,3e+6 280 nm O. perennis ET3 1,0e+6 7,8e+5 5,2e+5 2,6e+5 1 8,0e+4 CA1 0,0 6,0e+4 4,0e+4 2,0e+4 0,0 FL4 4,0e+4 0,0 0 1 4 5 0 O. grandis 7,2e+5 7,2e+5 4,8e+5 4,8e+5 2,4e+5 2,4e+5 0,0 0,0 FL4 CA2 5,0e+4 0,0 0,0 5,0e+5 1 2 3 4 O. heterophylla 4,0e+5 1 2 3 4 5 Retention Time 280 nm O. triangulata 4,8e+4 ET3 3,6e+4 3,0e+5 2,4e+4 2,0e+5 1,2e+4 1,0e+5 0,0 0,0 6,0e+4 4,0e+4 2,0e+4 0,0 349 nm 9,0e+4 FL14 FL4 349 nm 6,0e+4 3,0e+4 0,0 0 1 2 3 4 5 0 1 2 Retention Time B ET3 - Oxidized oenothein A 1,8e+6 Absorbance (au) FL4 CA2 A6 6,0e+4 280 nm 5 O. humifusa 349 nm 0 5 4 ET1 Retention Time A5 3 1,0e+5 6,0e+4 0 2 Retention Time 280 nm 1,5e+5 349 nm 1,2e+5 1 FL4 9,6e+5 ET1 1,8e+5 CA1 A4 1,2e+6 280 nm 9,6e+5 Absorbance (au) 3 349 nm Retention Time A3 1,2e+6 2 ET1 5,0e+5 2 349 nm ET3 1,0e+6 0,0 1,2e+5 O. longtituba 1,5e+6 ** *** ** ** 280 nm 2,0e+6 ET1 * * A2 2,5e+6 CA1 - Caffoyl tartaric acid 1,5e+6 1,0e+6 1,0e+6 6,0e+5 5,0e+5 5,0e+5 0,0 0,0 1,5e+6 300 400 ET1 Oenothein B 3e+6 300 400 CA2 - Feruloyl tartaric acid 200 1,5e+6 2e+6 1,0e+6 5,0e+5 1e+6 5,0e+5 0,0 0 300 400 5 0,0 200 1,0e+6 200 4 FL4 - Quercetin glycoside 1,5e+6 1,2e+6 200 3 Retention Time 300 400 FL14 - Flavonoid glycoside 0,0 200 300 Wavelength 18 400 200 300 400 Fig. S2 Ordinations from principal components analysis. Shown are the (A) species scores on PC 1 and PC 2 from the PCA on chemical traits, and (B) the trait loadings on these same axes. We also performed PCA on all chemical and non-chemical traits combined, where (C) species’ scores and (D) trait loadings are similarly displayed. The trait clusters identified by hierarchical cluster analysis (see Fig. 4), and the corresponding species with these traits, are shown by “i”, “ii”, and “iii” in C and D. To show the loadings on a comparable scale to that of the scores we multiplied the vectors by 5.5 and 9.5 in B and D, respectively. 33 33 acaulis acaulis acutissima acutissima serrulata serrulata elata elata 313 1 PC 2 202 0 triangulata triangulata sufffulta ulta suf 11 -1 -1 PC 2 00 -2 -2 triangulata triangulata suf sufffulta ulta -1 -1 -3 -3 -2 -2 -4 -4 -3 -3 ruticosa ffruticosa acaulis acaulis longituba longituba speciosa rhombipetala speciosa rhombipetala biennis biennis grandif lora grandif lora ersicolor vversicolor heterophylla lla heterophy gaura gaura perennis perennis af f inis af f inis acutissima acutissima serrulata serrulata elata elata berlandieri berlandieri rosea rosea ffruticosa villaricae illaricae sandiana vruticosa sandiana clelandii clelandii speciosa speciosa rhombipetala rhombipetala biennis biennis grandif grandif lora vvlora ersicolor ersicolor heterophy heterophylla lla perennis perennis af afdrummondii ffinis inis drummondii laciniata laciniata berlandieri berlandieri rosea rosea vvillaricae illaricae sandiana sandiana clelandii clelandii humifusa usa humif drummondii drummondii grandis grandis laciniata laciniata -4 -4 -2 -2 B gaura gaura humif humif usa 00usa -2 -2 11 -1 -1 00 -2 -2 -1 -1 -3 -3 -3 -3 22 00 22 -2 total querc -2 00 FG 4-quercetin 22 -4 -4 PC 11 PC -4 -4 -2 -2 00 22 PC PC 11 C D iiii 44 total FG ii -2 -2 00 -4 -4 -2 -2 serrulata serrulata clelandii clelandii elata ggrandiflora elata randiflora biennis biennis gaura gaura acaulis acaulis heterophylla heterophylla speciosa speciosa acutissima acutissima affinis affinis rhombipetala rhombipetala versicolor versicolor rosea rosea fruticosa fruticosa berlandieri berlandieri serrulata serrulata humifusa laciniata clelandii humifusa laciniata villaricae villaricae clelandii elata elata ggrandiflora drummondii drummondii randiflora biennis biennis gaura gaura grandis grandis perennis perennis affinis affinis sandiana sandiana versicolor versicolor rosea rosea fruticosa fruticosa berlandieri berlandieri humifusa humifusa laciniata laciniata villaricae villaricae drummondii drummondii grandis grandis perennis perennis iii -4 -4 -4 -4 iii -2-2 sandiana sandiana 00 22 44 PC 2 22 44 ii longituba longituba PC 2 triangulata triangulata suffulta suffulta 00 22 peaks 349nm acaulis acaulis heterophylla speciosa heterophylla speciosa acutissima acutissima rhombipetala rhombipetala 22 44 iiii 44 triangulata triangulata suffulta suffulta PC 2 oeno B total CA PC PC 11 PC 2 ox oeno A total ET oeno A peaks 349nm total other FGs peaks 280nm total FG -4 -4 PC 11 PC -4 -4 total kaemp 202 0 -2 -2 -4 -4 grandis grandis -4 -4 22 313 1 PC 2 22 PC 2 A longituba longituba i -2 -2 22 44 -4 -4 -2 -2 66 i sla total querc FG 4 oeno B total CA total ET iii -4 -4 iii -2 -2 00 22 44 66 22 44 66 PC 11 PC -2 -2 00 PC PC 11 19 ox oeno A PPC trichomes -4 -4 PC PC 11 oeno A toughness % water -2 -2 00 longituba longituba 66 ii total kaemp i -4 -4 00 other FGs 00 22 PC 11 PC -4 -4 peaks 349nm