Supporting Information

Winkler et al. 10.1073/pnas.0904852106 SI Methods ‘‘ main lineage (PML clade)’’ of Phytomyza. To obtain Datasets and Phylogeny Estimation. The across- data- meaningful estimates, these nodes with strong previous support set of Scheffer et al. (1) totaled 2,965 base pairs from 86 were also constrained as monophyletic. exemplars and three genes: cytochrome oxidase I (COI), 28S ribosomal RNA, and the nuclear protein-coding CAD, or rudi- Host Effects on Diversification. We tested for host effects on mentary. It was augmented by COI and partial CAD sequences diversification by comparing independent clades of Phytomyza for an additional 13 species of Phytomyza reported in ref. 2, plus associated with Ranunculaceae (n ϭ 5) and asterids (n ϭ 5). We 28S data from these same species, newly generated following the chose the most inclusive clades available that (i) fed entirely or methods of ref. 1. The unpartitioned dataset was analyzed with predominantly on one or the other host clade, (ii) had strong maximum likelihood (ML) in GARLI v.0.951 (3) using the support on the molecular phylogeny, and (iii) supported credible GTRϩIϩG model. Monophyly for the Phytomyza s.str. clade (2), estimates of both crown group age (from the PL analysis) and not initially recovered, was enforced in subsequent analyses, minimum species diversity (from ref. 2) from our sampling. Each because this grouping was strongly established by previous of the asterid-feeding clades represents an independent shift results (1, 2). A bootstrap analysis (500 replicates) performed in from Ranunculaceae feeding with the possible exception of the GARLI to gauge support for monophyly of Phytomyza sensu lato, nigra clade. In total, these 10 clades represent Ͼ440 species, 69% as newly defined (ref. 2; including Napomyza, Ptochomyza, and of the known diversity of Phytomyza s.str. Our selection criteria ), resulted in strong (90%) support for this node, probably bias the test against the predicted trend, since many of which defines the ingroup for the present study. The maximum the taxa excluded belong to apparently species-poor clades likelihood tree of ref. 2 was based on 3,065 base pairs from COI, feeding on Ranunculaceae (see Fig. 2). CAD, and PGD (phosphogluconate dehydrogenase), sequenced The geiger package (12) was used to calculate diversification from 108 species of Phytomyza and 5 outgroup taxa. rates for Phytomyza overall and for each of the 10 focal clades given observed present diversities and clade crown ages from the Divergence Time Calibration. For penalized likelihood analysis of PL analysis. Expected clade sizes, given this overall rate, were divergence times in r8s (4), the logarithmic penalty was used. then calculated for the focal clades (equation 5 of ref. 13), and Identification of the optimal smoothing parameter (s) by cross probabilities of observed clade sizes were calcuated in geiger. validation analysis was not straightforward because calculations For comparison, absolute diversification rates were also esti- failed for some values of s. However, as the remaining calculations mated for each clade. The minimum diversity of each clade was implied an optimal value near 103, suggested in the r8s manual as estimated by summing the numbers of described species in each an upper bound for s in usual cases, the smoothing parameter was component species group as listed in ref. 2. A likelihood ratio test set to 1,000. Penalized likelihood (PL) analysis of the Phytomyza was performed to compare a model with separate speciation only phylogeny used the same parameters. rates for the two groups vs. a model with a single combined rate. The BEAST (5) analysis was performed using the GTRϩIϩG For this test, log likelihoods for individual clades were computed model (partitioned by gene), with a Yule (pure birth) model using equation A18 of ref. 14 (the corresponding equation 1b in prior on speciation rates, implementing the uncorrelated log- ref. 13 is missing one term) and summed over each clade to normal relaxed clock (6) and using an output tree from r8s as a obtain an overall likelihood score (15, 16). An iterative script in starting topology. The weights of several parameter operators R was used to maximize this likelihood score by varying the rate were increased from default values to increase mixing of the parameter r in this equation separately for each group of clades, Yule prior and the frequency of topology changes (swap oper- then under a single combined rate. The ratio of observed to ator on branch rate categories and wide exchange and Wilson– expected clade size was also compared for the two host groups Balding operators to 100; uniform operator on internal node by using a Mann–Whitney U test (one-tailed). Both the Mann– heights and narrow exchange operator to 50). To reach station- Whitney U test and the likelihood ratio tests were repeated for arity, it was also necessary to constrain the following groups as three values of relative extinction rate: ␧ ϭ 0, ␧ ϭ 0.9, and ␧ ϭ monophyletic: Agromyzinae, , and ( ϩ 0.99. ϩ robiniae ϩ Phytomyza group). While the simple birth-death model is widespread in studies of The final analysis was run for 100 million generations, sampling diversification, the temporal constancy of speciation and extinction the chain every 1,000 generations after the initial burn-in period. rates it assumes has recently been rejected for many groups, in part The prior was unstable in this analysis, not reaching stationarity, because the correlation it implies between (log) size and age of but inspection of log files in TRACER v.1.4 (7) showed that age clades is only rarely observed (17). Instead, diversification may estimates for Phytomyza and its component clades were station- often be density dependent or otherwise limited by ecological ary and not affected by this anomaly. opportunity (17, 18). Inspection of the plot in Fig. 3 suggests that Two secondary fossil calibrations were used as minimum ages for our 10 focus clades, the relationship of diversity to age, even for for corresponding stem groups: Palaeophytobia prunorum (Ͼ48 clades on the same host, is neither simple nor obvious. We Ma) (8, 9), a trace fossil in Eocene wood identical to traces of the performed rank correlation analysis on clade age and log of species extant agromyzid genus Phytobia, and the Florissant fossil ‘‘Agro- number, for all 10 clades together as well as separately for asterid myza’’ praecursor (34 Ma) (10, 11), which exhibits the expanded and Ranunculaceae feeders. The results are summarized in Table third antennal flagellomere diagnostic of subgenus S4. The correlation coefficient was positive in every case, but never Dizygomyza. To facilitate comparison of estimates between approached statistical significance. While small sample size may be analyses, the root height (fixed at 64.4 Ma in PL analysis) was part of the explanation, it seems plausible that constant diversifi- tightly constrained in the BEAST analysis with a tight normally- cation does not accurately describe clade dynamics for Phytomyza distributed prior (mean ϭ 64.4 Ma and standard deviation ϭ 0.5 species groups. Ma). In addition to nodes used for calibration, divergence times Given our small sample sizes and the strong host-related were estimated for the Phytomyza group of genera and for the apparent rate heterogeneity, we did not attempt to directly

Winkler et al. www.pnas.org/cgi/content/short/0904852106 1of8 compare the fit of alternative, time-dependent diversification (tm for the second comparison is greater than the crown group models. As a test of the sensitivity of our conclusions to model age, so k3 ϭ 0). In each case, single-rate vs. two-rate models were assumptions, however, we conducted a ‘‘model-independent’’ compared by the Akaike information criterion (AIC) and by comparison of net diversification on the two hosts, consisting of likelihood ratio tests. a Mann–Whitney U test on clade sizes uncorrected for age. This We implemented the method of Rabosky (20) using the procedure is justified by the closely similar age distributions of LASER package v.2.1 (21) to compare single-rate vs. three-rate the two host-related sets of clades, as seen in Table S4. This test, Yule models by using the AIC and a likelihood ratio test. All like those based on the simple birth-death model, showed times were again adjusted using t0 ϭ 19 Ma, and shift times were significantly greater diversities for asterid-feeding clades (P ϭ constrained to 33 Ma and 24 Ma as in the previous analysis by 0.016, one-tailed; Table S4). Thus, while uncertainty about the modifying the yule3rate function of LASER. Because the shift model hinders identification of the underlying mechanism (see times were not estimated from the data (as they are generally by Discussion), and precludes strong inferences regarding actual using this function in LASER), we adjusted the AIC accordingly diversification rates during host-associated radiations, the fun- to incorporate only three free parameters. damental conclusion of greater net diversification by asterid An assumption of the branching time tests is complete sam- feeders seems robust to model details. An analogous tradeoff of pling of lineages up to 19 Ma. We believe this assumption to be detail for robustness is implicit in the often used replicated sister approximately true for our data, we know of no definite excep- group approach to diversification patterns. tions. However, some excluded taxa may belong to unidentified, early-branching lineages. A rough estimate of the possible Temporal Shifts in Diversification Rate. For tests of temporal shifts number of such taxa follows, derived from ref. 2. Our initial in diversification rate across major climatic transitions, the phylogenetic study left ϳ150 of 700 Phytomyza species unplaced equations of Paradis (19) were modified to model only periods to species group. Fifty of these species have not been examined of interest in the following manner. The minimum age of the for traits (male terminalia) defining species groups; nearly all interval (t0) was subtracted from all branching times, then probably belong to one of the major species groups. Approxi- additional terms were included representing survival of a certain mately eighty species (10 of which were sampled by our study) number of lineages (k3) past the maximum interval age (tm). The cannot be securely assigned to an identified species group, but final, modified equations are as follows (not including terms show general affinity to one or more of the four major clades representing censored, or minimum ages, which were not used within the PML clade; for example, nearly half potentially belong here; t and t are also adjusted by t ): m c 0 to the aquilegiae clade. It is likely that among these unsampled (a) Model A: constant speciation rate species, several lineages are represented which branch early k1ϩ2 within the PML clade, before 19 Ma. Ten to 15 species of ϭ ␦ Ϫ ␦ ͸ Ϫ ␦ Phytomyza show no obvious affiliation to any of the major clades log L k1ϩ2 log t͑1ϩ2͒i k3 tm [1] iϭ1 or other species groups; these are the most likely candidates for additional, early-branching lineages, but could also be anoma- (␦ in middle term was erroneously omitted in ref. 19) lous members of more typical lineages. Therefore, a liberal estimate of the number of missing lineages at 19 Ma would be k1ϩ2 between 10 and 20; the actual number may be much less. ␦ ϭ k ϩ /ͩ͸ t͑ ϩ ͒ ϩ k t ͪ [2] To simulate the possible effect of incomplete lineage sampling max 1 2 1 2 i 3 m iϭ1 on our finding of no significant effect of major Oligocene climate changes on Phytomyza diversification, we generated 1,000 ran- (b) Model C: rate shift at time tc dom phylogenies by using the Yule (pure birth) model in geiger

k1 for each of four possible numbers of missing taxa (m): 0, 6, 12, ϭ ␦ Ϫ ␦ ͸ ϩ ␦ ϩ ͑␦ Ϫ ␦ ͒ and 24. As our phylogeny shows 25 extant Phytomyza lineages at log L k1log 1 1 t1i k2log 2 k2tc 2 1 ϩ ϭ 19 Ma, each random tree contained 25 m taxa, after which i 1 lineages were pruned randomly until 25 lineages remained. All k2 trees were subjected to the above test in laser comparing a Ϫ ␦ ͸ Ϫ ␦ Ϫ ␦ ͑ Ϫ ͒ 2 t2i k3 1tc k3 2 tm tc [3] single-rate to a three-rate Yule model. LASER calculations iϭ1 failed for 5–10% of simulated phylogenies because these con- tained intervals without any branching times; these trees were k1 not included in the simulation results. The distribution of the ␦ ϭ ͸ ϩ ͑ ϩ ͒ likelihood ratios for remaining phylogenies under each value of max k /ͫ t i k k t ͬ [4] 1 1 1 2 3 c iϭ1 m was used to obtain an adjusted P value for the original comparison. These P values (Table S3) were all Ͼ0.25, and k2 increased monotonically with increasing numbers of missing ␦ ϭ ͸ ϩ ϩ ͑ Ϫ ͒ taxa. Therefore, our inability to find significant climate-related max k /ͫ t i k t k t t ͬ [5] 2 2 2 2 c 3 m c iϭ1 shifts in diversification rates is unlikely to be a methodological artifact of incomplete lineage sampling. It remains possible, of This test was repeated using the Phytomyza chronogram for the course, that sampling more actual lineages, if such exist, would periods 19–24 Ma vs. 24–33 Ma and for 24–33 Ma vs. 33–40 Ma provide more power to detect such shifts.

1. Scheffer SJ, Winkler IS, Wiegmann BM (2007) Phylogenetic relationships within the 3. Zwickl DJ (2006) GARLI: Genetic algorithm approaches for the phylogenetic analysis of large leaf-mining flies (Diptera: Agromyzidae) inferred from sequence data from multiple biological sequence datasets under the maximum likelihood criterion. PhD dissertation (Univ genes. Mol Phylogen Evol 42:756–775. of Texas, Austin). Available at www.nescent.org/wg࿝garli. 2. Winkler IS, Scheffer SJ, Mitter C (2009) Molecular phylogeny and systematics of 4. Sanderson MJ (2006) r8s, Analysis of Rates (‘‘r8s’’) of Evolution (Univ of California, leaf-mining flies (Diptera: Agromyzidae): Delimitation of Phytomyza Falle´ n sensu lato Davis) Version 1.71. Available at http://loco.biosci.arizona.edu/r8s. and included species groups, with new insights on morphological and host-use evo- 5. Drummond AJ, Rambaut A (2007) BEAST: Bayesian evolutionary analysis by sampling lution. Syst Entomol 34:260–292. trees. BMC Evol Biol 7:214. 6. Drummond AJ, Ho SYW, Phillips MJ, Rambaut A (2006) Relaxed phylogenetics and dating with confidence. PLoS Biol 4:699–710.

Winkler et al. www.pnas.org/cgi/content/short/0904852106 2of8 7. Rambaut A, Drummond AJ (2007) TRACER, BEAST software, Version 1.4. Available at 16. Ricklefs RE (2007) Estimating diversification rates from phylogenetic information. http://beast.bio.ed.ac.uk. Trends Ecol Evol 22:601–610. 8. Su¨ssH,Mu¨ ller-Stoll WR (1980) 100 Jahre Arboretum (1879–1979), ed Vent W (Hum- 17. Rabosky DL (2009) Ecological limits on clade diversification in higher taxa. Am Nat boldt Univ, Berlin), pp 343–364. 173:662–674. 9. Smedes HW, Prostka HJ (1972) Stratigraphic framework of the Absaroka volcanic 18. Rabosky DL (2009) Ecological limits and diversification rate: Alternative paradigms to supergroup in the Yellowstone National Park region. US Geol Surv Prof Pap 729-C. explain the variation in species richness among clades and regions. Ecol Lett 12:735– 10. Melander AL (1949) A report on some Miocene Diptera from Florissant, Colorado. Am 743. Mus Novit 1407:1–63. 19. Paradis E (1997) Assessing temporal variations in diversification rates from phylog- 11. Evanoff E, McIntosh WC, Murphey PC (2001) Stratigraphic summary and 40Ar/39Ar enies: Estimation and hypothesis testing. Proc R Soc Lond B 264:1141–1147. geochronology of the Florissant Formation, Colorado. Proc Denver Mus Nat Sci 4:1–16. 20. Rabosky DL (2006) Likelihood methods for detecting temporal shifts in diversification 12. Harmon LJ, et al. (2009) The geiger package: Analysis of evolutionary diversification, Version 1.2–14. Available at http://cran.r-project.org. rates. Evolution 60:1152–1164. 13. Magallo´ n S, Sanderson MJ (2001) Absolute diversification rates in angiosperm clades. 21. Rabosky DL (2006) LASER: A maximum likelihood toolkit for detecting temporal shifts Evolution 55:1762–1780. in diversification rates from molecular phylogenies. Evol Bioinf Online 2006:257–260. 14. Raup DM (1985) Mathematical models of cladogenesis. Paleobiology 11:42–52. Available at http://cran.r-project.org.. 15. Bokma F (2003) Testing for equal rates of cladogenesis in diverse taxa. Evolution 22. Wilf P, Labandeira CC, Johnson KR, Ellis B (2006) Decoupled plant and diversity 57:2469–2474. after the end-Cretaceous extinction. Science 113:1112–1115.

Winkler et al. www.pnas.org/cgi/content/short/0904852106 3of8 Agromyza sp. 1 Agromyza frontella Agromyza pseudoreptans Agromyza sp. 2 Agromyza ambrosivora Japanagromyza viridula Ophiomyia sp. 2 Ophiomyia phaseoli Melanagromyza obtusa Melanagromyza chalcosoma Melanagromyza cleomae Melanagromyza minimoides Melanag. virens/splendida Tropicomyia theae Hexomyza schineri Ophiomyia nasuta Ophiomyia cornuta A Ophiomyia quinta Ophiomyia sp. 1 Ophiomyia lantanae Cerodontha (C.) dorsalis Cerodontha (X.) atronitens Cerodontha (I.) capitata Cerodontha (Po.) incisa Cerodontha (Po.) muscina Cerod. (Ph.) flavocingulata Cerodontha (B.) angulata Cerodontha (D.) fasciata malvae C Caycomyza solidaginis Calycomyza majuscula Calycomyza flavinotum Calycomyza hyptidis Calycomyza lantanae Phytoliriomyza arctica Phytoliriomyza felti Metopomyza flavonotata Metopomyza scutellata Phytoliriomyza sp. 1 Galiomyza violivora philadelphivora Liriomyza fricki Liriomyza baptisiae Liriomyza chinensis Liriomyza huiobrensis Liriomyza trifoliearum Liriomyza brassicae Pseudonapomyza sp. 1 Pseudonapomyza lacteipennis Nemorimyza posticata Nemorimyza maculosa Phytobia sp. 1 Phytobia setosa Amauromyza flavifrons Amauromyza monfalconensis Amauromyza pleuralis Phytoliriomyza robiniae Aulagromyza luteoscutellata Aulagromyza orbitalis B Gymnophyt. heteroneura Aulagromyza nitida Aulagromyza discrepans Phytomyza scolopendri Phytomyza (N.) montanoides Phytomyza (N.) plumea Phytomyza (N.) lateralis Phytomyza (N.) cichorii Aulagromyza tridentata Phytomyza minuscula Phytomyza aquilegivora Phytomyza Phytomyza continua Phytomyza fuscula Phytomyza spinaciae Phytomyza syngenesiae Phytomyza lactuca Phytomyza ilicicola Phytomyza s.str. Phytomyza glabricola Phytomyza loewii Phytomyza angelicae Phytomyza angelicastri Phytomyza erigerophila Phytomyza Phytomyza primulae Phytomyza aprilina Phytomyza vitalbae main lineage Phytomyza aconiti Phytomyza davisii Phytomyza tetrasticha Phytomyza aquilegioides Phytomyza aquilegiana Phytomyza fallaciosa Phytomyza plantaginis Phytomyza trivittata Phytomyza sp. 'North Carolina' Phytomyza evanescens Phytomyza costata 60 50 40 30 20 10 0 Million Years Ago Paleocene Eocene Oligocene Miocene Plio

Fig. S1. Time-calibrated (PL) phylogeny of Agromyzidae, from ML analysis of family-wide dataset based largely on ref. 1 with additional species of Phytomyza added. Fossil calibrations are (A) leaf mine on Platanus (22), 64.4 Ma; (B) Palaeophytobia prunorum, borings in fossil wood (8), Ͼ48 Ma (9); (C) ‘‘Agromyza’’ praecursor, compression fossil from Florissant deposit (10), 34.0 Ma (11).

Winkler et al. www.pnas.org/cgi/content/short/0904852106 4of8 Table S1. Focal clades of Phytomyza, with crown group ages, minimum diversities, expected clade sizes, and estimated net diversification rates calculated using the geiger package (12) Minimum Expected clade diversity size P value Rate per Ma

Crown Main Clade Taxa included Hosts group age Total host ␧ ϭ 0 ␧ ϭ 0.9 ␧ ϭ 0 ␧ ϭ 0.9 ␧ ϭ 0 ␧ ϭ 0.9

Phytomyza (all) — 39.1 702 — — — — — 0.150 0.108 R1 minuscula group Isopyreae (Ran.) 8.33 3 3 6.98 16.9 0.918 0.886 0.049 0.014 R2 hendeli, loewii groups Ranunculaceae (3 tribes), 3 other 19.38 15 10 36.6 76.1 0.824 — 0.104 0.042 families R3 notata group Ranunculeae/Anemoneae (Ran.) 13.32 16 15 14.75 35.3 0.377 0.651 0.156 0.065 R4 albipennis, Ranunculeae/Anemoneae (Ran.) 8.41 26 26 7.06 17.1 0.003 0.222 0.305 0.143 ranunculella groups R5 aquilegiae group s.l. Isopyreae (Ran.) 15.68 37 34 21.01 48.1 0.131 0.470 0.186 0.094 A1 nigra clade Asteraceae, some Poaceae 8.9 63 55 7.6 18.5 Ͻ0.001 0.031 0.388 0.216 A2 albiceps, spondylii, mostly Asteraceae, Apiaceae 14.61 130 130 17.9 41.9 Ͻ0.001 0.044 0.286 0.177 angelicae groups A3 atomaria group Plantaginaceae, Orobanchaceae, 11.64 57 47 11.5 28.0 Ͻ0.001 0.130 0.288 0.158 few Ranunculaceae, Fabaceae A4 agromyzina clade Aquifoliaceae, Caprifoliaceae, 10 20.72 80 46 44.8 89.3 0.127 0.411 0.178 0.103 other families A5 obscura group Lamiaceae, Boraginaceae 8.78 18 18 7.5 18.2 0.036 0.383 0.250 0.107

Crown group ages are taken from the PL analysis, and minimum diversity estimates (total described species and with species on alternative host clades subtracted) are taken from ref. 2. Expected clade sizes assume a constant diversification rate for the genus; this assumption was removed to obtain rate estimates individually for each clade. Ran., Ranunculaceae

Winkler et al. www.pnas.org/cgi/content/short/0904852106 5of8 Table S2. Results of likelihood ratio tests comparing birth-death diversification models with separate rates for Ranunculaceae- and asterid-feeding Phytomyza clades vs. single-rate models ␧ value Rate (Ran.) Rate (Ast.) Rate (comb.) Likelihood ratio P value

0 0.175 0.277 0.240 8.162 0.004 0.9 0.062 0.145 0.113 5.249 0.022 0.99 0.007 0.026 0.017 7.186 0.007

Relative extinction rates of ␧ ϭ 0, ␧ ϭ 0.9, and ␧ ϭ 0.99 were used. Ran., Ranunculaceae-feeding clades; Ast., asterid-feeding clades; comb., all clades combined.

Winkler et al. www.pnas.org/cgi/content/short/0904852106 6of8 Table S3. Results of branching time analyses testing for temporal diversification rate variation from 40–19 Ma using survival analysis (SA) (19) and standard likelihood model fitting (ML) (20) Analysis Rate (A) Rate (B) Rate (C) Single rate Likelihood ratio P value ⌬AIC

SA (A:B) 0.274 0.094 0.128 2.455 0.117 Ϫ0.455 SA (B:C) 0.094 0.177 0.140 1.796 0.180 ϩ0.204 ML (A:B:C) 0.145 0.082 0.166 0.130 2.279 0.320 ϩ1.72 ML (m ϭ 0) 0.267 ML (m ϭ 6) 0.277 ML (m ϭ 12) 0.299 ML (m ϭ 24) 0.387

Both methods here assume a pure birth (Yule) model (i.e., ␧ ϭ 0) and compare likelihoods of models with single rates vs. those incorporating multiple rates. Adjusted P values for the ML test obtained by parametric simulation of phylogenies with several different levels of incompleteness due to missing lineages (m) are also given (see Text). Time periods tested correspond to major climatic transitions, and are as follows: 40–33 Ma (A), 33–24 Ma (B), and 24–19 Ma (C).

Winkler et al. www.pnas.org/cgi/content/short/0904852106 7of8 Table S4. Analysis of rank correlation between crown group age and clade size for focal clades of Phytomyza and a nonparametric test of difference in clade size between Ranunculaceae vs. asterid feeders Clade Taxa included Crown group age Rank age No. of species Log no. of species Rank no. of species

R1 minuscula group 8.33 1 3 0.477 1 R2 hendeli, loewii groups 19.38 9 15 1.176 2 R3 notata group 13.32 5 16 1.204 3 R4 albipennis, ranunculella groups 8.41 2 26 1.415 5 R5 aquilegiae group s.l. 15.68 7 37 1.568 6 Mean, Ranunculaceae feeders 13.04 4.8 19.4 3.4 A1 nigra clade 8.9 4 63 1.799 8 A2 albiceps, spondylii, angelicae groups 14.61 6 130 2.114 10 A3 atomaria group 11.64 5 57 1.756 7 A4 agromyzina clade 20.72 10 80 1.903 9 A5 obscura group 8.78 3 18 1.255 4 Mean, Asterid feeders 12.93 5.6 69.6 7.6

Correlation analysis, clade age/log clade size Spearman rank correlation P value

Ranunculaceae feeders only 0.17 0.72 Asterid feeders only 0.72 0.14 All 10 clades, ranked together 0.41 0.22 All 10 clades, host groups ranked separately 0.47 0.16

Mann–Whitney U test on species number (uncorrected), Ranunculaceae vs. asterid feeders. U ϭ 2; P ϭ 0.016 (one-tailed).

Winkler et al. www.pnas.org/cgi/content/short/0904852106 8of8