Wilson Island Upward Bound Academy Case Study


Understanding the processes underlying the origin of species is a fundamental goal of biology. It is widely accepted that speciation requires an interruption of gene flow between populations: ongoing gene exchange is considered a major hindrance to population divergence and, ultimately, to the evolution of new species. Where a geographic barrier to reproductive isolation is lacking, a biological mechanism for speciation is required to counterbalance the homogenizing effect of gene flow. Speciation with initially strong gene flow is thought to be extremely rare, and few convincing empirical examples have been published. However, using phylogenetic, karyological, and ecological data for the flora of a minute oceanic island (Lord Howe Island, LHI), we demonstrate that speciation with gene flow may, in fact, be frequent in some instances and could account for one in five of the endemic plant species of LHI. We present 11 potential instances of species divergence with gene flow, including an in situ radiation of five species of Coprosma (Rubiaceae, the coffee family). These results, together with the speciation of Howea palms on LHI, challenge current views on the origin of species diversity.

Speciation with strong gene flow is controversial among evolutionary biologists (1⇓–3). Unlike speciation without gene flow (e.g., allopatric speciation and polyploid speciation) (4), it requires both divergent natural selection and a mechanism to promote nonrandom mating (5⇓–7). Theoretically, these conditions might coincide via a pleiotropic magic trait, through linkage disequilibrium between genes involved in local adaptation and assortative mating, or because habitat differences within a species’ range produce divergent genetic adaptations as well as plastic responses that confer reproductive isolation (e.g., environmentally controlled shifts in flowering time) (3, 5, 7–10). The most controversial incarnation of speciation occurs in sympatry, when gene exchange is high. Numerous definitions of sympatric speciation have been proposed since it was first outlined by Charles Darwin (11). Population genetic definitions focus on random mating with respect to location or habitat of the mating partners. For the biogeographic definition, which we adopt here, the absence of geographic isolation is the key criterion. The relative merits and disadvantages of both views continue to be debated (1–3, 7, 8, 12, 13).

In plants, evidence for the influence of sympatric speciation with strong gene flow remains sparse, with a single study on the Howea palms of Lord Howe Island (LHI) presenting the only conclusive evidence (14, 15). On the other hand, sympatric speciation via polyploidization is well known in plants and is thought to have contributed significantly to species diversity but is not thought to involve ongoing gene flow (16). Very few studies have attempted to quantify the frequency of speciation without geographic isolation, and those that have indicate that it is exceptionally rare (17⇓–19). Kisel and Barraclough (12) demonstrated that the geographic scale required for speciation is related to the spatial scale of gene flow, and so, for organisms to speciate on small islands, they must have very restricted dispersal. In animals, speciation at small spatial scales has been ruled out for some taxa [e.g., in island birds (18) and Caribbean Anolis lizards (19)].

In this study, we conduct an empirical assessment of the frequency of sympatric speciation in plants. We use the flora of the remote LHI as a model system and a set of strict criteria to classify speciation events. Coyne and Orr (8) proposed four criteria for confirmation of a sympatric speciation event: (i) species must be sister taxa; (ii) an allopatric phase in their divergence must be highly unlikely; (iii) species must occur in sympatry; and (iv) species must demonstrate reproductive isolation. Additionally, the sister relationships recovered in phylogenetic reconstructions must not be an artifact of hybridization (8). These criteria, which we apply here, provide a consistent framework for diagnosing sympatric speciation events.

LHI presents an ideal setting to test the frequency of sympatric speciation in plants, because the geography and isolation of the island renders an allopatric phase highly unlikely (14, 15, 20). The product of a shield volcano which erupted 6.9 Mya (21), this small (<16 km2), subtropical island is located 600 km east of Australia. Apart from Ball's Pyramid (a sea stack 551 m high and 0.2 km2 at its base) situated 24 km southeast of LHI, there are no other islands in the vicinity (21, 22). It has been proposed that the small size of the island relative to probable rates of gene flow and the lack of any physical barriers mean that speciation events occurring within the confines of the island fulfill the second and third criteria of Coyne and Orr (8, 14, 15, 20). Currently, the topography of the island is heterogeneous. In the south, Mt. Lidgbird (777 m) and Mt. Gower (875 m) dominate the skyline and support various habitats (21, 23). This variation allows LHI to accommodate a remarkable diversity of species, given its size: 242 vascular plant species have been recorded, 90 of which are endemic (23, 24). The LHI flora has been subject to a thorough taxonomic treatment (24), and current species delimitations are likely to be indicative of complete, or nearly complete, reproductive isolation between close relatives.

Here we combine phylogenetic, karyological, and ecological evidence to assess the contribution of alternative modes of speciation to the composition of the LHI flora. In particular, we test the extent to which the hypothesis of sympatric speciation with gene flow in Howea palms on LHI may be generalized to other native flora. We conduct molecular phylogenetic analyses of genera with more than one native species on the island, through which sister species groups (i.e., potential sympatric speciation events) can be identified. We then evaluate sister species against a range of evidence from molecular dating, chromosome counts, and ecology to support or refute the sister species as products of speciation in the face of gene flow. Previous studies investigating macroevolutionary patterns of sympatric speciation often used co-occurrence of congeneric species as proxies for speciation events (e.g., 12, 18); our study, however, directly assesses the evolutionary relationships of closely related species within an entire plant community.


The native flora of LHI comprises 242 species in 179 genera. In 139 genera only a single species is present on LHI; of these species, 42 are endemic to the island. The remaining 40 genera include between two and six LHI species. In 13 cases, all species are endemic; in 15, all are non-endemic; and the remaining 12 genera contain a mixture of endemic and non-endemic species (Dataset S1).

Biogeographic Analyses of Source Regions for LHI Flora.

Several genera occurring on LHI are species rich and widespread. Biogeographic analyses were used to determine the most likely source regions for LHI species. In turn, this information established which regions are likely to harbor the sister species of LHI plants. The results (Dataset S1) show that dispersal from Australia is likely to have had a strong influence on the composition of LHI's flora (pAustralia = 0.38), with New Zealand, Norfolk Island, New Caledonia, and the Kermadec Islands also acting as significant source regions (pi = 0.15, 0.14, 0.10, and 0.03, respectively). Dispersal to LHI from these focal regions accounts for the majority of LHI species (pfocalregions = 0.81). The probability of a double colonization by two closely related species from a single region outside the focal regions is negligible ([pexcludingfocalregions]2 ≤ 0.0009). As a result, we focused our phylogenetic sampling on these focal regions (SI Appendix, Table S1).

Evolutionary Relationships of LHI Species.

To assess the evolutionary relationships of co-occurring congeneric species on LHI, DNA sequence-based Bayesian analyses (25) were used to generate phylogenetic trees for 32 of the 40 genera with more than one species on LHI. Phylogenetic reconstruction for the remaining eight genera was not possible because material for the LHI species was unavailable. For the 32 genera, we analyzed 2,456 DNA sequences from GenBank and 294 new DNA sequences (SI Appendix, Tables S2–S31). This sampling included 518 species from the focal biogeographic regions (49% of the total for the 32 genera) and 545 species from other regions; all LHI species except Blechnum geniculatum were represented (SI Appendix, Table S1).

For seven genera (Coprosma, Geniostoma, Korthalsella, Melicope, Metrosideros, Peperomia, and Rytidosperma) data from both plastid (cpDNA) and nuclear ribosomal (nrDNA) genomes were available. In these cases BEAST analyses were applied to each data set separately. No hard incongruences in the placement of LHI taxa were evident (SI Appendix, Figs. S1–S7), and the data sets were combined for the final analyses. In the final analyses 18 sister relationships between LHI taxa were identified in 14 genera (SI Appendix, Figs. S8–S38). Molecular dating indicates that four of these divergence events (in Adiantum, Ophioglossum, Cheilanthes, and Pterostylis) are likely to have occurred before the formation of LHI. In three genera (Doodia, Macropiper, and Xylosma) the relationships between LHI species were not fully resolved. We found evidence for one instance of hybrid speciation (26) among endemic species. In the genus Myrsine, cloned DNA shows that M. mccomishii carries sequences derived from M. myrtillina and another species, potentially M. platystigma. The non-endemic species Calystegia affinis, a species found only on LHI and Norfolk Island, also possessed mixed nrDNA sequences similar to C. soldanella (native to both islands) and C. marginata.

Species sampling within the focal regions for phylogenetic reconstruction ranged from 7 to 100% (SI Appendix, Table S1). We argue that phylogenetic reconstructions in which LHI taxa are not recovered as sister species pairs (n = 17) probably represent a true picture of non-sister species relationships, regardless of their level of sampling. The median of the sampling for these phylogenetic trees was 60%. As a result, in phylogenetic trees containing <60% of the focal region species, the recognition of LHI taxa as sister species may be unreliable.

Contributions of Speciation Modes to LHI Flora.

Each native species on the island was assigned to one of the following six categories (Materials and Methods): (i) sympatric speciation (Fig. 1A); (ii) allopatric speciation (Fig. 1C); (iii) colonization without speciation (Fig. 1D); (iv) hybrid species; (v) equivocal cases (Fig. 1 EH); and (vi) unknown mode of divergence. The sympatric speciation results are reported in two categories of phylogenetic sampling (Fig. 2). We have calculated that at least 4.5% of the flora (seven in situ cladogenetic speciation events; Table 1) may be the product of sympatric speciation, that is, with sampling of congeneric taxa from the focal regions ≥60% (Fig. 2). Second, we have evaluated that as many as 8.2% of LHI species (12 in situ speciation events; Table 1) could be the result of sympatric speciation, i.e., for any level of phylogenetic sampling. Colonization followed by no speciation accounts for 55.4% of the species on the island, and allopatric speciation has contributed to the evolution of 24.8% of the island's plant species (Fig. 2). Equivocal cases represented 4.5% of the flora (Fig. 2).

Fig. 1.

(A and CH) Possible speciation scenarios on islands. Triangles represent native island species, with green for endemic species and red for non-endemics. Circles represent species not found on the island. Divergences are shown relative to the age of the island (AOI). Posterior probabilities (pp) are indicated for nodes of interest. (A) Sympatric speciation. (C) Allopatric speciation. (D) Colonization without speciation. (EH) Equivocal scenarios. (B) Metrosideros nervulosa growing on Mt. Gower, LHI.

Fig. 2.

Origins of the LHI flora. Half of the colonization events did not lead to speciation. When speciation did occur (i.e., among endemic species), 12–22% of the resulting species are the products of sympatric divergences. At least 3.7% of all native LHI species are derived from speciation with gene flow events, a subset of sympatric speciation events, equivalent to 10% of the endemic flora (not shown on chart). Phylogenetic trees are given in SI Appendix, Figs. S8–S38.

Table 1.

Putative products of sympatric speciation

Gene Flow and the Geological History of LHI.

We evaluated the potential for geographic isolation to occur in angiosperms on LHI by reconstructing the spatial extent of LHI in the past. At its largest, when sea levels were 100 m lower than they are now, LHI measured 690 km2, and Ball's Pyramid measured 231 km2. During these periods of low sea level the maximum distance from one point on LHI to another on Ball's Pyramid was 57 km, and the minimum distance between LHI and Ball's Pyramid was 4 km.

At the maximum scale of LHI and Ball's Pyramid in the past (i.e., maximum distance of 57 km), we found that the average fixation index (FST, mean = 0.11, median = 0.07) for wind-dispersed plants falls well below the level considered necessary for neutral divergence (FST = 0.20) (12). For such plants, speciation within the island pair may reasonably be considered sympatric. For plants dispersed by other means, average FST is high enough at the same scale to support isolation by distance (FST, mean = 0.24, median = 0.20), suggesting that speciation could have been facilitated by spatially restricted gene flow.

Genetic and Ecological Variation in Metrosideros and Coprosma on LHI.

Two genera (Metrosideros and Coprosma) with well-sampled phylogenetic trees (100% and 80%, respectively) were investigated in more detail. A single sympatric speciation event was found in Metrosideros (Fig. 3B), whereas an in situ radiation with four speciation events was detected in Coprosma (Fig. 3A). Multiple accessions of each LHI species of Metrosideros and Coprosma were sequenced, confirming that are they distinct not only morphologically (24) but also genetically (Coyne and Orr's criterion 4), as shown by the parsimony phylogenetic reconstructions of these samples in SI Appendix, Figs. S39 and S40. Conventional cytological techniques were used to determine the chromosome numbers of endemic Metrosideros and Coprosma species (Table 1), indicating polyploidization was not involved in these speciation events. Metrosideros sclerocarpa occurs predominantly in wet environments surrounding creeks, whereas M. nervulosa (Fig. 1B) grows at higher elevations on exposed ridges and in the cloud forest (mean altitudes for these species are significantly different; P < 0.0001; Welch's t test; Fig. 3C). The flowering period for M. nervulosa (October–January) precedes that of M. sclerocarpa (December–February) (24). In Coprosma, species distributions are staggered along an altitudinal gradient (P < 0.0001; one-way ANOVA; Fig. 3C). Field observations indicate that flowering periods in these Coprosma are largely nonoverlapping: C. huttoniana flowers first (in May and June), C. lanceolaris second (in July and August), and C. inopinata flowers in September and October, with C. putida spanning its more distant relatives, flowering from August–November. Within each Metrosideros and Coprosma species, individuals at lower elevations flower slightly earlier in the season.

Fig. 3.

Ecological divergence in Coprosma and Metrosideros. (A and B) Phylogenetic subtrees with endemic LHI species underlined (full trees are shown in SI Appendix, Figs. S14 and S26). Numbers above branches are the upper 95% CI for the node ages. Numbers below branches are posterior probabilities. (C) Box plot depicting altitudinal distributions, showing the median (in bold type), interquartile range (box), and 1.5 times the interquartile range (bars); circles represent outliers (C. lanceolaris, n = 101; C. inopinata, n = 33; C. huttoniana, n = 46; C. sp. nov., n = 16; C. putida, n = 159; M. nervulosa, n = 99; M. sclerocarpa, n = 78). Flowering periods for each species are given in brackets.

Finally, to determine if sympatric speciation events may have played a widespread role in the evolution of Metrosideros and Coprosma, the age–range correlation approach of Barraclough and Vogler (17) was applied to the Metrosideros and Coprosma phylogenetic trees. In both genera the patterns recovered are consistent with occurrences of both sympatric (e.g., on LHI) and allopatric speciation at the geographic scale applied (SI Appendix, Fig. S41).


Based on the relatedness of endemic LHI species and the timing of their divergences, it is plausible that at least seven sympatric speciation events have taken place on the island. Accounting for 4.5% on the island's flora, these events include four speciation events in the genus Coprosma and one each in Metrosideros, Howea, and the fern genus Polystichum. Polyploid speciation was not evident in Metrosideros and Coprosma, making them good candidates for speciation with gene flow; however, polyploid speciation could not be ruled out in Polystichum. For Metrosideros and Coprosma more detailed phylogenetic and ecological data were collected, including evidence of shifts in elevational ranges and flowering times which may have played important roles in the divergences observed. Ecological divergence and phenological differences have been described previously in the speciation of Howea forsteriana and H. belmoreana (14), and theoretical models support the potential for speciation with strong gene flow in this case (27). In addition to the seven well-supported cases, a further five speciation events (two in angiosperms and three in ferns) also may have occurred recently on LHI. Limited phylogenetic sampling within the latter genera leaves the possibility that these LHI species may not be each other's closest relatives. However, their consideration provides an upper bound to the prevalence of sympatric speciation and speciation in the face of gene flow within the groups studied. If one is willing to accept that these events have occurred on LHI, it is evident that sympatric speciation has played a greater role in the evolution of the LHI flora than may have been expected from studies of other organisms (18, 19, 28). Nevertheless, other processes account for the larger proportion of the islands species, leaving sympatric events in the minority (Fig. 2).

The identified speciation events represent beguiling cases for potential sympatric speciation and speciation with gene flow. However, a number of issues must be considered before accepting them as such. Hybridization, within-island allopatric divergence, multiple colonizations, and extinction all present potential problems when interpreting the data presented here.

Detecting Hybridization.

A caveat of Coyne and Orr's criteria is that the phylogenetic pattern defining sister species pairs must be genuine and not the result of hybridization between distant relatives. The multiple nuclear sequences found in Myrsine mccomishii point to a role for hybrid speciation in the evolution of Myrsine on LHI. This pattern was not observed in any other endemic species group. However, for the majority of genera, either nuclear or plastid DNA regions were available, but not both, and so it is not possible to rule out hybridization or chloroplast capture (8). When data from both genomes were available (i.e., for Coprosma, Geniostoma, and Metrosideros), congruence between nuclear and plastid genomes indicates that postcolonization hybridization is unlikely to have generated these patterns, although we cannot eliminate this possibility completely.

Geographic Isolation Within LHI.

Concerns have been raised that the LHI system does not rule out geographic isolation of populations within the island, because the island may have been larger in the past (29). Studies of the geology and bathymetry indicate that the volcanic areas of LHI currently above sea level were eroded rapidly into their current state within 1–2 million years, after of the initial eruption and since then have been buffered from significant wave erosion by the presence of coral reefs (22, 30, 31). These studies also indicate that neither LHI nor Ball's Pyramid has subsided, but changes in sea level during glacial periods would have increased the terrestrial extent of both to the limits of the sea mounts upon which they sit (22, 30, 31). It is questionable whether these increases in size would have generated greater ecological diversity (20) or habitats suitable for many of the species discussed here, because these species are largely restricted to volcanic soils. If sea level fluctuations during the Pleistocene had led to within-island allopatric speciation, we would expect to observe a correspondence between these climatic events and the divergence times of species pairs (Table 1 and SI Appendix, Figs. S8–S38). Although estimates of divergence time indicate these speciation events took place within the lifetime of LHI, the confidence intervals (CIs) are too large (commonly >2 million years) to confirm whether the speciation events coincided with glacial periods.

Kisel and Barraclough (12) evaluated the scale at which two populations can diverge into separate species with limited opposing gene flow. Our analyses of their data indicate that species with wind-mediated seed or pollen dispersal are likely to have higher gene flow over large distances than plants exploiting other modes of dispersal. This result has important implications for speciation on LHI, because it is possible that populations of plants without wind dispersal may have diverged as the result of geographic isolation. Among our potential cases of sympatric speciation on LHI, the well-sampled angiosperm groups possess some form of wind dispersal (pollen for Howea and Coprosma and seed for Metrosideros). In these groups, geographic distance between individuals is unlikely to have restricted gene flow. The dispersal mechanism of several other taxa remains unknown, with the exception of Alyxia (insect pollinated and bird dispersed) and Geniostoma (bird-mediated seed and pollen dispersal). Therefore, divergence with gene flow restricted by geographic distance between populations cannot be excluded in Alyxia and Geniostoma.

The two mountains on LHI may act as habitat islands for montane species, a possible source of geographic isolation within the island. Among the potential cases of sympatric speciation (Table 1), all the montane species have been observed on both Mt. Lidgbird and Mt. Gower; an indication that this form of isolation is unlikely to have driven these speciation events. However, at this stage it is unclear whether mountaintop populations of the same species (e.g., Alyxia squamulosa) have differentiated genetically.

Currently, vegetation on Ball's Pyramid is sparse. The exposed, sheer cliff faces of the rock spire are inhospitable to most LHI species. Plant life is limited to the endemic shrub Melaleuca howeana (the host plant of the recently rediscovered LHI stick insect, Dryococelus australis), two herb species (Achyranthes aspera and Tetragonia tetragonioides), a sedge (Cyperus lucidus), and a grass (Sporobolus virginicus), all of which also occur on LHI (24, 32). Genetic analyses of these populations would provide direct evidence for the level of isolation of LHI and Ball's Pyramid and shed further light on the potential for allopatric speciation in this system.

Multiple Colonizations as an Alternative to Sympatric Speciation.

Double colonization, rather than in situ speciation, could potentially explain the existence of endemic sister species on LHI. Given enough genetic data, it should be possible to distinguish the phylogenetic pattern generated by in situ speciation, i.e., [(LHI endemic A, LHI endemic B), source population C] from that arising from two independent colonizations from the same source population, i.e., A (B, C). Unfortunately, this distinction may be disrupted by extinction of the source population, failure to sample the source population, or postcolonization hybridization. None of these factors can be ruled out conclusively. Nevertheless, there is some evidence that multiple colonizations by close relatives are unlikely (i.e., double colonization by very close relatives is apparently infrequent among non-endemic species; only the highly vagile Paspalum species have diverged within the last 6.9 million years).

It is possible to assess whether multiple colonization may have had a significant impact on our results. It is reasonable to assume that all non-endemic LHI species have been introduced via separate colonization events. Each endemic species could have resulted from either a unique colonization or an in situ divergence. For the two groups (endemic species versus nonendemic species), we calculated the number of species that co-occur with at least one congeneric species on LHI. If we adopt a null hypothesis that in situ divergence has not taken place, we would expect the frequency of non-endemic congeneric species (FCON) to be similar to the frequency of endemic congeneric species (FCOE). If sympatric speciation has occurred, FCOE should be higher than FCON because of the additive effects from multiple colonizations and in situ speciation events. We found that FCOE is significantly higher than FCON (P < 0.01; two-way z-test): 34.2% of the 152 non-endemic species but 56.7% of the 90 endemic species are accompanied by a congeneric species. We suggest that the higher FCOE is the result of sympatric speciation events among endemic species.


Extinction could have two possible consequences for our results. As mentioned previously, the sister species of an LHI endemic may have existed elsewhere and subsequently gone extinct, resulting in the spurious detection of sister relationships and overestimation of the rate of sympatric speciation. The effect of this occurrence probably is limited, because double colonization of the island is unlikely. Second, species that evolved via sympatric speciation may have gone extinct subsequently, leaving an underestimate of this mode of speciation. Unfortunately, without direct evidence (e.g., fossils), it is impossible to quantify these effects precisely.

Speciation with Gene Flow in Coprosma and Metrosideros.

Polyploidization is well documented in plants and may have acted as an isolating mechanism in the LHI taxa (Table 1) (16, 26). Polyploidization has not been recorded in Metrosideros but had been confirmed in 12 Coprosma species (33) before this study. We were able to exclude polyploid speciation in both genera on LHI (Table 1), suggesting that speciation in these groups has occurred despite high levels of gene flow. Polyploidization has been ruled out previously in the Howea palms (14).

Ecological and altitudinal separation of species is observed in Metrosideros and Coprosma (Table 1 and Fig. 3). The genus Coprosma possesses low rates of concerted evolution, leaving hybrid species with copies of nuclear genes from both parent species (34). A single accession identified morphologically as C. huttoniana growing at 600 m (the lower limit of this species’ range) does possess both C. huttoniana and C. putida copies of nrDNA, suggesting hybridization between these species does occur in sympatric zones but is unlikely to be responsible for the origin of C. huttoniana as a species.

For Metrosideros and Coprosma, we mapped species distributions along transects throughout the island and confirmed that species are found in close proximity. However, species within each genus display some habitat and phenological differentiation. The difference in ecology between Metrosideros species has interesting parallels with the bog and mountain ecotypes found within the Hawaiian Metrosideros polymorpha complex (35). For Metrosideros and Coprosma the available data are consistent with a scenario of ecological speciation under which colonization and local adaptation to new habitats leads to postzygotic reproductive isolation through reduced fitness of migrants and hybrids. Prezygotic isolation via altitudinal shifts in flowering time, either as a plastic or genetic adaptation, may precipitate population differentiation (9, 14, 15, 36). Although the populations are physically in parapatry along this gradient, unrestricted pollen and seed dispersal means that the frequency of encounters between the populations will be high, characteristic of sympatric speciation (3). Ecological (Table 1) and phenological shifts also are found in Alyxia (A. lindii, November–February; A. squamulosa, October–January) and Geniostoma (G. huttonii, January–March; G. petiolosum, September–December). The coincidence of ecological and phenological shifts is common in plants and is an indication that the factors promoting sympatric speciation in plants on LHI may not be unique to this system. Rather, they may be typical of any ecologically complex, isolated island.

The age–range correlation data indicate that allopatry played a role when Coprosma and Metrosideros colonized numerous Pacific islands and presumably speciated via anagenesis, whereas sympatric speciation has occurred within islands and continents. Age–range correlation methods have a number of inherent problems (37), and the results of this analysis should be approached with caution. However, this analysis further indicates that sympatric speciation is likely to have been a feature of the evolution of Metrosideros and Coprosma beyond LHI.


After decades of debate as to whether speciation can occur in the absence of geographic isolation (4–6, 10, 38), we have shown that it may occur frequently in plants on LHI. The speciation events in Metrosideros and Coprosma provide compelling evidence for at least five cases of speciation despite strong ongoing gene flow. The association between flowering time and altitude that is likely to have driven speciation in LHI's Metrosideros and Coprosma species may be widespread in other angiosperms and locations. For the other examples we present here (Table 1), more information is desirable. Nevertheless, we have set an upper bound for the frequency of sympatric speciation on LHI and a lower bound for the frequency of speciation with gene flow. Our results for plants are contrary to those found in insular animals (18, 19), suggesting that speciation in the face of strong gene flow may be a botanical specialty.

Materials and Methods

Biogeographic Analyses.

We coded the presence or absence of all native LHI species in 24 biogeographic regions (Dataset S1). Regions were defined as described by van Balgooy (39) with minor modifications (SI Appendix, Table S32). Endemic species (n = 90) or those present in more than 10 regions (n = 30) were considered as uninformative. Data for 122 indigenous species were used to calculate the probability that each region (i) is the source of a single species selected at random from the LHI species pool (pi, Eq. 1).

where Ns is the number of species included, NR is the number of source regions, and oi,,j is the presence (1) or absence (0) of species j in region i.

Molecular Phylogenetics.

In total 2,456 DNA sequences were downloaded from GenBank, and an additional 294 sequences were generated for this study using standard PCR and sequencing protocols (SI Appendix, Materials and Methods). Applying specific models of evolution for each gene region [assessed using MrModeltest v2.2 (40)], phylogenetic tree searches were conducted using BEAST v1.5.2 (25). Monte Carlo Markov chains for each tree search were run until the effective sample of all estimated parameters exceeded 200 (calculated using Tracer v1.5). When sister relationships were found among LHI species (SI Appendix, Figs. S8–S38), divergence times were estimated using molecular dating (25), calibrated with fossils and/or previously published divergence estimates (SI Appendix, Table S33). Details of tree search parameters are given in SI Appendix, Table S34, and calibration points are given in SI Appendix, Table S33. Maximum parsimony analyses were carried out as detailed in SI Appendix, Materials and Methods.

Categorization of Speciation Events.

Each species was assigned to one of six speciation categories using strict criteria. (i) Sympatric speciation, dependent on fulfillment of the following criteria: (a) LHI species are sister species in the phylogenetic tree; (b) both species are endemic; (c) the species relationship is strongly supported, i.e., by ≥0.9 posterior probability (41); (d) the upper 95% CI for the divergence time falls within the age of the island, i.e., 6.9 million years (Fig. 1). (ii) Allopatric speciation, i.e., endemics with no sister species on LHI. (iii) Colonization without speciation, i.e., non-endemic with no sister species on LHI. (iv) Hybrid species, i.e., species possessing mixed nrDNA sequences. (v) Events meeting a subset of the sympatric speciation criteria (c, d plus either criterion a or that at least one species is endemic) were considered as equivocal. (vi


Although much evidence suggests that the plasma membrane of eukaryotic cells is not homogenous, the precise architecture of this important structure has not been clear. Here we use transmission electron microscopy of plasma membrane sheets and specific probes to show that most or all plasma membrane-associated proteins are clustered in cholesterol-enriched domains (“islands”) that are separated by “protein-free” and cholesterol-low membrane. These islands are further divided into subregions, as shown by the localization of “raft” and “non-raft” markers to specific areas. Abundant actin staining and inhibitor studies show that these structures are connected to the cytoskeleton and at least partially depend on it for their formation and/or maintenance.

In eukaryotes, the plasma membrane serves to segregate the cell from its environment and to serve as the principal interface for communication between cells. Thus, its structure and properties are likely to impact many biological processes. For many years, the “fluid mosaic” model of Singer and Nicolson (1) has shaped our view of the plasma membrane. In this model, proteins diffuse freely in a homogenous lipid environment. This model found support in the work of Frye and Edidin (2), who showed that surface proteins could diffuse throughout a plasma membrane. But subsequent results showed that protein diffusion is 5–50 times slower in the plasma membrane than in artificially reconstituted membranes or liposomes, suggesting that there are significant barriers to movement (3).

Another clue suggesting that the plasma membrane has a more complex architecture was the finding that it was not homogenous with regard to protein and lipid composition, leading to the “lipid raft” model of van Meer and Simons (4). This model suggests that rafts have a distinct lipid composition that requires cholesterol and renders them resistant to certain detergents (5, 6). The partitioning of specific proteins into these lipid rafts has been suggested to be important in many cases of cell surface receptor signaling.

Another type of analysis that has indicated that plasma membranes have distinct compartments is single-particle tracking, which has shown that a number of transmembrane proteins and lipids are restricted in their movement to “confinement zones” that vary in size from 30 to 700 nm, depending on the cell type, protein, or lipid (7, 8). Within these compartments, proteins can diffuse with coefficients similar to those in synthetic membranes or liposomes (7). These results and others have led to the “picket-fence” model, in which transmembrane proteins, like pickets, are anchored to and lined up along a fence of cytoskeletal proteins surrounding the confinement zones (9). Lastly, recent results using single-molecule imaging have shown that GFP-labeled molecules associated with the plasma membrane move within confined and, in at least some cases, nonoverlapping regions (10).

Recently, we became interested in using transmission electron microscopy of membrane sheets to try and approach the problem of plasma membrane structure (11). We adhered T cells and other cells to coated EM grids by a variety of procedures and “ripped” the adherent plasma membrane away from the rest of the cell. This procedure exposed the cytoplasmic face of the plasma membrane to antibodies and other specific markers. By using a variety of probes, we found that all membrane-associated proteins in the cells that we examined are clustered into what we refer to as “protein islands” that can be subdivided further into regions that can be labeled with a “raft” marker versus a “non-raft” marker. Furthermore, all of these protein-rich islands contain actin, which may provide a direct link to the cytoskeleton of the cell. We find the same results with other different cell types as well, suggesting that this type of organization is general and, thus, provides us with a new framework for understanding plasma membrane heterogeneity, function, and intercellular communication.


Plasma Membrane Preparations from Activated and Nonactivated T Cells.

Short-term cultures of lymph node cells from 5c.c7 T cell receptor (TCR) transgenic mice represent an abundant source of physiologically normal antigen-specific T cells. T cells were allowed to bind to EM grids coated either with poly-l-lysine (PLL) or the relevant peptide-MHC (I-EK/MCC), plus costimulatory B7.1 molecules to mimic an antigen-presenting cell surface. T cells were bound to the PLL surfaces for 60 min at 37°C or alternatively preincubated with 50 μM PP2, a src kinase inhibitor that inhibits activation through the TCR and adhesion molecules, for 10 min at 37°C. Otherwise, T cells were activated for 3 min at 37°C on surfaces coated with I-EK/MCC and B7.1.

The activation efficiency of the different surfaces with or without PP2 treatment was analyzed by using video microscopy to assess calcium signaling (12). Untreated T cells interacting with the PLL surfaces adhere and spread strongly. These surfaces also induce sporadic calcium fluxes of very low intensity. T cells adhere well to the activating surface, although spreading is significantly reduced compared with the PLL surface, suggesting that the latter involves Focal Adhesion Kinase. Calcium signaling by T cells on the activating surface is comparable in strength and profile with T cells activated by antigen-presenting cells (data not shown). PP2 treatment completely inhibits cell spreading and calcium signaling in cells interacting with any surface.

For the EM studies, T cells were bound to the coated grids as described above. A coverslip coated with PLL is attached to the tops of the adhered cells. While slight pressure is applied to the coverslip using a rubber cork, excess liquid is removed by suction. The cells are then ripped by manual separation of the coverslip and the EM grid (11, 13–17), which leaves resting (on PLL) or activated (on I-EK/MCC + B7.1) membranes bound to the grid [this procedure is illustrated in supporting information (SI) Fig. 5A]. The membranes are large (≈7–10 μm) sheets of plasma membrane with the cytoplasmic side exposed (SI Fig. 5B). After paraformaldehyde or paraformaldehyde/glutaraldehyde fixation, proteins were labeled with antibodies and other reagents bound to gold particles. The standard EM staining reagents osmium tetroxide, tannic acid, and uranyl acetate were applied sequentially to visualize membrane structures and add contrast. The quality of the fixation was examined by using fluorescent correlation spectroscopy of GFP-tagged raft and non-raft markers (described in Visualizing Raft and Non-Raft Regions) in “live” membrane sheets, demonstrating that membrane-attached proteins are rapidly immobilized upon addition of fixative (data not shown). The quality of the membrane sheets was also demonstrated by the preservation of cytoskeletal structures, clathrin coated pits, and vesicles (SI Fig. 5B). Examples of membrane sheets that were not used for analyses due to imperfections, like ruffles and holes, are shown in SI Fig. 5C.

All Membrane-Associated Proteins Are Clustered.

In our analysis of both activated and resting T cell membranes, we saw a patchwork of dark and light staining regions, with the former occupying ≈20–50% of the plasma membrane depending on cell type and adhesion conditions (Fig. 1A). This pattern in T cells is very similar to that reported previously for mast cell, B cell, and fibroblast membranes (13–17). To better visualize these different regions, we used software that uniformly colors the EM images (Fig. 1A). All original images of membrane sheets are still on hand in either each figure or the supporting information. In our analyses and in those previously reported (13–17), antibodies directed at over a dozen cell surface and signaling proteins invariably labeled their targets within the darker contrast regions. These findings raised the possibility that all membrane-associated proteins would localize to these areas.

Fig. 1.

Localization of membrane-associated proteins in T cells. (A) Membrane sheets from resting T cells (PLL surface after PP2 treatment) were biotinylated on SH groups. Biotinylated proteins were detected by using streptavidin-conjugated 5-nm gold particles. (Left) An unmodified image of a typical membrane sheet. (Center) The same image in pseudocolor with the gold particles shown in filled black circles. (Right) Graph showing clustering by Hopkins analysis. (B) Membrane sheets from activated T cells (I-Ek/MCC + B7.1 surface) were labeled with 5-nm gold particles for tyrosine phosphorylation (Left) and ubiquitinylation (Right). Images are shown in pseudocolor with the gold particles shown as filled black circles. Hopkins analyses for tyrosine phosphorylation (Left) and ubiquitinylation (Right) show significant clustering. The relationship between the EM stain and pseudocolor is shown in the false-color bar.

To test this hypothesis, we first analyzed membrane sheets from resting T cells. Proteins within the membrane sheets were biotinylated on sulfhydryl or carboxyl groups and labeled with streptavidin-gold (Fig. 1A and SI Fig. 6A). However, this labeling procedure did not label proteins that were exclusively on the extracellular side of the plasma membrane (e.g., glycosylphosphatidylinositol-anchored molecules) or had no available sulfhydryl or carboxyl groups in their cytosolic domains. In each case, the distribution of gold particles was analyzed for statistically relevant clustering by Hopkins analysis (18). In this analysis, a Gaussian distribution shows that the label was distributed randomly, whereas a shift to the right indicates clustering. The marked rightward shift in the Hopkins analysis and the localization of the gold label in these experiments indicates that in resting T cell membranes, proteins as a whole were clustered within the darker contrast regions (Fig. 1A and SI Fig. 6A, graphs) and that clustering was independent of the modification chemistry.

The results of these experiments are consistent with the hypothesis that the darkly staining regions in these membrane preparations contain most or all membrane-associated proteins. In >50 membrane sheets analyzed for protein localization by amino acid biotinylation, we did not detect any biotinylated proteins in low contrast areas. These results were repeated and confirmed in membrane sheets attached to PLL at 4°C and 37°C in the presence and absence of PP2 (data not shown).

The same biotinylation patterns were observed in membrane sheets from MDCK (dog kidney), RBL-2H3 (rat basophil), and CHO (Chinese hamster ovary) cells adhering to PLL-coated EM grids (SI Fig. 7). Consequently, this finding that membrane-associated proteins are clustered within the darker staining regions is most likely true for most, if not all, eukaryotic cell types.

We wanted to investigate this phenomenon in activated T cells. However, biotinylation of SH groups in membrane sheets is possible only on EM grids coated with PLL, because the SH groups within the immobilized ligands on an activating surface cause strong background labeling. Therefore, we stained membrane sheets from activated T cells with antibodies specific for posttranslational modifications. These results show that all detectable tyrosine phosphorylation, ubiquitinylation, and symmetrical or asymmetrical dimethylation localized exclusively to the dark areas (Fig. 1B and SI Fig. 6B). Hopkins analyses for all of these stains show strong clustering (Fig. 1B and SI Fig. 6B, graphs). We also assessed the same protein modifications in membrane sheets from nonactivated T cells on PLL (with or without PP2 treatment) and obtained similar results (data not shown). As expected, tyrosine phosphorylation was strongly reduced in nonactivated T cells and undetectable after PP2 treatment.

All of the above results together show that all proteins associated with the plasma membrane were clustered in regions of higher contrast, which we propose to call protein islands.

Visualizing Raft and Non-Raft Regions.

The existence of lipid rafts in plasma membranes has been an area of considerable controversy. To label lipid rafts, we used the N-terminal 10 aa of the tyrosine kinase lck (containing a N-terminal myristoylation site and two S-palmitoylation sites) plus five additional lysine residues and either a tandem HA- or Myc-tag as a raft marker (19, 20). This marker was transformed into a non-raft marker with a “myristate plus basic” signal by mutating its S-palmitoylation sites from cysteine to alanine (20). Both constructs were expressed and localized in membrane sheets from resting and activated T cells (Fig. 2A and SI Fig. 8A). All experiments were repeated with inverted gold sizes and tags. Each tag was detected with antibodies from different species. Noninfected T cells did not show any label with the antibodies specific to the tags. Together, the two markers occupied 45–60% of the protein islands, and neither was detectable in the lighter staining regions. As shown by Hopkins analysis, each marker was highly clustered with respect to itself (Fig. 2A and SI Fig. 8A, top and middle graphs). Colocalization was analyzed by using bivariate Ripley's K statistic (21, 22). The bivariate Ripley's K analysis is shown by a plot of L(t) − t (y axis) versus distance in nanometers (x axis). L(t) − t values represent the number of differently sized neighbors to any gold particle within a certain distance. The black dotted line shows the theoretical values of L(t) − t for two randomly colocalizing gold labels with the same staining intensities as in the image. A data curve (represented by a red line) is based on the actual localization of gold particles in an image and has to lie above the 99% confidence envelope (represented by black lines) to show statistically significant attraction or association of two differently sized gold labels or below it to show their repulsion or explicit separation. The two markers are clearly separated when analyzed in this way (Fig. 2A and SI Fig. 8A, bottom graphs). Interestingly, we find that in the activated membrane sheets the raft and non-raft clusters are more aggregated, with contacts between raft and non-raft regions becoming more frequent. Therefore, the common border increases in length (Fig. 2A and Fig. 8A), indicating functional influences on the protein island morphology. The small number of raft markers in non-raft areas is consistent with previously published results and presumably reflects continuous depalmytoylation (19, 23). These results show that both raft and non-raft proteins are clustered and occupy distinct areas within the protein islands.

Fig. 2.

Localization of raft and non-raft markers. (A) Membrane sheets from resting (Left) and activated (Right) T cells expressing tagged non-raft and raft markers, which are shown in filled red circles (5-nm gold particles) and filled black circles (10-nm gold particles), respectively. Hopkins analyses show clustering for the non-raft marker (top graphs) and raft marker (middle graphs). The Ripley's K analyses (bottom graphs) show repulsion or explicit separation. (B) Membrane sheets from activated T cells infected with either a non-raft (Left) or raft (Right) marker were stained for the expressed marker with specific antibodies (10-nm gold particles, filled black circles) and for cholesterol with perfringolysin O labeled with 5-nm gold particles (filled red circles). Hopkins analyses for cholesterol show slight clustering (top graphs), whereas the raft and non-raft markers are strongly clustered (middle graphs). Ripley's K analyses show colocalization for raft and non-raft markers with cholesterol (bottom graphs). The relationship between EM stain and pseudocolor is shown in the false-color bar.

Another characteristic of lipid rafts is thought to be their enrichment for cholesterol and their loss of “detergent resistance” after depletion of cholesterol with methyl-β-cyclodextrin (6). We detected cholesterol in the inner leaflet of membrane sheets from resting and activated T cells with monomeric perfringolysin-O conjugated to 5-nm colloidal gold (24). Cholesterol distribution was specifically compared with that of the raft or non-raft markers described above. Remarkably, cholesterol is present throughout the protein islands, where it colocalizes with both raft and non-raft markers visually and by Ripley's K statistics (Fig. 2B and SI Fig. 8B, bottom graphs). In contrast, cholesterol is dramatically reduced in the protein-free areas. Hopkins analyses show that the distribution of cholesterol is non-random (Fig. 2B and SI Fig. 8B, top graphs) but less ordered than either marker (Fig. 2B and SI Fig. 8B, middle graphs). We also treated T cells with the cholesterol-depleting reagent methyl-β-cyclodextrin and could no longer detect the perfringolysin-O gold labeling (data not shown). However, depletion of cholesterol has no significant influence on the localization or clustering of the raft and non-raft markers in the membrane sheets of resting or activated T cells, suggesting that cholesterol is not required for localization of membrane-associated proteins to the protein islands (data not shown), although it may well be for functional properties.

Actin Anchors the Protein Islands.

Numerous studies have shown that the cytoskeleton plays an important role in membrane compartmentalization. Therefore, membrane sheets from resting and activated T cells were stained with antibodies to either actin (all six isoforms) or β-tubulin. Actin is detected abundantly in most protein islands and is excluded from the light staining areas (Fig. 3A and SI Fig. 9A). This finding is in contrast to β-tubulin, which can be detected in only a subset (SI Fig. 9A). In additional experiments, both actin and β-tubulin colocalize with the raft and the non-raft markers (SI Fig. 9A).

The heavy actin staining that we see associated with these islands suggests that actin polymerization might play an important role in the structure of these entities. A technical problem with using actin-depolymerizing reagents is that they prevent cells from adhering to surfaces (data not shown). Thus, we surface-biotinylated amine groups on T cells and then incubated them with either cytochalasin D or with the more potent actin-depolymerizing agents latrunculin A or B for 90 min at 37°C. Cells were then attached to EM grids coated with streptavidin for an additional 60 min at 37°C in the presence of the drugs. A coverslip coated with streptavidin was applied to the tops of the cells bound to the EM grid, followed by ripping and paraformaldehyde fixation. EM grids were further treated as described above. Untreated and drug-treated T cells do not spread as much on this surface, and the density of the protein islands was significantly increased compared with the previous ripping techniques (Fig. 3B and SI Fig. 9B). Remarkably, all drug treatments led to a dramatic reduction in the density of the protein islands, with latrunculin A having the greatest effect (Fig. 3B and SI Fig. 9B). The raft and non-raft markers were still clustered and localized to the dark staining regions (Fig. 3B and SI Fig. 9B, top and middle graphs). Raft and non-raft regions were distinct according to Ripley's K statistic and comparable in size to protein islands in untreated cells (Fig. 3B and SI Fig. 9B, bottom graphs). Interestingly, the protein islands in the membrane sheets from drug-treated cells still contained actin (SI Fig. 9C). This result is most likely due to short actin filaments or G-actin monomers.

Fig. 3.

Cytoskeletal association of protein islands. (A) Membrane sheets from resting T cells were stained for actin (all six isoforms) with 5-nm gold particles (filled black circles). Hopkins analysis shows clustering (graph). (B) T cells were surface-biotinylated on NH2 groups and cultured for 90 min with (Upper) or without (Lower) 20 μM latrunculin A. T cells were adhered to streptavidin-coated EM grids for an additional 60 min in the continued absence or presence of the drug and ripped by using a streptavidin-coated coverslip. In both cases, Hopkins analysis shows clustering for raft and non-raft markers in the untreated and latrunculin A-treated cells (top and middle graphs, respectively). Because of the high density of protein islands in the membrane sheets from untreated T cells, no explicit separation can be detected with Ripley's K function (bottom graphs). However, the markers still show separation in the latrunculin A-treated cells (bottom graphs). An obvious reduction in the number of protein islands after latrunculin A treatment is visible. The relationship between EM stain and pseudocolor is shown in the false-color bar.

In contrast to these results, disrupting microtubule polymerization with nocodazole under a variety conditions had no obvious effect on rafts, non-rafts, or the protein islands in general (data not shown).


Our results using membrane sheets derived from short-term cultured T cells and three other cell lines indicate that the plasma membrane is divided into compartments that contain membrane-associated proteins surrounded by protein-free regions. The latter is supported by atomic force microscopy studies of membrane sheets, showing similar sized regions with the thickness of a lipid bilayer (25). (These studies also show that most membrane sheets generated with this method have a continuous lipid bilayer.) The protein islands described here have diameters of ≈30–300 nm, which are very similar to the sizes reported for membrane confinement zones analyzed by using single-particle tracking. Structures of this size are just at or below the limits of fluorescence detection, which explains why they have not been seen previously. We are able to subdivide these protein islands into raft and non-raft regions by using lck-derived raft and non-raft markers. However, not all protein-containing regions were labeled with these two probes. This result and the localization of other raft and non-raft resident proteins suggest a greater complexity to the protein islands than just these designations.

We find that protein islands are enriched for cholesterol, with very little in the protein-free regions, indicating that cholesterol enrichment is not a unique feature of only detergent-resistant membranes or raft structures but is a feature of all protein-containing compartments. Cholesterol could, of course, be used differently in raft regions versus non-raft regions.

The heavy actin staining of these protein islands and their reduced density in membrane sheets after actin depolymerization suggests a close linkage to the cytoskeleton, although given the rapidity with which proteins diffuse in the plasma membrane, it is unlikely that this connection is very rigid. That actin might have a major role in the structure of these regions was indicated in previous single-molecule studies by an increase in the size of confinement zones and therefore a reduction of “hop-diffusion” rates (molecule movement from one confinement zone to another) and a slight increase in macroscopic diffusion rates after the depolymerization of actin (e.g., refs. 3, 7, and 26). However, several studies failed to see an effect of these agents on the movement of membrane-associated molecules (e.g., refs. 26 and 27).

The pattern of high- and low-contrast staining regions using this approach has previously been seen in a wide variety of other cell types (e.g., mast cells, B cells, fibroblasts, ovarian cancer cells, and lymphocytes) (13–17). Because membrane proteins tagged by antibody probes invariably localize to areas of high contrast, the protein island structures that we describe here are likely to be a general characteristic of all or most cell types.

The results described here were obtained by using paraformaldehyde- or paraformaldehyde/glutaraldhyde-fixed membranes. Therefore, one potential artifact is that the observed clustering is due to cross-linking of proteins during fixation. However, the molecules cluster in distinctly non-random ways, such as in the differential clustering of raft and non-raft markers on activating surfaces versus nonactivating surfaces, despite the fact that they cannot interact with the surfaces themselves (Fig. 2A). Furthermore, published data found by using this same protocol has shown that Fc Receptor I on mast cells clusters independently of the adaptor protein LAT (linker for activation of T cells) and associated molecules and that they then move together (but do not interdigitate) upon activation (14). We have very similar results in the T cell system described here showing that TCR/CD3 complexes clustering independent from LAT and that they move together in an activated cell (unpublished data).

Other possible artifacts due to the interaction of cells with the surface of the EM grid were excluded by using three different adhesion methods (immobilized ligands, PLL, and streptavidin) with T cells. On all three surfaces, very similar or identical protein islands were detected.

The data presented here suggest a different model of plasma membrane organization. The original fluid mosaic model of Singer and Nicolson (1) has already been modified over the years into several different models of plasma membrane compartmentalization, driven by evidence suggesting the existence of lipid rafts (4–6) and confinement zones, the detection of hop diffusion (7), and the influence of actin on compartmentalization (7, 26) as well as recent single-molecule studies (10). Additionally, T cell activation induces the formation of TCR-containing microclusters, which can fuse to form larger clusters (28). The latter supports our results showing aggregation of raft and non-raft islands (Fig. 2A and SI Fig. 8A) and TCR/CD3 complexes and LAT clusters in T cells adhered to an activating surface (unpublished data). These phenomena and our results can all be encompassed into the protein island model (Fig. 4), in which proteins localize to “proteinphilic” membrane compartments, probably because of protein–protein interaction and/or their affinity for certain lipids. These protein islands are separated and surrounded by a “sea” of protein-free membrane and are linked to the actin cytoskeleton. These actin anchors and the larger actin polymers in the cortical cytoskeleton most likely play a key role in protein island formation and/or maintenance, as suggested by the inhibitor studies described here. The dynamics and flexibility of the actin cytoskeleton would allow these compartments to be mobile with a certain degree of restriction. Raft regions occupy distinct regions within these protein islands, and previous work has shown that they are uniquely sensitive to cholesterol depletion, suggesting that they have a distinctive structure and function. There may be multiple types of raft regions, as some have suggested, or it may be that different raft resident proteins can localize to different regions on a cell surface by segregating within protein islands (16). There may be protein species that do not congregate in these islands, given that our analyses do not include glycosylphosphatidylinositol-linked proteins or molecules with no available sulfhydryl or carboxyl groups in their cytosolic domains.

Fig. 4.

The protein island model. In this model all membrane-associated proteins are clustered in protein islands (green lipids) that are surrounded by a sea of protein-free membrane (yellow lipids). The islands can be subdivided into raft and non-raft islands, which is also illustrated by their lipid composition (bright-green and dark-green lipids, respectively) and protein contents (gray and red proteins, respectively). Molecules move with high diffusion rates within the islands, and the islands themselves can move with significant restrictions in the membrane. The protein islands are connected to the cytoskeleton (orange), most likely by actin because it plays a crucial role in island formation and/or maintenance. We propose that theses islands can exchange proteins and lipids by hop diffusion when in physical contact.

Our findings have implications for many cellular events involving the plasma membrane. Activation thresholds, segregation, and transduction of signaling pathways at the plasma membrane are all likely to be influenced by the compartmentalization of membrane-associated proteins into protein islands. Protein islands may also play a role in cell–cell communication, membrane trafficking, and membrane fusion. Therefore, although many aspects of the protein islands are unknown at this time, we hope that the results described here will be a useful framework for further investigation and thinking about the structure and function of the plasma membrane.

Materials and Methods

T Cell Culture.

T cells were isolated from lymph nodes of 5c.c7 TCR-transgenic mice and stimulated with MCC peptide with or without retroviral infection with the Phoenix system (29). Cells were maintained in RPMI medium 1640 plus 10% FCS. IL-2 at 30 units/ml was added to the cells after 24 h in culture. Cells were harvested on day 6 or 7.

Preparation of Plasma Membrane Sheets and Gold Labeling.

Formvar- and carbon-coated nickel EM grids were coated with PLL as described (13) or sequentially with biotinylated PLL, streptavidin, and biotinylated I-EK/MCC and B7.1. Cells were allowed to attach to the differently coated grids. The grids were rinsed, placed on an acetate disk, and covered with PLL- or streptavidin-coated coverslips. Pressure was applied to the coverslip by bearing down with a cork. The coverslips were lifted, leaving the lower cell surface adherent to the coated EM grid. Membranes were fixed immediately after the ripping procedure. Membrane sheets were labeled with the appropriate detection reagents and postfixed followed by sequential OsO4, tannic acid, and uranyl acetate staining as described (13). Grids were air-dried and examined by transmission electron microscopy.

EM Data Analysis.

Negatives of membrane sheets were digitized and machine-based colored. The positions of the gold particles are indicated by circles (Figs. 1–3) according to their positions determined during the mapping of gold particles for the statistical analysis. Mapping gold particle distribution and statistical analyses were performed as previously described (21, 22).

For additional details, see SI Materials and Methods.


We thank Dr. Art Johnson (Texas A&M University, College Station, TX) for the monomeric perfringolysin-O construct and John Perrino and Jon Mulholland (Stanford University) for support. Spatial statistics methods were developed by Drs. S. Steinberg, J. Zhang, K. Leiderman, and D. Roberts at the University of New Mexico/Sandia National Laboratories Center for Spatiotemporal Modeling under National Institutes of Health Grant P20 GM66283. This work was supported by National Institutes of Health Grants AI225511 (to M.M.D.) and AI051575 (to B.S.W.) and a grant from the Howard Hughes Medical Institute (to M.M.D.). B.F.L. was supported by a postdoctoral fellowship from the Human Frontier Science Program Organization.


  • To whom correspondence should be addressed at: Howard Hughes Medical Institute, Beckman Center, Room B221, 279 Campus Drive, Stanford, CA 94305. E-mail: mdavis{at}pmgm2.stanford.edu
  • Author contributions: B.F.L., B.S.W., and M.M.D. designed research; B.F.L. performed research; B.F.L., J.R.P., Z.S., and B.S.W. contributed new reagents/analytic tools; B.F.L. analyzed data; and B.F.L. and M.M.D. wrote the paper.

  • The authors declare no conflict of interest.

  • This article contains supporting information online at www.pnas.org/cgi/content/full/0609009103/DC1.

  • Abbreviations:
    T cell receptor;
  • Freely available online through the PNAS open access option.

  • © 2006 by The National Academy of Sciences of the USA

One thought on “Wilson Island Upward Bound Academy Case Study

Leave a Reply

Your email address will not be published. Required fields are marked *