Abstract
A contemporary interpretation of Dollo’s Law states that the evolution of a specialized structure is irreversible. Among land plants, reproductive specialization shows a trend toward increasing complexity without reversion, raising questions about evolutionary steps and the irreversibility of reproductive complexity. Ferns exhibit varied reproductive strategies; some are dimorphic (producing separate leaves for photosynthesis and reproduction), while others are monomorphic (where one leaf is used for both photosynthesis and spore dispersal). This diversity provides an opportunity to examine the applicability of Dollo’s Law in the evolution of reproductive leaf specialization. We analyzed 118 species in Blechnaceae and Onocleaceae, applying quantitative morphometrics and phylogenetic comparative methods to test the pillars of a modernized interpretation of Dollo’s Law. The evolution of dimorphism in Blechnaceae is neither stepwise nor irreversible, with direct transitions from monomorphism to dimorphism, including several reversions. In contrast, Onocleaceae exhibits an irreversibility to monomorphism only upon further specialization of fertile leaves for humidity-driven spore dispersal; this suggests that additional specialization, not dimorphism alone, may facilitate irreversibility. These results provide insight into the canalization of fertile-sterile leaf dimorphism in seed plants, where the addition of traits like heterospory and integuments lead to further specialization and potential irreversibility. These findings suggest that as new specialized traits evolve alongside preexisting ones, reversion may become increasingly unlikely.
Introduction
The evolution of complex phenotypes relies on the specialization of distinct organs. For instance, the origin of eyes in animals was facilitated by the specialization of light-sensing structures into complex eye spots (Land & Fernald, 1992; Schwab, 2018); the swim bladder in fish evolved from the specialization of an ancestral lung (Liem, 1988); and flowers evolved from the specialization of reproductive leaf series in a determinate shoot system (Specht & Bartlett, 2009). This mode of extreme specialization (e.g., the evolution of an organ into a highly specific and narrow function) allows for a division of labor, which can lead to unique morphological features, and even complex social behaviors (Cooper & West, 2018). To understand the evolution of diverse body plans and phenotypes, it is imperative to examine the processes underpinning organ specialization.
In some cases, however, the origin of highly specialized structures may lead a lineage down an irreversible evolutionary trajectory (i.e., Dollo’s Law). As originally formulated, Dollo’s Law states that a “complex” structure is never reevolved in an identical form after it has been lost (Gould, 1970). While historically coopted to support orthogenesis (directed evolution), over time Dollo’s Law has been refined to explain the broader concept of irreversibility in complex traits, often in a macroevolutionary framework (Goldberg & Igić, 2008; Simpson, 1953). There are several evolutionary expectations under a newly refined Dollo’s Law. For instance, transitions involving complex traits are expected to be directional and stepwise, with evolutionary changes toward increasing specialization being more common than reversals. As a result, transitions from specialized to generalized states are predicted to be unlikely, rare, or absent. Additionally, the rate of trait evolution is expected to slow in highly specialized lineages due to “constraints” imposed by trait complexity. While sometimes controversial, Dollo’s Law has played a role in understanding the evolutionary processes that shape organ specialization and the conditions under which reversals might, or might not, occur (Bull & Charnov, 1985; Goldberg & Igić, 2008; Kohlsdorf & Wagner, 2006; Lynch, 2023; Visser et al., 2020).
In plants, broadly irreversible reproductive specialization (here focused on the production of sporangia on specialized structures) is a key feature of their evolutionary history (Leslie et al., 2021). For instance, the earliest vascular plants such as Cooksonia produced sporangia directly on the terminal tips of vegetative axes (Edwards & Feehan, 1980). In contrast, seed plants produce sporangia wrapped in maternal tissue (e.g., an ovule) born on wholly distinct structures for reproduction. While some functional exceptions occur (Bazzaz & Carlson, 1979; Dickinson & Sattler, 1975; Dufaÿ et al., 2003), evolutionary specialization has entirely separated assimilation and reproduction in seed plants (e.g., leaves photosynthesize and ovules/cones/flowers reproduce). The progressive and incremental specialization of reproductive structures across land plants, coupled with the absence of reversions of reproductive leaf dimorphism to ancestral states among seed plants, raise questions about the steps involved in the evolution of fertile-sterile leaf dimorphism and whether lineages can revert from this condition. Specifically, does the evolutionary trajectory toward reproductive leaf dimorphism follow a stepwise and irreversible pattern, in line with a modern interpretation of Dollo’s Law?
While it is difficult, if not impossible, to answer this question in seed plants given their full separation of reproductive and photosynthetic functions, ferns offer a contrasting system where this division is not so clear. Most ferns are “monomorphic,” meaning they balance both photosynthesis and reproduction on a single organ (Goebel & Balfour, 1900; Vasco et al., 2013); many, however, are reproductively “dimorphic,” meaning they produce distinct forms of spore-bearing (fertile “sporophylls”) and nonspore-bearing (sterile or vegetative “trophophylls”) leaves. With their diversity in reproductive leaf strategies (Wagner & Wagner, 1977), ferns present an ideal system for exploring the evolution of fertile-sterile leaf dimorphism and the steps involved in organ specialization.
One major clade of ferns with diverse reproductive leaf morphology includes the families Onocleaceae and Blechnaceae. The Onocleaceae has four genera and five species (PPG, 2016), restricted mostly to mesic temperate and subtropical regions. Nearly all species of the Onocleaceae are holodimorphic; that is, have conspicuously dissimilar sterile and fertile leaves. Importantly, reproductive leaf dimorphism in Onoclea sensibilis (Onocleaceae) is so specialized that it has led to the differential investment in the hydraulics and biomechanics of fertile compared to sterile leaves, mirroring investment strategies in seed plants (Suissa et al., 2024). Moreover, fertile fronds in O. sensibilis evolved additional specialization with hygromorphic properties (i.e., movement in response to changes in humidity) that help facilitate spore dispersal (Suissa, 2021), akin to seed-dispersal strategies in some pine cones (Dawson et al., 1997). Sister group to the Onocleaceae is the Blechnaceae, which contains about 265 species. The group is most diverse in the American tropics as well as Southeast Asia and Australia (de Gasper et al., 2017; Testo et al., 2022). The Blechnaceae include species that are monomorphic and others that have varied degrees of dimorphism, including holodimorphism. Together, the functional specialization of fertile fronds in the Onocleaceae and the diversity of reproductive leaf strategies in the Blechnaceae provide an ideal system to investigate the evolution of fertile-sterile leaf dimorphism, as well as whether further specialization of fertile fronds alters these evolutionary dynamics and the ability to revert to monomorphism.
Here, we examine the evolutionary transition rates and directionality of reproductive leaf dimorphism in the Blechnaceae and Onocleaceae to test whether Dollo’s Law applies to the evolution of fertile-sterile leaf dimorphism. Leveraging natural history collections and quantitative morphometrics, we generated information on fertile-sterile leaf dimorphism for 118 species. Integrating this dataset with a recent time-calibrated phylogeny and stochastic character mapping, we estimate the directionality, rate, and evolutionary lability of reproductive leaf dimorphism. Finally, we determine the presence of further functional specialization of dimorphic fronds in the Onocleaceae by examining the presence of humidity-driven pinnule movement for spore dispersal. By determining the pathways and mechanisms underlying fertile-sterile leaf dimorphism, our study offers insights into the evolution of reproductive leaf specialization, informing broader patterns of reproductive evolution across land plants. Moreover, our insights suggest that the irreversibility described by Dollo’s Law may not stem from the evolution of a single specialized trait alone. Instead, the accumulation of multiple specialized traits, interacting and reinforcing each other, likely makes reversals to simpler states increasingly improbable.
Materials and methods
Sampling
Species were selected based on their presence in a recent phylogeny of Blechnaceae and Onocleaceae (de Gasper et al., 2017). For each species in the tree, we searched for high quality digitized herbarium specimens on the Global Biodiversity Information Facility (GBIF, 2024; https://www.gbif.org/) and PteridoPortal (2024). Images of specimens that most accurately matched the typical description of each species were selected manually and downloaded onto a local drive (https://github.com/Suissajacob/blechnaceaeDimorphism). Synonymy was addressed by cross-referencing species names against a list of accepted species names on the Checklist of Ferns and Lycophytes of the World (Hassler & Schmitt, 2018). For analyses of humidity-driven pinnule movement, we destructively sampled fertile pinnae from vouchered herbarium specimens from the Harvard University Herbaria.
Quantitative leaf analysis
We chose the highest-quality digitized herbarium specimen to use in our analyses. This was based on the presence of nonoverlapping fertile and sterile leaves. Individual specimens were uploaded into imageJ (Schindelin et al., 2012), and a scale was set using the scale bar mounted on the digitized specimen. For each species, at least five pinnae from the middle of the frond were examined. Fertile pinnae, sterile pinnae, and sori were isolated in imageJ using the circle tool. The area outside of each segment was cleared, and the image was turned into an 8-bit gray-scale for processing. A threshold was set to extract the mask of each pinna or sorus, respectively. Total area, length, and perimeter were documented for each set of pinnae and soral masks. Area, perimeter, and length were then averaged for each species across the five total measurements per pinna and sorus, respectively. We were interested in relative measurements of fertile and sterile pinnae as well as soral size to pinna size across species. To quantify this, we measured the difference in fertile and sterile pinna area, perimeter, and length (Figure 2). We also quantified total pinna dissection as the perimeter divided by the area. Lastly, we calculated the soral area to lamina ratio by dividing soral area by fertile pinna area. All analyses were conducted in R (R Core Team., 2024).
Distinct types of reproductive leaf dimorphism in Blechnaceae + Onocleaceae. Monomrphic species produce isomorphic leaves with sporangia on the underside of leaves used for photosynthesis. Extra laminate dimorphic species produce visually distinct fertile fronds, with <50% fertile lamina covered by sori. Nonextra laminate dimorphic produce visually distinct fertile fronds with >50% of fertile lamina covered by sori.
Vouchered specimen (left) with measured leaf traits shown on the right, including the area and perimeter sterile pinnae, fertile pinnae, and sori. Below, a phylo-PCA illustrates the separation between monomorphic species (blue) and dimorphic species (yellow). Monomorphic species are primarily distinguished from dimorphic species based on their increased dissection of sterile pinnae relative to fertile pinnae and minimal differences between pinna area of sterile and fertile fronds. Dimorphic species tended to have increased dissection of their fertile pinnae and higher soral area to laminar area ratios.
Phylogenetic tree and time calibration
The phylogenetic tree used in our analysis was derived from de Gasper et al. (2017). It is often best practice to use chronograms—where branch lengths are in units of time—when estimating ancestral character states using a model-based approach (O’Meara, 2012; Suissa et al., 2024). As such, we generated a time-calibrated phylogeny using four secondary node calibrations. These points were generated from a recent time-calibrated fern phylogeny (Testo & Sundue, 2016) as well as Vicent et al., (2017). Calibrated nodes correspond to the base of Athyriaceae (87.1065–91.06 million years), the most recent common ancestor of Onocleaceae (78.1175–85.9663 million years), the most recent common ancestor of Onoclea and Mattueccia (56.0–66.0 million years), and the most recent common ancestor of Lomaridium (21.2–67.2 million years). We time-calibrated the phylogeny with these points using the chronos function in APE (Paradis et al., 2004). We fit a set of models to the data, including a relaxed, strict, and correlated clock model. Based on the information criterion outlined by Sanderson (2002), the relaxed clock model best fits the data. To time calibrate the phylogeny and generate a chronogram, we used a relaxed clock model with 10 rate categories and a smoothing parameter of 0. This model corresponds to the saturated (SAT) model (Sanderson, 2002), allowing rate heterogeneity across the tree (Paradis, 2013).
Phylogenetic PCA and K-means clustering
We used phylogenetic corrected principal component analysis (phylo-PCA) to determine the number of discrete forms of fertile-sterile leaf dimorphism across our species (Revell, 2009). Phylogenetic corrected PCA aims to remove correlations between measurements and to find the orthogonal axes of variation between the different reproductive forms while accounting for the shared ancestry of each species in the phylogeny. All quantitative measurements were included in a phylogenetic corrected PCA using the phyl.pca function in the Phytools package (Revell, 2023). Measurements were standardized to unit scale, and we used the lambda parameter to optimize values.
We then used K-means clustering to determine how points cluster in morphological space. To determine the optimal number of clusters for the dataset, we employed the K-means clustering algorithm and calculated the total within-cluster sum of squares (WSS) for varying numbers of clusters. First, we defined a function which computes the total within-cluster sum of squares for a specified number of clusters. The wrapper function utilizes the kmeans function from base R, and was applied to our dataset scores. The K-means algorithm was executed with 10 random starting points (nstart = 10) to ensure robustness and avoid local minima. We computed the WSS for a range of clusters from 1 to 15. We then applied the function to each value in this sequence using the map_dbl function, extracting the WSS for each cluster count. This range of WSS values was subsequently used to create an elbow plot, which aids in identifying the optimal number of clusters by visual inspection of the plot, specifically looking for a point where the rate of decrease sharply slows down (“elbow” point).
Discrete character coding
Using the quantitative character coding and phylogenetic PCA, we observed two major clusters in our dataset, which roughly correspond to monomorphic and dimorphic. We reexamined the species coded as dimorphic and observed that there were two morphological groupings within the category. These morphological groupings included those with sori covering the full lamina of fertile fronds and those where sori were not covering the full lamina. These were only present for roughly five species, which made quantitative analysis in the PCA difficult to accomplish. From these observations, we decided to split dimorphic into two states: extra laminate dimorphic (ELD) and nonextra laminate dimorphic (NELD) in order to ask whether ELD is an intermediate state in the evolution of NELD (Figure 1). ELD species produced fertile fronds that were visually distinct from the vegetative leaves, with sori on fertile fronds covering <50% of the lamina. In contrast, NELD species produced fertile fronds that were also visually distinct from the vegetative leaves but with sori on fertile fronds covering >50% of the lamina (in nearly every case fertile fronds of NELD species had essentially no laminar area). Percent of soral area was coded for each species by visually examining at least five herbarium fertile fronds per species (Figure 1).
Model fitting and stochastic character mapping
Under Dollo’s Law, we would expect a precursor model to be the best supported, that is, directional evolution with no (or limited) reversions. Moreover, we would expect rates of evolution to be lower in lineages with more specialized reproductive structures. To test these predictions, we used our discrete coding scheme and time-calibrated phylogeny. We fit a series of 11 models of character evolution using the corhmm function in the corHMM package (Beaulieu et al., 2017; Boyko & Beaulieu, 2021, 2022). These included transition matrixes with a single rate regime: equal rates, all rates different (ARD), symmetrical rates (SYM), and three ordered transition matrixes which force a state to move from monomorphic to NELD with ELD as a precursor state. We then accounted for rate heterogeneity across the tree by implementing models of character evolution with hidden rates using corHMM. We recognize that root node trait specification can have drastic impacts on determining trait evolution (Goldberg & Igić, 2008). As such, we specified the root node as monomorphic based on broader macroevolutionary data across all ferns (unpublished data) and given the outgroups of the Blechnaceae and Onocleaceae (Athyriaceae and Woodsiaceae) are monomorphic. We specified the root state using the pi flag in the corhmm function. We then tested the relative fitness of each model using the corrected Akaike Information Criterion (AICc).
To reduce potential bias from model misspecification, we quantified ancestral states by estimating model averaged marginal scaled likelihoods of states from each model based on their AIC weights. Using the best-fitting hidden rate model, we then used stochastic character mapping (Bollback, 2006) to simulate 1,000 realizations with a burnin of 100, specifying the root state to be monomorphic using the pi function in makeSimmap in Phytools. The resulting character maps were visualized using the plotSimmap function in Phytools. We summarized the results of the 1,000 stochastic character maps using the summary function in Phytools (Revell, 2024).
State-dependent diversification rates
We implemented character state speciation extinction (SSE) models to examine the potential impact of fertile-sterile leaf dimorphism on lineage diversification. Using the hisse function in the hisse package (Beaulieu & O’Meara, 2016), we tested a total of five models: a standard BiSSE model (not distinct from the original BiSSE analysis), a null BiSSE model, a character-independent diversification model with two (CID-2) or four (CID-4) hidden states, and a full HiSSE model with two hidden states. We compared the goodness of fit of each model using AICc. In order to incorporate uncertainty in model selection, we estimated model averages for all tips and nodes of the phylogeny from the five models using AIC weights. We compared net diversification rates between states by plotting model averaged rates by tip states.
Humidity-driven movement
For each species, we sampled three to four pinnae from five different herbarium specimens. Pinnae were placed in two distinct treatments: 100% relative humidity and 0% relative humidity. Given that these data were generated during Covid lockdown, we were unable to use temperature- and humidity-sensitive chambers. As such, we had to use simple home-made techniques. To simulate 100% humidity, pinnae were placed in zip-lock bags with wet paper towels and left for 1 hr. Humidity levels were tested using an Acurite wireless temperature and humidity monitor inside the zip-lock bags. Pinnae were removed and immediately imaged using a Bausch and Lomb 10X triplet hand lens mounted on an iPhone 8 (Supplementary Figure S1). To simulate 0% humidity, pinnae were placed in a standard kitchen oven at 100 °F for 1 hr and imaged using the same method. Previous experiments found that there are no transient effects of temperature on fertile fronds in Onoclea sensibilis (Suissa, 2021). As such, we were confident that changes in frond morphology were due to changes in humidity. For all 40 pinnae, 20 measurements were taken of pinnae openness, or the horizontal length from the costa to the outer tip of the pinnule (Supplementary Figure S1). The specimen segments of which photos were taken were chosen based on their ability to be accurately measured (i.e., the pinnule laid flat so angles did not inhibit the measurement, and measurements were taken of segments consistent in width to each other; Supplementary Figure S1). Data were then imported into R, and the length between costa and pinnule tip was compared for each pinnule within a single pinna. For each species we performed a t-test with Bonferroni correction to compare the means between pinnule length under each humidity treatment within species.
Results
Quantitative circumscription of reproductive leaf dimorphism
In total, we generated quantitative data from herbarium specimens of 66 species (Supplementary Table S1). The first three axes of the phylogenetic PCA accounted for 91.49% of the variance in reproductive leaf dimorphism across all species (Figure 2; Supplementary Figure S2). Increases in PC1 correlate most strongly with increases in area and perimeter differences between fertile and sterile pinnae and decreases in soral area to laminar area ratios (Supplementary Figure S2). Increases in PC2 correlate most strongly with decreases in differences in pinna length and perimeter between fertile and sterile leaves and decreases in soral area to laminar area ratios. K-means clustering of the phylogenetically corrected loadings suggested a total group of 2 (Supplementary Figures S3 and S4), recovering two major groups distinguished primarily by PC1 (Figure 2; Supplementary Figure S2). These include monomorphic and dimorphic broadly. In the PCA, monomorphic species are primarily distinguished from dimorphic species based on their increased dissection of sterile pinnae relative to fertile pinnae and minimal differences between pinna area of sterile and fertile fronds. Dimorphic species tended to have increased dissection of their fertile pinnae and higher soral area to laminar area ratios. Differences in fertile and sterile pinna perimeter and length do not seem to broadly separate clusters substantially along PC1. The loadings, including standard deviation, of the Phylo-PCA loadings were strongest among PC 1–3 (Table 1). The lambda correction for the Phylo-PCA was 0.23. Lambda acts as a scaling factor that adjusts the degree to which phylogenetic information is incorporated into the analysis. A lambda of 0 indicates that the analysis does not account for phylogenetic relationships, essentially treating all species as independent and reflecting low phylogenetic signal. In contrast, a lambda of one signifies a full incorporation of phylogenetic relationships, highlighting a high phylogenetic signal and indicating that species are analyzed with respect to their shared evolutionary history.
Phylo-PCA loadings of the top 6 PCs.
| PC axis | Loadings | Stdev |
|---|---|---|
| PC1 | 49.11 | ±1.72 |
| PC2 | 22.04 | ±1.15 |
| PC3 | 20.35 | ±1.11 |
| PC4 | 5.31 | ±0.56 |
| PC5 | 1.68 | ±0.32 |
| PC6 | 1.51 | ±0.30 |
| PC axis | Loadings | Stdev |
|---|---|---|
| PC1 | 49.11 | ±1.72 |
| PC2 | 22.04 | ±1.15 |
| PC3 | 20.35 | ±1.11 |
| PC4 | 5.31 | ±0.56 |
| PC5 | 1.68 | ±0.32 |
| PC6 | 1.51 | ±0.30 |
Phylo-PCA loadings of the top 6 PCs.
| PC axis | Loadings | Stdev |
|---|---|---|
| PC1 | 49.11 | ±1.72 |
| PC2 | 22.04 | ±1.15 |
| PC3 | 20.35 | ±1.11 |
| PC4 | 5.31 | ±0.56 |
| PC5 | 1.68 | ±0.32 |
| PC6 | 1.51 | ±0.30 |
| PC axis | Loadings | Stdev |
|---|---|---|
| PC1 | 49.11 | ±1.72 |
| PC2 | 22.04 | ±1.15 |
| PC3 | 20.35 | ±1.11 |
| PC4 | 5.31 | ±0.56 |
| PC5 | 1.68 | ±0.32 |
| PC6 | 1.51 | ±0.30 |
Models of character evolution
Using our quantitative framework, we coded fertile-sterile leaf dimorphism for a total of 118 species (Supplementary Table S2). We evaluated 11 models of character evolution, including seven single and four hidden rate models, using the corhmm function in the corHMM package. We excluded ordered precursor models with hidden rates given that they did not converge on optimal results based on unreasonably high transition rates that were close to the upper bounds of parameter space. The models were compared based on their log-likelihood (logL), Akaike Information Criterion (AIC), and the difference in AIC (delta.AIC) relative to the best fitting model (Table 2).
corHMM model parameters and fitness.
| Model | Rate category | log.l | df | AIC | delta.AIC | Weight |
|---|---|---|---|---|---|---|
| Restricted ARD/Restricted ARD | 2 | −51.02485 | 14 | 122.0497 | 0 | 8.34E-01 |
| SYM/SYM | 2 | −55.08468 | 18 | 126.1694 | 4.119663 | 1.06E-01 |
| ER/ER | 2 | −59.91327 | 18 | 127.8265 | 5.776837 | 4.64E-02 |
| ARD/ARD | 2 | −51.14233 | 18 | 130.2847 | 8.234957 | 1.36E-02 |
| Restricted ARD | 1 | −66.01664 | 4 | 140.0333 | 17.983584 | 1.04E-04 |
| Precurser-ARD | 1 | −66.75803 | 4 | 141.5161 | 19.466358 | 4.94E-05 |
| ARD | 1 | −66.01664 | 6 | 144.0333 | 21.983584 | 1.40E-05 |
| SYM | 1 | −71.26359 | 6 | 148.5272 | 26.477479 | 1.48E-06 |
| ER | 1 | −76.47775 | 6 | 154.9555 | 32.905792 | 5.96E-08 |
| Precurser-SYM | 1 | −80.79705 | 4 | 165.5941 | 43.544399 | 2.92E-10 |
| Precurser-ER | 1 | −99.64231 | 4 | 201.2846 | 79.23491 | 5.19E-18 |
| Model | Rate category | log.l | df | AIC | delta.AIC | Weight |
|---|---|---|---|---|---|---|
| Restricted ARD/Restricted ARD | 2 | −51.02485 | 14 | 122.0497 | 0 | 8.34E-01 |
| SYM/SYM | 2 | −55.08468 | 18 | 126.1694 | 4.119663 | 1.06E-01 |
| ER/ER | 2 | −59.91327 | 18 | 127.8265 | 5.776837 | 4.64E-02 |
| ARD/ARD | 2 | −51.14233 | 18 | 130.2847 | 8.234957 | 1.36E-02 |
| Restricted ARD | 1 | −66.01664 | 4 | 140.0333 | 17.983584 | 1.04E-04 |
| Precurser-ARD | 1 | −66.75803 | 4 | 141.5161 | 19.466358 | 4.94E-05 |
| ARD | 1 | −66.01664 | 6 | 144.0333 | 21.983584 | 1.40E-05 |
| SYM | 1 | −71.26359 | 6 | 148.5272 | 26.477479 | 1.48E-06 |
| ER | 1 | −76.47775 | 6 | 154.9555 | 32.905792 | 5.96E-08 |
| Precurser-SYM | 1 | −80.79705 | 4 | 165.5941 | 43.544399 | 2.92E-10 |
| Precurser-ER | 1 | −99.64231 | 4 | 201.2846 | 79.23491 | 5.19E-18 |
corHMM model parameters and fitness.
| Model | Rate category | log.l | df | AIC | delta.AIC | Weight |
|---|---|---|---|---|---|---|
| Restricted ARD/Restricted ARD | 2 | −51.02485 | 14 | 122.0497 | 0 | 8.34E-01 |
| SYM/SYM | 2 | −55.08468 | 18 | 126.1694 | 4.119663 | 1.06E-01 |
| ER/ER | 2 | −59.91327 | 18 | 127.8265 | 5.776837 | 4.64E-02 |
| ARD/ARD | 2 | −51.14233 | 18 | 130.2847 | 8.234957 | 1.36E-02 |
| Restricted ARD | 1 | −66.01664 | 4 | 140.0333 | 17.983584 | 1.04E-04 |
| Precurser-ARD | 1 | −66.75803 | 4 | 141.5161 | 19.466358 | 4.94E-05 |
| ARD | 1 | −66.01664 | 6 | 144.0333 | 21.983584 | 1.40E-05 |
| SYM | 1 | −71.26359 | 6 | 148.5272 | 26.477479 | 1.48E-06 |
| ER | 1 | −76.47775 | 6 | 154.9555 | 32.905792 | 5.96E-08 |
| Precurser-SYM | 1 | −80.79705 | 4 | 165.5941 | 43.544399 | 2.92E-10 |
| Precurser-ER | 1 | −99.64231 | 4 | 201.2846 | 79.23491 | 5.19E-18 |
| Model | Rate category | log.l | df | AIC | delta.AIC | Weight |
|---|---|---|---|---|---|---|
| Restricted ARD/Restricted ARD | 2 | −51.02485 | 14 | 122.0497 | 0 | 8.34E-01 |
| SYM/SYM | 2 | −55.08468 | 18 | 126.1694 | 4.119663 | 1.06E-01 |
| ER/ER | 2 | −59.91327 | 18 | 127.8265 | 5.776837 | 4.64E-02 |
| ARD/ARD | 2 | −51.14233 | 18 | 130.2847 | 8.234957 | 1.36E-02 |
| Restricted ARD | 1 | −66.01664 | 4 | 140.0333 | 17.983584 | 1.04E-04 |
| Precurser-ARD | 1 | −66.75803 | 4 | 141.5161 | 19.466358 | 4.94E-05 |
| ARD | 1 | −66.01664 | 6 | 144.0333 | 21.983584 | 1.40E-05 |
| SYM | 1 | −71.26359 | 6 | 148.5272 | 26.477479 | 1.48E-06 |
| ER | 1 | −76.47775 | 6 | 154.9555 | 32.905792 | 5.96E-08 |
| Precurser-SYM | 1 | −80.79705 | 4 | 165.5941 | 43.544399 | 2.92E-10 |
| Precurser-ER | 1 | −99.64231 | 4 | 201.2846 | 79.23491 | 5.19E-18 |
The best-fitting model included two rate categories and a restricted ARD transition matrix allowing transitions from monomorphism to extra laminate dimorphism as well as from monomorphism to nonextra laminate dimorphism, but not between extra laminate dimorphism and nonextra laminate dimorphism. These results indicate that the evolution of leaf dimorphism is best fit by a model that accounts for unordered and distinct transition rates between each state. Most of the phylogeny was in rate category 1 (slow rate), where most transitions occurred between monomorphic and ELD. In this rate category 1 transitions between monmorphic and NELD were low. In contrast, the base of Blechnum and the base of Neoblechnum plus Doodia were in the rate category 2 where most transitions between NELD to monomorphic occurred. The best-fitting model recovered transition rates between ELD and NELD to be 0 transitions per million years, suggesting that there are no direct observations of transitions between these two states in the Blechnaceae and Onocleaceae. In rate category 1, transition rates were highest between ELD and monomorphic, while in rate category 2, transition rates were highest between NELD and monomorphic (Figure 3).
Summary of 100 sampled stochastic character maps showing the transition in reproductive leaf dimorphism from monomorphism (blue), extra laminate dimorphic (green), and nonextra laminate dimorphic (yellow). Stochastic character mapping was used to estimate ancestral states. The probability of each ancestral state is represented by a pie chart, which depicts the model averaged marginal posterior probability from all tested models. Branches are painted based on their estimated character state summarized across 100 randomly selected posterior samples of character realizations along the tree. Colored boxes at the tips represent species-level character states. The best-fitting transition matrix is in the upper left depicting transitions in both rate categories. Arrows depict directionality of allowed transitions, and colors indicate the rate of transition, with dark red colors indicating fast and blue indicating slow rates. Family or subfamily denoted as bars along the phylogeny. Inset phylogeny depicts branches and tips in either rate category (blue: rate 1 and red: rate 2).
Evolutionary direction and rate of reproductive leaf dimorphism
Overall, the timing of cladogenic events across the tree fell within the confidence intervals of a time-calibrated phylogeny including the Blechnaceae and Onocleaceae (Schuettpelz & Pryer, 2009; Testo & Sundue, 2016; Testo et al., 2022; Vicent et al., 2017). Across all 1,000 simulations of the stochastic character map using the single rate ARD transition matrix, there was an average of 195.87 state changes across the tree. The majority of changes were between ELD to monomorphic (88.09 total transitions) and monomorphic to ELD (87.66 total changes). There were a total of 12.80 transitions between monomorphic to NELD and a total of 7.32 changes between NELD to monomorphic. No changes occurred between ELD to NELD or the reverse. The total time spent in each state varied across the tree. For 54.05% of the time, the states of the Blechnaceae and Onocleaceae were NELD. This was followed by monomorphic (38.36%) and ELD (7.59%).
The entire Onocleaceae was recovered as NELD with a root node state as likely NELD. The base of the Blechnaceae was reconstructed as most likely monomorphic, with 12 transitions to NELD across the family. The prominent transitions to NELD from monomorphic occurred at the base of Stenoclaenoideae and the base of Lomaridium, plus the rest of the Blechnoideae (Figure 3). There was a transition to NELD at the base of Struthiopteris and Oceanopteris, both from ancestrally monomorphic states. There are four major monomorphic clades within the Blechnaceae, including the Woodwardioideae, Sadleria, Blechnum ss., and Doodia. Sadleria and Woodwardia are likely ancestrally monomorphic, but Blechnum and Doodia were reconstructed as reversions from ancestrally NELD states. ELD occurs in five species, including Anchistea virginica, Salpichlaena volubilis, Telmatoblechnum serrulatum, Bainea insignis, and Doodia caudata, all representing independent origins of this character state.
State-dependent diversification rates
The best-fitting SSE model was a full hidden rate model (HiSSE), accounting for two rate categories and state-dependent diversification (Supplementary Table S3; Supplementary Figure S5). In this model, net turnover rates (speciation rate plus extinction rate) were allowed to vary according to the state and rate category. Turnover rates (speciation plus extinction) range from 0.1318 to 1.2247 across different states, while the extinction fraction (eps) is consistently 0.7499 for the hidden states. Transition rates between hidden states are relatively low, indicating rare transitions between states. There is an overall decoupling of states and diversification rates (Supplementary Figure S5). The violin plots showed a large overlap of model averaged net diversification values between leaf dimorphism, with mean values being nearly identical. The model averaged tip net diversification rates for monomorphism were 0.0583 and dimorphism was 0.0575. There are three major clades diversifying at fast rates, including Woodwardioideae (monomorphic), Blechnum/Austroblechnum (dimorphic and monomorphic), Doodia/Oceanopteris (dimorphic and monomorphic), and a subset of Parablechnum (monomorphic). The other major monomorphic clade (Sadleria) is evolving under a slow rate category. This suggests that there is likely an additional factor correlated with diversification that is decoupled from fertile-sterile leaf dimorphism.
Humidity-driven movement
All observed species in the Onocleaceae develop fertile pinnae that are hygromorphic (Figure 4). Based on a Student’s t-test, pinnule length was significantly different (p < 0.001) under 100% compared to 0% relative humidity within each species. The starkest difference in pinna length under the humidity treatments were Pentarhizidium orientale and Onoclea sensibilis (Figure 4), the latter previously observed (Suissa, 2021).
Boxplots depicting the length of costule-pinnule distance under 0% or 100% relative humidity. Significance tested based on a Student’s t-test. Each point represents a single pinnule. *** indicates significance where p < 0.001.
Discussion
Reproductive specialization is a key feature in the evolutionary history of plants (Gifford & Foster, 1989; Willis & McElwain, 2014). Given these trends and the ubiquity of reproductive leaf dimorphism in the seed plants, we initially anticipated that transitions from monomorphism to fertile-sterile leaf dimorphism in ferns would be unidirectional and irreversible, following Dollo’s Law (Bull & Charnov, 1985; Goldberg & Igić, 2008; Gould, 1970; Simpson, 1953). While this pattern of irreversibility finds support across major clades of land plants (Leslie et al., 2021), ferns do not follow a progressive sequence toward increasing reproductive complexity and specialization. We reveal that the evolution of reproductive leaf dimorphism in ferns is not a stepwise progressive increase in specialization. Instead, transitions can occur directly from monomorphism to dimorphism, and in the Blechnaceae we observe several instances of reversions from dimorphism back to monomorphism, indicating trait reversibility. However, this is not the whole story. In the Onocleaceae, once fertile fronds evolved further specialized traits (specifically humidity-driven movement for spore dispersal), we observed no reversions to monomorphism. This suggests that further functional specialization of dimorphic fronds—not only the transition to reproductive leaf dimorphism—may make it more difficult to revert to a monomorphic state.
The morphological circumscription of fertile-sterile leaf dimorphism
While commonly thought of as “simple” or “primitive” land plants, ferns have diverse reproductive strategies in terms of their fertile-sterile leaf dimorphism. Historically, ferns have been classified as monomorphic, holodimorphic, and hemidimorphic, relating to their division of labor and functional specialization of spore-bearing and photosynthetic leaves (Wagner & Wagner, 1977). This classification system has largely remained subjective, with limited exploration into the quantitative characters that underpin fertile-sterile leaf dimorphism in ferns.
Before discussing evolutionary patterns, we must first understand the traits that differentiate reproductive leaf strategies. By leveraging quantitative image analysis and phylogenetic principal components analysis, we found that there is, in fact, a quantitative morphological framework for broadly circumscribing species as monomorphic or dimorphic (Figure 2), bolstering historical categorization schemes (Wagner & Wagner, 1977). This quantitative approach to circumscribing states is a useful abstraction, analogous to other biologically complex traits including pollination syndromes in bell flowers (Lagomarsino et al., 2017) or feeding type in waterfowl (Li & Clarke, 2016). The major morphological traits that separate dimorphic from monomorphic species include dissection of fertile pinnae, the ratio of soral to laminar area in fertile fronds, and a difference in area between fertile pinnae and sterile pinnae. Specifically, monomorphic species tend to have nearly identical laminar area between fertile and sterile pinnae. In contrast, dimorphic species tend to have less dissected pinnae and a much higher ratio of soral area to pinna lamina—in many cases covering the entire fertile lamina (Figure 1; NELD).
While these leaf traits are similar to those observed in other dimorphic fronds such as Polybotrya (Moran, 1987), these are likely not all the traits that separate fertile-sterile leaf dimorphism from monomorphism in ferns. For instance, frond height, angle, phenology, and soral patterning may also be important distinctions (Moran, 1987). Our quantitative framework of circumscribing character states provides the context to understand how fine-scale developmental transitions of these traits can lead to the evolution of reproductive leaf specialization. While we do not find any evidence of hemidimorphic Blechnaceae or Onocleaceae, we do observe slight differences in some of the holodimorphic species. For instance, some species have sori covering less than 50% of the laminar area on their fertile fronds. We have termed these species extra laminate dimorphic (ELD). While we did not observe enough species in this state to parse out these taxa in a phylo-PCA, we coded this character to examine whether it acted as a precursor (Marazzi et al., 2012) in the evolution of nonextra laminate dimorphism (NELD; Figure 1).
Fertile-sterile leaf dimorphism is labile and evolutionarily reversible in the Blechnaceae
Under a refined Dollo’s Law, trait evolution is expected to be directional and stepwise, with limited reversions from specialized to generalized structures. However, our findings in the Blechnaceae indicate that the evolution of reproductive leaf dimorphism is highly labile, lacking a stepwise progression, and with many reversions from a specialized to a generalized state.
Given its “intermediate” morphology, we initially hypothesized that extra laminate dimorphism would be a precursor state (Marazzi et al., 2012). However, we did not find support for an ordered character transition model (Figure 3; Table 2). Even more strikingly, our best fitting transition models (both hidden and single rate) indicated no transitions from ELD to NELD. The best fitting transition models suggest that nonextra laminate dimorphism evolved independently several times within the Blechnaceae, including in the most recent common ancestor of Lomaridium and the rest of the Blechnoideae, Stenochlaena, Struthiopteris, and Oceanopteris, as well as several terminal tip-level transitions (Figure 3). In each of these cases, the ancestral state prior to this transition was recovered as monomorphic, meaning that transitions from monomorphism to nonextra laminate dimorphism can occur without an intermediate step between extra laminate dimorphism. We do not interpret this macroevolutionary pattern to imply that transitions from monomorphism to dimorphism occur instantaneously—this is of course biologically unlikely. Rather, we do suspect that an ELD state is in fact a developmental intermediate, but not a long-term evolutionarily stable state. For instance, within many dimorphic species, developmental aberrations can lead to intermediate leaf forms between NELD and ELD states, even within an individual (Supplementary Figure S6; Beitel et al., 1981). Notably, some of the morphological differences between leaf forms in these developmental mutants overlap with the traits that differentiate dimorphic from monomorphic species (Figure 2), including laminar area, pinna dissection, and soral to laminar ratio. While developmental teratologies are not necessarily indicators of evolutionary shifts, these observations provide a potential developmental hypothesis for understanding how nonextra laminate dimorphism can evolve from monomorphism (Supplementary Figure S1).
This developmental process does not necessitate that the transition from monomorphism to dimorphism was stepwise or gradual, as hypothesized under Dollo’s Law; quite the contrary. It suggests that several developmental shifts in laminar traits can facilitate the full transition from a monomorphic state to a holodimorphic state. These transitions can potentially occur without long-term stability in an ELD state (Figure 3). Evidence suggesting this evolutionary instability includes minimal observations of ELD across species in the phylogeny, high transition rates with nearly 80 transitions from ELD to monomorphism across the tree (Figure 3), and population-level observations suggesting that morphological intermediates showing ELD phenotypes (Supplementary Figure S6) are extremely rare relative to the state of most other individuals. While limited data are available to test this hypothesis, our observed evolutionary instability of ELD may result from lower potential fitness of taxa conforming to this state. For instance, ELD species have fertile lamina that is partially, but not entirely, covered in sori meaning fertile fronds are not photosynthesizing nor are they as fecund as they could potentially be if they were either monomorphic or NELD. Given that fertile fronds can be large carbon sinks (Britton & Watkins, 2016; Chiou et al., 2005; Watkins et al., 2016), it may not be optimal to develop a leaf that is partially photosynthesizing and partially reproductive in the way ELD species are (this logic may not extend to monomorphic or hemidimorphic species, however).
In terms of reversibility, within the Blechnaceae, we observe several independent reversions from dimorphism to monomorphism. These reversions occur at the base of Blechnum, Neoblechnum, Oceanopteris cartilaginea, and Doodia (Figure 3), corroborating reversion observed in other lineages, like Cyclodium (Moran & Labiak, 2015). These findings suggest that the evolution of specialized reproductive leaf dimorphism in this clade does not conform to irreversibility predicted by Dollo's Law. Notably, the clades with numerous reversions to monomorphism also exhibit high rates of morphological evolution (Figure 3), indicating that shifts in evolutionary rates may accompany these reversions, aligning with some expectations of rate changes under a modernized Dollo’s Law.
Reversal in reproductive specialization is an intriguing observation because fertile-sterile leaf dimorphism is, for the most part, an irreversible trait in the seed plants—meaning all seed plants produce distinct structures for reproduction and others for photosynthesis; none are monomorphic. There are a few functional exceptions to this rule, including epiphyllous flowers in Helwingia and Phyllonoma (Dickinson & Sattler, 1975; Stork, 1956) or photosynthetic flowers in Ambrosia trifida (Bazzaz & Carlson, 1979). However, even in these cases, the development of an ovule which houses the sporangium as well as the flower are still fully differentiated axes with modified reproductive leaf homologs, and this is distinct from producing spores directly on a leaf. In the gymnosperms, however, we do see intermediary analogs such as the more leaf-like fertile leaves in Cycas and development teratologies like the leaf-bearing ovules in Ginkgo biloba (Sakisaka, 1929; Soma, 1999). It is uncertain why Dollo’s Law has applied to fertile-sterile leaf dimorphism in seed plants; however, it likely has to do with the further evolution of adaptations reinforcing dimorphism. For instance, the evolution of heterospory and the integument at the base of seed plants surely relate to the irreversibility of dimorphism. As an analogy, the heterosporous ferns are all holodimorphic with no known reversions, and each fertile frond resembles a seed in most cases (Haig & Westoby, 1989). Moreover, within the angiosperms, the evolution of flowers and associations with animals likely added additional layers of reproductive complexity, making it difficult to transition from a highly specialized dimorphic state. While flowers are quite different from fern leaves, within the Blechnaceae at least, Dollo’s Law does not apply to fertile-sterile leaf dimorphism, and reversions to monomorphism have occurred several times.
Functional specialization of fertile fronds and its evolutionary implications
The evolution of specialized organs should facilitate the canalization of distinct functions where natural selection can act on each organ individually. Indeed, specialization is often facilitated by the intensification of one function of the ancestral organ—for instance, in the lungs of vertebrates (Liem, 1988). Under Dollo’s Law, this intensification should eventually lead to the irreversibility of specialized traits, with low rates of morphological evolution in specialized clades. While we find that the Onocleaceae have low rates of trait evolution, most other clades also have low rates of morphological evolution, regardless of reproductive state, creating difficulty in drawing broader conclusions on the relationship between rates and organ specialization (Figure 3). However, in contrast with the Blechnaceae, we do find the evolution of fertile-sterile leaf dimorphism within the Onocleaceae to be irreversible; that is, nonextra laminate dimorphism evolved a single time at the base of the Onocleaceae with no reversions.
Coinciding with the evolution of dimorphism is the observation that all examined species in the Onocleaceae (four of the total five) have further evolved humidity-driven movement of their fertile fronds, where fertile pinnae, pinnules, and segments open in dry air to facilitate the release of spores (Figure 4). Previous work on Onoclea sensibilis demonstrated that this movement is highly specialized and facilitates effective spore dispersal during the early spring, which is the most propitious time of dispersal given limited canopy cover and the longer growing season for gametophytes and subsequent sporophytes (Suissa, 2021). The anatomical structure facilitating humidity-driven movement in O. sensibilis is brought about by the differential thickening of cell walls and differences in the orientation of cellulose microfibrils in the cell walls of the pinnules—strikingly similar to structures in pine cones (Dawson et al., 1997), wheat awns (Elbaum et al., 2007), and Geranium fruits (Abraham & Elbaum, 2013). While we did not examine the anatomy of the cell walls of other members of the Onocleaceae, it is likely that movement of these fertile pinnae is governed by similar anatomical principles. Given that all species in the Onocleaceae have hygromorphic fertile pinnae, we interpret this as a complex reproductive synapomorphy of the group. Importantly, given that humidity-driven laminar movement cannot occur in monomorphic species, it is axiomatic that leaf dimorphism evolved first, which facilitated hygromorphic specialization.
We hypothesize that this functional specialization of fertile fronds may enhance the fitness of lineages in this state because they may increase relative spore output or more efficiently regulate the timing of spore dispersal (Suissa, 2021). Indeed, increased fitness of a state is one hypothesized pillar in the processes of irreversibility (Bull & Charnov, 1985). This form of hyper-specialization is also hypothesized to lead to trait irreversibility as occurred with specific feeding types in stem-group ants (Jouault et al., 2024), large North American mammalian predators (Van & Hertel, 1998), and in some capacity self-fertilization in flowering plants (Barrett, 2013; Igic & Busch, 2013). While phylogenetic evidence alone can be methodologically biased (Goldberg & Igić, 2008), the fossil evidence further indicates that Onoclea has been holodimorphic for nearly 70 million years (Rothwell & Stockey, 1991), suggesting that this state has been fixed and potentially irreversible for much of the evolutionary history of the Onocleaceae.
Conclusions
Within a hyper-diverse clade of ferns, we find two contrasting patterns in the evolution of fertile-sterile leaf dimorphism, which provide insight on the application of Dollo’s Law in the evolution of reproductive leaf specialization in plants. In the Blechnaceae, the evolution of reproductive leaf dimorphism is neither stepwise nor unidirectional, with multiple reversions from dimorphism to monomorphism. However, in the Onocleaceae, the further evolution of humidity-driven fertile leaf specialization has led to the irreversibility of dimorphism. The irreversibility of fertile leaf dimorphism in the Onocleaceae may have been facilitated by additional adaptations, increased fitness, and the long persistence in a specialized state (Rothwell & Stockey, 1991). In this way, reversions of the primary state are even more difficult because multiple different states must change over time, each with a seemingly different fitness threshold. The canalization of fertile-sterile leaf dimorphism in Onocleaceae may be analogous to evolutionary processes observed in seed plants. For instance, the continued evolution of heterospory, integuments, cones, flowers, and fruits likely compounded on top of the dimorphic habit, further enhancing specialization and making it statistically improbable for reversals to occur. Our findings in these ferns provide broader insights into Dollo’s Law, suggesting that it is not merely the evolution of a single complex trait that is evolutionarily irreversible, but rather the accumulation of multiple specialized traits that interact and reinforce each other, which makes reversion to a prior state evolutionarily challenging or statistically improbable.
Supplementary material
Supplementary material is available online at Evolution.
Data availability
All data and associated code are publicly available at GitHub (https://github.com/Suissajacob/blechnaceaeDimorphism) and Zenodo (https://zenodo.org/records/14017651).
Author contributions
JSS designed the project, analyzed the data, and wrote the manuscript; MS collected data, analyzed data, and provided feedback on the manuscript.
Funding
MS was funded by the Harvard University Department of Organismic and Evolutionary Biology Research Experiences for Undergraduates program funded by The National Science Foundation (1757780).
Conflict of interest: The authors declare no conflict of interest.
Acknowledgments
The manuscript has been greatly improved by suggestions from Christina Caruso and two anonymous reviewers. We thank Corey Reese for assistance with quantitative image analysis. We also thank James E. Watkins, Li- Yaung Kuo, Pei-Jun Xie, Michael Kessler, and Daniela Francisca Aros Mualin for helpful discussion on fertile-sterile leaf dimorphism. We acknowledge Robbin Moran for his helpful comments and suggestions to the manuscript. We also want to recognize that this work would not be possible without the funding, support, and effort of herbarium staff, plant collectors, and taxonomists around the world who have made the use of digitized herbarium specimens possible.
