Genera are an example of a higher taxon whose circumscription is manmade (Clayton 1983, Entwisle and Weston 2005). Therefore, the criteria for generic distinction are arbitrary. Several criteria have been used in the past to define them. One is the ‘kind’, that genera are groups of organisms that are clearly distinct from other genera (Stevens 1985). Another is that genera are groups of reciprocally monophyletic species assemblages (e.g., Garnock-Jones et al. 2007). Over time there have also been trends in ‘lumping’ and ‘splitting’ species assemblages (Humphreys and Linder 2009) with often no stable consensus reached, because of the arbitrary nature of genera. With the advent of molecular phylogenies the circumscription of genera has become more complicated. It seems that a criterion that is more commonly used in algae over the past few years, is that a genus can be designated based on any ‘supported’ phylogenetic clade (whether reciprocal monophyly is found or not, nor whether there are clear diagnostic characters). In a bifurcating tree, the number of supported branches will increase with the number of taxa used and the number of genes sequenced (more data), practically ad infinitum, leading to more and more splitting of groups. This trend in red algal taxonomy to name genera based on this minimal criterion, especially with the advent of molecular-assisted systematics, is reflected in the increase in genera. For example, Schneider and Wynne’s “synoptic review of red algal genera” series (Schneider and Wynne 2007, 2013, 2019, Wynne and Schneider 2010, 2016, 2022) found a steady increase, and possibly rate, of new genera proposed since the mid-1950s (21 new genera in 2010, 27 in 2013, 40 in 2016, 58 in 2018, and 31 in 2022) (Kylin, 1956). While this could reflect a better understanding of the diversity of higher taxa in red algae, there is also a component of splitting based on a supported lineage, often nested in an unsupported grade.

A good example of this back and forth of generic nomenclature is the red algal family Gracilariaceae (order Gracilariales), another commercially important group, that has had a complex nomenclatural history recently due to generic circumscriptions being based largely on molecular phylogenies and, correctly, attempts at finding diagnostic characters. The genus has been split into several new genera (most recently in Gurgel et al. 2018) based on diagnostic morphological differences and molecularly supported clades. It has been shown though that the diagnostic morphological characters do not correspond to the new genera, and all the recently recognized new genera have been placed back into the original genus Gracilaria (Lyra et al. 2021). This has led to a genus that is more easily distinguishable, has a set of exclusive characters, is genetically quite divergent, and is reciprocally monophyletic. This seems to be a fair approach.

In this study, we tried to reinvestigated the data presented in Yang et al. (2020) using additional and alternative phylogenetic methods, support measures, as well as careful selection of molecular models and partitioning of genes and codons to see if the splitting of the genus Pyropia into several genera is supported by more comprehensive phylogenetic analyses.

A criterion for generic distinction that is used in studies that employ molecular data, is that at a minimum, phylogenies should provide strong support on a branch of either a distinct species or set of species. Sometimes this can be supported by diagnostic characters, even cryptic ones (Díaz-Tapia et al. 2017). These phylogenies and support values should be based on more than one gene, and various methods of tree reconstruction. While maximum parsimony and distance methods have lost favor in tree reconstruction, methods that incorporate realistic models of DNA (or amino acid) evolution and probabilistic estimates of tree topologies are mostly used. Maximum-likelihood (ML) methods, in programs such as RAxML (Stamatakis 2014) or IQ-TREE (Minh et al. 2020), and Bayesian reconstruction methods are the most common. For support values, the non-parametric bootstrap (Felsenstein 1985) has a long history and for Bayesian analysis, posterior probabilities as implemented in programs like MrBayes (Ronquist et al. 2012) are common. These probabilistic methods require an estimate of molecular evolution (a model or models) to reach a maximum or high probability topology. It is an empirical fact that mutation frequencies accumulate differentially between genes (more conserved versus less so), within and between regions of genes (e.g., codon positions in protein-coding genes), and even gene location (nuclear versus organellar). These variable changes require a careful partitioning of the data to arrive at models that more accurately reflect these variations in mutations. Different models can be estimated for these different genes and partitions using various model testing programs (Santorum et al. 2014, Kalyaanamoorthy et al. 2017).

Sutherland et al. (2011) used a concatenated data set of the plastid rbcL (the large subunit of ribulose bisphosphate carboxylase / oxygenase) and nrSSU (nuclear-encoded small subunit ribosomal RNA) genes in their comprehensive analysis of the diversity of the Bangiales. They used one molecular model for the entire data set, even though they partitioned the data by genes, and implemented both ML and Bayesian reconstruction. Results showed that a strongly supported group of bladed Bangiales was sister to filamentous Bangiales (‘Bangia’ 3), although this sister relationship was not supported using three support methods. They resurrected the genus Pyropia J. Agardh (1899) for this clade. Pyropia contains the most commercially important Bangiales (e.g., P. haitanensis, P. tenera, P. yezoensis). There were other supported groups, with various levels of branch support within this genus but they did not believe this warranted further splitting. This work was recently continued by Yang et al. (2020) with the same data set (rbcL and nrSSU), plus the addition of some new species and they made the decision to further split Pyropia. In their analysis, they used one molecular evolution model (GTR + I + G) for all gene partitions and implemented only one method of tree reconstruction (Bayesian analysis).

Bayesian support values (as posterior probabilities) are known to over-estimate ‘confidence’ of branches (Suzuki et al. 2002, Simmons et al. 2004), with potentially only high values (≥0.95) predicting true branches. Plus all measures of tree reconstruction and support measures are differently sensitive to model violations, priors selected, etc. (Douady et al. 2003, Simon 2022). Yang et al. (2020) did not implement any other methods. Even so, their support for the new generic clades was variable and posterior probabilities were: Neopyropia values were 0.92, Neoporphyra values were 1.00; Pyropia values were 0.90; Calidia (an illegitimate name now Phycocalidia – Santiañez and Wynne 2020) values were 0.99; Porphyrella gardneri was a singleton sister to the previous taxon but in a polytomy (unsupported sister relationship – 0.60); Uedaea onoi was also a singleton, sister to the reminder of the taxa but only with support of 0.93. Many of these values do not indicate high confidence in the genera (even though it was stated in the paper that they all had ‘strong support’). They also argue that there are some biogeographic trends, in the distribution of genera. For example, the new genus Neopyropia is from cold temperate zones, whereas Pyropia is ‘mainly’ subpolar to cold temperate. The distinction is hard to fathoms, especially with patterns of sampling (several well sampled areas, many poorly sampled), and the difficulty (impossibility) of species identification in new locations making distribution patterns very tentative. Thus, these new generic assignments are solely based on support values in molecular phylogenies. This practice is becoming more common in algae with simple morphologies (Verbruggen 2014), like the Bangiales that lack complex thallus structure or complex post-fertilization patterns. Making decisions of nomenclatural changes, especially in groups that are intensively studied or commercially important, needs a more rigorous analysis and clearer and unambiguous support for branches using all available methods—at least more than one.

While nomenclature should be governed by the best science available, the consequences of changes should also be considered. Most taxonomic changes have no detrimental consequences and are never further remarked on. But in an alga that is commercially exploited and for which regulations and contracts are in place, a taxonomic change, at least leads to a plethora of legal contracts and government regulatory laws needing to be amended.

We reinvestigated the same rbcL and nrSSU data presented in Yang et al. (2020) using additional phylogenetic methods, support measures, careful selection of molecular models and partitioning of genes and codons to see if the splitting of the genus Pyropia was supported by more comprehensive analyses.

We used two reconstruction methods: ML analyses (not used in Yang et al. 2020) were implemented using IQ-TREE 2.2.0 (Minh et al. 2020) and Bayesian inference analysis with MrBayes v. 3.2 (Ronquist and Huelsenbeck 2003). Alignments were partitioned by gene and, for rbcL, by codon. Model selection (Kalyaanamoorthy et al. 2017) and partitioning to select the best partitions (Chernomor et al. 2016) were implemented in IQ-TREE, potential merging of partitions was also implemented. Three ML branch support methods were used: non-parametric bootstrapping (500 replicates) (Felsenstein 1985); Ultrafast bootstrapping (UF) (3,000 replicates) (Hoang et al. 2018) and the approximate likelihood ratio test (SH-aLRT) (3,000 replicates) (Guindon et al. 2010). These support methods provide different approaches in determining branch reliability and have different confidence thresholds in determining branch support. Bayesian inference analysis also partitioned genes and codon positions. The model partitions were unlinked, with variable rates, and six rate categories (Supplementary Fig. S1). Two parallel runs of Markov chain Monte Carlo were performed for 3,000,000 generations, sampling every 1,000 generations. Estimated samples size, split frequencies, and stationarity were checked after each run. Post-analysis, 10% of generations were removed as a burn-in and posterior probabilities visualized in Figtree v1.1.4 (Rambaut 2009) and Canvas X Draw (Canvas GFX, Inc., Boston, MA, USA) were used to manipulate trees (e.g., collapse branches) for presentation.

Different evolutionary models were selected, for the different partitions, reflecting their different patterns of change (Supplementary Fig. S1). The bladed red algae, of the previous genus Pyropia comprise a strongly supported clade, but the new genera have variable support (Fig. 1, Supplementary Fig. S2). While Phycocalidia (88% bootstrap [BS] / 1.0 posterior probability [PP] / 96% approximate likelihood-ratio test [aLRT] / 97% UF) and Neoporphyra (100% BS / 1.0 PP / 100% aLRT / 100% UF), plus the singleton Porphyrella gardneri have strong support with all methods, the other genera showed less support. Neopyropia was supported only by PP analyses (as in Yang et al. 2020) and with aLRT, but not with BS or UF (50% BS / 1.0 PP / 90% aLRT / 78% UF). The newly circumscribed Pyropia is not supported at all (49% BS / 0.81 PP / 72% aLRT / 68% UF), again highlighting its unsupported status in Yang et al. (2020) (0.90 PP). The relationship between the genera was also often unsupported. The clade of Phycocalidia and Porphyrella is unsupported (Fig. 1), rendering reciprocal monophyly of the genera as unsupported, the sister relationship of Neopyropia to the two previous genera is also not well supported with all methods (56% BS / 1.0 PP / 91% aLRT / 78% UF). The clade containing Neoporphyra and Pyropia is supported (90% BS / 1.0 PP / 97% aLRT / 97% UF).

To determine if the support values changes with outgroup selection, we used the “Bangia 3” sequences of Sutherland et al. (2011), suggested as a sister group to their Pyropia. The results did not vary much (Supplementary Fig. S2), with support of genera slightly lower (e.g., Neopyropia (44% BS / 0.98 PP / 89% aLRT / 63% UF) and Pyropia (68% BS / 0.97 PP / 95% aLRT / 88% UF) and relationships between genera also not well supported. For example, the clade of Phycocalidia and Porphyrella (41% BS / 0.84 PP / 71% aLRT / 61% UF) and this clade with Neopyropia (59% BS / 1.0 PP / 94% aLRT / 85% UF) are not supported with all methods. In all analyses the clade of Pyropia as defined by Sutherland et al. (2011) is fully supported. Examining the relationship within this phylogeny in detail (Supplementary Fig. S3) also shows that there are many supported clades (only using BS and PP) within all these proposed new genera, some supported by bootstrap analysis, some by posterior probabilities, some by both. Should they be considered future genera?

Based on our findings we propose a circumscription of the genus Pyropia sensu Sutherland et al. (2011).

Pyropia J. Agardh 1899: 149–153; following Sutherland et al. (2011).