We selected utricles from 11 species (Fig. 3; Suppl. material 1): all four of C. phalaroides gr. (representatively covering its entire known morphological and geographical variation), and six from sect. Abditispicae (80% of the known species; Roalson et al. 2021). Due to the scarcity of sect. Abditispicae herbarium collections (and in particular of specimens bearing ripe utricles), we had to limit our sampling to the few mature specimens we located. After studying in situ the full collections of K, QCA, QCNE, and UPOS herbaria, we only managed to obtain 15 utricles: two from C. acaulis; three from C. collumanthus; one of C. humahuacaensis; two of C. macrosolen, four from C. ruthsatzae, and three of C. subantarctica. For C. phalaroides gr. we managed to obtain 32 utricle images: 14 of C. gibertii, three of C. moesta, three of C. paraguayensis, and 12 of C. phalaroides (Suppl. material 1). Detailed pictures of the utricles were taken with a Nikon stereoscopic microscope (Nikon SMZ745) and an Olympus stereoscopic microscope (Olympus SZX16). We compiled a set of 49 2D-scaled images.

Figure 3. 

Representative utricles images used for GM analyses. From the top to the bottom, left to right A problematic species: Carex herteri (Herter, W.G.F., 19091, S), C. hypsipedos (Weberbauer 2617, G) B C. phalaroides gr.: C. gibertii (Arechavaleta s.n., US), C. paraguayensis (L. Pereira-Silva 350, FLOR) and C. phalaroides (G. Rodríguez-Palacios 23GERP15, UPOS) C C. sect. Abditispicae: C. acaulis (DM Moore 1240, K), C. collumanthus (PJ Grubb 339, K), C. humahuacaensis (S. Martín-Bravo et al., 178SMB21, UPOS), C. ruthsatzae (G. Rodríguez-Palacios 46GERP15, UPOS), C. subantarctica (Marcia Waterway, MW2015.020, UPOS), and C. macrosolen (S. Martín-Bravo et al., 11SMB10, UPOS). Scale bar: 4 mm.

Nine fixed landmarks were placed on homologous points within utricles following Jiménez-Mejías and Martinetto (2013) approach to the carpological features of Carex. All these fixed landmarks were limited to the beak and utricle base, as homologous structures between species (Suppl. material 2: fig. S1). In addition, to collect as much shape information as possible from the utricle body, we placed eight semi-landmarks equidistantly, on non-homologous points of the utricle body margins. These defined two curves on the right and left side of the utricle, from the base of the beak to the beginning of the substipitate base. Landmarks were digitalized, on images using TPSDig2 (Rohlf 2015) and datasets were created in a .tps format. All fix and sliding landmarks were positioned in the same order in all images.

Main GM analysis was conducted with a complete dataset including all 49 utricle images (from hereinafter referred as “complete GM dataset”). We split this dataset into sect. Abditispicae species (15 utricle images) and C. phalaroides gr. species (32 utricle images), creating Abditispicae GM dataset and C. phalaroides GM dataset two and three, respectively. Datasets two and three excluded problematic species thus we obtained the consensus utricle configuration for each group to allow visual shape comparison with the problematic species. GM analysis was performed four times, first for identifying potential outliers through a PCA scatter plot, second for the complete GM dataset, and third and fourth for shape exploration of sect. Abditispicae and C. phalaroides gr. datasets, respectively. The GM analysis of the complete dataset revealed that the highly deviant utricle of C. macrosolen induced a strong bias to the analysis generating a substantial deviation to the PCA, due to its large peak size (Suppl. material 2: fig. S2), thus, this was removed from all subsequent GM analyses, leaving the complete dataset with only 47 utricle images.

We proceeded to landmark analysis with geomorph, R package v.4.0.2 (Adams et al. 2021). Semilandmarks were set as sliding points with the geomorph::define.sliders function. We subsequently performed generalized Procrustes analysis (GPA) using the minimized squared distances method. GPA calculates the consensus configuration of the dataset, along with its shape variation, and makes a separation within size and shape components of the datasets (Viscosi and Cardini 2012). It is here used as a superimposition method because it has been shown to be accurate in sample means estimation (Rohlf 2000a, b, 2003). GPA was performed with the geomorph::gpagen function and a maximum of 1000 iterations. The studied utricle morphospace was visualized by performing a principal component analysis (PCA) only for the complete dataset, and the shape variation within the morphospace was depicted with geomorph::picknplot.shape function. Subsequently, a Procrustes multivariate analysis of variances MANOVA was carried out with the geomorph::procD.lm function to assess statistical patterns of shape variation for a set of Procrustes aligned coordinates between the considered groups (Adams et al. 2021). Accounting for the small size of our dataset, the significance of shape variation between the two major groups was performed against a null model generated by permuted resampling, which uses a residual randomization permutation of 999 replicates (Collyer et al 2015; Renner et al 2018). The mean shape of all sampled utricles for every dataset was visualized with geomorph::mshape and geomorph::plotRefToTarget functions. To achieve an objective attribution of the problematic species to either sect. Abditispicae or C. phalaroides gr. we performed a Discriminant Function Analysis (DFA), more precisely the Linear Discriminant Analysis (LDA) with MASS R package (Venables and Ripley 2002). For this analysis we set a train dataset only with sect. Abditispicae and C. phalaroides gr. species and prepared two tests, one for C. herteri and another for C. hypsipedos. These datasets were created from coordinates calculated in the GPA. We checked the coordinates of C. herteri and another for C. hypsipedos correctly and incorrectly assigned to each of the groups of train dataset through confusion matrices.

Characters to be measured for C. phalaroides gr. initially followed the review of Carex in Rio Grande do Sul (Hoff-Silveira and Longhi-Wagner, 2012) with slight modifications according to our own observations. The different structures to be measured were selected from the middle zone of each organ, aiming for homogeneity within the data and to facilitate the comparison between individuals, with the following exceptions (1) the bracts, in which we selected the upper spike bract and the lower spike bract and (2) the length and width of the leaves, for which the longest and the widest leaves of each individual were selected, respectively. A list of 38 potentially diagnostic characters was established, although exploratory PCA analyses retrieved that only 24 characters were diagnostic. Our final dataset was constituted by 24 morphological characters measured on 56 individuals (Suppl. material 3).

Sampling of sect. Abditispicae relied on literature data in order to take into account the entire variation span known for each taxon (Suppl. material 3) after contrasting that our own measurements felt within the reported variation intervals.

Principal component analysis was carried out on Rstudio v. 1.4.1717 (R Core Team 2022) using 11 morphological variables: two vegetative characters and the rest reproductive-related characters (Table 1) as these were the only characters available for every taxa. For every character in all the considered taxa of sect. Abditispicae we included three independent data as detailed on the corresponding taxonomic description (Suppl. material 3): maximum, minimum and the mean. This way we ensured the consideration of the maximum possible span of each species within the morphospace.

The significance of those characters that allowed the best separation of each of the two morphogroups (C. phalaroides gr. and sect. Abditispicae; see results) was evaluated by non-parametric Kruskal–Wallis test with Rstudio v. 1.4.1717 (R Core Team 2022). After discarding that the dataset had a normal distribution, we employed Kruskal-Wallis test as a non-parametric alternative to ANOVA test. As visual support for the test, we performed violin graphic plots to present the comparison between the problem species and the morphogroups we tested them against.

Procrustes analyses performed for the different datasets recovered the consensus utricle configurations and deviations for every landmark and semilandmark coordinate (Suppl. material 2: fig. S3). The low utricle shape affinities of the two tested groups were compared and shown at Suppl. material 2: fig. S4. Subsequently, MANOVA results show shape variation is significantly different between the two considered groups (Table 2A) denoting that these are well delimited and can be successfully differentiated using the proposed configuration of landmarks and semilandmarks. For the PCA of the complete dataset, the first three principal components accumulated up to 81.2% of the variance (54.3%, 17%, 10%). PCA scatter-plot for the complete dataset displays the position of the problematic species, C. herteri and C. hypsipedos, within the morphospace (Fig. 4), illustrating a higher proximity to the sect. Abditispicae cluster than to C. phalaroides gr one. In the case of C. herteri, its affinity was much clearer than for C. hypsipedos. LDA model only obtained one linear discriminant and prior probabilities for the main groups were 0.347 for sect. Abditispicae and 0.653 for C. phalaroides gr. The model obtained a 0.959 of accuracy meaning 95.9% of samples were correctly classified (Table 2 B), only four coordinate samples were misclassified. When testing C. herteri dataset in the trained model, both of its coordinates were placed within sect. Abditispicae with an accuracy = 1, while C. hypsipedos shape information was not clearly positioned in either of the two groups, so the accuracy of the model in this case was only 0.5. Visually, the four graphs comparing the consensus configurations of the sect. Abditispicae GM and C. phalaroides gr. GM datasets with the problematic species shapes (Fig. 5) assist the results revealed by PCA and DFA analyses of higher shape affinity of C. herteri with sect. Abditispicae than with C. phalaroides gr., while shape resemblance of C. hypsipedos remained uncertain.

Figure 4. 

PCA scatter-plot of the geometric morphometric analysis excluding Carex macrosolen. Squares represent sect. Abditispicae taxa, triangles represent C. phalaroides gr. taxa, and circles represent C. herteri and C. hypsipedos taxa. Utricles shapes at the margins of the graph display the extreme shapes of the morphospace for a better visualization of the utricle morphological features with greater weight within the principal components.

Figure 5. 

Comparative figures of consensus utricle shapes of both Carex sect. Abditispicae (triangles and blue line) and C. phalaroides gr. (triangles and green line) with C. herteri (circles and orange line) and C. hypsipedos (circles and purple line).

PCA performed to assess C. herteri and C. hypsipedos proximity to major groups included all the available morphological characters (Table 1), since these already allowed the best separations of morphogroups. We only retained principal components with eigenvalues>1. The first two principal components accumulated the 65.9% of the variance and the 75.5% on the first three: PC1 43.1%; PC2 22.8%; PC3 9.7%. PCA values for every character are shown at Table 3. PCA scatter-plot PC1–PC2 (Fig. 6) revealed the existence of two separated clusters, one for sect. Abditispicae and another for C. phalaroides gr. The problematic species were placed close to the first one, indicating morphological similarities for the analyzed characters. The morphological characters which contributed the most to the first principal components were leaf length, utricle length and width and achene length and width, while to the second component important characters were the lower spike bract width, leaf width, spike width and inflorescence length (Table 3). For the Kruskal–Wallis test, eight of the eleven analyzed characters obtained a significant p–value < 0.05 (Table 3) . Only two of them were vegetative characters (leaf and lower spike bract lengths) while the remaining were reproductive-related characters. Most significant p-values were scored by leaf length (3.513e-11), utricle width (4.062e-09), utricle length (1.987e-09), achene length (2.935e-08) and achene width (1.218e-07). Violin plots at Fig. 7 illustrate the distribution differences of the measurement dataset supporting the use of non–parametric tests to assess mean differences between groups.

Figure 6. 

PCA scatter-plot of the traditional morphometric analysis. Squares represents sect. Abditispicae taxa, circles represent C. herteri and C. hypsipedos, and triangles represents C. phalaroides gr. taxa.

Figure 7. 

Violin plots; Violin plots illustrating distribution and mean differences for the analyzed characters with traditional morphometrics for the problematic species, C. phalaroides gr. and sect. Abditispicae.

Dwarf species Carex herteri and C. hypsipedos were assigned to C. phalaroides gr. by Wheeler (1996) and Poindexter et al. (2017) respectively. This designation was based on morphological affinities not tested under statistical approaches. Given the morphological complexity of the group, its adscription was in need of a revision. Carex sect. Abditispicae was a major candidate to incorporate these species as these concur on distribution and morphological characteristics (Wheeler 1987, 1989, 2002).

Our GM, DFA and traditional morphometrics results reveal a high statistical support and a close utricle shape resemblance among sect. Abditispicae and C. herteri for all the analyses performed (Figs 2–4), therefore this species may be better considered as part of this section based on its morphological features. Contrastingly, C. hypsipedos does not display evident statistical affinities, or shape resemblance with either sect. Abditispicae or C. phalaroides gr. so its affiliation persists unsolved, though it can be excluded as a member of the C. phalaroides gr.

Nevertheless, it would be desirable confirmation from procedures such as DNA barcode for two main reasons: (1) The frequent morphological homoplasy that affects the delimitation of infrageneric units within the genus (Jiménez-Mejías et al. 2016a), exacerbated in this particular case by the extreme reduction of such plants (dwarfism, see Jiménez-Mejías et al. 2021), that might further confound their morphological affinities; and (2) The extremely low sampling size of the problematic species (known only from their type collections), thus perhaps no representative of the entire species variation. Ripe utricle morphology is usually fairly constant, as supported by its recurrent use in identification keys (e.g. see keys in Egorova 1999 or Ball and Reznicek 2002). However, variation in utricle size, and also moderately in shape, is also known in Carex (Jiménez-Mejías et al. 2017, 2018). Accordingly, we cannot entirely rule out that the included problematic taxa samples were outliers and so could be somehow biasing the inferred affinities of the two problematic taxa, although it would be certainly unexpected because of the consistency in shape variation within each of the detected groups. Dwarfism, acaulescency, and character reduction should be considered the principal cause of the deficiency of herbarium collections and the absence of field citations of these two problematic species, due to their inconspicuousness. Due to the impossibility of performing a destructive sampling on the already poor type collections from which C. herteri and C. hypsipedos were described, the knowledge of these two taxa would benefit from a focused fieldwork sampling.

The adscription of C. herteri to sect. Abditispicae would imply an area extension of a thousand km from the Patagonian steppes and high mountainous Andean habitats of the section to the Uruguayan Pampa. This, in turn, implies a much wider ecology for the group, from the cold-dry steppes and high-altitude habitats of the known species to the warm dryness of the pampa. As a common factor, Carex sect. Abditispicae ecology seems to be linked to stressful environments and may behave as pioneers in colonization processes.

Our approach using GM has assessed fruit shape variation in a non–qualitative way, as it is commonly studied on traditional morphometrics (Chen et al. 2018). Some examples of systematic and taxonomic implications derived from GM have been previously done in Liu et al. (2018) with Chinese oaks leaves, Terral et al. (2012) with the seeds of Phoenix genera species, or Van der Niet et al. (2010) assessing flower shape variation. These studies agreed in the application of GM as a useful approach for providing detailed information on the morphological variation of the plant structures with taxonomic value. In addition, research on plant organ shapes and its relationship with other organisms or environmental factors might shed additional light on other fields such biogeography, ecology, and genetics, as we also do when we assign C. herteri to sect. Abditispicae.

Our study supports the utility of GM on testing systematic affinities in species with graminoid morphology, particularly for Cyperaceae. To this end we used carpological characters, which have been often ignored in sedges, despite the useful characters for group delimitation residing in such organs (Jiménez-Mejías and Martinetto 2013). Successful differentiation between and within complicated groups such as sect. Abditispicae and C. phalaroides employing utricle shape sets a landmark for future taxonomic studies in a genus where its general morphology is typically affected by homoplastic processes.

No conflict of interest was declared.

No ethical statement was reported.

Funding received from the Regional Government of Madrid, Spain (Macondo, Award# SI1/PJI/2019-00333 to P.J.-M.), Spanish Research Agency of the Ministry of Science and Innovation (DANZ, Award# PID2020-113897GB-I00 to P.J.-M.), and Kew Royal Botanic Gardens, United Kingdom (Bentham-Moxon Trust grant 2021).

A. M.-A., T.V., P. J.-M. contributed to study conceptualization. Data preparation was carried out by A. M.-A., P. J.-M. Morphological study was carried out by A. M.-A. Analyses were carried out by A. M.-A. First manuscript draft was written by A. M.-A., P. J.-M. All authors contributed to the following and final version writing, review and editing. All authors have read and agreed to the published version of the manuscript.

Ana Morales-Alonso https://orcid.org/0000-0003-1497-1070

Tamara Villaverde https://orcid.org/0000-0002-9236-8616

Pedro Jiménez-Mejías https://orcid.org/0000-0003-2815-4477

All of the data that support the findings of this study are available in the main text or Supplementary Information.