Using geometric morphometrics to disentangle Didymosphenia hullii and D. geminata (Bacillariophyceae) from Connecticut, USA, and two congeneric relatives

ABSTRACT Two morphologically similar populations of Didymosphenia were reported from the West Branch of the Farmington River, Connecticut, USA in 2014 and 2016. We described one of them as a new species, D. hullii. The other was observed subsequently in late 2016 and resembled D. geminata, but given the worldwide nuisance characteristic of this species, the identification required confirmation. In this work we used geometric morphometrics analysis to test for quantitative distinctions between the two Connecticut populations, along with two morphologically similar taxa from the literature, D. laticollis and D. pumila. We successfully separated the four entities and confirmed the identity of the second blooming diatom in the Farmington River, Connecticut, as D. geminata, which is the first confirmed report of this species for the state. We conclude that geometric morphometrics, in combination with features viewed with scanning electron microscopy and light microscopy, enhances the ability to distinguish among these morphologically similar species of Didymosphenia. We summarize these findings and pinpoint morphological distinctions that can be used to separate the taxa during routine light microscopy analyses. HIGHLIGHTS Geometric morphometrics was used to differentiate four Didymosphenia taxa. Morphological comparisons of D. hullii and D. geminata are shown. Didymosphenia geminata was recently reported in Connecticut, USA.


Introduction
Didymosphenia geminata (Lyngbye) Mart. Schmidt and D. hullii Khan-Bureau are known to produce thick mats in cold, oligotrophic streams. The former is an important nuisance freshwater diatom with an expanding geographic range in North America. The mucilaginous stalks of both Didymosphenia taxa can persist in the environment even after the cell is no longer viable.
Two purportedly non-indigenous, nuisance stalkforming diatoms, Cymbella janischii (A.W.F. Schmidt) De Toni and D. hullii were first observed in 2013 and reported in the West Branch of the Farmington River in 2014 and 2016, respectively, although a fisherman reported long mucilaginous tufts growing on rocks in 2011 (Khan-Bureau et al., 2014. The West Branch of the Farmington River is within the Connecticut River Watershed in the state of Connecticut (USA) and is important for recreational fishing. The channel where the river is most fished is typically less than 75 cm deep (Khan-Bureau et al., 2014). The oligotrophic conditions, coupled with shallower and stable flows, are favourable for the growth of D. hullii (Khan-Bureau et al., 2014, and would also support D. geminata (Spaulding & Elwell, 2007;Bothwell & Kilroy, 2011), although this latter species was not observed at that time.
During 2015-2017 the state of Connecticut, USA, experienced a moderate, then extreme, drought (US Army Corp of Engineers, https://waterdata.usgs.gov/ usa/nwis/uv?01186500). The West Branch Reservoir has a minimum release of 50 cfs thus providing the majority of the flow below the confluence with the Still River. Typically, the West Branch of the Farmington River and the Still River converge under normal flow conditions (Khan-Bureau et al., 2014). However, in the drought conditions the Still River had very low discharge rates and minimally flowed into the West Branch. Because there was negligible mixing of the warmer and higher nutrient water of the Still River, the cold oligotrophic water was largely the only source discharging into the main branch of the West Branch of the Farmington River creating favourable conditions for the growth of D. hullii.
Monitoring activities during late 2016 revealed a different taxon, which looked similar to D. geminata. The D. geminata-like, large-celled diatom was blooming where the smaller D. hullii had previously been prolifically blooming above the confluence of the West Branch and the Still River (Khan-Bureau et al., 2014). Didymosphenia hullii was observed in low abundance with the D. geminatalike diatom at the Riverton Cemetery site and was also found one mile downstream (Whittemore Recreation Area, Barkhamsted) with C. janischii tufts and in smaller mucilaginous tufts by itself. Analysis of sequences of the V4 18S rDNA gene region was unable to distinguish between D. hullii from Connecticut and published data of D. geminata, because of a lack of variation using this barcode marker (Khan-Bureau et al., 2016). However, there were differences in cell ultrastructure between the type population of D. hullii from the West Branch and published morphological data for D. geminata (Khan-Bureau et al., 2016).
If D. geminata is now present in the West Branch, it becomes important to understand the distinguishing features of the two species recorded from this river system for management purposes. However, D. geminata and D. hullii intergrade in their diagnostic measurements and their distinction is subtle, prompting our closer examination. In this study we present an analysis of the larger-celled taxon that bloomed in the West Branch. For this, we use light microscopy (LM) and scanning electron microscopy (SEM), as well as published micrographs (Spaulding, 2010;Metzeltin & Lange-Bertalot, 2014). This time we used geometric morphometrics (GM) analysis of the abovementioned micrography data.
GM is used as a quantitative analysis of diatom form, shape and size, and can evaluate slight cell differences using several landmarks of diatom valves where SEM often cannot make those distinctions (Beszteri et al., 2005;Cristóbal et al., 2020). GM has been used increasingly in diatom research because it can help distinguish between very closely related species especially when combined with conventional morphological methods; LM and SEM (Pappas et al., 2014;Siver et al., 2015;Uyua et al., 2016;Blanco et al., 2017;Morales et al., 2017;Cristóbal et al., 2020).
SEM does provide clearer resolution of diatom morphology and has provided increased scrutiny of closely related species (Khan-Bureau et al., 2016, but see counter example in Beszteri et al., 2005) figure S1). Samples of mucilaginous growth from rocks and submerged vegetation were taken and placed in Whirlpac® bags, stored in ice, and transported to the lab for processing.

Light microscopy analysis
Fresh preparations of live material were observed at 200× and 400× magnifications using a BX 60 Olympus microscope. Images were digitally captured using an Olympus DP 25 camera and cellSens software (2560 × 1920 pixels). For permanent slide preparation the river samples were centrifuged to concentrate diatom cells. The supernatant was then poured into a beaker and distilled water and 68% nitric acid were added in a 1:1 relation. The mixture was then simmered on a hot plate for 1 h or until the organic matter was visibly oxidized and allowed to cool for several minutes. Samples were then rinsed 4-5 times using deionized water to eliminate the acid by decantation, and then centrifuged to concentrate the diatom frustules (Khan-Bureau et al., 2014. After air-drying overnight aliquots of each diatom sample, they were placed on coverslips, the latter were mounted on glass microscope slides using Naphrax mounting medium, heated on a hot plate and then cooled. The diatom frustules were examined at 600× and 1000× magnifications using the microscope and camera specified above.

Scanning electron microscopy analysis
Aliquots of cleaned samples were dried onto aluminium foil. The aluminium foil was adhered to SEM stubs with double-sided tape. Diatom samples on the stubs were coated for 1 min at 1.8 kV with gold/ palladium using a Polaron sputter coater following the methodology of Morales et al. (2001). The stubs were viewed with a field emission FEI Nova Nano 450 scanning electron microscope located at the University of Connecticut Electron Microscopy Laboratory. Image plates of LM and SEM images were created using Adobe® Creative Suite® 6 Photoshop. A SEM stub of the unknown large-celled Didymosphenia was deposited as a permanent voucher at the George Safford Torrey Herbarium at the University of Connecticut, accession number CONN00227584.

Valve shape analysis
In order to discriminate the identified population from similar taxa, a comparative morphological study was performed by means of geometric morphometry. Valve shape analysis was based on a total of 76 valves photographed under LM. Compared taxa included the larger celled Didymosphenia geminatalike population (n = 7) and D. hullii (n = 47) from Connecticut samples, the latter correspond to the population encountered in type material. Also, D. geminata (n = 11), D. laticollis (n = 7), both from type material, and D. pumila (n = 4) were scanned from the literature (Metzeltin & Lange-Bertalot, 2014) and formed part of the 76 valves analysed. Individual images were binarized using Fiji software (Schindelin et al., 2012). Valve outlines were segmented and vectorized with the Shape ver. 1.3 package (Fig. 1), which describes object shapes using Elliptical Fourier Analysis (EFA) by fitting a closed curve to an ordered set of data points, then decomposing the outline into a sum of harmonically related ellipses (Kuhl & Giardina, 1982). The resulting dataset of normalized Fourier descriptors was subsequently analysed by means of Principal Component Analysis (PCA). After visualization of these components, we chose the first three PCs as representing biologically meaningful phenotypic variation in valve shape (Hirsch et al., 2014). The null hypothesis of no differences in shape between the compared populations was tested by means of a oneway Analysis of Similitude test (ANOSIM, Anderson, 2001) using Euclidean distances between the PCA scores. Statistical analyses were performed using Statistica 10 (StatSoft inc., 2011).

Field and preliminary laboratory observations
The third stalk-forming diatom, morphologically similar to D. geminata, was found growing prolifically in the West Branch of the Farmington River, Connecticut from late 2016-2017, in the same area where D. hullii was blooming, and was first observed in 2013 and reported in 2014 (Khan-Bureau et al., 2014). In this period 2016-2017, D. hullii was growing further downstream and in some instances it was mixed with C. janischii. Examination of live material revealed that the mucilaginous stalk growth in the undetermined taxon was very similar in texture, colour and length to that of D. hullii. Under LM, fresh preparations showed that the valves of the undetermined taxon were much larger than those of D. hullii and had a cell shape and ultrastructure consistent with the descriptions of D. geminata in Metzeltin & Lange-Bertalot (2014) and Spaulding (2010) (Table 1).

LM and SEM analysis
LM and SEM images of D. hullii (Figs 2, 3, 6, 7) and the large-celled Didymosphenia (Figs 4,5,8,9) from the West Branch of the Farmington River were compared (Table 1, Figs 2-9), which showed that both taxa differ greatly in valve length and width. Regarding areolar density there was a superposition of ranges, but the undetermined taxon was inclined to have a higher density (8-10 in D. hullii vs. 9-13 in the undetermined taxon). Regarding stria density and number of stigmata, although the undetermined taxon tended to have a slightly higher number of striae per 10 µm and number of stigmata, a clear difference could not be appreciated (Table 1). As observed in Table 1, most measurements and ranges for D. hullii were different from the four populations of D. geminata included in the table. As presented by Khan-Bureau et al. (2016), the head-pole of D. hullii is round and blunt (Figs 2, 3, white arrows), the constriction between the central inflation and the head-pole is poorly defined (Figs 2, 3, black arrows), and the footpole is bulb-shaped, but short (Figs 2, 3, white arrowheads). In contrast, valves of the undetermined Didymosphenia had a distinctly capitate head-pole (Figs 4, 5, white arrows), distinct constrictions between apices and central region (Figs 4, 5, black arrows), and rounded, slightly subcapitate foot-poles (Fig. 4, white arrowhead and Fig. 5, black arrowhead).
Valves of the D. geminata-like taxon (Figs 4,5,8,9) were 80-135 µm in length, being larger than valves in type material of D. geminata from Streymoy, Faroe Islands (Metzeltin & Lange-Bertalot, 2014), but fitting well within the range given for the populations of this same taxon from the state of Colorado (Spaulding et al., 2020, who reported two populations, one from South Border Creek and another from the Blue River) ( Table 1). Valves of the D. geminata-like taxon had a wider range in width than those in Colorado, and some cells were up to 4 µm wider than those in type material (Table 1), while the stria density was slightly higher than that in both Colorado and 9-11 9-13 7-9 8-9 7-9 8-10 type populations. Regarding the number of stigmata, the D. geminata-like population fit well within the ranges calculated for both the type and Colorado specimens, while the areolar density in the undetermined taxon only showed a higher range of variation (9-13 vs. 11-12.5 in both Colorado and type populations) ( Table 1). The ultrastructure of the unknown Didymosphenia taxon and D. hullii showed deep inclined walls that are surrounded by spine-like projections and dendritic slits below the spine (Figs 6-9) (Metzeltin & Lange-Bertalot, 2014;Khan-Bureau et al., 2016). However, three clear distinctions were made. Internal views of the valves showed that the D. geminata-like taxon had (1) a more expanded polar nodule where (2) a more elevated and prominent helictoglossa sits (compare Figs 7 and 8, black arrows), and the external openings of the stigmata were round in D. hullii while in the D. geminata-like taxon they vary from elliptical to slit-like (compare Figs 6 and 9, white arrows). At ultrastructural level, the D. geminatalike taxon had similar features to those of the type and Colorado populations, except for the external openings that in the two latter populations were more elliptic and less variable than those observed in Fig. 9 (marked by white arrows).

Morphometric analysis
For this analysis, D. laticollis and D. pumila were integrated for comparative purposes and because they are the most morphologically similar taxa to D. hullii and the D. geminata-like taxon. After the segmentation, vectorization and the EFA, the resulting outlines were obtained and diagrammed (Fig. 10). Only four average valve outlines were obtained since the D. geminata-like taxon from the West Branch of the Farmington River and the type material data for D. geminata were grouped together by the process. The average valve outlines shown in Fig. 10 clearly differentiate the four taxa, showing that the shape of the head-pole, foot-pole and valve centre were critical for this distinction.
The PCA analysis of the EFA dataset also showed a distinction among the compared populations. Since the taxa taken into account, considered all at the same time, produced a busy graph that was difficult to interpret, we processed information in three steps. In a first step, we graphed the Fourier descriptors for D. hullii and D. geminata from type material (Fig. 11). In this case, the three axes that represented the biologically meaningful phenotypic variation, and that account for 95% of total variance, completely separated two groups. The group represented by red dots in Fig. 11 corresponds to the D. geminata from the type population; D. hullii appears represented by blue dots. Second, the Fourier descriptors for the large-celled Didymosphenia from Connecticut were plotted against D. geminata from the type (Fig. 12). In this case there was no separation into groups and the three axes represent 83.4% of the total variance. Three data points were recovered aside from a more cohesive group. However, also within the type population, at least one of the valves points grouped away from the rest (Fig. 11). In a third step, we plotted the Fourier descriptors for populations of D. hullii, D. laticollis and D. pumila. In this case the axes represented 58% of the variance and the three populations appeared separated in the graph (Fig. 13).
The ANOSIM analysis performed on PCA scores, based on Euclidean distances, indicated significant differences between the Fourier descriptors obtained for valves of D. hullii when compared with D. geminata of the type, D. laticollis and D. pumila (Figs 11,12,13, Table 2). We considered that the amount of variance and the LM and SEM analyses we presented above were sufficient to determine that the D. geminata-like taxon from Connecticut and D. geminata of the type are sufficiently similar as to be regarded as one taxon, this being the reason for not including the North American population in the ANOSIM analysis.

Discussion
The coexistence of D. hullii with the larger-celled Didymosphenia cast hesitation about their recognition as distinct species, prompting further analysis to verify their distinction and to provide diagnostic characteristics that might be useful at the bench for their quick discrimination under routine LM analyses. Upon SEM analysis, it became clear that two separate populations were present in the West Branch of the Farmington River, but this separation required critical consideration of valve features.
As presented in Khan-Bureau et al. (2016: table 1), traditional diagnostic features such as valve measurements, stria and areola densities, and number of stigmata greatly overlap between D. hullii and D. geminata, but these species can be differentiated by the valve outline, especially the shape and features of the head-and foot-poles, and their variation in a diminution size series. Thus, using the head-pole features alone is insufficient to separate these two species whose blooms might have serious impacts on their habitats.
Similarly, distinguishing D. laticollis and D. pumila from the other two taxa is based on subtle differences of head and foot-pole shape and features or absence of spine-like projections at the valve face periphery (as is the case of D. pumila) for which SEM is required (Metzeltin & Lange-Bertalot, 2014;Khan-Bureau et al., 2016). Also in the case of these two additional taxa, reliable, easily observable diagnostic features are needed for their distinction during routine analyses.
Distinguishing among closely related diatom species is increasingly relying on multiple lines of information or approaches. Although more weight is now given to molecular information (e.g. Poulíčkova et al., 2010;Souffreau et al., 2013;Lefebvre et al., 2017;Kahlert et al., 2019;Kollár et al., 2019;Pinseel et al., 2020) as an additional, and even alternative, source of information for species delimitation, methods such as GM, used for a deeper analysis of morphological information may also be informative and decisive in this endeavour (e.g. Beszteri et al., 2005;Potapova & Hamilton, 2007;Fránková et al., 2009;Pappas et al., 2014;Blanco et al., 2017;Rusanov et al., 2018;Cristóbal et al., 2020). In this study, GM reliably separated all of the Didymosphenia taxa considered in this analysis, supporting the distinction of D. hullii from D. geminata of the type and of both taxa from D. laticollis, and D. pumila as presented in the literature (Metzeltin & Lange-Bertalot, 2014;Khan-Bureau et al., 2016). The fact that we used published micrographs of type material of D. geminata, D. laticollis, D. pumila and our own type material of D. hullii, made all our determinations more reliable and provided a basis for comparison of other populations with our own results. In fact, this is what we did by incorporating the population of D. geminata growing in Connecticut and finding that it shares statistically comparable features with type material of D. geminata from Streymoy, Faroe Islands type material.   Thus, traditional morphometry, LM, SEM, and now GM support the separation of D. geminata from Connecticut from the morphologically closely related D. hullii, D. laticollis and D. pumila, therefore, representing the first confirmed report of the species D. geminata in the state of Connecticut and for the West Branch of the Farmington River (Khan-Bureau et al., 2016). This species had been reported once before, under the older synonym Gomphonema geminatum (Lyngbye) C.Agardh by Terry (1907), as a 'common' diatom in an undisclosed Connecticut locality, but no supporting information was provided (Khan-Bureau et al., 2014).
Shape analysis tools offer precise and accurate descriptions, enable rigorous statistical analysis, and allow visualization interpretation and communication of the results (Wishkerman & Hamilton, 2018). Shape analysis attempts to quantify differences in shape within sets of given specimens, allowing the application of a usual statistical test to discriminate morphometrically between populations . Particularly, GM has been widely used in phycological studies in order to disentangle species complexes (Poulíčková et al., 2017). Specifically used in diatom taxonomy, cell outline is numerically encoded resulting in a high dimensional vector subjected to multivariate analysis that provides the arguments for shape comparisons, assessment of variability, etc. (Fodor & Hâruța, 2020).
Within the framework of GM, two main approaches are currently in use: landmark-based extraction of shape, and coefficients of shape functions fitted to object outlines. The widespread use of landmark analysis is constrained by the necessity of using many landmarks to ensure good discrimination between populations (Cerisier et al., 2019). This method has limitations where no clear observable and repetitive homologous structures exist among compared individuals, as is the case of many diatom outlines (Wishkerman & Hamilton, 2018). Also, landmarks need to be digitized individually and manually, hindering high throughput analyses (Kloster et al., 2014). These limitations led us to use the alternative Elliptical Fourier Analysis coupled with Principal Components in order to differentiate D. hullii from similar taxa. We observed that even though the general frustule outline is more or less preserved in the size diminution series of the Didymosphenia species considered in this study, the morphologies are less pronounced when senescent cells are compared. Despite this, our results demonstrate the efficiency of shape analysis associated with multivariate statistics in order to disentangle morphologically similar taxa along the whole diminution series.
Didymosphenia geminata is an important nuisance diatom species that until now had not been formally reported in Connecticut, USA. In the north-eastern USA, D. geminata was reported for the states of Vermont and New Hampshire in the Connecticut River in 2007 Kumar et al., 2009;Khan-Bureau et al., 2014). Reports of D. geminata from locations without previous records of this species have been used to test hypotheses of new spread or recent increases in density of an otherwise rare species due to changes in the environment (Valéry et al., 2009;Bothwell et al., 2014;Jones et al., 2019). Presently, we do not know if D. geminata has a historical record in this river in Connecticut (Khan-Bureau et al., 2014, as the characteristic stalklike growths had not been reported until 2011 (Khan-Bureau et al., 2014).
Future work will examine the molecular diversity of D. geminata from this location in order to help reveal its relationship to other Didymosphenia taxa, to determine its divergence from other D. geminata accessions, and possibly pinpoint locations of related populations. Khan-Bureau et al. (2016) obtained sequences of the V4 18S rDNA region from D. hullii and compared these with published sequences of taxa in Didymosphenia. Unfortunately, the lack of variation for this single barcode marker along with a lack of published sequences from other Didymosphenia taxa at the time prevented us from resolving relationships among them. It has been reported that certain diatoms, including D. geminata, are difficult to amplify and to obtain DNA barcode data from them (Zimmermann et al., 2011;Jones et al., 2019). Future work to characterize the Didymosphenia taxa from the West Branch of the Farmington River in Connecticut will require a multi-locus approach, as was recommended by Poulíčková et al. (2017) and Jones et al. (2019), work that we look forward to. We also are incorporating palaeolimnology methods to search for Didymosphenia in historical river sediments, which will help determine if this genus is indigenous and rare, or non-native in Connecticut. 00004DFIN.AALP/2017 integrated within the Operational Program for Sustainability and Efficiency in the Use of Resources 2014-20, POSEUR-03-2013-FC000001. We thank the reviewers for their helpful comments.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
We thank Trout Unlimited Connecticut and The Eightmile River Committee for funding this project and Rivers Alliance of Connecticut for support of this project. EAM was co-funded by the Portuguese Foundation for Science and Technology (FCT) project UIDB/04683/2020 -ICT (Institute of Earth Sciences), and the Agência Portuguesa do Ambiente, APA-000004DFIN.AALP/2017 integrated within the Operational Program for Sustainability and Efficiency in the Use of Resources 2014-20, POSEUR-03-2013-FC-000001. LE was also partly funded in the framework of the DIATOMS project (LIST-Luxembourg Institute of Science and Technology).

Supplementary information
The