Quantitative Chemical Composition, Anti-Oxidant Activity, and Inhibition of TNF Release by THP-1 Cells Induced by Extracts of Echinodorus macrophyllus and Echinodorus grandiflorus

This study investigated the similarities between Echinodorus macrophyllus and Echinodorus grandiflorus, plant species that are traditionally used in Brazil to treat rheumatism and arthritis, whose anti-inflammatory effects are supported by scientific evidence. The contents of cis- and trans-aconitic acid, homoorientin, chicoric acid, swertisin, caffeoyl-feruloyl-tartaric acid, and di-feruloyl-tartaric acid were quantified by UPLC-DAD in various hydroethanolic extracts from the leaves, whereas their anti-oxidant activity and their effect on TNF release by LPS-stimulated THP-1 cells were assessed to evaluate potential anti-inflammatory effects. The 50% and 70% ethanol extracts showed higher concentrations of the analyzed markers in two commercial samples and a cultivated specimen of E. macrophyllus, as well as in a commercial lot of E. grandiflorus. However, distinguishing between the species based on marker concentrations was not feasible. The 50% and 70% ethanol extracts also exhibited higher biological activity, yet they did not allow differentiation between the species, indicating similar chemical composition and biological effects. Principal component analysis highlighted comparable chemical composition and biological activity among the commercial samples of E. macrophyllus, while successfully distinguishing the cultivated specimen from the commercial lots. In summary, no differences were observed between the two species in terms of the evaluated chemical markers and biological activities.

The first and second editions of the Brazilian Pharmacopeia contain a monograph of E. macrophyllus [5,6], while a monograph of E. grandiflorus is found in the fifth and sixth editions [7,8]. These two species are utilized for similar medicinal purposes in various regions of Brazil as teas or herbal preparations [2]. The interchangeable use of E. macrophyllus and E. grandiflorus may be attributed to difficulties in distinguishing them based on morphological characteristics. Both species are aquatic plants with large leathery leaves, which have led to their popular name chapéu-de-couro (leather hat) [9]. E. macrophyllus and E. grandiflorus share inflorescences with delicate white flowers, which are also used for decorative purposes, and their leaf and floral anatomy do not exhibit noticeable differences to non-specialists. The presence of phenotypic plasticity and the striking similarity in morphological structures [10] makes it challenging to differentiate these species unequivocally.
The chemical composition of leaves from E. macrophyllus and E. grandiflorus has been extensively investigated, revealing similarities that include the presence of diterpenes, flavone C-glycosides, hydroxycinnamoyl tartaric acid derivatives, alkaloids, saponins, and phenolic acids [11][12][13][14]. Moreover, the observed biological activities of these species have been linked to their chemical composition. For instance, a flavonoid-rich fraction from E. macrophyllus exhibited a more potent anti-inflammatory effect than the extract in the air pouch model in mice, reducing leukotriene B 4 release and neutrophil migration in vitro [11]. The anti-arthritic activity of E. grandiflorus leaves has been attributed to flavone C-glycosides, as evidenced by the potent effect observed when administering a fraction enriched in these compounds in an antigen-induced arthritis model in mice [13]. Additionally, the flavonoid-rich fraction elicited the most potent inhibition of tumor necrosis factor (TNF) release by lipopolysaccharide (LPS)-stimulated THP-1 cells, along with its constituents isovitexin and isoorientin [12]. Other classes of metabolites present in these species, such as diterpenes, hydroxycinnamoyl tartaric acid derivatives, and cisand transaconitic acid, have also been reported for their anti-inflammatory activity by inhibiting TNF release in LPS-stimulated THP-1 cells [12].
Although the leaves of E. macrophyllus and E. grandiflorus are consumed locally for medicinal purposes, there are no clear-cut differences based on morphological characteristics or chemical composition that can be used to unambiguously differentiate them. Therefore, the aim of this study was to undertake molecular, chemical, and biological investigations of different extracts prepared from E. macrophyllus and E. grandiflorus leaves in order to highlight similarities and differences between these species. The extracts were analyzed by UPLC-DAD (ultra-efficiency chromatography coupled to diode array detector) for the quantification of chemical markers (cis-and trans-aconitic acid, homoorientin, chicoric acid, swertisin, caffeoyl-feruloyl-tartaric acid, and di-feruloyl-tartaric acid), using a method developed and validated in the present work. The extracts had their anti-oxidant activity assayed in vitro, along with their effect on the release of TNF by LPS-stimulated THP-1 cells, to access their potential anti-inflammatory effect. Finally, a multi-variate exploratory principal component analysis (PCA) model [15,16] was built, aiming to investigate the possible existence of correlations between the quantitative chemical composition of the extracts and their biological effects.

Plant Drug Samples
The leaves of E. macrophyllus and E. grandiflorus were obtained from Indústria Farmacêutica Catedral (Belo Horizonte, Brazil). Two lots of the first species were purchased and labeled as DV1 (April 2014) and DV2 (March 2019), along with a lot of the second species identified as EG (November 2019). These plant drugs were provided dried and underwent quality control checks as indicated in the supplier's reports. Upon receipt, the leaves were pulverized using a knife mill (Marconi, São Paulo, Brazil) and stored in glass flasks for further use. A third sample of E. macrophyllus aerial parts was collected at the Natural History Museum of the Universidade Federal de Minas Gerais (UFMG) and a voucher was incorporated into the BHCB herbarium at UFMG, under the number BHCB 28557. After drying in a ventilated oven at 40 • C, this plant (drug-coded DVMus) was pulverized in a knife mill and stored in a glass flask.

DNA Isolation, Amplification, Sequencing, Alignment, and Pairwise Distance Calculation
Total genomic DNA was extracted from fresh leaves using a modified version of the 2 × CTAB protocol of Ref. [17]. For PCR amplifications, we used primers described in the Supplementary Table S1 and the parameters previously described [10,18], with minor modifications. The amplified and purified products were sequenced on both DNA strands using the same PCR primers by Macrogen Inc. (Seoul, Korea). The DNA sequence electropherograms were edited and a consensus generated through the Staden program [19]. The sequences were aligned using the MUSCLE program [20] and thereafter manually adjusted using the MEGA7 program [21], according to the procedures described by Ref. [22]. Pairwise distances of the combined dataset (47 terminals and 3182 base pairs) between the major groups recovered in the phylogenetic analyses were calculated using the p-distance method in MEGA7 [21]. Gaps and missing data were treated with the pairwise deletion option.

Molecular Markers, Taxon Sampling, and Molecular Phylogenetic Analyses
Nucleotide sequences from three nuclear genome regions, the nrITS consisting of ITS1, ITS2 and the intervening 5.8S gene, the 5S non-transcribed region (5S-NTS) and the second intron of LEAFY, and two plastid regions (matK-trnK and psbA-trnH) were used in the analysis. Marker selection was based on previous molecular phylogenetic analyses of Echinodorus [10,18]. The taxon sampling was based on a selection of taxa available in GenBank selected to cover the taxonomic diversity of Echinodorus and included 38 taxa and 47 terminals. Voucher information, geographic origins, and GenBank accession numbers are provided in Supplementary Table S2. To root the trees, we used Echinodorus berteroi, which was recovered as sister to the other species of the genus according to the results of Lehtonen and [10,18]. Most species of Echinodorus, except E. berteroi (the type species of the genus), have recently been transferred to the genus Aquarius Christenh. & Byng [23]. An evaluation of this proposal was beyond the objectives of this study and we followed the previous classification here, keeping all species in Echinodorus.
To compare the outcomes of different methodologies in phylogenetic analysis, we employed both maximum parsimony (MP) and Bayesian inference (BI) techniques. Our investigation focused on evaluating a model-free nucleotide substitution approach (MP) in comparison with a model-based approach (BI). We conducted an extensive examination of individual markers and a merged dataset to detect any potential inconsistencies. For the maximum parsimony analysis, we utilized PAUP* version 4 [24] and implemented Fitch parsimony with equal weights for unordered characters (Fitch, 1971) as the criterion for determining optimality. Each search involved 2000 replicates, with random taxon additions; branch swapping was carried out using the tree-bisection and reconnection (TBR) algorithm.
To mitigate excessive swapping on suboptimal islands, we stored a maximum of 10 trees per replicate. To evaluate the internal support, we employed character bootstrapping [25], with 2000 replicates using simple addition, and TBR branch swapping. Throughout the bootstrapping process, we retained up to 15 trees per replicate. The assessment of bootstrap support levels relied on bootstrap percentages (BP), where values falling between 50% and 70% were classified as weak, 71% and 85% as moderate, and values exceeding 85% as strong [26].
Bayesian analysis using MrBayes 3.2.7a [27] was implemented in the Cyberinfrastructure for Phylogenetic Research (CIPRES) Portal 2.0 (San Diego Supercomputer Center, La Jolla, CA, USA [28], treating each DNA region as a separate partition. An evolutionary model for each DNA region was selected in MrModeltest 2 [29] using the hierarchical likelihood ratio tests (hLRTs). To estimate model parameters for each partition separately, we utilized the unlink command. Our analysis involved conducting two independent runs, each with four chains, spanning a total of 10,000,000 generations. During the Markov chain Monte Carlo (MCMC) sampling, we sampled one tree every 1000 generations, employing a temperature parameter of 0.2. We assessed convergence between the runs using multiple criteria. The average standard deviation of split frequencies was evaluated and confirmed to be less than 0.01, indicating convergence. Additionally, we calculated the potential scale reduction factor (PSRF), which yielded a value of 1.0, further validating convergence. Convergence was achieved after 1,970,000 generations. We discarded the initial 2500 trees (25%) as burn-in and utilized the remaining trees to evaluate the topology and to calculate posterior probabilities (PP) through a 50% majority-rule consensus. It is worth noting that, in Bayesian analysis, posterior probabilities cannot be directly compared with bootstrap percentages (BP), as PP values tend to be higher [30]. To evaluate support, we applied criteria similar to a standard statistical test. Groups with a PP value exceeding 0.95 were considered strongly supported, while PP values ranging from 0.90 to 0.95 were categorized as moderately supported. Groups with PP values below 0.90 were considered weakly supported.

Preparation of Extracts
Portions of 1.0 g of the dried leaves of E. macrophyllus (DV1, DV2, DVMus) and E. grandiflorus were sonicated with 150 mL of 96 • GL EtOH or with hydroethanolic solutions at 90% (90% EtOH v/v), 70% (70% EtOH v/v), and 50% (50% EtOH v/v) in 3 cycles of 10 min each. The ethanol content of the hydroethanolic solutions was adjusted with the aid of an alcoholmeter. The obtained extracts were filtered, pooled, and concentrated in a rotary evaporator, under reduced pressure, at an average temperature of 55 • C. Subsequently, the hydroethanolic extracts were lyophilized for complete elimination of residual water. The extracts were transferred to glass flasks and kept in a desiccator until constant weight for yield calculation.

Quantification of Chemical Markers by UPLC-DAD
The contents of the chemical markers cisand trans-aconitic acid, homoorientin, chicoric acid, swertisin, caffeoyl-feruloyl-tartaric acid, and di-feruloyl-tartaric acid were quantified in the extracts described in Section 2.5, using an UPLC-DAD method developed and validated by us.

Identification of Chemical Markers by LC-MS Analysis
The identity of the chemical markers found in the extracts was confirmed by ultraperformance liquid chromatography with electrospray ionization and triple quadrupole mass spectrometry (UPLC-ESI-MS) analysis, carried out in a Waters Acquity UPLC system (Waters, Milford, MA, USA), composed of a binary pump, an auto sampler, an in-line degasser, a photodiode array detector (Waters), and a mass spectrometer Xeco™ Triple Quadrupole MS (Waters). Data were processed using MassLynx4.1 (Waters). The chromatographic conditions were those described for the UPLC-DAD method (Section 2.6.1). The mass spectrometer was operated in the exploratory mode (scan), both at the negative and positive ionization modes using the following conditions: capillary voltage of 3.54 kV; 10-70 V ramp cone voltage; source temperature 120 • C; desolvation temperature 450 • C. Mass/charge ratios (m/z) from 100 to 900 were evaluated. The samples were injected at 1 mg/mL, obtained by solubilizing 1 mg of each extract in 1.0 mL of methanol HPLC grade. Extracts were sonicated in an ultrasound bath for 10 min and centrifuged at 8400× g for 10 min; the supernatants were filtered through a PVDF membrane (0.22 µm) and automatically injected into the UPLC-ESI-MS system.

Validation of the UPLC-DAD Method
The method was validated according to the guidelines of Refs. [31][32][33]. The 70% EtOH extract was selected for validation studies because of its higher complexity compared with other extracts.
Selectivity and system suitability. Purity of peaks from the analytes was evaluated by UV spectra recorded by DAD at different points, in chromatograms of sample solutions, and analyzed by the relationship between the purity angle and the threshold. System suitability was assessed by the outcomes for resolution, retention factor, tailing, number of theoretical plates, and RSD (relative standard deviation) of the retention time of peaks, calculated by Empower 3 software (Waters, Milford, MA, USA).
Linearity, precision, limit of quantification (LOQ), and limit of detection (LOD). Standard solutions were prepared for each compound using a 1:1 mixture of cisand transaconitic acid (6.25, 12.5, 25. Triplicate aliquots (5 µL) of these prepared solutions were injected onto the UPLC-DAD system using the developed method. The calibration curves were constructed by analyzing the injected mass of each compound and quantifying the corresponding peak areas.
The analysis was performed on two independent days and linear regression analysis was employed to determine the calibration curves using Prism 6.01 software (GraphPad, San Diego, CA, USA). The calibration curve of chicoric acid was also used to quantify the contents of caffeoyl-feruloyl-tartaric acid and di-feruloyl-tartaric acid. The intra-day precision was evaluated by calculating the relative standard deviation (RSD) values of sample solutions analyzed on the same day (n = 6) at 100% of the analyte concentration. Similarly, the inter-day precision was assessed by analyzing the sample solutions on two different days, performed by two different analysts (n = 12). The limits of quantification (LOQ) and detection (LOD) were determined by injecting standard solutions at progressively lower concentrations, in quintuplicate. The LOQ was determined as the concentration at which the peak area exhibited an RSD below 3%, while the LOD was defined as the concentration that yielded a signal-to-noise ratio (S/N) of 3.
Recovery and robustness. Accuracy of the method was accessed by recovery studies. Recovery was assessed by adding known aliquots of reference compound solutions to the extracts at three concentration levels falling within the calibration curve. To evaluate robustness, six sample solutions were prepared and analyzed using both the established conditions and by altering the following parameters: column temperature (38 and 42 • C), wavelength (352 and 348 nm to flavonoids and phenolic acids; 218 and 222 nm to aconitic acids), and formic acid content in the mobile phase (0.08 and 0.12% v/v). The concentrations of the analytes were compared by ANOVA, followed by Tukey's test (p < 0.05).

Effect on the Release of TNF In Vitro
The effect of the extracts on the release of TNF was evaluated in vitro in THP-1 cell culture (monocyte lineage derived from human acute monocytic leukemia: ATCC TIB-202) stimulated by LPS. Cell viability was evaluated by the MTT (3-4,5-dimethyl-thiazol-2-yl-2,5-diphenyltetrazolium bromide) assay, as previously described, with modifications [34]. A cell suspension was prepared at 1.0 × 10 6 cells/mL, obtained by centrifuging a culture flask and sequentially resuspending the pellet with RPMI 10% FBS and 2 µL PMA (1 mg/mL). Then, 100 µL aliquots of this suspension were transferred to the wells of a microplate, which were incubated at 37 • C in a 5% CO 2 atmosphere for 6 h to allow cell differentiation and adhesion. After this period, the cells were starved with RPMI medium, supplemented with 2% FBS, and incubated again under the same conditions for 12 h. The cells were transferred to a 96-well plate at 1 ×10 6 cells/well and incubated for 24 h in an oven at 37 • C and a humidified atmosphere of 5% CO 2 . Subsequently, 100 µL of complete medium with 2% of FBS containing different concentrations of the extracts (60 µg/mL) were added in triplicate and the plate was incubated for 3 h under the conditions described above. Then, 20 µL of LPS (1 ng/mL) were added to the wells and the plate was incubated again for 24 h under the same conditions. After incubation, the microplate was centrifuged (1800× g for 5 min) and 100 µL of the supernatant was collected to quantify cytokines by the ELISA method, according to the protocol described in the sequence. The cell pellet was used to assess cell viability by MTT and 28 µL/well of tetrazolium salt (MTT) was added at a concentration of 2 mg/mL in phosphate buffer (PBS). After 90 min of incubation with the MTT, the entire contents were aspirated by a Pasteur pipette, followed by the addition of DMSO to each well (100 µL/well) to solubilize the formed formazan crystals. The microplates were read in a spectrophotometer at 510 nm. Cell viability was calculated according to the equation (A − B/C − B) × 100, where A, B, and C are the absorbance of the samples, blank, and negative control, respectively.

ELISA Protocols for In Vitro Assays
The assay was conducted using an ELISA kit from R&D Systems (DY210-KIT TNF), following the manufacturer's instructions and guidelines. An amount of 100 µL/well of TNF capture antibody (4 µg/mL), solubilized in PBS, was added to a 96-well microplate and the plate was kept at 4 • C for 12 h. After incubation, the plate was washed (PBS/Tween 20, 0.1% v/v) and completely dried. Thereafter, 200 µL of the blocking solution (1% bovine serum albumin in PBS) was added. After 2 h, the plate was washed and dried as described above. Then, 100 µL of samples or TNF standard solution (1000, 500, 250, 125, 62.5, 31.25, and 15.6, pg/mL) was added. The microplate was incubated for 18 h at 4 • C and then the plates were washed and dried. In the sequence, 100 µL of TNF detection antibody solution (500 ng/mL) was added. After 2 h, the plates were washed and dried and then 100 µL of a solution containing streptavidin bound to peroxidase (R&D system) was added to the microplate wells. The plate was incubated for 30 min and washed and dried as described above. Subsequently, 100 µL of o-phenylenediamine (OPD) in citrate buffer (0.4 mg/mL) and 2.4 µL of hydrogen peroxide solution (35% v/v) were added to the wells. The reaction was stopped after 30 min by adding 50 µL of 1 mol/L hydrochloric acid. The OPD oxidation product was detected in a microplate reader at 490 nm. The concentration of TNF was calculated using the calibration curve.

DPPH Radical Scavenger Activity
The assessment of anti-oxidant activity by the 2,2-diphenyl-1-picrylhydrazyl radical (DPPH) assay was carried out as previously described [35], with adaptations. All extracts were initially tested at 200 µg/mL; subsequently, the active samples were evaluated at 1-200 µg/mL and concentration-response curves were constructed. To carry out the experiments, 250 µL of each sample, methanol (negative control), and pyrogallol (50 µg/mL, positive control) were added to a 96-well plate. Then, 100 µL of methanol (blank wells) and DPPH (120 µg/mL, reaction wells) were added to the microplate. The measurements were carried out using a microplate reader (EL808IU-Biotek, Winooski, VT, USA) at 515 nm, with readings taken every 5 min for a total of 25 min. The percentage of radical scavenging activity (% RSA) was determined using the following formula: % RSA = (AC − AS)/AC) × 100, where AC represents the absorbance of the control and AS represents the absorbance of the samples taken at 35 min. The results were expressed as the EC 50 value, which was the effective concentration at which 50% of DPPH radicals were scavenged. The EC 50 was determined through non-linear regression analysis using GraphPad Prism, version 6.0.

β-Carotene/Linoleic Acid Co-Oxidation Assay
The assays were carried out as previously described [36], with minor adaptations. The extracts were solubilized in methanol (2.2 mg/mL) and subsequently diluted from 2.47 to 200 µg/mL (dilution factor 1:3). For the experiments, 25 µL aliquots of the solutions were added to a 96-well plate. A mixture of 25 mg of linoleic acid and 100 mg Tween 20 (Sigma, St. Louis, MO, USA) was added to a round-bottom flask containing 1 mL of a β-carotene solution in chloroform (1 mg/mL, Sigma) and transferred to a rotary evaporator for complete evaporation of the solvent. Then, 50 mL of aerated water was added to the flask, affording the β-carotene emulsion. A blank emulsion was prepared similarly, except for the addition of the β-carotene solution. Aliquots of 250 µL from both emulsions were added to the even and odd columns of the microplate. Methanol was used as a negative control and quercetin (20 µg/mL, Sigma) was used as a positive control. Readings were taken immediately at a wavelength of 470 nm and subsequently at 15 min intervals up to 120 min. Throughout the reading period, the microplate was maintained at an incubation temperature of 45 • C. The anti-oxidant activity was expressed as % of inhibition of lipid peroxidation (%I), using the equation I% = control absorbance (Ac) (initial abs − final abs) − sample absorbance drops (Aam) (final abs − initial abs)/Ac × 100). Assays were performed in triplicate and IC 50 values were determined by non-linear regression using GraphPad Prism, version 6.0 (GraphPad Software, San Diego, CA, USA).

ROS Activity in THP-1 Cells
The effect of the extracts on the production of reactive oxygen species (ROS) by THP-1 cells was performed according to [37] with adaptations. THP-1 cells were cultured in RPMI medium supplemented with 10% fetal bovine serum (FBS). A cell suspension was prepared at a density of 6 × 10 5 cells/mL, obtained by centrifuging a culture flask and sequentially resuspending the pellet with RPMI 10% FBS and 4 µL PMA (100 nM), following incubation for 72 h in an oven at 37 • C and 5% CO 2 . Thereafter, PMA was washed with RPMI 10% FBS, with subsequent incubation of the cells without adding PMA and 2% of FBS in an oven at 37 • C and 5% CO 2 for 24 h. Subsequently, 1 × 10 6 cells per well were placed (250 µL) in a 24-well plate and incubated for 24 h in an oven at 37 • C and 5% CO 2 . After incubation, the cells were treated with the extracts (60 µg/mL in RPMI with 0.2% DMSO) and the plate was incubated for another 3 h in the previous conditions. Then, the cells were treated with 20 µL of a LPS solution (10 µg/mL) at 15 min intervals up to 75 min. The following controls were added to the plate: wells containing only RPMI medium without LPS, wells containing RPMI medium with LPS, and wells containing non-differentiated THP-1 cells. Quercetin (100 µmol/L in methanol) was used as a positive control. Following LPS stimulation, the cells were incubated with 10 µM H 2 DCFDA (2 ,7 -dichlorodihydrofluorescein diacetate) for 15 min at 37 • C and 5% CO 2 . The supernatants were collected and fluorescence was read in a microplate reader (Varioskan™ LUX, Thermo Scientific, Waltham, MA, USA) at 493 nm for excitation and at 522 nm for emission.

Statistical Analysis
The concentrations of the analytes were compared using analysis of variance (ANOVA), followed by Tukey's test for post hoc multiple comparisons, with a significance level set at p < 0.05. Statistical analysis was performed using GraphPad Prism software, version 6.0 (GraphPad Software Inc.). The results were expressed as mean ± standard deviation (SD) and differences were considered statistically significant when p < 0.05.

Molecular and Phylogenetic Analysis of E. macrophyllus
The BOLD (Barcode of Life Data System) database of Echinodorus is completely based on GenBank sequences and therefore we used BLASTN (nucleotide blast) on the NCBI BLAST homepage [38] to search in the nucleotide collection database and to identify the sequences/taxa with higher identity with the sequences we generated from E. macrophyllus. The best match varied between markers (Table 1). For psbA-trnH, matK-trnK, and Leafy, the best match was with E. paniculatus, while, for ITS, the best match was with E. osiris, but the differences between the additional hits were very small.
Since an unequivocal match was not possible, we performed phylogenetic analysis to verify which taxa were most related to the sequenced E. macrophyllus sample using the ITS, 5S-NTS, Leafy, matK-trnK, and psbA-trnH markers. Initial analysis with each marker individually did not detect any case of strongly supported incongruence, so a combined matrix was used for the final analysis. Table 2 presents the general features of the datasets and parsimony statistics, along with a summary of the models implemented for each partition. The strict consensus tree from the parsimony analysis and the Bayesian majority-rule consensus tree were for the most part congruent and, as the latter was more fully resolved and had stronger overall support, it was chosen for presentation and discussion (Figure 1). Our analysis recovered three main clades with high support within Echinodorus: clade A (PP 1.00, BP 100%), clade B (PP 1.00, BP 98%), and clade C (PP 1.00, BP 62%) (Figure 1). Clade C in turn was divided into three main clades: a clade formed by E. grisebachii and E. heikobleheri (clade D: PP 1.00, BP 100%), another formed by E. trialatus, E. emersus, and E. scaber (clade E: PP 1.00, BP 100%), and clade F (PP 1.00, BP 100%), which included our sample of E. macrophyllus and 20 other taxa. Relationships within clade F were poorly resolved and poorly supported. Our sample of E. macrophyllus was recovered as sister to a clade containing E. osiris and six other taxa, but with low support (PP 0.85, BP < 50%). By using neither analysis, our E. macrophyllus sample was recovered as closest to the E. macrophyllus sample available in GenBank. This result was probably due to a combination of factors. First, sequence divergence between Echinodorus species was very low. Mean pairwise distances within and between the major clades identified in the molecular phylogenetic analysis (Figure 1) showed low values, particularly within the clades (Table 3). In clade F, for example, which included the E. macrophyllus samples, the mean pairwise distance among the sequences of all markers was only 0.007; for some markers, such as matK-trnK, these values were even lower (Table 4). These results suggested a recent origin and diversification of some species groups of the genus and indicated that the combination matK and rbcL, for example, (proposed as the standard for plant DNA barcoding [39]) in most cases would not be sufficient for the discrimination of closely related species of Echinodorus. Second, the sampling of Echinodorus in nucleotide collection databases was still small relative to the breadth of geographic distribution and morphological variability of the species. For E. macrophyllus, for example, distributed throughout Brazil and also in Bolivia [40], there was only one sample available in GenBank. Finally, it was also not possible to rule out the possibility of conflicting identifications between our sample and those of the GenBank.

Development and Validation of an UPLC-DAD Method for the Analysis of Chemical Markers
Aiming to investigate the chemical composition of E. macrophyllus and E. grandiflo  In general, the identification of related and morphologically similar species of Echinodorus is difficult and requires specialized taxonomic expertise. The use of DNA sequences can represent a great differential in this sense, particularly when it comes from the certification of dried samples processed for the industry, when the identification based on morphological characters is not possible. However, this will require increased population and geographic sampling of species and the use of more polymorphic and more informative markers, such as the nuclear markers 5S-NTS and Leafy, which proved to be the most diversified among those tested (Table 4) and thus may be good candidates for an initial examination.

Development and Validation of an UPLC-DAD Method for the Analysis of Chemical Markers
Aiming to investigate the chemical composition of E. macrophyllus and E. grandiflorus, different extracts were prepared by sonicating the dried leaves with distinct solvents. The extracts prepared with 50% EtOH and 96 • GL EtOH showed the highest and lowest yields, respectively (Table 5). These findings indicate the predominance of hydrophilic compounds in both species. UPLC-DAD profiles were initially recorded for all extracts using an exploratory run and pointed out the 70% EtOH extract as the one with the most complex matrix. Therefore, this extract was employed for development and validation of the analytical method to quantify chemical markers. A segmented gradient (concave and linear) of water and acetonitrile, intercalated with isocratic elution steps, was used for UPLC analysis. The elution conditions were set and optimized to allow the adequate separation of compounds 1-6 in extracts of leaves from E. grandiflorus and E. macrophyllus. Two different wavelengths were used for quantifying the analytes: at 220 nm for the analysis of cisand trans-aconitic acid (Figure 2A) and at 350 nm to quantify flavonoids and tartaric acid ( Figure 2B).
The identification of the chemical markers in the extracts was performed by the analysis of reference compounds (cis-and trans-aconitic acid, homoorientin, chicoric acid, and swertisin) in the established chromatographic conditions and comparison of their retention times and UV data recorded by DAD with those of peaks registered in the chromatograms. The identity of the chemical markers was further confirmed by UPLC-ESI-MS analysis. Two other constituents of the extracts were assigned by UPLC-ESI-MS analysis: caffeoyl-feruloyl-tartaric acid (5) and di-feruloyl-tartaric acid (6). To the best of our knowledge, compounds 5 and 6 have not been isolated so far from E. macrophyllus. On the other hand, several caffeoyl-feruloyl-tartaric acid derivatives along with cinnamoyl-tartaric acid derivatives were obtained from E. grandiflorus [41].
Among the compounds identified in the extracts of E. macrophyllus, only homoorientin [42], swertisin [43], and caffeoyl-feruloyl-tartaric acid [11] have been previously reported; the occurrence of all other constituents is herein described for the first time. On the other hand, all of them have been previously reported as constituents of E. grandiflorus [12,13,41], corroborating the similar chemical composition of these species.
In the sequence, the developed UPLC-DAD method was validated for the quantitation of the six chemical markers identified in the extracts (Figure 2). The selectivity was assessed by spectral homogeneity of the peaks, evaluated by DAD. The ascending, upper, and descending regions of the spectrum of each peak matched exactly, indicating that other compounds were not co-eluted, i.e., the threshold angle was higher than the purity angle. The reliability of the chromatographic conditions was assured by the results of system suitability tests, as depicted in Table 5. Relative standard deviation (RSD) values of retention time (≤1%), number of theoretical plates (N > 2000), retention factor (k > 0.5), USP tailing factor (≤2.0), and resolution (Rs > 1.5) obtained for all analytes fell within the limits recommended by [31,44] for quantitative analysis.
The calibration curves constructed for the reference compounds were linear along the established ranges, with r 2 values above 0.99 (Table 6). Curves obtained in two consecutive days were statistically similar (p > 0.05). Furthermore, statistical analysis showed that calibration data presented normal distribution, with randomly distributed residuals. The method showed good precision with RSD values below 4.3% for all analytes (Table 6). Additionally, the mean concentrations determined for the inter-day precision experiment did not differ statistically (p > 0.05) for all analytes. LOQ values between 0.04 and 1.92 µg and LOD between 0.01 and 0.58 µg (Table 6) demonstrated the high sensitivity of the method. The method demonstrated robustness (p > 0.05) under all tested conditions, as the introduced variations did not significantly alter the concentrations of the analytes when compared with the results obtained using the established conditions. The method was also accurate, as indicated by the recovery rates ranging from 95.8% to 110.8% for all tested compounds. The results of the method validation are summarized in Table 6. Among the compounds identified in the extracts of E. macrophyllus, only homoorientin [42], swertisin [43], and caffeoyl-feruloyl-tartaric acid [11] have been previously reported; the occurrence of all other constituents is herein described for the first time. On the other hand, all of them have been previously reported as constituents of E. grandiflorus [12,13,41], corroborating the similar chemical composition of these species.

Quantitation of Chemical Markers in Extracts of E. macrophyllus and E. grandiflorus
The developed UPLC-DAD method was applied to the analysis of 12 extracts (96 • GL EtOH, 90% EtOH, 70% EtOH, and 50% EtOH) prepared from leaves of three different samples of E. macrophyllus (commercial lots DV1 and DV2; cultivated sample DVMus), along with four extracts of E. grandiflorus (96 • GL EtOH, 90% EtOH, 70% EtOH, and 50% EtOH) prepared from a single commercial lot. The quantitative results expressed as percentage mean (w/w) ± SD of triplicates are depicted in Table 7. As expected, all extracts prepared with 50% EtOH and 70% EtOH showed high concentrations of the polar compounds cisand trans-aconitic acid, analyzed jointly (1), with variable contents in the samples. The 70% EtOH extracts obtained from DV1 (8.09 ± 0.02% w/w) and DV-Mus (7.47 ± 0.03%) showed similar contents of 1, whereas the highest concentration was measured in the 50% EtOH of E. grandiflorus (20.12 ± 0.16%) ( Table 7).
The quantification of chemical markers in the extracts of E. macrophyllus and E. grandiflorus did not provide sufficient differentiation between these species.
Up until now, the chemical composition of E. grandiflorus has been investigated more than E. macrophyllus; previous reports have included the quantitation of homoorientin by HPLC-DAD in 70% EtOH extracts [45], along with the determination of cisand transaconitic acid and compounds 2, 3, and 4 in different extracts using a distinct HPLC-DAD method [12,13]. The Brazilian Pharmacopoeia [8] establishes caffeic acid, isoorientin, and swertiajaponin as chemical markers of E. grandiflorus. Among these markers, homoorientin was also described and quantified for E. macrophyllus and E. grandiflorus in this work. To the best of our knowledge, the quantitative chemical composition of E. macrophyllus extracts has not been reported so far, although some of the compounds herein quantified (4, 5, and 6) have been detected in this species, along with several other flavone C-glycosides and hydroxycinnamoyl tartaric acid derivatives [11,43]. Differences between the mean concentrations of the extracts were analyzed by ANOVA, followed by Tukey test (p < 0.05). For the contents of chemical markers, different letters within the columns indicate significant differences. ND-not determined (bellow the LOQ).

In Vitro Inhibition of TNF Release
The potential anti-inflammatory activity of extracts from E. macrophyllus and E. grandiflorus leaves was assayed in vitro, by measuring their inhibitory effect on the release of TNF by LPS-stimulated THP-1 cells. The extracts were not cytotoxic for THP-1 cells at 60 µg/mL, showing cell viability above 90%, according to results from SRB and MTT assays. Furthermore, none of the extracts or dexamethasone (Dx) (used as a positive control) induced the release of TNF in the absence of the inflammatory stimulus (LPS), indicating an absence of a pro-inflammatory effect.
The extracts prepared from the three samples of E. macrophyllus induced a similar inhibition of TNF release by LPS-stimulated THP-1 cells (Figure 3). None of the extracts obtained with 90% EtOH induced a significant reduction of this cytokine. Among the 96 • GL ethanol extracts, only those prepared from DV2 exhibited a significant reduction in TNF levels (44.1 ± 2.1%). All extracts prepared from E. macrophyllus, with either 70% EtOH or 50% EtOH, reduced the release of TNF significantly, reaching 61.9 ± 3.0% (70% EtOH extract) and 46.6 ± 10.2% (50% EtOH extract) inhibition for the extracts obtained from DV1. Furthermore, 70% EtOH and 50% EtOH extracts of DV2 induced TNF inhibition rates similar to the extracts prepared with DVMus (p > 0.05), whereas significant differences were obtained for those 70% EtOH and 50% EtOH extracts prepared from DV1 (p < 0.05). Taken together, the obtained results indicated that extracts prepared with 50% EtOH and 70% EtOH induced a more potent inhibition of TNF release for the three plant drugs of E. macrophyllus. It is worth noting that the 70% EtOH and 50% EtOH extracts of DV1 and DVMus were the most active extracts, showing similar activity without significant differences among them. In general, the extracts showing high inhibitory activity of TNF release (70% EtOH and 50% EtOH) were those with the highest contents of aconitic acids and hydroxycinnamoyl tartaric acid derivatives. The data presented in this study supported the findings reported by [11], who observed a decrease in TNF and IL-1β levels in exudates when mice were treated with a fraction enriched in tartaric acid derivatives from E. macrophyllus, using an air pouch model of inflammation.
We have previously investigated the effect of E. grandiflorus extracts on the release of TNF by LPS-stimulated THP-1 cells, non-differentiated into macrophages. In the previous study [12,13], it was observed that the 50% and 70% EtOH extracts exhibited significantly higher inhibition rates of TNF release (respectively, 100 ± 0.0% and 66.7 ± 5.0%) than those herein reported. Such differences may be related to the lower contents of flavone Cglycosides in the extracts here evaluated, since the authors ascribed the anti-inflammatory activity mainly to these compounds. Moreover, they did not employ cells differentiated into macrophages, as herein reported.
In general, the extracts showing high inhibitory activity of TNF release (70% EtOH and 50% EtOH) were those with the highest contents of aconitic acids and hydroxycinnamoyl tartaric acid derivatives. The data presented in this study supported the findings reported by [11], who observed a decrease in TNF and IL-1β levels in exudates when mice were treated with a fraction enriched in tartaric acid derivatives from E. macrophyllus, using an air pouch model of inflammation.
Data reported here for the in vitro inhibition of THF release by LPS-stimulated THP-1 cells treated with different extracts of E. macrophyllus and E. grandiflorus showed no significant differences in the responses induced by these species. Therefore, it was not possible to differentiate them based on the assayed biological activity, as similarly observed for the chemical markers quantified in the extracts.

In Vitro Anti-Oxidant Activity and Effect on ROS Production
The anti-oxidant activity of the extracts was assayed in vitro by the DPPH and βcarotene/linoleic acid co-oxidation radical scavenging methods. The extracts derived from E. macrophyllus DV1 exhibited significantly higher activity in the DPPH assay compared with those obtained from DV2 and DVMus. The EC 50 values of the DV1 extracts were below 200 µg/mL, with the 50% EtOH extract demonstrating the highest activity (EC 50 = 109.6 ± 1.7 µg/mL), as shown in Table 8. As for E. grandiflorus, only the 70% EtOH extract presented an EC 50 value < 200 µg/mL (135.2 ± 2.0). The active extracts showed high contents of hydroxycinnamoyl tartaric acids (Table 7), suggesting the participation of these compounds in the observed anti-oxidant effect. In the same direction, E. macrophyllus fractions containing tartaric acid derivatives have been reported to promote significant reduction of DPPH radical by approximately 70% [11].
Regarding the β-carotene/linoleic acid co-oxidation assay, the most active extracts from E. macrophyllus were those prepared with 50% EtOH, either for DV1 (IC 50 40.1 ± 1.6 µg/mL), DV2 (IC 50 23.9 ± 1.8 µg/mL), or DVMus (IC 50 17.7 ± 1.5 µg/mL). In turn, all E. grandiflorus extracts were active, the 50% EtOH extract being the most active one (IC 50 6.8 ± 1.5 µg/mL) ( Table 8). As far as we know, this is the first report on the anti-oxidant activity of extracts from E. macrophyllus and E. grandiflorus in the β-carotene/linoleic acid co-oxidation assay. Notably, among all extracts prepared with different plant drugs of both species, the 50% EtOH extracts showed the highest anti-oxidant activity, probably due to their higher contents of flavone-C-glycosides and hydroxycinnamoyl tartaric acid derivatives. This supposition was supported by several reports on the anti-oxidant action of different phenolic derivatives [46,47].
The results from the β-carotene/linoleic acid co-oxidation assay pointed out the 50% EtOH extracts as the most active ones and, for this reason, their effect on the ROS production by LPS-stimulated THP-1 cells was assayed. None of the assayed extracts inhibited ROS formation at high rates, with inhibition values below 20% ( Table 8).
The anti-oxidant activity of a compound or extract can be more accurately estimated by using three experimental models that associate different oxidation mechanisms, including: (i) the formation, chain reactions, and action of free radicals; (ii) lipid peroxidation and degradation of defense cells systems; (iii) the formation of ROS [48]. In line with this recommendation, we utilized these different models to demonstrate the anti-oxidant potential of E. macrophyllus and E. grandiflorus extracts in our study. Table 8. Anti-oxidant activity of extracts of E. macrophyllus (DV1, DV2, and DVMus) and E. grandiflorus assayed in vitro by the DPPH and β-carotene/linoleic acid co-oxidation models and effect of selected extracts on ROS produced by LPS-stimulated THP-1 cells.

Plant Drug
Extract DPPH (EC 50 µg/mL ± SD) β-Carotene (IC 50  Phenolic compounds are known to inhibit ROS due to the presence of hydroxyl groups, capable of scavenging free radicals by electron donation [49,50]. ROS inhibition can also result from the increase of cell anti-oxidant defenses [51]. Extracts of Echinodorus prepared with 50% EtOH promoted high DPPH radical capture, decreased lipid peroxidation, reduced ROS production, and diminished the release of TNF. These extracts possessed high contents of hydroxycinnamoyl tartaric acids and flavonoids, which may explain their anti-oxidant and anti-inflammatory action. These results demonstrated the potential of these extracts for further investigation as potential agents for the treatment of inflammatory disorders [52].

PCA Model
An exploratory unsupervised classification PCA model was built to investigate the similarity between E. macrophyllus and E. grandiflorus, based on the contents of chemical markers and biological activities of the assayed extracts. The extracts were allocated at the rows and the chemical markers (cis-and trans-aconitic acids, CTAA; homoorientin, HOMO; swertisin, SWER; chicoric acid, CHA; caffeoyl-feruloyl-tartaric acid, CATA; diferuloyltartaric acid, DFTA) were put in the columns, along with data from the biological assays (lipid peroxidation in the β-carotene assay, Bcar; DPPH radical scavenging, DPPH; inhibition of TNF release, TNF), generating a matrix of dimensions 144 × 11. Data were preprocessed by autoscaling to provide equal weights for all investigated variables in both methods [15].
The obtained results for PCA model are depicted in Figure 4 as a biplot of PC1 × PC2. A biplot is a graphical representation simultaneously showing the relations among scores (describing samples) and loadings (describing variables). A model with two PCs accounted for 67.83% of the total variance data. In this model, none of the samples presented Q residues above the limit at a 95% confidence level, thus indicating the absence of outliers in the dataset. PC1 accounted for 39.86% of variance and discriminated DVMus (positive scores) from DV1/DV2 (negative scores) samples (these two clusters are indicated by red rectangles in Figure 4). EG samples were not clearly discriminated by this PCA model, even when PC3 was considered (10.8% of the variance), and it was not possible to ascribe its similarity to any of the evaluated samples of E. macrophyllus. Based on the PC1 of this model, we could conclude that the variation within different samples of the same species (DV, E. macrophyllus) was greater than the variation between the two species. In other words, the differences observed within E. macrophyllus samples were more pronounced than the differences between E. macrophyllus and E. grandiflorus. In addition, the anti-inflammatory (TNF) and anti-oxidant (DPPH/Bcar) activities were captured by PC2 (27.97%), respectively, with positive and negative loadings (green circles).  A heatmap (correlation map) was also generated to characterize the relationship between the analyzed variables ( Figure 5B). The inhibition of TNF release showed weak correlations with CTAA (rTNF-1 = 0.0198; rTNF-2 = 0.1543; rTNF-3 = 0.0732), which, in turn, correlated moderately with DPPH activity (r = 0.2149). DPPH activity showed weak/moderate correlations to all assayed chemical markers: HOMO (r = 0.1545), SWER (r = 0.1654), CHIA (r = 0.303), CATA (r = 0.2755), and DFTA (r = 0.2507). Conversely, these chemical markers were inversely correlated to Bcar anti-oxidant activity: CTAA, r = −0.2709; HOMO, r = −0.2344; SWER, r = −0.4801; CHIA, r = −0.3293; CATA, r = −0.2847; DFTA, r = −0.3321. The apparent contradictory results observed in the anti-oxidant assays could be attributed to the differences in the methods employed. The DPPH assay was based on the measurement of radical scavenging activity, while the β-carotene test evaluated the inhibition of lipid peroxidation. These assays targeted different mechanisms of anti-oxidant activity and may have elicited distinct responses from the constituents present in the extracts [53]. In Figure 4, samples were clustered as a function of the scores/samples in PC1. Aiming to better visualize interrelations, Figure 5A shows the same PCA biplot of Figure 4, but with marked clusters as a function of the loadings/variables. As depicted in Figure 5, the inhibition of TNF release was related to DV1, suggesting a positive correlation between these variables (see the black ellipse). Moreover, the anti-oxidant activity Bcar was associated to DV2 (red ellipse), while the DPPH activity was weakly correlated to EG (green ellipse). Chemical markers (CTAA, HOMO, SWER, CHA, CATA, and DFTA) were positively correlated to DVMus (grey circle in Figure 5A), which was consistent with the higher concentrations of these assayed constituents in this sample.
A heatmap (correlation map) was also generated to characterize the relationship between the analyzed variables ( Figure 5B). The inhibition of TNF release showed weak correlations with CTAA (r TNF-1 = 0.0198; r TNF-2 = 0.1543; r TNF-3 = 0.0732), which, in turn, correlated moderately with DPPH activity (r = 0.2149). DPPH activity showed weak/moderate correlations to all assayed chemical markers: HOMO (r = 0.1545), SWER (r = 0.1654), CHIA (r = 0.303), CATA (r = 0.2755), and DFTA (r = 0.2507). Conversely, these chemical markers were inversely correlated to Bcar anti-oxidant activity: CTAA, r = −0.2709; HOMO, r = −0.2344; SWER, r = −0.4801; CHIA, r = −0.3293; CATA, r = −0.2847; DFTA, r = −0.3321. The apparent contradictory results observed in the anti-oxidant assays could be attributed to the differences in the methods employed. The DPPH assay was based on the measurement of radical scavenging activity, while the β-carotene test evaluated the inhibition of lipid peroxidation. These assays targeted different mechanisms of anti-oxidant activity and may have elicited distinct responses from the constituents present in the extracts [53]. The inhibition of lipid peroxidation is regarded as a more complex and selective method for identifying anti-oxidant compounds than DPPH, although both methods are important to characterize the anti-oxidant response [54]. The DPPH assay has been known to exhibit nonspecific responses to various compounds and may yield misleading results when used in isolation. To ensure a more accurate assessment of anti-oxidant activity, it is recommended to analyze the DPPH assay in conjunction with other complementary The inhibition of lipid peroxidation is regarded as a more complex and selective method for identifying anti-oxidant compounds than DPPH, although both methods are important to characterize the anti-oxidant response [54]. The DPPH assay has been known to exhibit nonspecific responses to various compounds and may yield misleading results when used in isolation. To ensure a more accurate assessment of anti-oxidant activity, it is recommended to analyze the DPPH assay in conjunction with other complementary anti-oxidant tests [55]. The literature data have indicated low correlation between results from DPPH test and other anti-oxidant assays [56], as herein observed for the low positive correlation with the β-carotene assay (r = 0.2690).
PCA results clearly demonstrated the similarities in the quantitative chemical composition and biological activities of E. macrophyllus and E. grandiflorus extracts, which was in accordance with previous reports [57]. However, PCA was not able to discriminate E. grandiflorus from the analyzed samples of E. macrophyllus. The analyzed chemical markers showed a stronger correlation with DVMus, allowing for the differentiation between E. macrophyllus samples (DVMus vs. DV1 and DV2). However, these same chemical markers, which had positive loadings on PC1, did not show a significant correlation with E. grandiflorus extracts. Therefore, based on the quantification of these chemical markers and the assessed biological activities in this study, it was suggested that E. macrophyllus and E. grandiflorus could not be distinguished from each other.
Furthermore, our findings indicated that the two commercial lots of E. macrophyllus (DV1 and DV2) exhibited comparable contents of the analyzed chemical markers and presented similar biological activities. However, according to the PCA model, DVMus displayed distinct chemical marker contents compared with DV1 and DV2. This divergence could be attributed to the fact that DVMus was sourced from a single cultivated specimen, while DV1 and DV2 were composed of multiple specimens grown in different environments. These varying growth conditions could impact the chemical profiles of the plants [58].

Conclusions
Our study explored, for the first time, molecular analysis, quantitative chemical composition, and biological activity of various extracts from E. macrophyllus and E. grandiflorus leaves aiming to highlight the similarities and differences between these species. The molecular analysis of E. macrophyllus did not allow its genetic differentiation from other Echinodorus species, including E. grandiflorus, suggesting that there are only a few distinct genetic characteristics separating these species. The quantitation of six chemical markers in extracts prepared from commercial samples (DV1 and DV2) and a cultivated specimen (DV-Mus) of E. macrophyllus, along with a commercial lot of E. grandiflorus (EG), revealed higher contents of these markers in the 50% and 70% ethanol extracts. However, we were unable to differentiate between the species based on the concentrations of these markers. Moreover, the 50% and 70% ethanol extracts also elicited higher biological activity-inhibition of TNF release by LPS-stimulated THP-1 cells and anti-oxidant activity-in comparison with the 90% and 96 • GL ethanol extracts. This difference in activity was attributed to the higher contents of chemical markers in the former extracts. Nevertheless, the results from the biological assays did not allow us to discriminate between the two species. Overall, our data demonstrated that E. macrophyllus and E. grandiflorus have similar chemical composition and biological activity. Additionally, the PCA analysis revealed a comparable chemical composition and biological activity among the commercial samples of E. macrophyllus. However, the analysis successfully distinguished the chemical composition of the cultivated specimen from the commercial lots. Interestingly, the cultivated and commercial samples of E. macrophyllus exhibited a higher level of intra-specific variance compared with the inter-specific variance observed with E. grandiflorus. It is worth noting, though, that the quantified chemical markers and biological activities assessed in this study did not reveal any discernible differences between the two Echinodorus species.
As part of our future perspectives, we plan to conduct in vivo studies to investigate the anti-inflammatory activity of the 50% and 70% ethanol extracts from E. macrophyllus.
These studies will involve the use of proof-of-concept models as well as disease models, including acute and chronic antigen-induced arthritis. By evaluating the efficacy of these extracts in vivo, we aim to contribute to the development of an herbal preparation from E. macrophyllus that can be potentially utilized for managing arthritis and other inflammatory diseases. These studies will provide valuable insights into the therapeutic potential of E. macrophyllus and its application in the field of inflammatory disorders.