The TransPEGASO cruise was carried out on board the R/V Hespérides between 21 October and 21 November 2014. We sampled 42 oceanographic stations along a transect (Fig. 1) from the westernmost Mediterranean (Cartagena, Spain) and across the Atlantic Ocean to Punta Arenas (Chile), with a stopover in Buenos Aires between 8 and 16 November (Supplementary material Table S1). Based on geographic location and Chl a concentration, we assigned the stations to seven biogeographical provinces (Longhurst 2007): Mediterranean (MEDI), North Atlantic Subtropical Gyre East (NAST- E), Canary Current Coastal (CNRY), North Atlantic Tropical Gyre (NATR), Western Tropical Atlantic (WTRA), South Atlantic Tropical Gyre (SATL) and Southwest Atlantic Continental Shelf (SWAS) (Table S1).

Water samples were obtained twice a day, at 9 a.m. and 4 p.m. local time (Table S1), through an underway flow-through system that collected water from 4 m depth. The water was screened in situ (except for nutrients) through a 150-μm nylon mesh to remove the larger particles, and the total sampling time did not exceed 90 seconds. Temperature, salinity and conductivity were determined with an online SBE 21 SeaCAT thermosalinograph.

For determination of nitrate, nitrite, silicate and phosphate concentrations, 100 mL of seawater were introduced in Falcon vials that were kept frozen (−20°C) until analysis in the home laboratory. The nutrient concentrations were measured colorimetrically with an Alliance Evolution II autoanalyser, following the procedures described in Hansen and Koroleff (1999).

Samples for flow cytometric analysis consisted of 4.5 mL of water that were preserved with 1% paraformaldehyde plus 0.05 % glutaraldehyde (final concentrations), left to fix for 15 minutes, deep frozen in liquid nitrogen and stored at −80°C until analysis (Zamanillo et al. 2019). The samples were analysed six months after the cruise, in the home laboratory, with a Becton Dickinson FACScalibur flow cytometer provided with a laser emitting at 488 nm (Gasol and Del Giorgio 2000). Before analysis, the samples were thawed and 10 μL of a 10⁵ mL^–1 solution of yellow-green 0.92 μm Polysciences latex beads were added per 600 μL sample as an internal standard. Synechococcus, Prochlorococcus, picoeukaryotes and flow cytometric nanoeukaryotes (hereafter FC-nanoeukaryotes) were identified on the basis of their autofluorescence and scattering properties.

For fluorometric Chl a (Chl a_Fl) determinations, 250 mL of seawater were filtered through 25 mm Whatman GF/F glass fibre filters (nominal pore size 0.7 μm) using a <200 mm Hg vacuum. The filters were subsequently frozen at −20°C and, after 24 hours, introduced into 90% acetone and kept in the dark at 4°C for another day. The Chl a_Fl concentrations were determined fluorometrically (Yentsch and Menzel 1963) using a calibrated Turner Designs fluorometer. No phaeopigment correction was carried out.

Approximately 250 mL of seawater sample was placed in glass bottles, preserved with formaldehyde solution (4%) and stored in the dark until analysis on land. Phytoplankton examination was carried out by means of the inverted microscope method (Utermöhl 1958); 100 mL of water was introduced in composite chambers and left to settle for 48 h. Subsequently, the entire base of the chambers was scanned at 125× to quantify the larger, less abundant organisms of the microphytoplankton (>20 µm), and at least two transects were examined at 312× to enumerate the smaller and more frequent nanophytoplankton forms (<20 µm). When possible, phytoplankton was identified to the species level, but many organisms could not be adequately classified and were pooled in categories such as “Nanoflagellates 3-20 µm”, “Small dinoflagellates (<20 µm)” or “Unidentified small coccolithophores (<10 µm)”. Organisms in the picoplankton size range cannot be adequately quantified with this method.

Pigment analysis by HPLC, following the method of Latasa (2014), was carried out for three phytoplankton size classes obtained on board by SFF (Herbland et al. 1985): total phytoplankton, nano- plus microphytoplankton, and picophytoplankton. For total phytoplankton (Chl a_tot), we filtered the 0.35- to 2-L water samples onto 25 mm Whatman GF/F glass fibre filters. For nano- plus microphytoplankton ≥3 µm (Chl a_n+m≥3), we filtered 0.2 to 1.8 L of seawater at the SWAS province stations and 2.5 L in the other ones onto 25 mm Nuclepore polycarbonate filters (3 μm pore size). The resulting filtrate was filtered again through 25 mm Whatman GF/F filters to collect the picophytoplakton (Chl a_pico<3) class. Immediately after filtration, the filters were introduced into cryovials and stored frozen at -80°C until analysis in the Xixón/Gijón laboratory of the Instituto Español de Oceanografía. The procedure for pigment extraction began by wetting the filters with 2.5 mL acetone 90% that contained an internal standard of trans-β-apo-8’-carotenal (Sigma), followed by sonication (30 seconds, 8×10 cycle and 40 intensity). After keeping the filters in acetone at –20°C for 24 h, the samples were vortexed, filtered again through Whatman GF/F glass fibre filters to remove debris, and placed in amber cryovials. The extracts were analysed by HPLC using an Agilent series (Waldbronn, Germany) 1200 chromatographic system with a G1311A quaternary pump, a G1367C autosampler with a 100-µL capillary loop, a G1316B column thermostat and a G1315C diode array detector. The sample volume injected was 1200 μL and the sequence of sample withdrawal and injection was 30 μL sample-20 μL water-30 μL sample-20 μL water and ejection into seat, repeated 12 times; the composition of the eluents is detailed in Latasa (2014). A total of 36 pigments (Tables 1 and S2) were detected at 474 and 664 nm and identified by retention time with an online diode array detector. The peak area of the pigments was calculated with the Agilent ChemStation software. The Chl a concentration in total (Chl a_tot) and size-fractionated samples (Chl a_n+m≥3 for the ≥3 µm and Chl a_pico<3 for the <3 µm classes) was estimated as the sum of monovinyl chlorophyll a (MV-Chl a), divinyl chlorophyll a (DV-Chl a), chlorophyllide a and chlorophyll a allomers and epimers. DV-Chl a did not reach detection level in the SWAS region. To facilitate comparison with other works, the relationship between Chl a_tot and Chl a_Fl is shown in Figure S1.

An estimate of the size structure of the phytoplankton assemblages was obtained from pigment concentrations quantified by HPLC, by means of the diagnostic pigment algorithms (Table 2) of Vidussi et al. (2001), later refined by Uitz et al. (2006). In this approach, diatoms and dinoflagellates are considered as microphytoplankton, cryptophytes and flagellates containing 19’-Butanoyloxyfucoxanthin (19’But) and 19'-hexanoyloxyfucoxanthin (19'Hex) as nanophytoplankton, and cyanobacteria and “green flagellates” as picoplankton. An estimate of the total Chl a concentration can be obtained as a weighted sum of the seven diagnostic pigments considered (SDP, see Table 2). An additional evaluation of the three phytoplankton size fractions was obtained using the Chl a–based approach (Table 2) of Hirata et al. (2008, 2011).

A size index (SI) based on an expression derived by Bricaud et al. (2004) was used to synthesize the size structure of each algal assemblage:

where 1 µm, 5 µm and 50 µm are taken as representative sizes for the picophytoplankton (f_pico), nanophytoplankton (f_nano) and microphytoplankton (f_micro) proportions (expressed as a fraction of 1).

To assess the photoacclimation response to irradiance of the phytoplankton groups containing diadinoxanthin (Ddx) as the main light-protecting pigment, which include diatoms, dinoflagellates, haptophytes and pelagophytes, we calculated the ratio Ddx/LHC between the concentration of Ddx and the sum of the concentrations of four major light-harvesting carotenoids (LHC): fucoxanthin (Fuco), 19’But, 19’Hex and peridinin (Per).

Based on the main pigment markers, the contribution of different phytoplankton groups to total Chl a was calculated using version 1.95 of the CHEMTAX chemical taxonomy software (Mackey et al. 1996). Essentially, the CHEMTAX algorithm uses one or several initial matrices of pigment: Chl a ratios for the selected phytoplankton groups to derive the contribution of each pigmentary class to the total Chl a. The samples corresponding to Chl a_tot, Chl a_n+m≥3 and Chl a_pico<3 were clustered separately according to the contribution of the pigments 19’But, 19’Hex, alloxanthin (Allo), chlorophyll b (Chl b), chlorophyll c2 (Chl c2), Chl c2-monogalactosyldiacylglyceride ester [14/14] (Chl c2-MGDG [14/14]), divinyl chlorophyll a (DV-Chl a, Fuco, neoxanthin (Neo), Per, prasinoxanthin (Pras), violaxanthin (Viol) and zeaxanthin (Zea); DV-Chl a was removed in SWAS because it was below detection level. Each size class data set could be grouped into two clusters, one containing the SWAS samples (cluster 2) and one (cluster 1) containing the remaining ones (Table S2). CHEMTAX was run following the procedures of Latasa (2007) and Latasa et al. (2010). Briefly, we created 29 randomized copies of the initial ratio matrix and ran the program eight successive times. After the eighth run, a single average matrix was made and used again for a final run. Nine pigmentary groups were quantified for the whole and <3 µm size classes in MEDI, NAST-E, CNRY, NATR, WTRA and SATL provinces: chlorophytes, cryptophytes, diatoms, dinoflagellates, haptophytes, prasinophytes, pelagophytes, Prochlorococcus and Synechococcus. The same groups were considered for SWAS, except Prochlorococcus, which was not found there. Both Prochlorococcus and Synechococcus were excluded from the calculations for the ≥3 µm class.

Water samples of 37 stations were collected and filtered through pre-combusted (450°C, 4 h) 47-mm Whatman GF/F glass fibre filters using an acid-cleaned all-glass filtration system and under positive pressure with low N₂ flow. Immediately after filtration, CDOM absorbance was measured in the spectral range between 240 and 750 nm at 1-nm intervals in a Lambda 800 (Perkin-Elmer) dual-beam spectrophotometer equipped with a 10-cm quartz cell, with pre-filtered (0.2 µm) Milli-Q water used as a reference blank. The CDOM absorption coefficients at 254, 325 and 443 nm were calculated as follows:

where a_filtrate(λ) is the absorbance of the filtrate, ι is the path length (in m) and 2.303 converts between log10 and natural log.

Temperature and salinity effects were corrected by subtracting a null point (mean value between 700 to 750 nm) from the entire scan (Mitchell et al. 2000). The wavelengths chosen for the study were 443 nm because of its use in remote sensing, and 325 and 254 nm because of their applicability to estimating dissolved organic carbon (DOC) concentrations (Weishaar et al. 2003, Baker and Spencer 2004).

For onboard determination of total particulate matter absorption coefficients [a_p(λ), with m⁻¹ as units], seawater samples of 0.85 to 2 L were filtered through 25-mm-diameter GF/F filters. Immediately after filtration, absorbance was measured from 300 to 750 nm at 1-nm intervals in a Lambda 800 (Perkin-Elmer) dual beam spectrophotometer. Absorbance was checked to remain lower than 0.4 (Cleveland and Weidemann 1993). The a_p(λ) were determined by the quantitative filter technique using the simple transmittance method of Mitchell et al. (2000). The blank scans obtained from filters wetted with filtered (0.2 µm) seawater were subtracted. Absorption coefficients of non-algal particles [a_nap(λ)] (i.e. non-chlorophyllous, unpigmented particles) were determined using the methanol extraction method of Kishino et al. (1985). Absorption coefficients of a_p(λ) (first measurement) and a_nap(λ) (measurement after methanol extraction) were estimated according to the equation

where a_filter (λ) is the measured absorbance with the quantitative filter technique, S is the clearance area of the filter, V is the volume of filtered water, and β(λ) is the amplification factor vector (Mitchell and Kiefer 1988). The β(λ) factor was calculated following Bricaud and Stramski (1990) with the equation

In order to correct for residual offsets in the sample filter relative to the reference and for scattering artefacts due to particle loading (Mitchell et al. 2000), the optical density of all spectra was corrected by subtracting the readings at 750 nm, a wavelength where absorbance by particles was assumed to be negligible.

Phytoplankton absorption coefficients [a_ph(λ)] were obtained by subtracting a_nap(λ) from a_p(λ). The Chl a–specific phytoplankton absorption coefficient [a_ph*(λ)] was estimated as a_ph(λ)/Chl a_tot. Finally, total absorption coefficients [a_t(λ)] were calculated based on the equation

where a_w(λ), the absorption coefficient of pure water, was taken from Pope and Fry (1997) and from Morel et al. (2007) for the UV-blue range.

In general, a_ph*(λ) depends not only on the pigmentary composition of the cells, but also on the so-called package effect, which consists of a reduction in the absorption of pigmented particles relative to the absorption of the pigments in solution. The package effect is stronger when either cell size or intracellular pigment concentration increase, and can be estimated by means of the package index Qa^*(443). This index can be calculated as the ratio of the phytoplankton absorption a_ph(λ) and the absorption of pigments in solution a_sol(λ), following the approach of Bricaud et al. (2004), based on the weight-specific absorption coefficients proposed by Goericke and Repeta (1993) and the work by Bidigare et al. (1990):

a_pig(λ)=S C_i a^*_sol,i(λ)

a_sol(λ)=a_pig(λ) + a_miss(λ)

where a_pig(λ) is the sum of pigment-specific absorption coefficient of the i-th pigment [a^*_sol,i(λ)] multiplied by their concentrations (C_i, mg m^–3) in the medium, The a_miss(λ) expression is a so-called “missing term” that depends on total Chl a (Chl a_tot) concentration [a_miss(λ)=0.0525*Chl a_tot^0.85; Bricaud et al. 2004]. The Qa^*(λ) index can increase from 0 (maximum package effect) to 1 (no package effect).

In order to assess the effect of accessory pigment composition on phytoplankton absorption properties, we grouped the carotenoids into two categories: non-photosynthetic carotenoids (PPC), including Zea, Dax, Ddx and β-carotene, and photosynthetic carotenoids (PSC), comprising Fuco, Per, 19'But and 19'Hex (Bricaud et al. 2004).

The Chl a concentrations derived from the Uitz et al. (2006) and Hirata et al. (2011) estimates (hereafter referred to as VU and HI, respectively) for the different size categories were compared with the SFF results using the standardized major axis (sma) regression method (Falster et al. 2006), which is appropriate (Warton et al. 2006) for testing whether the slope fits a specific value (1 in this case).

In order to summarize the pigment composition of the samples, a principal component analysis (PCA) was performed on the correlation matrix among the concentrations of the 11 most abundant pigments plus the biomarkers Allo, Neo, Per, Pras and Viol (Légendre and Légendre 1998). The concentration data (x_i) were transformed according to the expression y_i= log(x_i + 0.1), where 0.1 is the minimum non-zero concentration measured in the data set.

The TransPEGASO transect crossed seven biogeochemical provinces between MEDI and SWAS (Fig. 1), encountering a wide range of hydrographical properties (Fig. 2, Table 3). Temperature (Fig. 2A) was about 21°C in MEDI, increased to mean values (Table 3) of 28.7°C and 28.2°C in NATR and WTRA, respectively, and decreased to 8-14°C in SWAS. Salinity (Fig. 2A, Table 3) ranged from 34.8 to 36.8 between MEDI and CNRY, showed a minimum of 34.8 in WTRA, increased again in SATL (36.3-37.4) and decreased to 32.6-33.6 in SWAS.

Nitrate, nitrite, and phosphate concentrations (Fig. 2B, C, Table 3)) were generally low from MEDI to NATL and relatively high in SWAS, where the mean concentrations of these nutrients were respectively 4.08 µM, 0.13 µM and 0.57 µM. Silicate (Fig. 2C) concentration was low (0.2-1.42 µM) and very variable across the whole transect. The N:P ratio [(nitrate+nitrite)/phosphate] was always lower than the Redfield value of 16 (Table 3).

Hereafter, unless otherwise stated, for simplicity Chl a concentrations will refer to the HPLC determinations. Chl a_tot and the sum of the HPLC-measured Chl a_pico<3 and Chl a_n+m≥3 classes (Chl a_pico+n+m) were correlated (Fig. S1B) (Chl a_pico+n+m=0.85+ Chl a_tot + 0.005; n=41, r²=0.95, p<0.0001; one outlier was excluded) but the slope was <1 (the 95% confidence limits were 0.79-0.92).

The highest Chl a_tot concentrations (Fig. 2D, Table 1) were recorded in the shelf waters of SWAS (mean±SD, 1.40±0.54 mg m^–3), followed by those of MEDI (0.47 mg m^–3 Chl a_tot). In the remaining regions, average Chl a_tot ranged from 0.06±0.02 mg m^–3 (SATL) to 0.20±0.02 mg m^–3 (NATR). The Chl a_pico<3 and Chl a_n+m≥3 classes followed a similar pattern (Fig. 2D), with the highest values in SWAS and MEDI. No significant correlations were obtained when the proportion of Chl a in the picoplankton size fraction (f_pico<3) was compared with Chl a_tot (Fig. S1C). The differences among average f_pico<3 (Fig. 3A) for the different provinces were not significant (Kruskal-Wallis test).

The main phytoplankton pigments determined by HPLC in each province (and the name abbreviations used hereafter) are presented in Table 1 for whole water and in Suplementary material Tables S3 and S4 for the Chl a_n+m≥3 and Chl a_pico<3 classes, respectively. Aside from MV-Chl a, the most important pigments present in the whole community (Fig. S2, Table 1) were Chl b, Fuco and 19’Hex in SWAS, and Fuco, 19’Hex and Chl c2 in MEDI. CNRY was characterized by the dominance of Zea, Phaeop, 19’Hex and DV-Chl a and the most important pigments in the other provinces were Zea, DV-Chl a and 19’Hex. The most abundant pigments (excluding MV-Chl a) in Chl a_n+m≥3 included Fuco (dominant in MEDI and SWAS), Phaeop and 19’Hex (Fig. S2, Table S3). In Chl a_pico<3 (Fig. S2, Table S4), Zea and DV-Chl a (in varying order of abundance) were the most important pigments (apart from MV-Chl a) in NAST-E, NATR, WTRA and SATL, followed by 19’Hex; Zea and Phaeop dominated in CNRY, Chl b and 19’Hex in MEDI, and Fuco, 19’Hex and Chl b in SWAS.

The PCA returned two principal components with >1 eigenvalues, which together explained 85.5% of the total variance of the 16 pigment descriptors (Table 4). The first principal component (PC1) had correlations exceeding 0.82 with all pigments except Zea and DV-Chl a, which were negatively correlated with it (Table 4). The second principal component (PC2) showed a strong positive correlation with Zea (0.96) and correlation coefficients <0.3 with all the other pigments (Table 4). The scores of PC1 (Fig. S1D) expressed the contrast between the high Chl a provinces (MEDI and SWAS) on the positive side and the other provinces in the negative one. PC2 appeared to be mainly related to the variability of Zea in the SWAS province, although it also singled out a CNRY sample with relatively low zeaxanthin concentration (Fig. S1D).

The ratio Ddx/LHC ranged from a low value of 0.09 in MEDI to a peak of 0.40 in SWAS and showed fluctuations that appeared to be linked to the morning-afternoon alternation of the sampling times (Fig. 4A); there was a weak but significant correlation (n=40, r²=0.14, p<0.05; Fig. S3A) between solar radiation and the Ddx/LHC ratio. Both the PPC/Chl a_tot (data not shown) and the PPC/PSC (Fig. 4B) pigment ratio peaked in the tropical and subtropical Atlantic and showed their lowest values in MEDI and SWAS.

All CHEMTAX groups except Synechococcus and Prochlorococcus showed generally low abundances in the tropical and subtropical Atlantic, and peaked in SWAS and to a smaller extent in MEDI (Figs 5, 6, 7, Table 5). Prochlorococcus showed maximum concentrations in WTRA and did not occur in SWAS (Fig. 5A), and Synechococcus peaked in NATR around 15.1°N, and at several SWAS stations (Fig. 5B). Haptophytes represented a substantial part of the phytoplankton community in all provinces and diatoms and prasinophytes were important in MEDI and SWAS (Fig. 7A). Superimposed on these general patterns, there was noticeable heterogeneity within stations of the same province, in particular in SWAS, which showed a haptophyte Chl a maximum at station 34 (39.56°S) (Fig. 6D) and a diatom peak evidenced by both cell counts and Chl a concentrations at station 36 (–43.43°S) (Fig. 6B).

In the n+m≥3 size class (Fig. 7B), the most important CHEMTAX groups were diatoms and haptophytes in MEDI, NAST-E, CNRY, NATR and SWAS; haptophytes, chlorophytes and dinoflagellates in WTRA; and haptophytes, dinoflagellates and prasinophytes in SATL. The pico<3 class (Fig. 7C) was dominated by prokaryotes in NAST-E, NATR, WTRA and SATL but also showed a high contribution of diatoms in CNRY, of diatoms and prasinophyes in MEDI and SWAS, and of haptophytes in all provinces.

There were significant correlations between the CHEMTAX-derived contribution to Chl a_tot and cell counts (microscopy for eukaryotes and flow cytometry for the prokaryotes and FC-nanoeukaryotes) of cryptophytes (n=37, r²=0.20, p<0.01), dinoflagellates (n=29, r²=0.24, p<0.01), Prochlorococcus (n=31, r²=0.70, p<0.0001) and Synechococcus (n=41, r²=0.44, p<0.0001, one outlier excluded). FC-nanoeukaryotes were also correlated with haptophyte Chl a (n=39, r²=0.90, p<0.0001). The global correlation was also significant for diatoms (n=29, r²=0.15, p<0.05), but there were marked discrepancies between the Chl a contribution and the cell counts for this group at some stations of CNRY and SWAS (Fig. 6B).

According to microscopy (Table S5), the most abundant diatom taxon in MEDI was Guinardia striata. The diatom peaks of stations 9 and 10 (19°02′W, 20°33′02″N and 19°39′26″W, 19°19′54″N; respectively) of CNRY were mainly contributed by Chaetoceros spp. <20 µm and Pseudo-nitzschia spp. (thin). In the other provinces of the Subtropical and Tropical Atlantic, diatoms were represented by low densities of Cylindrotheca closterium, Hemiaulus hauckii, Thalassiosira/Porosira spp. >20 µm, Proboscia alata and Pseudosolenia calcaravis. The most abundant diatom taxa in SWAS were Chaetoceros criophilus, C. lorenzianus, Chaetoceros <20 µm, Thalassiosira <20 µm and Eucampia spp. The CHEMTAX algorithm indicated that, on average, diatoms were dominant in SWAS (Fig. 7), but for same stations of this zone (Fig. 6B) this abundance was not associated with high cell counts. Flow cytometric measurements revealed a substantial increase in pico- and FC-nanoeukaryotes at several stations of SWAS (Fig. 5D); a high presence of cells in the picoplankton size range could also be seen by optical microscopy in some SWAS samples (data not shown), but their concentration could not be properly quantified. SEM examination of water from these samples revealed the presence of the diatom Minidiscus sp. and of Triparma laevis, a member of the Parmales (Fig. S4), a phytoplankton group genetically very close to the diatoms, with which it shares a similar pigmentary suite (Ichinomiya et al. 2010). The Parmales have small cells with siliceous plates and are ubiquitous in marine waters (Ichinomiya and Kuwata 2015, Ichinomiya et al. 2016). Both Minidiscus and the Parmales (Kaczmarska et al. 2009, Jewson et al. 2016) have sizes of 2-3 μm.

Most haptophytes large enough to be counted by microscopy were likely to be pooled into the “Nanoflagellates 3-20 µm” category (which comprised the majority of “Other” in Table S5), but part of them were identified as Phaeocystis colonies or coccolithophores. Colonial Phaeocystis occurred at stations 33, 35 and 36, in the northern part of SWAS, at concentrations ranging from 5 10⁶ to 89 10⁶ in terms of individual cells L^–1 (data not shown). Coccolithophore concentrations ranged between 70 and 30000 cells L^–1 throughout most of transect (Table S5), but reached a peak of 0.6 10⁶ cells L^–1 at station 36 (SWAS). A total of 18 coccolithophore taxa were recorded; the dominant group were the “Unidentified small coccolithophores (<10 µm)” (mostly Emiliania huxleyi and probably Gephyrocapsa spp.), but Calcidiscus leptoporus, Discophera tubifera, Syracosphaera pulcha and Umbellosphaera irregularis could be relatively abundant, in particular in the tropical Atlantic Ocean,.

The CHEMTAX dinoflagellate group (Fig. 7) was present in all the provinces but contributed less than 16% of the total Chl a in the three size classes. Apart from the ubiquitous “Unidentified small dinoflagellates (<20 µm)” and “Unidentified large dinoflagellates”, a high diversity of dinoflagellate taxa (89 in total) was identified by microscopy in the tropical and subtropical Atlantic provinces, including Karlodinium spp. and several species of Ceratium, Dynophysis, Gonyaulax, Oxytoxum and Protoperidinium (Table S5). In contrast, only Dinophysis acuminata, Karlodinium spp., Prorocentrum balticum and Protoperidinium pacificum were present in SWAS, but at relatively high abundances. As mentioned for haptophytes, other CHEMTAX groups comprising cells of generally small size such as chlorophytes, pelagophytes and prasinophytes are likely to have been included in pooled microscopy categories such as the “Nanoflagellates 3-20 µm”.

Size structure estimation of the phytoplankton community based on SFF and diagnostic pigment approaches

The proportion of Chl a_pico<3 and Chl a_n+m≥3 derived from the SFF (Fig. 3A) and the size distribution for each province resulting from the application of the VU and HI procedures can be seen in Figures 3B and C. The relationships between the proportion of Chl a in the various size classes, as determined by SFF and the VU and HI methods (data not shown), were only significant (Table 6) between the nano+microphytoplankton from Hirata (f_n+mHI) and from the SFF (f_n+m≥3). However, there were significant linear relationships among the various estimates of the Chl a content in each size class with respect to total Chl a (Fig. S5). The correlations between the VU and HI picophytoplankton (f_pico) and microphytoplankton (f_micro) proportions were significant, but the standardized major axis (sma) slope of the regression between f_micro from VU (f_microVU) and f_micro from HI (f_microHI) was lower than unity (Table 6). The SI index (log-transformed) was weakly correlated with Chl a_tot and with f_pico<3 (Fig. S6A, B).

Average CDOM across the transect ranged from 0.02 m⁻¹ (WTRA) to 0.16 m⁻¹ (MEDI) at 443 nm [a_CDOM(443)], from 0.09 m⁻¹ (SATL) to 0.43 m⁻¹ (MEDI) at 325 nm [a_CDOM(325)] and from 1.07 m⁻¹ (SATL) to 1.90 m⁻¹ (MEDI and CNRY) at 254 nm [a_CDOM(254)] (Fig. 8A, Table 7). There was a significant correlation between CDOM absorption and Chl a_tot (Fig. 8B) [n=31, r²=0.16, p<0.05 for a_CDOM(443); n=31, r²=0.55, p<0.0001 for a_CDOM(325); n=35, r²=0.47, p<0.0001 for a_CDOM(254)]. The ratio between a_CDOM and Chl a_tot showed the lowest values in MEDI and SWAS, as shown in Figure S8 for a_CDOM(254).

At 443 nm, CDOM was in general the largest component of total non-water light absorption, in particular at the MEDI station and in some of the low Chl a provinces (Fig. 9). The average CDOM contribution ranged from 74% in MEDI and 69% in NATR to 34% in WTRA, while phytoplankton absorption accounted for 47% of the total in WTRA and SWAS and for 18% to 24% in the other provinces (Fig. 9, Table 7).

When the whole data set was considered, phytoplankton absorption at 443 nm [a_ph(443)] showed a positive correlation with Chl a_tot [Fig. 10A; a_ph(443)=0.098 + Chl a_tot^0.88, n=37; r²=0.93, p<0.0001]. The Chl a–specific absorption by phytoplankton [a_ph*(443)] appeared to decrease with increasing Chl a_tot (Fig. 10B, Table 7) and increased with increasing package index Qa*(443) (Fig. 10C). However, these relationships are difficult to evaluate statistically due to shared measurements in the dependent and independent variables (Dunlap et al. 1997). On the other hand, Qa*(443) was not correlated with Chl a_tot (data not shown).

To investigate the potential effects of the phytoplankton size distribution on Qa*(443) and a_ph*(443), we related these parameters to two size indicators, the SI index and f_pico<3, but the correlations were not significant (Figs S6B, S7A, C, D). To explore the influence of the chemotaxonomic structure of the phytoplankton community on a_ph*(443), this parameter was related to two descriptors of pigment composition, the ratio PPC/PSC between the non-photosynthetic (PPC) and photosynthetic (PSC) carotenoid concentrations, and the first two principal components of the pigment analysis; a_ph*(443) was weakly positively correlated with PPC/PSC (Fig. 11A) and negatively correlated with PC1 (Fig. 11B; r_s²=0.13, p<0.05); the relationship with PC2 (data not shown) was not significant.

Based on the observed Chl a concentrations, the oceanographic provinces encountered during TransPEGASO were operationally classified into a low Chl a (<0.5 mg m^–3) and a high Chl a (≥0.5 mg m^–3) category. The first or “oligotrophic” group includes waters of CNRY and of the Tropical and Subtropical Atlantic gyres (NAST-E, NATR, WTRA and SATL), and the second or “eutrophic” one includes SWAS and the single MEDI station. Nutrient concentrations supported this categorization. Average nitrate (Table 3) approached or exceeded 1 µM only in MEDI (0.8 µM) and SWAS (4.1 µM), average silicate ranged between 0.3 and 0.7 µM except in WTRA (1.1 µM), and average phosphate was lower than 0.12 µM in all provinces except SWAS.

During our survey, the CNRY stations could be considered as oligotrophic, although the province may be affected by sporadic offshore filaments of nutrient-rich waters originating in the seasonal NW African coastal upwelling (Longhurst, 2007, Menna et al. 2016). The MEDI sample was taken in the Alborán Sea, a zone of strong mesoscale variability, in which the influence of the Atlantic jet coming through the Strait of Gibraltar and wind forcing cause coastal upwelling and other hydrographical structures associated with nutrient fertilization and relatively high Chl a concentrations (Tintoré et al. 1991, Macías et al. 2007). The eutrophic condition of SWAS, already described in previous studies (Gibb et al. 2000, Vega-Moreno et al. 2012), reflects the strong hydrodynamism of this region, subjected to tidal mixing and the interactions of the Brazil and Malvinas Currents with the waters over the continental shelf. During our study, the influence of the Sub-Antarctic Surface Waters associated with the Malvinas Current were reflected in the temperature and salinity minima (Fig. 2A) and the high concentrations of nutrients (Fig. 2B, C) found in the southern part of the transect.

As discussed below, the differences in composition and size structure of the phytoplankton community were mainly linked to changes in Chl a_tot concentration. However, a substantial part of the pigment variability encountered during the cruise was associated with photoacclimation responses. The morning-afternoon changes of Ddx/LHC (Fig. 4A) and the positive correlation of this ratio with solar radiation (Fig. S2A) may reflect the photoprotective role of Ddx, as timescales for pigment synthesis are of minutes to hours for photoprotective carotenoids (PPCs) and of hours to days for light-harvesting pigments (Ferris and Christian 1991, Bidigare et al. 2014). Both high light and low nutrient conditions (Eisner et al. 2003) were likely to be responsible for the relative increases in the proportion of PPC/PSC (and PPC/Chl a_tot, data not shown) found in the oligotrophic provinces of our transect (Fig. 4B).

Comparable results concerning pigment and phytoplankton distributions have been reported in other studies. Gibb et al. (2000) sampled the Atlantic Ocean between 50°N and 50°S and, based on HPLC analyses, recorded a dominance of PSCs to PPCs from 50-30°N, to the north of the start of our survey, to 35-45°S. Acevedo-Trejos et al. (2018) used in situ data and trait-based modelling to find an increase of average phytoplankton size and other parameters when moving from low to high latitudes.

The VU and HI algorithms (Table 2) have been widely used to infer the size structure of phytoplankton communities (Taylor et al. 2011, Zeng et al. 2018); however, verification has been limited (see Brewin et al. 2014). In our data set, there was no correlation between the proportion of picophytoplankton estimated by VU or HI, and that obtained by SFF (Table 6). It must be noted, however, that in spite of the lack of relationship of the Chl a proportions, there was a significant correlation between the Chl a concentrations allocated to each size class by the different methods (Fig. S5); this arises (Goodwin and Leech 2006) because, other things being comparable, the value of the correlation increases with increasing variability among the observations, and Chl_a concentrations are more variable than proportions (coefficients of variation ranged from 18% to 132% for proportions, and from 301% to 710% for Chl a concentrations). Both the VU and HI approaches predicted a lower proportion of small cells than that found in the filtration results (Fig. 3A), as found by Brewin et al. (2014). Apart from possible methodological issues (see below), this discrepancy may have arisen because our SFF used a limit of 3 µm between the small and large categories instead of the 2 µm threshold of the VU and HI approximations, so the f_pico<3 fraction of Figure 3A includes part of the nanophytoplankton size class of the other methods. This difference is not trivial, as suggested by the extensive list of phytoplankton taxa with sizes ranging from <2 µm to 3 µm (Vaulot et al. 2008). Indeed, some groups generally considered as nanoplankton, such as haptophytes and pelagophytes, may have most of their cells in the 2 to 3 µm category (Cabello et al 2016, Fig. S3) or, as found for the haptophytes in some samples, may even include forms <2 µm. In our work, for example, the proportion of haptophytes in Chl a_pico<3 was quite high in all the provinces, and pelagophytes and prasinophytes in this size class were also well represented in MEDI and SWAS (Fig. 7). These findings agree with those of other studies indicating the relevance of organisms of these groups in the <3 µm size range (Massana 2011) and highlight the need to reach a wide operational agreement concerning thresholds for the smallest phytoplankton size class, which we suggest could be made for a larger size limit, perhaps 5 µm, as adopted by Mousseau et al. (2001), based on the obervation that this is the size of the smallest particles that mesozooplankton can efficiently graze (Fortier et al. 1994). In addition, a larger size limit would contribute to minimize experimental artefacts (e.g. due to clogging of filters).

All size distributions (Fig. 3) showed a greater proportion of nano- and microphytoplankton cells in MEDI and SWAS (together with NATR for the SFF data) than in the other provinces. For SWAS, however, there was a clear discrepancy between size classes estimated by SFF and by pigment algorithms. The small contribution of nano- + microphytoplankton determined by size fractionation (Fig. 3A) contrasted with the higher proportions derived from the VU and HI models (Fig. 3B, C), and with the high contribution of diatom Chl a in all the size classes (Fig. 7). The total Chl a (SDP) estimated by the VU algorithm was in good agreement with Chl a_tot from HPLC (SDP=0.96*Chl a_tot - 0.01, n=41, r²=0.88, p<0.0001, Fig. S3B), indicating that the discrepancy stemmed from a misallocation of part of the Chl_a pool between the different size classes. Optical microscopy observations showed relatively low concentrations of nano- and micro-sized diatoms in several SWAS stations, in which the diatom contribution to Chl a_tot was high according to CHEMTAX (Tables 5 and S5). This discrepancy is very probably due to the presence of Parmales and of picoplankton-sized diatoms such as Minidiscus, which could have been part of the high picophytoplankton densities found in SWAS. As the VU approach to estimating taxonomic composition through diagnostic pigments is based on the premise that organisms with diatom-type pigments belong to the microplankton size class, picophytoplankton forms with this pigment signature will be incorrectly classified as microphytoplankton.

Estimation of phytoplankton size based on pigment composition is challenging because many pigments are shared by small and large phytoplankton forms (Roy et al. 2011). In this context, application of CHEMTAX to HPLC-determined pigment concentrations allows for a finer classification of phytoplankton chemotaxonomic groups than approaches based on simpler algorithms. Methods such as that of Hirata et al. (2011) based on the relationship of the proportion of functional groups or size classes with total Chl a are influenced by the phytoplankton composition in the data sets used to derive the model parameters. On the other hand, filtration procedures have their own problems (e.g. due to cell breakage), especially when Chl a concentrations are low (Brewin et al. 2014). It has also been noted that the size fractionation technique tends to overestimate the smaller fractions because of the defective retention of larger cells with non-spherical shapes (Murphy and Haugen 1985).

The dominance of Zea and DV-Chl a in the oligotrophic provinces agrees with the prominent contribution of Synechoccus and Prochlorococcus to Chl a_tot and with abundance and biomass estimates derived from flow cytometric counts (Fig. 5B; see also Zamanillo et al. 2019). However, our SFF results indicate that pore sizes of 2 or 3 μm may retain a part of the Synechococcus population, suggesting the need for caution when interpreting fractionation data. Due to their small size, both Prochlorococcus and Synechococcus have relatively high surface-to-volume ratios and can thrive in nutrient-poor situations, although Prochlorococcus seems to be associated with more oligotrophic conditions than Synechococcus spp (Latasa et al. 2010). This observation could explain the higher relative importance of Synechococcus in CNRY, which may be affected by occasional upwelling filaments (Menna et al. 2016). The most abundant eukaryotic pigments corresponded in general to the haptophytes, which include both nano- and picoplankton forms (Vaulot et al. 2008). However, diatoms were important in all the size classes of CNRY, although high cell counts at some stations were not reflected into comparatively high Chl a concentrations, an observation that could be due, at least in part, to the small cell size of the dominant taxa (Chaetoceros spp. and thin Pseudo-nitzschia spp.). Due to their elongated shape, the thin Pseudo-nitzschia spp. would likely pass through 3-μm pore filters, helping to explain the high contribution of diatom pigments in the f_pico<3 fraction.

CHEMTAX results for MEDI and SWAS indicated that the main contributors to all the size classes were diatoms, haptophytes and prasinophytes (Fig. 7). Prokaryotic phytoplankton, probably outcompeted by other groups (Partensky et al. 1999), accounted for only a small proportion of Chl a_tot in both provinces. The absence of Prochlorococcus in SWAS agrees with a lower temperature limit of 15°C for this genus (Johnson et al. 2006). As mentioned above, the strikingly high proportion of diatom Chl a in the pico<3 class and the discrepancy between the microplankton contribution determined by size fractionation and estimated by the VU and HI models in SWAS could be explained by the presence of Parmales and picoplankton-sized diatoms such as Minidiscus, which appear to be more common than previously considered (Leblanc et al. 2018). These findings support the observations of Ferreira et al. (2013), who noted that the size of phytoplankton cells in Patagonian waters tended to be smaller than that found in other areas with a similar Chl a concentration.

Our microscopic observations revealed that the most abundant diatom taxa were several Chaetoceros species, small Thalassiosira (<20 µm) and Eucampia spp., whereas haptophytes were partially represented by Phaeocystis colonies and coccolithophores. García et al. (2008), in their study of a spring cruise in the Patagonian Shelf, noted the occurrence of large diatom and dinoflagellate blooms, accompanied by nanoflagellates including Phaeocystis cf. antarctica. Similarly, Ferreira et al. (2013) mentioned the dominance of nanoplankton-sized Thalassiosira spp. in the same area in spring, while Souza et al. (2012) studied a summer bloom dominated by the coccolithophore Emiliania huxleyi, the haptophyte Phaeocystis antarctica and other microalgal taxa.

Except in SWAS, a_CDOM(443) accounted for more than 50% of total non-water absorption (Table 7), in agreement with previous studies (Nelson et al. 1998, Gonçalves-Araujo et al. 2018); the relatively high a_CDOM(443) contribution in the MEDI sample coincides with the findings of Pérez et al. (2016). The relatively high values of the ratio a_CDOM(254)/Chl a_tot in the oligotrophic provinces (Fig. S8) may be due, at least in part, to a reduction of the microbial processing of DOC caused by nutrient limitation (Thingstad et al. 1997, Romera-Castillo et al. 2013). The positive relationship between a_CDOM and Chl a_tot, as other authors have noted (Nelson et al. 1998, Xing et al 2014), suggests CDOM production by phytoplankton, which are the ultimate source of dissolved organic matter in the open sea. Likewise, the low a_CDOM values at WTRA stations, the most distant from the continents, could be indicative of a terrestrial contribution to the a_CDOM signal of regions closer to land.

Similarly to Bricaud et al. (2004) and others, we found that a_ph(443) could be expressed as a power function of Chl a (a_ph(443)=0.098*Chl a_tot^0.88, n=37; r²=0.93, p<0.0001; Fig. 10A); the presence of outliers could be due to natural variations or to analytical artefacts. The exponent of the relationship was similar to that determined by Bricaud et al (2010) for southern Pacific waters [a_ph(440)=0.0617*Chl a_tot^0:93]. The decrease in the Chl a specific absorption coefficient, a_ph*(443) with increasing Chl a and the positive relationship between a_ph*(443) and Qa*(443), as seen in Figure 10B and C, are a common finding in absorption studies because, apart from potential statistical artefacts due to common variables in the calculation of ratios (Dunlap et al. 1997), high Chl a values tend to be associated with a lower contribution of accessory pigments and with larger cells (and therefore a higher package effect due to size) (Bricaud et al. 1995). We found (Fig. 10C) a positive relationship between aph*(443) and the package effect index Qa*(443), suggesting the occurrence of package effects.

The lack of correlation of both Qa*(443) and a_ph*(443) with SI and with f_pico<3 (Fig. S7), and the relationships of a_ph*(443), negative with PC1 (Fig. 11A) and positive with PPC/PSC (Fig. 11B), suggest that the variation of Qa*(443) and a_ph*(443) in our data set was mainly related to the chemotaxonomic structure (pigment composition) and photoacclimation processes (changes in the intracellular pigment concentration) of the different phytoplankton communities and that any potential effects of cell size were ultimately masked. This contrasts with findings of other works that showed that the size structure of the phytoplankton algal population was a major factor explaining variability of a_ph*(443) in surface ocean waters (e.g. Ciotti et al. 2002, Bricaud et al. 2004, Ferreira et al. 2013, Wang et al. 2014). One of the reasons for the discrepancy may be simply that the range of total Chl a concentrations considered in these studies is much wider and includes many measurements with higher Chl a concentrations (e.g. exceeding 5-10 mg m^–3) than those in our data set (our maximum Chl a_tot was 2.4 mg m^–3); as discussed above, for a given amount of scatter of the dependent variable, the value of the correlation increases with increasing variability among the observations of the independent variable (Goodwin and Leech 2006). On the other hand, our outcomes are in line with those of other studies where, in addition to the major contribution of size structure, either changes in intracellular pigment concentration or variations in pigment composition were also found to be important factors shaping a_ph*(443) (Brunelle et al. 2012, Ferreira et al. 2017, Kheireddine et al. 2018); it can be noted also that these works used a Chl a concentration range rarely exceeding 5 mg m^–3.