A total of 492 samples were chiseled from outcrops along the section. Samples were selected so as to be as fresh and unaltered as possible. This included chipping off weathered surfaces while in the field. Each sample was calibrated to height (Fig. 4). Samples then were split, with one portion powdered in an agate ball mill, and subsequently freeze-dried.

Each of the powdered samples was analyzed for bulk sediment stable isotope composition at the Stable Isotope Laboratory, University of Southampton, UK. A known mass (∼ 80 µg) was placed into a headspace vial, dried overnight, and flushed with helium. To each sample, 10 mL of 100 % phosphoric acid was added and allowed to react. The liberated CO2 gas was measured using an EUROPA Scientific GEO 20–20 mass spectrometer fitted with a microCAPS for carbonate analysis. Results are reported in standard delta notation relative to Vienna Pee Dee Belemnite (VPDB). An in-house standard of Carrara marble, calibrated to NBS-19 limestone, was measured multiple times to evaluate accuracy and precision. The external analytical precision (1σ), based on these replicate analyses, was 0.028 ‰ for δ13C and 0.057 ‰ for δ18O.

The amount of CaCO3 in each sample was calculated from the beam height response during isotope mass spectrometer measurements (Spofforth et al., 2010). The liberated CO2 gas, when squeezed up in the bellows, is measured and generates a current, the beam height. The pressure of CO2 gas is directly proportional to the beam height and therefore the mass of carbonate in the sample. Over 100 samples of pure CaCO3, with masses between 200 and 480 µg, were analyzed to establish a linear relationship between beam height and carbonate content (CaCO3=mx+b; R2= 0.94–0.99). Results were validated by analyzing 30 samples on a C-H-N-O elemental analyzer.

The un-powdered sample split was examined for calcareous nannofossils. Raw sediments were processed to prepare standard smear slides (Bown and Young, 1998). To assess the reproducibility of our counting methods in every single sample, a pivotal sample was prepared 10 times by two different operators. Repeated counts of the identical sample performed by different analysts gave similar results (sd < 2–5 %). Particle density estimates (Baccelle and Bosellini, 1965) were not carried out because samples have a high range in the terrigenous content (22 to 90 %). An increase or decrease in the silicoclastic component is mainly related to the major or minor efficiency of the chemical and mechanical weathering on land (Agnini et al., 2009). In the studied sediments, the variation in the amount of the terrigenous content through time has modified the density of the allochemic particle component. Consequently, calcareous nannofossil absolute abundances could not be estimated correctly using a homogeneous/constant particle density or by weighing the same amount of sediment for each smear slide. However, the scope of semi-quantitative counts performed in this study is to recognize the precise position of biostratigraphic biohorizons rather than use these data as a proxy of the paleoproductivity of taxa. Essentially, the identification of the appearance or disappearance of any given taxon is not affected by its stratigraphic abundance pattern, which obviously reduces the negative effect of the variable abundance of the silicoclastic component throughout the section. Samples were examined under a Zeiss light microscope at 1250× magnification. Calcareous nannofossils were determined using taxonomy proposed by Aubry (1984, 1988, 1989, 1990, 1999b), Perch-Nielsen (1985) and Bown (2005).

A total of 200 samples were examined, providing an average time resolution of ca. 25 kyr. A preliminary qualitative estimate of the abundance and preservation state of calcareous nannofossil assemblages was performed for all samples. An initial large batch (185) was analyzed primarily to provide biostratigraphic control for the Cicogna section, and the basic results have been presented by Dallanave et al. (2009). We re-checked and refined the positions of some biohorizons by examining 15 additional samples, primarily across some of the CIEs, such as B1/B2, PETM, H1 and H2, and K/X (Cramer et al., 2003). The calcareous nannofossil biostratigraphic schemes used by Dallanave et al. (2009) were those of Martini (1971) and Okada and Bukry (1980). The new zonal scheme of Agnini et al. (2014) is also used here. Biohorizon nomenclature follows that given by Agnini et al. (2014): base (B), base common (Bc), top (T) and top common (Tc).

Calcareous nannofossil biostratigraphic results are based on semi-quantitative analyses, which is based on counts of the number of specimens of selected taxa present in a prefixed area, 1 mm2 or three long traverses (modified after Backman and Shackleton, 1983). Calcareous nannofossil paleoecological results are instead based on relative abundances of calcareous nannofossil taxa (percent of the total assemblage), calculated from counts of at least 300 specimens.

To capture changes in calcareous nannofossil assemblages we also use a statistical approach. Principal component analysis (PCA) was preferred to other methods, such as the non-metric multidimensional scaling (MDS) procedure, for which a small number of axes are chosen prior to the analysis and the data are fitted to these dimensions (Hammer et al., 2001). However, non-metric MDS results were performed and are available as Supplement (Fig. S2). Multivariate analysis of variance (MANOVA) was carried out on our data set to determine whether significative differences are present among the three groups of samples recognized with PCA.

PCA and MANOVA were performed on the percentages of 15 subgroups using the statistical software package PAST ver. 2.17c (Hammer et al., 2001). The former analysis is often used for examining paleontological data (e.g., Buccianti et al., 2006; Kucera and Malmgren, 1998; Watkins and Self-Trail, 1992; Thibault and Gardin, 2010; Marino et al., 2008; Bordiga et al., 2015), as it can point out hypothetical variables (components) that explain much of the variance in a multidimensional data set. The first principal component accounts for the most variability in any data set examined. Each succeeding component has the highest variance possible relative to the preceding components (Hammer et al., 2001). This method also increases the symmetry, homoscedasticity and linearity of the data set (Aitchison, 1986). The chosen subgroups were Chiasmolithus, Coccolithus, Ellipsolithus, Discoaster, Ericsonia, Fasciculithus, Girgisia, Octolithus, Prinsius, Sphenolithus, Toweius, Rhomboaster/Tribrachiatus, Zyghrablithus, reworked forms, and “others”.

The bulk rock δ13C record for the Cicogna section can be described, in a general sense, as a long-term decrease of approximately 3 ‰, punctuated by a series of negative CIEs (Fig. 4). The most prominent low in δ13C coincides with the CMU.

Previously established polarity chron boundaries and key calcareous nannofossil biohorizons at the Cicogna section (Dallanave et al., 2009) provide a very good stratigraphic framework. Once placed onto a common timescale, in this case WO-1 (Westerhold et al., 2008), the δ13C record at Cicogna is fairly similar to those records generated using upper Paleocene and lower Eocene marine carbonate at other locations (Cramer et al., 2003; Zachos et al., 2010; Slotnick et al., 2012). This includes, for example, bulk carbonate δ13C records at ODP Site 1262, and DSDP Site 577 (Fig. 5) The relatively high δ13C values near the base of the Cicogna section document the late stages of the PCIM, which was centered within C25r (Fig. 1). The overall drop in δ13C across the section marks the long-term global decrease in δ13C that lasted through Chron C24n (Fig. 1). The record contains multiple negative shifts in δ13C. There is, however, an intriguing difference: across the Cicogna section, the long-term 3 ‰ shift in bulk carbonate δ13C values is generally offset from that in bulk carbonate δ13C records at sites 1262 and 577 by approximately -1 ‰.

The superimposed CIEs are considered to correspond to CIEs found in δ13C records from elsewhere, some of which represent known or inferred hyperthermal events (Cramer et al., 2003; Lourens et al., 2005; Nicolo et al., 2007; Zachos et al., 2010; Slotnick et al., 2012). There are three pairs of CIEs below the CMU (Fig. 4), as well as during the initial upper Paleocene long-term decline in δ13C. These correspond to the B1/B2, C1/C2 and D1/D2 CIEs documented by others (Cramer et al., 2003; Zachos et al., 2010). Notably, at Site 1262, the B1/B2 CIEs occur during middle C25n, and the C1/C2 CIEs occur at the start of C24r (Fig. 5). The same is true at Cicogna. Interestingly, at Cicogna, the B2 and C2 CIEs show greater magnitudes than the B1 and C1 CIEs, and these paired excursions are more pronounced than at all other locations examined to date. An additional paired CIE occurs in the uppermost Paleocene (Fig. 4). This may correlate to a fourth set of late Paleocene CIEs documented at Site 1262 (Zachos et al., 2010).

The lower Eocene portion of the δ13C record at Cicogna (Fig. 4) begins at the CMU, which marks the PETM (Giusberti et al., 2007; Dallanave et al., 2009). As at many locations, the PETM is characterized by a prominent negative CIE. The shift in δ13C at Cicogna is approximately -2.5 ‰, a decrease that begins abruptly at 28.7 m and returns more gradually to near pre-excursion values by about 33 m. From approximately 33 to 54 m, the δ13C curve shows a relatively smooth trend. At 54 m, a pair of CIEs begin, with the first pair having a magnitude of about 1.0 ‰. These are the H1/H2 events (Cramer et al., 2003), which occurred in the upper part of Chron C24r (Lourens et al., 2005; Zachos et al., 2010; Dickens and Backman, 2013; Dallanave et al., 2015). Above the H1/H2 CIEs, and within Chron C24n, are a series of smaller (0.4 to 0.6 ‰) CIEs. Those at approximately 60, 65 and 72 m are correlated with the I1/I2, J and K/X events, respectively. In summary, the δ13C record at Cicogna correlates with that at ODP Site 1262 (Zachos et al., 2010) and DSDP Site 577 (Dickens and Backman, 2013; Fig. 5), as well as at several other locations (Cramer et al., 2003; Slotnick et al., 2012, 2015b). This is important because it enables comparison and discussion between widely separated sedimentary records within a firm temporal framework.

The δ18O values range from -1.08 to -3.64 ‰ with a mean value of -1.96 ‰ and a standard deviation (1σ) of 0.50 ‰ (Fig. 4). However, at the broad scale, δ18O increases upsection, with Paleocene samples averaging -2.10 ‰ and Eocene samples averaging -1.89 ‰. This trend is noteworthy because δ18O values should decrease upsection if the composition of the CaCO3 was principally reflecting rising global temperatures through the early Eocene. The 1σ of δ18O values also increases upsection, being 0.33 ‰ across Paleocene samples and 0.56 ‰ across Eocene samples.

There is virtually no correlation (r2= 0.014; r= 0.12) between δ18O and δ13C values across all samples (Fig. 6). However, most “short-term” CIEs do display decreases in δ18O (Fig. 4). An interval of anomalously low δ18O values occurs from 39.9 to 40.9 m, where the spatic calcite was observed.

The CaCO3 content varies between 9.4 and 77.7 % across the sample suite, with a mean value of 54.3 % and a 1σ of 8.2 % (Fig. 4). Two important findings emerge from the CaCO3 content record. First, from 39 to 54 m, where we find limited variance in the δ13C curve, CaCO3 content averages 52.1 % with a 1σ of 4.9 %. Thus, while the average is similar to that calculated for the entire section, the standard deviation is much less. At Site 1262, the corresponding time interval is also characterized by limited variance in δ13C values and carbonate contents, the latter inferred from the abundance of Fe counts in XRF scans (Zachos et al., 2010). Second, across all samples, the CaCO3 content covaries somewhat (r= 0.29) with δ13C (Fig. 6). This is because several lows in CaCO3 content coincide with minima in δ13C, as is obvious for the B1/B2, PETM and H1/H2 events (Fig. 4).

Calcareous nannofossils are generally abundant, diverse, and moderately well preserved. The sole exception is across a 10 cm interval from 28.75 to 28.85 m, which corresponds to the onset of the CIE that marks the PETM. The three samples from this interval are virtually barren of calcareous nannofossils.

Secondary overgrowth of calcite can partially or wholly blur species-specific morphological features. Such diagenetic alteration, however, only marginally influences the relative, semi-quantitative and absolute abundance of calcareous nannofossil taxa (Toffanin et al., 2013). Calcite dissolution, on the other hand, can significantly affect the relative abundances of various calcareous nannofossils within a given volume of sediment. This is because the removal of more dissolution susceptible taxa, such as Toweius and holococcoliths, necessarily increases the abundance of less dissolution susceptible taxa, such as discoasters (Roth and Thierstein, 1972; Adelseck et al., 1973; Roth, 1983; Bornemann and Mutterlose, 2008; Toffanin et al., 2013). In general, moderate to strong calcite dissolution also decreases the total abundance of calcareous nannofossils within a given volume of sediment (Adelseck et al., 1973; Toffanin et al., 2011). In the Cicogna section, calcite overgrowth on discoasters is the prevalent process affecting calcareous nannofossil assemblages (Plate I). Most assemblages display high abundances (> 1000 specimens mm-2) and a high diversity, which include more fragile taxa. It follows that dissolution has not severely altered most assemblages in samples from the Cicogna section. Rather, the calcareous nannofossil record is considered to represent a genuine paleoecological signal.

Nannofossil assemblages from the Cicogna section display several general trends (Figs. 7–9). At the most basic level, there is a decrease in the total number of nannofossils (N mm-2) with decreasing age. Paleocene samples average approximately 2600 specimens mm-2, whereas Eocene samples above the H1/H2 events average approximately 1200 specimens mm-2. This decrease in abundance broadly corresponds to a change in calcareous nannofossil composition, as supported through a series of additional observations at the Cicogna section (Figs. 7–9):

Coccolithus and Toweius constitute nearly half of the assemblages considering the entire section. However, these genera show a clear decrease in abundance upsection, with a mean value of 60 % in Paleocene samples and 35 % in Eocene samples.

All assemblage data were used for PCA. This indicates that PC1 (41.3 %) and PC2 (14.7 %) together account for 56 % of the variance in the data set. The PCA graph (Figs. 10a, S1) shows that samples can be subdivided into three subgroups. The first two populations of samples are distinguished because of their different positions along the x axis (PC1). The third population is much more dispersed but a possible discrimination from the other two seems to be hypothesized because of its different position along the y axis (PC2). The use of a different statistical procedure, such as non-metric MDS, does not substantially change these results (Fig. S2). To further support the subdivision of the study samples in three subgroup, we applied MANOVA to our data set (Fig. 10b). The result clearly confirmed that Paleocene, PETM and Eocene samples are in fact isolated one from each other.

Polarity chron boundaries and calcareous nannofossil biohorizons (Table 1; Fig. 4) provide a solid stratigraphic framework for the Cicogna section. Calcareous nannofossil biohorizons, including additional ones defined here, align in same stratigraphic order when compared to other locations, such as ODP Site 1262 and DSDP Site 577 (Table 1; Fig. 11). The Cicogna section represents sediment accumulation between 57.5 and 52.2 Ma on the WO-1 timescale (Dallanave et al., 2009). The average SR was ca.15.2 m Myr-1, although this must have varied (Figs. 3, 11). The CMU, which marks the “core” of the PETM and ca. 80–100 kyr, showing a higher sedimentation rate than much of the remaining record (Dallanave et al., 2009; Krishnan et al., 2015).

Once placed into the above stratigraphic framework, the bulk carbonate δ13C profile documented at Cicogna correlates well with that generated at ODP Site 1262 (Fig. 5). In fact, it is similar to δ13C profiles generated at multiple locations (Figs. 2, S4), as long as records have been properly calibrated in both the depth and time domains. This includes accounting for core stretching and core gaps at scientific drilling sites, such as at DSDP Site 577 (Dickens and Backman, 2013), and accounting for changing strike and dip along land sections, such as done at Cicogna (Fig. 3). During late Paleocene and early Eocene times, the Cicogna section records the long-term decrease in δ13C. Superimposed on this drop were multiple, often paired, negative CIEs. The PETM definitively represents the most prominent CIE, but several other CIEs occurred before and after. Importantly, the relative positions of polarity chron boundaries, key calcareous nannofossil biohorizons and CIEs at Cicogna align well with those found at other locations (Table 1; Figs. 5, 11).

A clearly recognizable δ13C pattern spans the late Paleocene through the early Eocene at several locations (Cramer et al., 2003; Nicolo et al., 2007; Galeotti et al., 2010; Zachos et al., 2010; Slotnick et al., 2012, 2015b), although the total number of CIEs remains uncertain. At Cicogna, the problem lies in the interval surrounding the K/X event, which broadly corresponds to the start of the EECO (see discussion in Slotnick et al., 2012). We cannot confirm with our sample resolution whether a series of short-term, small-amplitude CIEs mark this time, an idea suggested from δ13C records of the Clarence Valley sections (Slotnick et al., 2012, 2015b). However, as at other locations, such as Site 1262, no significant CIEs occurred within the 1.6 Myr between the PETM and the H-1/ETM-2 event (Fig. 5).

The time-correlative δ13C template implies changes in the mean ocean δ13C of dissolved inorganic carbon (DIC). In turn, these compositional changes very likely represent variations in fluxes of highly 13C-depleted carbon to and from the ocean or atmosphere, such as changes in the release and storage of organic carbon (Shackleton, 1986; Dickens et al., 1997; Kurtz et al., 2003; Deconto et al., 2010; Komar et al., 2013). The δ13C record at Cicogna offers no direct insight on the location of this carbon (e.g., seafloor methane, permafrost, peat). However, it does support an important concept: the magnitudes of given CIEs appear somewhat related to one another and to the long-term δ13C record. In particular, the PETM occurred about halfway between the long-term high and low in δ13C, and heralded a relatively long time interval lacking CIEs. A generic explanation is that a very large mass of 13C-depleted carbon was injected from some organic reservoir into the ocean or atmosphere during the PETM, and that the reservoir needed to recharge for considerable time before the next injection (H-1/ETM-2) could occur (Dickens, 2003; Kurtz et al., 2003; Lunt et al., 2011; Komar et al., 2013).

The overall -1 ‰ offset of the δ13C curve between the records at Cicogna and at sites 577 and 1262 (Fig. 5) warrants brief discussion. It probably does not reflect wholesale diagenesis and resetting of the primary signal at any of these sections. Otherwise, a recognizable correlative δ13C record and well-preserved nannofossils (Plate I) would not be found at all three locations. In fact, it is difficult to modify the original δ13C composition of carbonate over appreciable distance (greater than several meters) in marine sedimentary sequences dominated by fine-grained calcite, even those now exposed on land as lithified rock, such as at Cicogna or in the Clarence Valley. This is because the carbon water / rock ratio remains low, because almost all carbon exists in carbonate, and because temperature minimally influences carbon isotope fractionation (Matter et al., 1977; Scholle and Arthur, 1980; Frank et al., 1999). Instead, the offset in the δ13C curves probably relates to differences in the composition of the original carbonate, a concept that we return to later.

However, local dissolution and re-precipitation of carbonate definitely has occurred in the Cicogna section. This can be observed in the overgrowths of secondary calcite on discoasters and Rhomboaster/Tribrachiatus (Plate I). This process should dampen the original CIEs, because on the meter scale, dissolution and re-precipitation of carbonate would involve δ13C gradients in the DIC of surrounding pore water (Matter et al., 1977; Scholle and Arthur, 1980). This may explain, in part, why the magnitude of early Paleogene CIEs in bulk carbonate records are often muted relative to those found in other carbon-bearing phases (Slotnick et al., 2015b).

The δ18O record at Cicogna is intriguing because many of the CIEs are characterized by negative excursions but absolute values of δ18O generally and unexpectedly increase upsection (Fig. 4). Similar results have been documented in bulk carbonate stable isotope records at other locations, such as ODP Site 1215 (Leon-Rodriguez and Dickens, 2010) and Mead Stream (Slotnick et al., 2012). Even the δ18O record of bulk carbonate at ODP Site 1262 shows minimal long-term change from the late Paleocene to the early Eocene (Zachos et al., 2010), the time when high-latitude surface temperatures and deep ocean temperatures presumably increased by 5–6 ∘C, and one might expect a > 1 ‰ decrease in the δ18O of marine carbonate.

Like previous workers, we cannot discount the notion that temperatures at low and high latitudes responded differently across the early Paleogene (Pearson et al., 2007; Huber and Caballero, 2011). Unlike for carbon isotopes, however, local dissolution and re-precipitation of carbonate should significantly impact the δ18O of marine carbonate. This is because the oxygen water / rock ratio would be high before lithification, and because temperature strongly influences oxygen isotope fractionation (Matter et al., 1975; Scholle and Arthur, 1980; Frank et al., 1999). In general, as calcite-rich sediments and surrounding pore water are buried to higher temperatures along a geothermal gradient, local dissolution and re-precipitation of carbonate shifts carbonate δ18O to lower values (above references; Schrag et al., 1995). It is likely that, during sediment burial, the bulk carbonate δ18O records in many lower Paleogene sections, including at Cicogna, have been modified. We suggest that a signal of surface ocean temperature changes remains in the Cicogna section, which gives rise to short-term δ18O excursions that coincide with CIEs and several known or suspected hyperthermal events. However, the entire δ18O record at this location likely has shifted to more negative values preferentially with increasing burial depth and age. This partly explains the observed relationship between bulk carbonate δ13C and δ18O, which lies along a trajectory expected for diagenesis (Fig. 6). A potential test of this idea would be to show that the overgrowths on nannofossils (Plate I) have a significantly lower δ18O than the primary core carbonate of nannofossil tests.

A detailed stable carbon isotope curve provides a powerful tool to place past changes in calcareous nannofossil assemblages into a highly resolved framework. This is because, as implied above, truly global changes in the δ13C composition of the ocean should occur within the cycling time of carbon through ocean, which is < 2000 years at present day and presumably for the entire Cenozoic (Broecker and Peng, 1982; Shackleton, 1986; Dickens et al., 1997).

Across the study interval at Cicogna, several calcareous nannofossil taxa appear or disappear (Table 1). Moreover, their abundances also change between these horizons (Figs. 7–9). One might hypothesize that these changes in nannofossil assemblages were related to the established (e.g., the PETM, H1/ETM-2 and K/X) and potential (e.g., the B1/B2, I1/I2) hyperthermal events that occurred during the late Paleocene and early Eocene (Figs. 1, 5). However, the timing between recorded evolutionary appearances and extinctions of calcareous nannofossils and perturbations in δ13C is variable (Figs. 7–9). For instance, several significant calcareous nannofossil changes observed close to H1/H2 hyperthermals (e.g., B T. othostylus, B S. radians, B S. villae, Tc D. multiradiatus, T T. contortus) predate these events. By contrast, several biotic changes observed close to the B1/B2 CIEs (e.g., B D. delicatus, Tc S. anarrhopus, B D. multiradiatus, T Ericsonia robusta) happened at the end of these events. The PETM seems to provide the only case when a negative CIE precisely corresponds to major changes in calcareous nannofossil assemblages.

Profound changes in calcareous nannofossil assemblages occurred across the PETM in several locations (Fig. 2), both in terms of relative abundances and increases in origination and extinction rates (Aubry, 1998; Bown et al., 2004; Raffi et al., 2005; Gibbs et al., 2006a; Agnini et al., 2007a; Self-Trail et al., 2012). At Cicogna, the assemblages show remarkable, though mostly transient, relative abundance variations across the PETM, including an increase in Coccolithus, a decrease in Zygrhablithus, Sphenolithus, Toweius and Prinsius, and an extinction of most fasciculith species (Fig. 8). Not surprisingly, these changes are very similar to those in the Forada section, which is also located in the Belluno Basin (Agnini et al., 2007a).

Although these changes in relative abundance of taxa alone represent a notable difference with respect to background conditions, most of the changes are transient and/or local when compared with other data sets (Bralower, 2002; Gibbs et al., 2006b; Agnini et al., 2007b; Angori et al., 2007; Mutterlose et al., 2007). For instance, an increase in abundance of Discoaster and Fasciculithus was reported for some of the PETM section studied (e.g., Bralower, 2002; Tremolada and Bralower, 2004; Raffi et al., 2009), but these assemblage variations were not observed in other sections (e.g., Gibbs et al., 2006; Agnini et al., 2007a; Self-Trail et al., 2012). The only global calcareous nannofossil assemblage features of the PETM are represented by the evolutionary appearance of Rhomboaster/Tribrachiatus lineage, the presence during the CIE of short-lived species such as Discoaster araneus, and the disappearance of several species of fasciculiths (Raffi et al., 2005; Agnini et al., 2007a).

While changes in calcareous nannoplankton assemblages during the PETM have been investigated at high resolution at different locations (e.g., Bralower, 2002; Gibbs et al., 2006b; Agnini et al., 2007a), the longer-term perspective in which such changes occurred during the early Paleogene has remained uncertain (Gibbs et al., 2012). The record at Cicogna provides this opportunity.

The PCA of calcareous nannofossil census data (%) indicates that two principal components (PC1 and PC2) account for most (56.0 %) of the variability in our 15 selected subgroups. Such analysis also permits the studied samples to be subdivided into two populations and a possible widely dispersed group (Fig. 10a). The first two populations are distinguished because of a major difference along the x axis representing PC1, whereas the third population seems to stand out because of a difference along the y axis representing PC2. Importantly, each of these three populations constitutes a homogeneous group in the time domain: group 1 includes all upper Paleocene samples (Paleocene samples and B1/B2 events); group 2 consists of almost all lower Eocene samples (Eocene samples, H1/H2 events and K event); and group 3 comprises samples that span the PETM (both core and recovery), and two samples that come from sediment deposited during the core of the H1 and B2 events (Fig. 10). These results indicate that late Paleocene calcareous nannofossil assemblages are statistically different in their composition from those of early Eocene samples. To check whether calcareous nannofossil assemblages across the PETM are statistically different from those of either the late Paleocene or the early Eocene, we performed MANOVA, which pointed out that ellipses containing 95 % of the data points for each group (late Paleocene, early Eocene and PETM) are virtually not overlapping one to each other, suggesting that three statistically different populations are recognized across the PETM, the late Paleocene and the early Eocene background assemblages, and the PETM fossil associations.

The general shift in the relative abundance of placoliths (i.e., Coccolithus, Toweius and Prinsius), the major component of the late Paleocene assemblages, to nannoliths/holococcoliths (i.e., Sphenolithus and Zygrhablithus), the major component of the early Eocene assemblages, largely explains the PC1 component or axis 1 (Fig. 10). By contrast, the dramatic shift toward negative values in the PC2 component or axis 2 during the PETM happens because of the increase in Ericsonia and reworking and the presence of Rhomboaster–Tribrachiatus plexus. Presumably, this relates to peculiar paleoenviromental conditions that developed during the event. One can hypothesize that this may have been a major difference in the physicochemical parameters of sea surface waters, such as higher temperatures, higher nutrient concentration or reduced carbonate saturation state.

Statistical analysis of our data from Cicogna does not highlight any prominent short-term changes in calcareous nannofossil assemblages, other than across the PETM and perhaps the B2 and H1 events. However, several biohorizons occur around the B1/B2 events. Specifically, these are the Bc Z. bijugatus, the brief high abundance of Octolithus spp., the evolutionary onset of the D. delicatus/D. multiradiatus lineage, the presence of the short-ranged E. robusta, the final radiation of late Paleocene fasciculiths (i.e., F. richardii group, F. hayi, F. lilianae, F. alanii), and the Tc of S. anarrhopus. All these happened at Cicogna and at ODP Site 1262 within Chron C25n (Agnini et al., 2007b; Dallanave et al., 2009; Fig. 11), which spanned only 0.54 Myr (Westerhold et al., 2008). These near-synchronous events are intriguing because while the various nannofossils represent only minor components of late Paleogene assemblages, they were destined to become either an abundant constituent of Eocene populations (e.g., Z. bijugatus and the D. delicatus/D. multiradiatus lineage) or extinct after having been a distinctive element of Paleocene assemblages (e.g., Fasciculithus spp. and S. anarrhopus). Following the PCIM, the long-term increase in temperature and decrease in δ13C (Fig. 1) coincided with a series of minor changes in nannofossil assemblages, which subsequently became important, presumably for evolutionary reasons.

Similar to the late Paleocene, calcareous nannofossil assemblages after the PETM do not show major rearrangements of common taxa during the early Eocene. Instead, minor components of these assemblages exhibit a sequence of closely spaced biohorizons. The sequence of these biohorizons is T Fasciculithus, B D. diastypus, B T. contortus, T T. bramlettei, Tc D. multiradiatus, T T. contortus, B T. orthostylus, B S. radians, T D. multiradiatus, B D. lodoensis, B G. gammation and Bc D. lodoensis (Table 1). Within the resolution of available paleomagnetic and δ13C data, all these biohorizons are virtually synchronous between the Cicogna section and ODP Site 1262 (Fig. 11). They also almost all occurred in near synchrony at DSDP Site 577 (Dickens and Backman, 2013), although the precise correlation remains uncertain, given problems with coring disturbance and subtleties in age models at this location.

Importantly, for stratigraphic purposes, the B and Bc of D. lodoensis are approximately coeval at all three locations and spaced apart by about 750 kyr. Unless one examines samples in detail, these two biohorizons can be confused and result in an erroneous assignment of early Eocene ages.

The evolutionary appearances and extinctions amongst early Eocene nannofossil assemblages may suggest the presence of uneven communities living in an extreme climate in which alterations of environmental conditions, even minor, might trigger evolutionary changes or prominent variations in abundances of a limited number of taxa that typically do not represent the dominant component of assemblages. A possible explanation is a generally higher tolerance of cosmopolitan taxa to variations in environmental conditions (Boucot, 1975; Winter et al., 1994). In contrast, highly specialized taxa that are adapted to a particular ecological niche may display greater sensitivity to modifications in the photic zone environment (MacArthur and Wilson, 1967; Pianka, 1970; Baumann et al., 2005).

In summary, several genera of calcareous nannofossils, such as Rhomboaster, Tribrachiatus, Sphenolithus, Discoaster and Zygrhablithus were, at least to some extent, affected during the late Paleocene–early Eocene transition, because they show an increased rate of taxonomic evolution (Fig. 11). However, these genera are all minor groups in terms of overall abundance, at least in most lower Paleogene sediment sequences, and they all belong to nannoliths and holococcoliths. It appears that these organisms were more sensitive to environmental changes than heterococcoliths, for example the cosmopolitan genera Coccolithus and Toweius.

Any comprehensive paleoenvironmental interpretation involving early Paleogene calcareous nannofossils remains tentative because many taxa, such as Rhomboaster/Tribrachiatus, Discoaster, Sphenolithus and Zygrhablithus, are extinct. Still, some single species or species groups are considered to be useful for reconstructions of paleoenvironmental conditions (Geisen et al., 2004). With that viewpoint, and with an understanding of modern holococcolith/nannolith ecology and classical biogeographical model, we provide a scenario regarding late Paleocene–early Eocene calcareous nannofossil evolution.

Modern holococcolithophores have numerous tiny rhombohedral calcite crystallites, and are considered as haploid stages of certain heterococcolithophores, which can live in just about any marine photic zone environment, although higher abundances and diversity are typical in oligotrophic settings (Billard and Inouye, 2004). The most common Paleogene holococcolith was Zygrhablithus bijugatus. This taxon has been interpreted as a K specialist more adapted to stable environments and oligotrophic conditions (Aubry, 1998; Bralower, 2002; Tremolada and Bralower, 2004; Agnini et al., 2007a; Self-Trail et al., 2012). Nannolith is a term used to describe peculiar morphotypes usually observed in association with coccoliths but lacking the typical features of heterococcoliths or holococcoliths. Ceratolithus cristatus, a modern nannolith, has been observed on the same cell together with Neosphaera coccolithomorpha (Alcolber and Jordan, 1997), suggesting that the nannolith stage (C. cristatus) corresponds to the holococcolith stage in other taxa (Young et al., 2005). Paleogene nannoliths include taxa with peculiar morphologies such as Discoaster, Fasciculithus and Sphenolithus. These genera often have been associated with warm waters and oligotrophic environments and are almost unanimously interpreted as K specialists (Haq and Lohmann, 1976; Backman, 1986; Wei and Wise, 1990; Bralower, 2002; Gibbs et al., 2004; 2006a, b; Agnini et al., 2007a). K specialists fluctuate at or near the carrying capacity (K) of the environment in which they thrive (MacArthur and Wilson, 1967), and are usually characterized by long individual life cycles and low reproductive potential. The K-specialist strategy is advantageous in highly stable, typically oligotrophic environments, which allows the evolution of stenotopy and where organisms compete by specialization and habitat partitioning (Hallock, 1987; Premoli Silva and Sliter, 1999). The narrow range of adaptability to changes in habitat or ecological conditions stimulates a rapid speciation.

At present, it is commonly accepted that modern holococcoliths and nannoliths are not produced by autonomous organisms; rather, they are stages in the life cycle of coccolithophores. Moreover, the passage between the two stages may be triggered by environmental factors (Billard and Inouye, 2004).

Hence, though Paleogene holococcoliths/nannoliths have no direct descendants in present-day oceans, they may very well have shared similar physiological features and life cycles with modern taxa. Assuming this is the case, the increase in the relative abundance of holococcoliths and nannoliths at the expense of heterococcoliths as well as the higher rates of evolution shown by holococcoliths and nannoliths may suggest conditions in which highly specialized taxa could flourish and rapidly evolve. This scenario is consistent with the idea, based on laboratory and modern ocean data, that the calcareous nannoplankton response to environmental change is species- or group-specific rather than homogeneous across the entire assemblage (Riebesell et al., 2000; Langer et al., 2006; Iglesias-Rodriguez et al., 2008; Lohbeck et al., 2012). Variations in the thermal and chemical structure of photic zone waters may thus account for the observed changes in the early Paleogene calcareous nannofossil assemblages.

Like at Cicogna, well-preserved calcareous nannofossils dominate bulk sediment carbonate contents of early Paleogene strata at sites 577 and 1262 (Backman, 1986; Zachos et al., 2004; Dickens and Backman, 2013). Given that the nannofossil assemblages are fairly similar (Fig. 11), a very basic question returns: why is the overall early Paleogene bulk carbonate δ13C record at Cicogna less by approximately 1 ‰?

A variety of explanations for the δ13C offset can be offered. For example, sediments at Cicogna had greater amounts of organic matter, and during burial diagenesis, a fraction of this carbon was consistently added so as to decrease the δ13C of pore water DIC. We note, though, that Corg contents (wt %) at the proximal Forada section generally have values less than 0.1 wt % (Giusberti et al., 2007). Similar Corg contents are found at ODP Site 1262, where values range from 0.0 to 0.3 wt % (Zachos et al., 2004).

A cursory examination of early Paleogene bulk carbonate δ13C records from other sites of the North Atlantic/western Tethys region (e.g., sites 550 and 1051; Fig. 2) shows a commonality: these locations also display negative 0.5 to 1 ‰ offsets relative to correlative records at sites 577 and 1262 (Cramer et al., 2003). The δ13C of DIC in modern surface waters (< 100 m) ranges by about 2 ‰, because of the differences in temperature, primary productivity and water mass mixing (Kroopnick, 1985; Tagliabue and Bopp, 2008). Notably, however, gradients in δ13C of surface water DIC are gradual, such that large regions have fairly similar values. It is possible that bulk carbonate δ13C values in early Paleogene North Atlantic sections record lower values than locations near the Equator or in southern latitudes because of past ocean circulation.