Plant life form, function, and metabolic pathway have been found to dictate the chain length distribution of leaf waxes (ACL) in terrestrial plants32–34. Several studies have also provided evidence that climate factors (i.e., temperature and moisture) may modulate the ACL, consistent with the role of the leaf wax layer in regulating water loss of the plant. Positive (negative) relations between growth temperature (moisture availability) and n-alkane chain length (ACL) have been reported (e.g.,35–40). In our GL-1090 record, we can readily argue that temperature does not exert a dominant control on leaf wax chain length because higher ACL values are observed during cold glacial climates and not during warm intervals (Figure 2a).
The possible control of moisture on the ACL can be explored by considering long-term variations in rainfall patterns. In SE Brazil, rainfall is strongly controlled by the South American Summer Monsoon (SAMS)41,42, the intensity of which is primarily driven by precession-forced changes in insolation24,43. Accordingly, high austral summer insolation during late MIS6 and Termination II would strengthen SAMS activity and rainfall in SE Brazil (Figure 3b), while the opposite would occur during MIS 5e (Figure 3b and c). Further, abrupt hydroclimate change occurred in the tropics during glacial Terminations, and a strengthening of SAMS rainfall44–46 is largely reported, also linked to a southern placement of the Intertropical Convergence Zone (ITCZ) and decreased Atlantic cross-equatorial heat transport47,48. Data-model assessments indicate drier austral tropics during the Last Interglacial, with an accentuated decline of summer and spring and almost unchanged autumn/winter rainfall in South America25. Likewise, a speleothem δ18O record from the Amazon region (and also under the influence of SAMS) shows a relatively weak monsoon during MIS 5e in phase with low austral summer insolation49.
In core GL-1090, terrigenous inputs traced with soil-derived brGDGTs (Supplementary material; Supplementary Figure 1) are ultimately controlled by changes in the rainfall rate in SE Brazil. Their pattern indicates stronger runoff (intensified SAMS) during Termination II and lower rainfall (weakend SAMS) during the Last Interglacial (Fig 3d), supported by complementary multi-proxy magnetic and isotopic studies50,51. Terrigenous supply from runoff is thus consistent with changes in rainfall rates dictated by austral summer insolation patterns (Figure 3b). The inferred decrease in monsoonal rainfall from the end of Termination II to MIS 5e is accompanied by a decrease in ACL values in core GL-1090 (Figure 3b), suggesting that a precipitation deficit is not pressuring plant life forms. This pattern strongly suggests that rainfall changes are unlikely the leading factor behind the observed ACL variation.
Having ruled out temperature and precipitation as the primary controls on ACL variations, we argue that the ACL record likely reflects vegetation changes in the adjacent hinterland. Among C3 plants, deciduous and evergreen broadleaf trees (angiosperms) show overall lower ACL values (ca. ≤ 29-30) compared to herbaceous and conifer vegetation (ACL > 30), and C3 savannah plants produce longer-chain n-alkanes (higher ACL) than the C3 rainforest plants33,52. We thus propose that the ACL in core GL-1090 traces changes in the Atlantic Forest in SE Brazil with lower (higher) ACL indicating expansion (retraction) of the forest. A previous record in the study area has suggested using ACL as a proxy for changes in the Atlantic Forest53. In order to further constrain the ACL as a tracer of vegetation (Atlantic Forest) changes, we compared our results with available pollen data from the Colônia crater in SE Brazil located only a few kilometers from the shoreline30. The higher (~ 30.4) ACL values during late MIS 6 are consistent with the dominant presence of grassland vegetation composed of an association of Poaceae, conifers (Araucaria and Podocarpus), and shrubs30 as the ones noted today at high elevation in the state of Paraná (southern Brazil)54. Toward the Last Interglacial, the vegetation composition fluctuated with the same species as during the penultimate glacial, except for the Poaceae, which decreased sharply, allowing shrubs to expand55. During MIS 5e, trees replaced shrubs, as shown in the high arboreal pollen frequencies30, and this trend is mirrored in the concurrent smooth decreasing of ACL (Figures 3c and e).
Such a coherent variation between ACL and arboreal pollen data, already observed in other studies (e.g.56), strengthens our interpretation of ACL as a proxy of Atlantic Forest variations in SE Brazil. In this sense, the Last Interglacial likely shows the largest expansion of the arboreal cover of the Atlantic Forest over the studied period. Upon the demise of the Last Interglacial and towards MIS 4, a linear increasing trend is embedded in our ACL, with values reaching close to those of MIS 6 (Figure 3c). Throughout this interval, an orbital pattern is clearly distinguished following MIS 5 substages, with MIS 5d and b (MIS 5c and a) presenting a pattern toward higher (lower) ACL. Superimposed on this orbital scale, noticeable millennial-scale ACL negative excursions through MIS 5 substages likely indicate periods of short-term Atlantic Forest expansions briefly interrupting this long-term retraction.
Our data require a mechanism to account for the inferred variations of the Atlantic Forest vegetation distribution on orbital and millennial time scales. Here, we propose that these variations were primarily driven by the global atmospheric concentration of carbon dioxide, whereby elevated atmospheric CO2 enhances forest plant growth through the “fertilization” effect (increased photosynthesis) and promotes decreased stomatal conductance that reduces transpiration (water loss)8–10. The role of this mechanism and its effect on biomass gains, biodiversity, and mortality rates, mainly over long-term periods, is still uncertain57. Modeling results with a potential vegetation model for Brazilian biomes indicated an increase in humid forest cover for future high-emission scenarios, although this tends to be counteracted by decreasing precipitation and rising temperatures57. Based on our record and the complementary pollen data, we argue that raising atmospheric CO2 during the penultimate deglaciation promoted the expansion of the Atlantic Forest in SE Brazil during MIS5e (Figure 3c). Interestingly, a similar conclusion is drawn by a pollen record combined with leaf wax data from the SE African margin. According to the authors, the development of interglacial woodlands would be primarily favored by higher CO2 levels that allow decreased stomatal conductivity, relieving drought stress17.
Aside from orbital variation, the ACL in core GL-1090 also varies on millennial time scales, revealing rapid changes in the SE Brazil biomes. During glacial intervals, the abrupt climate oscillations between the so-called Dansgaard-Oescheger (DO) stadial and interstadial conditions (Figures 3a and f) are linked with precipitation changes in (sub)tropical South America24 that have the potential to promote a vegetation response. However, the wetter climate during Dansgaard-Oescheger (DO) stadials, shown in the Botuvera cave δ18O record, does not involve lower ACL (i.e., forest expansion) (Figures 3a and c). In contrast, short-term forest expansion, denoted by lower ACL, is observed in DO interstadials (e.g., 19-21 and, to a lesser extent, 23-24 (Figures 3a and c)). The negative ACL excursions are nearly synchronous with millennial-scale CO2 increments of ~10 – 20 ppm (Figures 3a and c), implying the role of CO2 in driving a rapid Atlantic Forest response. Furthermore, the link between ACL (Atlantic Forest) and CO2 is endorsed by the isospectral correlation, which yields a strong significant correlation between them (R > -0.80, p < 0.01) (Figure 3g).
Further, wavelet coherency analysis between ACL and CO2 shows a well-defined anti-phased (180 º) signal taking place at precessional variability (~ 20 ka), which means that on orbital time scales, CO2 and arboreal Atlantic Forest are synchronous. A wide coherency region at ~ 5 – 10 ka time scales could be related to large background changes of global atmospheric CO2 since they occur at main transition periods (i.e., Termination II and MIS 5/4 boundary) (Figure 3g). The phase-angle during Termination II could still indicate a slight leading of atmospheric CO2 related to the Atlantic Forest, which aligns with our interpretation.
Biome changes may impact the stability of the underlying soils as increasing vegetation cover (forest expansion) contributes to stabilizing and reducing soil erosion. The main peaks in soil material (brGDGT concentration) occur in periods of reduced arboreal cover (higher ACL) and enhanced precipitation (Figures 3a, c, and d), with this being conditioned by the core stratigraphy, disregarding chronological uncertainties (Figure 2). In this sense, our data highlight the dynamic response of the Atlantic rainforest and underlying soils in SE Brazil induced by global CO2 changes.
Aiming to quantify the effects of CO2 over biomass gains, we ran potential vegetation simulations with the CPTEC-PVM2 initialized with the 1961–1990 climatology61, which reasonably reproduces the main South American biomes29 (Figure 4). We evaluated the distribution of net primary productivity (kgC/m²yr) in experiments where only CO2 is changed from penultimate glacial (180 ppm) to Last Interglacial (280 ppm) levels (Figure 4a). Simulations indicate that regions of tropical evergreen forest, which under the current climatology occur in the Amazon region and SE Brazil, suffer a loss in net primary productivity under penultimate glacial CO2 concentration compared with the present (Figure 4b). Otherwise, when atmospheric CO2 concentration is set to the Last Interglacial level, it favors a biomass gain of roughly 0.16 – 0.24 kgC/m²yr relative to glacial conditions (Figure 4b).
Future projections of vegetation cover in a range of conditions where only CO2 or CO2 and climate (temperature and precipitation) are considered suggest that tropical biomes are maintained or increase their productivity if the long-term CO2 fertilization effect plays a role29. This positive feedback caused by atmospheric CO2 is only countered if the dry season is longer than four months29. However, no observational evidence suggests that the CO2 fertilization effect will likely play an important role in tropical forests on longer time scales29. In this sense, our data will work as long-term evidence that the CO2 fertilization effect is a pervasive mechanism operating on millennial to orbital time scales. Evidently, the role of precipitation and temperature cannot be neglected29. Monitoring Brazilian non-Amazon tropical forests shows that recent drying and warming trends in the southeastern region have increased the annual carbon loss, turning carbon sink areas into carbon sources62. Considering this view, we hypothesize that past millennial-scale decreases in precipitation or even the drier condition of the Last Interglacial for the southern tropics25 still yielded a sufficient rainfall supply to allow the CO2 fertilization effect to occur. In this scenario, dry seasons likely do not exceed four months, and the range of CO2 variability is not large enough to cause saturation within enzymatic scales29. Our study sheds light on the role of humid forests as carbon sinks during past periods of natural warming, working likely as negative feedback to slow down global warming and capture carbon released from the deep ocean. Deforestation of remaining Atlantic Forest areas seriously threatens any future carbon-induced biomass gain, urging protection and reforestation initiatives to preserve its natural long-term capacity in carbon storage.