Growth of the Marine Carbon Reservoir

Fig. 1A displays the carbon-isotopic record in Meishan, China (5). Immediately preceding the extinction event, the isotopic composition of carbonate carbon declines by about 7‰ during a period of about 100 thousand years (Kyr), first slowly, and subsequently rapidly. At the same time, the isotopic composition of organic carbon also changes. We seek the physical fluxes in the carbon cycle that predict the chemical signals and . Of particular interest is whether the apparent downward acceleration of the carbonate signal can provide a quantitative indication of the underlying biogeochemical dynamics.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Carbon isotopic signals immediately preceding the end-Permian extinction in Meishan, China, indicate superexponential growth of the marine inorganic carbon reservoir. (A) Inorganic (, blue) and organic (, green) carbon isotopic changes (5). The dates at both ends of the time axis are resolved within Myr (2); in between, time is interpolated linearly in proportion to stratigraphic height. The peak extinction activity occurs at approximately Myr (2). (B) Reconstruction (solid blue line) of the dimensionless mass added to the marine inorganic carbon reservoir, assuming initial condition and an influx of isotopically light carbon with isotopic composition . (Inset) The same curve plotted on log-linear axes; the concave upward curvature suggests that grows superexponentially. The dashed red line compares to a superexponential growth law proportional to , where the critical time is coincident with the extinction peak.

To obtain such understanding, we consider the carbon cycle to be a simple exchange between globally mixed reservoirs of inorganic and organic carbon, via photosynthesis and respiration (20), and assume that the isotopic changes represent a perturbation from a preexisting steady state. We assume that the system is perturbed by an influx of isotopically light inorganic carbon of constant isotopic composition , forcing to decline out of its steady state. Because organic carbon is isotopically light, a decrease in its rate of burial is effectively indistinguishable from such light inputs. However, changes in organic burial can account for only a negligible fraction of the peak perturbed flux (SI Text). Changes in the burial flux of carbonate carbon are limited to an even smaller impact because its isotopic composition is identical to that of the inorganic reservoir. We therefore hold both burial rates constant to maintain simplicity. The mass of the inorganic reservoir then grows with time t in response to the increased inputs, by an amount , where is the initial, steady-state, size of the reservoir. The normalized perturbation is then straightforwardly related to changes in the geochemical signals by (SI Text)where determines the scale of the perturbation and the “force” F is a weighted sum of two nonsteady-state effects: the nonzero derivative and the departures of and from their unperturbed values. The solution of Eq. 1 for provides the normalized time-dependent perturbation of the marine reservoir of dissolved inorganic carbon (DIC). The way in which the DIC reservoir grows with time is a function of how it is forced out of its steady state; consequently, knowledge of can be used to test models of the end-Permian carbon cycle.

The form of Eq. 1 provides an immediate clue. Because accelerates sharply downward, is increasingly negative. Consideration of the unforced equation then suggests that M grows exponentially or faster. Numerical solutions of Eq. 1 confirm this view. Fig. 1B plots for the case , representative of the isotopic composition of remineralized organic carbon, assuming a constant sediment accumulation rate between the known dates (2). The curvature of the log-linear plot in Fig. 1B, Inset suggests that the growth of is faster than exponential. (Although the total quantity of light carbon required for the isotopic excursion depends on (1, 2, 6, 9), the shape of is independent of when, as here, is much smaller than .) Superexponential growth of the marine inorganic carbon reservoir implies that the carbon cycle behaves nonlinearly. As we show later, a simple dynamical mechanism containing a leading-order nonlinearity predicts superexponential growth proportional to . Such a growth law, where is coincident with the extinction peak, is given by the red-dashed line in Fig. 1B. The burst near —an incipient singular blow-up—can be traced back to the rapid downward acceleration of the carbonate signal, a feature that is not exhibited by the linearly decreasing isotopic excursions of the Early Triassic (21).

Fig. 2 combines this analysis with the analogous carbon isotopic signal in the Gartnerkofel-I core drilled in the Carnic Alps, Austria (3, 4). Cyclostratigraphic analysis of the Gartnerkofel core indicates that the sediment accumulation rate over the corresponding interval is approximately constant (22). When the accumulation rate is set to 22.5 cm/Kyr, approximately within a factor of 2 of the earlier estimate (22), reconstructions of for Gartnerkofel are qualitatively similar to that for Meishan. Fig. 2, which superposes both reconstructions in linear, log-linear, and log−log plots, confirms the inferences already drawn from the Meishan data. Moreover, the similarity of the Meishan and Gartnerkofel reconstructions validates our assumption of a constant accumulation rate at Meishan. We conclude that grows no slower than exponentially, and likely superexponentially.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Rescaled reconstructions of the mass of the marine inorganic carbon reservoir for both the Meishan (red) and Gartnerkofel (blue) data, as a function of rescaled time , where Kyr and time advances to the right. (A) Plots of M vs. show that both curves behave similarly with respect to a critical time . (B) Plots of on log-linear axes suggest that both curves grow superexponentially. (C) Plots of on log−log axes compare well to a straight line (dashed black) with slope −1, suggesting that both reconstructions grow like .

These observations impose constraints on interpretations of end-Permian environmental change. For example, CO₂ released to the atmosphere from a single, massive Siberian coal−basalt eruption (10) would be mostly transferred to the oceans after about 10²–10³ y (23). The uptake rate would decrease with time as the oceans acidify, reducing their capacity to take up more CO₂ (23). would then grow sublinearly, qualitatively different from what is shown in Figs. 1 and 2. Alternatively, such an eruption could be more gradual, limited by the rate at which vents form to release overpressured gas arising from contact metamorphism in intruded sills (11). The pressure released by each vent decreases overpressurization. Wherever the pressure is lowered, the rate of vent formation would decrease, and would again be sublinear. It is possible, however, that vents were sufficiently dispersed in space so that their formation times were essentially independent and randomly distributed over some (possibly small) interval of time; in this case, one expects a roughly constant rate of CO₂ emission, and would grow linearly. Likewise, the release of methane from methane hydrates (1, 9) could be clustered in time, but there is no reason to expect total methane emissions to grow much faster than linearly, in part because any resulting warming—a potential positive feedback—would be only logarithmically sensitive to increased CH₄ and CO₂ levels (24). Each of these scenarios would produce geochemical signals that qualitatively differ from those predicted by exponential or superexponential growth (SI Text). The constraints provided by this reasoning derive only from the shape of , not the quantity of isotopically light carbon required for the event.

The Carbon Source and Its Remobilization

We seek a mechanism that can result in growth of that is exponential or faster. Because such an upheaval of the carbon cycle implies substantial changes within the microbial biosphere, we hypothesize that the perturbation arises from the emergence of a new regime of microbial metabolic activity. We show below how such a mechanism leads to the observed dynamics. Before doing so, we first identify two important ingredients of our hypothesis: a suitably large source of degradable organic carbon and the new metabolic pathway that would be favored for its consumption.

In terms of the modern carbon cycle (25), our reconstructions of require about 7,000–14,000 Gt of remineralizable organic carbon. In the Late Permian, a low-O₂ marine environment (5, 12, 13) would have increased the concentration of organic matter in sediments. The high values of immediately before the perturbation suggest that organic carbon was sequestered in sediments at a rate at least 50% greater than usual (SI Text), which, in modern terms (25), corresponds to excess sequestration of at least 0.08 Gt C yr⁻¹. Integrated over the first 175 Kyr of Fig. 1A, this corresponds to 14,000 Gt C. We suggest that a substantial fraction remained remineralizable and relatively labile compared with organic carbon in modern sediments, thereby overcoming the apparent limitations (2) posed by the relatively small size of modern pools of isotopically light organic carbon (26) (e.g., methane hydrates, fossil fuels, soil organic carbon, and peat).

The accumulation of sedimentary organic matter would have been especially sensitive to changes in the biosphere’s ability to metabolize the products of fermentation. Among these products, acetate provides a major growth substrate for methanogens. The conversion of acetate to methane by methanogenic archaea—acetoclastic methanogenesis (27)—begins by activating acetate to acetyl coenzyme A (acetyl-CoA). Carbon monoxide dehydrogenase (CODH) then catalyzes the cleavage of acetyl-CoA, after which, in common with the utilization of all methanogenic substrates, methyl-coenzyme M reductase (MCR) catalyzes the reduction of a methyl group to methane (27). The activation to acetyl-CoA within methanogens occurs via two different pathways in two distinct groups of organisms: Members of the family Methanosaetaceae use a single-step acetyl-CoA synthase (ACS) pathway, whereas some members of the genus Methanosarcina use a two-step acetate kinase (AckA)-phosphoacetyl transferase (Pta) pathway (27). The AckA/Pta pathway is more energy efficient, requiring only one ATP molecule per acetate molecule activated, whereas the ACS pathway requires two (28). Growth on acetate using the low-efficiency ACS pathway within Methanosaeta is thermodynamically possible because of unique, poorly understood innovations in their electron transport chain (29); this limitation may be responsible for their observed slow rate of growth (30).

Fournier and Gogarten (17) have recently shown that the high-efficiency pathway in Methanosarcina evolved via a single horizontal gene transfer event, probably from a clade of cellulolytic bacteria belonging to the class Clostridia, after the mid-Ordovician evolution of vascular land plants (450–500 Ma). This is the only methanogenic pathway shown to have evolved via gene transfer. It also appears to be a conspicuously recent event within the evolution of methanogenesis, as all other methanogenic pathways have a broader phylogenetic distribution implying much more ancient origins. Methanosaeta may be more widespread in modern low-acetate marine environments (31). However, the dominance of the high-efficiency AckA-Pta pathway at high acetate concentrations (32) combined with the greater growth potential of Methanosarcina suggests that conditions for the emergence of acetoclastic Methanosarcina—specifically, a low-O₂ marine environment (5, 12, 13) and the accumulation of sedimentary organic matter—would have been favorable in the Late Permian. Moreover, reports of significantly reduced marine sulfate concentrations (33⇓⇓–36) suggest that competition from sulfate-reducing bacteria would have been diminished, thereby amplifying the importance of methanogenesis in Late Permian marine sediments.

Phylogenetic Analysis

The relevance of acetoclastic Methanosarcina to the end-Permian event depends crucially on the timing for the ancestor of this group. To obtain an estimate for this date, we reconstructed archaeal phylogenies from 50 representative genomes and constructed relaxed molecular clock chronograms using PhyloBayes 2.3 (SI Text). Fig. 3A illustrates our results. To estimate the time of the last common ancestor of known acetoclastic representatives of Methanosarcina, we generated chronograms using four independent ribosome-based datasets, separately containing 29 concatenated universally conserved ribosomal proteins, 12 concatenated archaeal-specific ribosomal proteins, 16S ribosomal RNA, and 23S ribosomal RNA. The phylogeny of these ribosomal components reflects the vertical cellular history of the Methanosarcinales, including Methanosarcina and its descendants, the recipient lineage of the ackA/pta transfer. We further assume that ribosomal sequences evolve in a relatively clock-like manner, providing more reliable dates than most other genes in the absence of internal calibration. Our four independent age estimates, shown by the gray bars in Fig. 3A, are consistent with each other. Combining them together yields the joint estimate depicted by the black bar in Fig. 3A, strikingly close to the end-Permian extinction (SI Text). The discrepancy with a previous estimate (37) derives from the earlier use of an autocorrelated clock model that is less reliable for the estimation of deep-time phylogenies than the approach used here (SI Text).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

(A) Subtree of calibrated archaeal chronogram showing Methanosarcinales and related groups. Acetoclastic Methanosarcina (bold) groups with other members of Methanosarcinales, including the distantly related acetoclastic Methanosaetaceae (M. thermophila, M. concilii). Shaded bars indicate the estimated acetoclastic Methanosarcina ancestor age ranges (±1 SD) for each dataset. The combined estimate of Myr is indicated by the black bar. The tree was generated from the Universal rProtein dataset, with the age of acetoclastic Methanosarcina adjusted to match the combined estimate. (B) Members of the genus Methanosarcina shown to grow on acetate all descend from the ancestor of sequenced clade representatives containing the transferred ackA/pta genes (*), congruent with the age-estimated node in A. The relationships between known taxa are represented by their 16S phylogenetic tree.

Additional phylogenetic character analysis of 16S sequences from 33 species within the genus Methanosarcina lends increased precision to the placement of the gene transfer event within the chronogram of Fig. 3A. As shown in Fig. 3B, all Methanosarcina strains that grow readily on acetate diverge within the M. acetivorans/M. mazei/M. barkeri clade, whereas all strains shown not to grow on acetate diverge more deeply within the tree. This result supports a relatively recent horizontal transfer of ackA/pta within the Methanosarcina genus (SI Text), at a time consistent with the Late Permian. The combined results of Fig. 3 therefore support the hypothesis that the emergence of the acetoclastic pathway in Methanosarcina provided the microbial instigation of the end-Permian burst in the carbon cycle.

Methanogenic Expansion

Wherever sulfate was limiting—because of a widespread drawdown of sulfate levels (33⇓⇓–36), a localized depletion by sulfate reducers, or both—the introduction of the high-efficiency acetoclastic pathway would have diminished a thermodynamic barrier (38) to greater acetate production by fermenters. Not only would acetate be converted more quickly to methane, but also sedimentary organic matter would be fermented more rapidly to acetate. Methanosaeta would have played a supporting role, but a preexisting steady state excludes the possibility that Methanosaeta itself would have excited the perturbation.

The resulting methane burst would have been oxidized to CO₂, either by anaerobic methanotrophs at the expense of any remaining sulfate or aerobically. O₂ levels, which were likely already low (5, 12, 13), would have been depressed further. Given the assumptions of an effectively unlimited substrate and a preexisting steady state, the size of the acetoclastic methanogenic niche—i.e., the carrying capacity K—would therefore have increased, at a rate proportional to the rate at which nearby sulfate and O₂ were depleted. Taking the depletion rate proportional to the methane flux, we then have , where is the total methane production and is a conversion constant. Integrating both rates yieldswhere is the initial (unperturbed) carrying capacity. At the time scale of the geochemical signals (>10³ y), the methanogenic population is always at carrying capacity. The methane production rate is thereforewhere β is the individual metabolic rate. Eqs. 2 and 3 describe unstable growth: Increasing methane production (A) increases the carrying capacity (K), which in turn increases the methane production rate . Then if β is constant, grows exponentially.

However, several mechanisms may have acted to increase β. These include the aforementioned interactions with other microbial communities; warming due to higher CO₂ and methane levels; adaptive radiation (39) as subtaxa evolved to more efficiently use acetate in particular environments; and greater access to any limiting mineral nutrients as the anoxic, methanogenic niche rose upwards. Each of these mechanisms implies a functional dependence on the total methane production A. We hypothesize a linear response:where is the initial metabolic rate and is a constant. Inserting Eqs. 2 and 4 into Eq. 3, we obtainwhere , , and . For nonzero , the solution blows up at a time determined by initial conditions. Near , the dynamics are dominated by the nonlinearity. To leading order,which is drawn as the dashed red curve in Fig. 1B and the dashed straight line in Fig. 2C (as , using the best-fitting scale factor). The good fit supports the nonlinear model (Eq. 5), but it raises an important question: As t approached —the time of peak extinction activity—how could the growth of the methanogenic community have been sustained?

Nickel Limitation and the Meishan Nickel Record

Methanogens require nickel (18). The active site of the enzyme MCR, used by all methanogens, is the nickel cofactor ; moreover, the CODH enzyme complex used by acetoclastic methanogens also contains a nickel cofactor (40). Seawater concentrations of nickel have likely been beneath the limiting threshold for methanogens for roughly the last 2 billion years (19). Accordingly, we suggest that methanogenesis in Late Permian sediments was limited by nickel. If so, the methanogenic expansion would have required increased access to nickel.

Kaiho et al. (41) have reported a sharp increase in the concentration of nickel in Meishan sediments coincident with the carbon isotopic spike. Because their analysis extends only through the last of the carbon isotopic spike and does not consider the effects of lithologic changes, we have performed new measurements of nickel concentrations at Meishan, not only over the entire interval shown in Fig. 1 but also well above it, and have corrected our measurements for variable concentrations of carbonate (SI Text). Fig. 4 displays our results. Nickel concentrations before the carbon isotopic spike are typically at least twice as high as after, and are up to seven times greater just before the abrupt downturn. With such elevated concentrations, nickel would no longer limit methanogenic activity. All methanogens would have prospered, but the successful evolution and rapid expansion of acetoclastic Methanosarcina would have been especially favored in the substrate-rich end-Permian environment.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Nickel concentrations (red) in the Meishan sediments, compared with the isotopic composition of carbonate carbon (blue). (A) Approximately 9 m of sedimentary section. (B) Close-up of the changes, over a range of less than 1 m. The largest nickel concentration precedes the smallest value of . Nickel concentrations are reported in terms of a carbonate-free basis (SI Text).

The nickel likely originated from Siberia. Earth’s largest economic concentration of nickel is in the Noril’sk region, deposited during the emplacement of the Siberian traps (42). The Meishan nickel signal may therefore represent changes in ocean chemistry induced by the Siberian eruptions, thereby linking the growth of acetoclastic Methanosarcina to massive Siberian volcanism. The apparent insensitivity of nickel to redox processes (43⇓–45) supports this interpretation. However, elemental redistribution during diagenetic redox cycling could represent an alternative cause of nickel enrichment in the sediment (46); further analysis of redox-sensitive elements will help evaluate that possibility.