Table 1 and Figure 2 summarize the whole-rock major element data; the data are shown in Supplementary Table 2. Independently of the stratigraphic subdivisions of the Meishan section, the data justify division of the section into beds lying above 24e, those lying below 24e, and 24e separately (Figure 2). Overall, SiO₂ content ranges up to 57 wt%, and is highly variable, especially in beds below 24e. Bed 24d has increased SiO₂ compared with most of the stratigraphy below 24e. CaO ranges up to 55 wt%, and decreases upsection as SiO₂ increases. Above bed 24e, Al₂O₃ ranges up to 20 wt%, while remaining below 9 and 0.8 wt% in beds below 24e and bed 24e, respectively. Except for anomalous concentrations of Fe₂O_3T in bed 24f (up to 30 wt%), Fe₂O_3T above Bed 24e is below 8 wt%, and below 2 wt% in all other samples. The section below bed 26 contains less than 2 wt% MgO; MgO increases upsection. Above bed 24e, K₂O content ranges up to 5.1 wt%, but is below 0.2 and 2.5 wt% in bed 24e and beds below 24e, respectively.

The major element data show correlations (Table 2; Supplementary Figure 5; Supplementary Table 4) that are consistent with dominantly carbonate and siliciclastic lithologies. We consider statistically significant correlations to have r² > 0.5. In beds above 24e, where the section is dominated by mudstones and marls, SiO₂ is correlated with Al, Na, K, and Ti; SiO₂ is negatively correlated with Mn and Ca. Neither Fe nor Mg is correlated with other major elements. In bed 24e, SiO₂ is correlated with Al, Na, K, Ti, and P, and is negatively correlated with Mg and Ca. Fe shows no correlation with other major elements, but Mg is strongly correlated with Ca, and negatively correlated with K and Ti. Below bed 24e, SiO₂ is correlated with Al, Fe, Na, K, and Ti, and negatively correlated with Ca. Mg is correlated with K and Ti, whereas Ca is negatively correlated with Na, K, and Ti. The geochemical data thus reflect mixing between limestone and siliciclastic components. The absence of statistically significant major element correlations with Fe₂O_3T in bed 24e and those above 24e is unexpected, and suggests that Fe₂O_3T content is controlled by something other than detrital, siliciclastic input.

Table 3 and Figures 3 ,4 summarize the whole-rock trace element data; the complete data set is Supplementary Table 3. Elements analyzed but not reported in Table 3 are excluded because there are numerous missing values, or because other elements are adequate proxies (REE). Figure 3 shows the trace element data as a function of stratigraphy, whereas Figure 4 shows trace element data against Al₂O₃ content, the best indicator of siliciclastic input.

Trace element concentrations vary on the scale of individual sediment laminae. The data indicate that the units we sampled are best treated as groups: beds below 24d can be grouped, beds 24d, 24e, 24f, and 26, in some cases, need to be treated separately, and beds above 26 can be grouped. Most notably, bed 24f, the beginning of the extinction horizon, shows enrichments, compared with the other parts of the stratigraphy, in As, Cr, Mo, Pb, Sb, Th, Tl, U and W. The overlying black mudstone, bed 26, shows enrichments in As, Ba, Co, Ga, Mo, Pb, Rb, Sb, Th, Tl, and V. In contrast with bed 24f, bed 26 does not show enrichment in Cr, U or W. In both beds, there are individual samples with Ni up to 170 ppm, but there is no overall pattern of Ni enrichment. Two samples in bed 24f have V content of 109 ppm, but the other seven samples of 24f have V below 25 ppm. The majority of the trace elements that are enriched in beds 24f and 26 are redox sensitive, or have important metabolic pathways.

Al₂O₃, Sc, Th, and Zr best characterize the detrital component (Figure 4). In bed 24e and below, Zr content is variable, but always below 100 ppm. In beds 24f and 26, Zr is between 100 and 300 ppm, whereas the section above bed 26 contains between 100 and 200 ppm Zr. Similarly to Zr, Y shows an increase from bed 24e (always <20 ppm) to 20–60 ppm in beds 24f, 26, and 27, with a “baseline” of 10–30 ppm above bed 27. The distribution of Sc parallels that of Y. Ce (and other REE) content also follows a pattern similar to Zr, in that it is below 25 ppm and variable in beds below 24e; in beds 24f and 26, Ce is enriched (34–98 ppm), whereas in beds above 26, it has a range of 30–80 ppm.

Below bed 24e, V content is variable between 8 and 160 ppm, with four samples being below the limit of detection (BLOD) of 5 ppm. In bed 24e, V is below 32 ppm, with 5 of 10 samples being BLOD. Above 24e, two samples are BLOD, whereas the others range up to 210 ppm, with a median value of 65. For Cr, many samples are BLOD (detection limit, DL, is 20 ppm). The highest value, 280 ppm, is in bed 24f, and the “baseline” above bed 27 is around 50 ppm. For Co, also, there are numerous missing values with a DL of 1 ppm. Notably, all of bed 24e is BLOD, whereas bed 24d shows up to 16 ppm. Beds 24f and 26 range up to 35 ppm Co, whereas beds 27 and above mostly contain <10 ppm Co. The high DL of Ni, 20 ppm, limits our data. Below bed 24 d and in 24e, Ni is BLOD for most samples, whereas in bed 24d, Ni spikes to 170 ppm. There is a similar spike to 170 ppm in beds 24f and 26. Fewer than half of samples above bed 26 have detectable Ni. The highest Cu content, 100 ppm, is in bed 26. Most samples below 24f are BLOD, and the scattered data for beds above bed 26 suggest a “baseline” of about 30 ppm.

The distribution of Zn contrasts with that of the other transition metals. In beds above 24e, Zn content is below 200 ppm, and averages about 60 ppm, with 21 of 34 samples being BLOD (DL is 30 ppm). By comparison, 6 of 10 samples in bed 24e returned values with a median of 430 ppm. There is a distinct enrichment of Zn in bed 24d; 6 of 9 samples have reportable values with a median near 300 ppm and a high value of 1,180 ppm. In the beds below 24d, 6 of 11 samples are BLOD, but there is a single high value of 1,000 ppm. The pattern for Pb is similar to Zn: bed 24d is enriched (median value of ∼30 ppm), with spikes to 125 ppm in some beds below 24d. Pb is BLOD in most of bed 24e, but shows enrichment in beds 24f and 26. Above bed 26, Pb is mostly below 20 ppm, with a few values reaching 50 ppm. Th and U are above the DL in all samples. Below bed 24d, Th values are erratic, varying from 0.3 to 12 ppm; bed 24d is not distinct from the units below it. In bed 24e, Th is low (0.4–1.2 ppm). In bed 24f, Th is slightly higher, from 0.4 to 20 ppm, whereas in bed 26 it ranges from 15 to 38 ppm. Above bed 26, Th values are between 3 and 15 ppm. Beds below 24e are not distinct in their U contents, varying from 2.2 to 15 ppm. Bed 24e is also in this same range, with U between 2.1 and 4.8 ppm. There is a distinct spike in U content in bed 24f, with U ranging from 4.4 to 22 ppm. Bed 26 is also somewhat elevated, with U between 6 and 12 ppm. Above bed 26, U content is generally low, between 1.5 and 4.5 ppm, with the exception of two samples in bed 27, with 5.4 and 7.3 ppm.

In beds above 24e, REE patterns are almost flat when normalized to PAAS (Post-Archean Australian Shale; Taylor and McLennan, 1985), with some samples showing negative Ce anomalies and a slight enrichment in the MREE (Figure 5). Overall REE abundances vary by more than a factor of 10. In Bed 24e, the REE patterns show depletion of LREE, and generally low REE abundances. Beds below 24e show variable REE abundances, and all samples have patterns that are depleted in LREE. In the limestones of bed 24e and below, there is no distinct negative Ce anomaly. Several samples throughout the section show positive Y anomalies.

Because there are numerous missing values in the trace element matrix, the statistical significance of the correlation coefficient for trace elements varies more than it does for major elements. As a rough guide, we have chosen r² > 0.5 to indicate correlation; the correlation data are shown in Supplementary Table 4. In beds below 24e, an expected suite of lithophile trace elements (Sc, Ga, Rb, Nb, Cs, Ba, Zr, Hf, Pb, Th) is strongly and positively correlated with Al₂O₃ (Figure 4), Fe₂O_3T, K₂O, and TiO₂, but is only moderately correlated with SiO₂. This correlation is less pronounced in bed 24e and the beds above 24e. A second suite of elements, including V, Co, Mo, the REE, Ta, and ⁸⁷Sr/⁸⁶Sr, is less strongly correlated with the same suite of major elements, whereas a third suite of trace elements, including Cr, Cu, Tl, and U, has weak correlations with Al₂O₃, Fe₂O_3T, K₂O, and TiO₂. Sr is strongly to moderately anti-correlated with SiO₂, Al₂O₃, Fe₂O_3T, K₂O, and TiO₂. CaO is moderately to strongly correlated with Sr, but is anti-correlated with the lithophile trace elements listed above, as well as with ⁸⁷Sr/⁸⁶Sr. There is a weak anti-correlation between CaO and Cr, Mo, Ta, and Tl. The same patterns generally hold in bed 24e, except that there is a weak to moderate anti-correlation between MgO and CaO and Cr, Zn, Sb, Th, Tl, MREE, and HREE. Above bed 24e, MnO, MgO, and CaO are anti-correlated with most of the lithophile elements. Also above bed 24e, there is a weak anti-correlation between Mo and SiO₂, MnO, MgO, and P₂O₅.

Correlations among the trace elements are consistent with suites of trace elements that are controlled by siliciclastic or carbonate components. Thus, Sc, Ga, Rb, Zr, Nb, Cs, Ba, Pb, Th, and U are mutually and mostly strongly correlated, and are related to the siliciclastic component. Cr, Cu, Co, Ni, Sb, Mo, and W constitute a second suite that shows mutual correlations; this group is not distinctly associated with either lithological component. The REE are correlated within themselves, and with Zr, Ta, Hf, Pb, Th, and U. Sr is the only trace element that is negatively correlated with Sc, V, Cr, Ga, Rb, Y, Nb, Cs, Ta, Pb, Zr, Ba, Th, and the REE; it is not positively correlated with any other trace element.

Table 4 and Figures 6 ,7 present trace element data for the acetic acid extracts. Major elements were not analyzed in the acetic acid samples. The correlation matrix for the acetic acid extract data is Supplementary Table 5. The data are reported as ppb in solution, and are mostly from beds above 24e, where we expect the most pronounced signal of changes in environmental conditions. Because we have only poor constraints on the amount of sample dissolved during acetic acid leaching, conversion of these data to ppm in the solid introduces large uncertainties. Comparison of concentrations in the acid solution provides a qualitative measure of the trace element content of the carbonate fraction in each sample. The low Ga, Zr, and Th content in these solutions is consistent with our assumption that the siliciclastic component made little or no contribution to these data, though Rb and Cs data suggest that some exchange with phyllosilicates probably occurred. Elements that were sought but were below detection limits in more than 75% of samples include Cr, Ga, Ge, As, Zr, Nb, Mo, Ag, In, Sn, Sb, Hf, Ta, W, Tl, and Bi. The absence of reportable Cr, As, Zr, Mo, Sb, and W contrasts with data from the bulk rock samples.

Of the transition metals that were above detection limits in these analyses (V, Co, Ni, Cu, Zn, Pb), all but Co show a distinct concentration peak in bed 24d, just below the extinction horizon in bed 24e (Figure 6). The high concentration of Zn continues into bed 24e, but not into bed 26. Ba is also elevated in bed 24d. U shows a similar distribution, with an additional peak in bed 24f. Some lithophile elements (V, Rb, Y, Th, Ce) show a slight increase upsection. By comparison with the top of bed 24d, the extinction horizon, bed 24f, is not particularly enriched in trace metals, despite the extreme enrichment of Fe, which is present as pyrite. U and Ba have elevated values in 24f. Throughout the section, Th is below detection limit in the carbonate-rich samples, whereas U is enriched in beds 24d, 24e, and 24f, compared with the extracts from beds above 24f. Despite obvious problems in quantitation of the HREE, REE patterns are distinguishable (Figure 7), with beds above 24e having higher REE concentrations and distinct MREE enrichment, whereas both 24e and the beds below 24e show LREE depletion. Negative Ce anomalies are present in most samples below bed 30, and negative Eu anomalies characterize most samples above bed 24e.

Major and trace element data track the major change of lithology between the carbonate-dominated section in beds 24e and below to the siliciclastic-dominated section in beds 24f and above (Figure 2). There are variations on the scale of single laminae within the carbonate section, where SiO₂ content can reach above 45 wt% (sample 24c6a) in individual samples. Elevated SiO₂, when paired with elevated Al₂O₃, reflects detrital input deriving from turbidite events (Li G. et al., 2016). Detrital input appears to have SiO₂/Al₂O₃ ∼ 5, the average value in beds above bed 26, and SiO₂/Al₂O₃ above eight indicates SiO₂ content independent of detrital input (Figure 2), either as sponge spicule accumulations (Zheng et al., 2013) or as diagenetic silicification. Whole rock compositions thus can be interpreted in terms of three components - carbonate, detrital siliciclastics, and biogenic silica. In the trace element data, the signature of biogenic silica cannot be separately identified, and only the carbonate and detrital components can be recognized. Because the siliciclastic section above bed 24e is dominated by detrital inputs, it is difficult to infer ambient seawater conditions from data in that part of the section.

The non-deposition of carbonates in the extinction horizon (beds 24f–26) may reflect an acidification event in the surface ocean. If the break in carbonate deposition in beds 24f–26 was due to acidification, it was a transient condition lasting less than about 15 kyr, which was followed by carbonate deposition for ∼26 kyr in beds 27–29. Evidence for acidification is tied to the latter part of the extinction interval in the Early Triassic (Clarkson et al., 2015), and not to the latest Permian, and is also presumed to be short-lived, not exceeding 10 kyr. Furthermore, the acidification event is modeled to have decreased oceanic pH from ∼8 to ∼7.3 (Clarkson et al., 2015), a pH at which there would not be extensive dissolution of existing carbonate sediments in marine settings. Alternately, the cessation of carbonate deposition in beds 24f–26 could be due to voluminous detrital input that suppressed carbonate deposition, input that is coincident with deposition of the volcanic ash in bed 25. Carbonate deposition depends on carbonate saturation, i.e., the extent to which the activity product of Ca²⁺ and CO₃ ²⁻ exceeds the solubility product; detrital input that is rich in clays will drop the activity of Ca²⁺ by providing a large increase in adsorption sites, sequestering Ca²⁺ in smectite and montmorillonite. Increased detrital input could also reflect increased erosion, as suggested by δ⁷Li data (Sun et al., 2018), suggesting that the extinction on land was coincident with the marine extinction event (Sephton et al., 2005; Shen S. Z. et al., 2011; Zhang et al., 2015). Negative Δ¹⁹⁹Hg in Meishan beds 25–26 is interpreted to indicate that the Hg budget was dominated by local terrigenous inputs (Grasby et al., 2017), also pointing to increased erosion contemporaneously with the extinction. Major and trace element data are insufficient to distinguish between these alternatives, but the absence of evidence for carbonate dissolution at the top of bed 24e is consistent with the latter interpretation.

The exceptional enrichment of Fe₂O_3T in samples 24f7 and 24f7b, to 30 wt%, signals the pyrite-rich layer in bed 24f. The continuity of this pyrite layer in several sections at Meishan indicates that it is a primary depositional feature, not a diagenetic feature. The pyritic horizon is thin, being <5 mm at the top of bed 24f (though all of bed 24f is pyrite-rich in comparison to the rest of the stratigraphy) that has a total thickness of 24 mm (Supplementary Figure 3), and represents initiation of an euxinic event. Fossils and burrows in bed 24e (Chen Z. Q. et al., 2015) show that bed 24e was deposited under suboxic conditions, so that the euxinic event associated with the end-Permian extinction post-dates bed 24e. Deposition of bed 24f occurred over less than 2 kyr, and the euxinic event could be restricted to a fraction of this interval. There is no clear evidence, in our data, of where termination of euxinic conditions might have occurred upsection. The deposition of the pyrite horizon requires not only the availability of sulfide, but also of Fe in a quantity that is not reflected in any other part of the Meishan section. We interpret the cessation of pyrite precipitation as reflecting the exhaustion of the Fe reservoir rather than the cessation of euxinic conditions: the data of Xiang et al. (2020) and Wei et al. (2020) both show the persistence of anoxia from bed 24f upward through at least bed 26, though euxinia in this interval may have been episodic. There is evidence for trace metal depletion in the water column by precipitation in euxinic waters (Takahashi et al., 2014), but the abundance of Fe in their samples does not show much change across the extinction interval; their data, from an open ocean setting in the Panthalassic Ocean, may not track Fe abundances in shallow water sections like Meishan. The source of Fe - from the overlying water column that includes fresh weathering inputs, or diagenetically mobilized Fe in pore waters - is unclear. Fe enrichment in bed 24f7 is accompanied by enrichments in As, Mo, and Tl, but not in Co, Cu, or Ni.

Previous studies have documented a change in mineralogy in the Meishan section, going from dominantly calcite in the limestones in bed 24e and below, to dominantly Fe- and Mg-bearing carbonates (ankerite or dolomite) in the carbonate beds above bed 24e (Liang, 2002; Li and Jones, 2017). Figure 2 shows the calculated carbonate mineralogy of the Meishan section. The extent of MgO, MnO, and Fe²⁺ partitioning into the carbonate component is not clear from geochemical data, so that the mineralogical proportions in Figure 2 are estimates, based on the assumption that all CaO is partitioned into calcite, all MgO is partitioned into dolomite (ignoring Mg in chlorite and smectite), and all MnO is present as rhodochrosite. In the absence of analytical data on Fe³⁺/Fe²⁺ and S²⁻, we assumed that Fe²⁺ is present as pyrite, and have excluded siderite from our calculations. Xiang et al. (2020) measured carbonate associated Fe but do not discuss these data separately. In their data, pyrite represents 0.4–0.8 of highly reactive Fe in the bed 24–28 interval, so that roughly half of the reactive Fe could be present as siderite. Analyses of dolomite grains in beds 25 and above show that Fe-rich zones occur as narrow (<5 µm) rims on some dolomite grains (Li and Jones, 2017).

Trace element data suggest that beds 24e and below have a different detrital component than those above 24e. Thus, for example, there are two different trends of Cr and V abundance with respect to Al₂O₃ (Figure 4), one trend for beds above 24e showing low Cr/Al and V/Al and a second trend for beds 24e and below with elevated Cr/Al and V/Al. Similar patterns are evident for Y and La (for the REE in general), where data for beds 24e and below show a trend with low Y and La abundance, with a second trend of higher Y and La abundance for beds above 24e.

Multiple factors could contribute to the change in trace element abundances between the section below bed 25 and that above bed 25. A change in the provenance of the sediments is possible, if there is a re-ordering of drainage patterns on land that is coincident with the extinction beginning at the top of bed 24e. A change in the bulk composition of material being transported into the marine environment is possible, if volcanic material deposited on, and subsequently eroded from land comes to comprise a sufficiently large proportion of the total detritus. A change in the proportion of mechanical to chemical weathering could produce different major and trace element patterns as a function of time. Finally, a change in patterns of winnowing of heavy minerals during transport could affect some element distributions, if there were a significant change in the volume of rainfall.

There is an up-section increase in trace element abundances. If, for beds 24e and below, the sediment source area was predominantly carbonates with low trace element content, a change in provenance to a dominantly siliciclastic source area for the beds above bed 24e could account for the increase in trace element abundances. Reconstruction of paleo-drainage patterns would be required to evaluate this possibility.

Numerous volcanic ashes were deposited across all of South China in the latest Permian and early Triassic. If the proportion of ash-derived detritus increased in going from bed 24e to the beds above 24e, and the detrital input in the lower part of the section had a low trace element inventory, this could account for increased trace element abundances up-section. Volcanic ash is relatively quickly eroded, and would produce a transient spike in trace element patterns, unlike the persistent increase in trace element abundances we observe in our data for beds 26–30. Elevated trace element abundances could persist if multiple ash events were involved in this erosional scenario. Because present models suggest global dispersion of some trace elements as aerosol (e.g., Ni and Hg) from the Siberian Traps, identifying those trace elements that were aerosolized and therefore should correlate with Ni and those that were dispersed mainly in the ash beds in South China would be a priority.

A progressive increase in atmospheric CO₂ (Saunders and Reichow, 2009) characterized the Late Permian. Elevated atmospheric CO₂ is implicated in acidification of rain and the surface ocean, in increasing atmospheric and surface ocean temperatures, and in the rate of silicate weathering (Broecker and Sanyal, 1998; Winnick and Maher, 2018). Compared to an early Changhsingian baseline, increases in acid rain, in atmospheric temperature, and in the rate of silicate weathering all point to an increase in the overall rate of chemical weathering (Algeo and Twitchett, 2010) in the late Changhsingian. The increase of chemical weathering will increase the dissolved load of rivers, and this ultimately will be reflected in increased trace element abundances in the strata above bed 24e. The timing of these effects is uncertain and is debated, but the change in trace element patterns in our data occurs at the very top of bed 24e, and into beds 24f and 26.

Abundances of trace elements with low solubility in aqueous systems (e.g., Ti and Zr) are tied to the abundances and distributions of accessory minerals (rutile, titanite, zircon). The change of Ti abundance from bed 24e to the strata above 24e potentially reflects a change in the winnowing of heavy minerals during riverine transport (c.f., Bennett and Canfield, 2020; Cole et al., 2017).

Li isotopic and abundance data (Sun et al., 2018) show a change between the section below bed 25 and that above, that Sun et al. interpret to indicate increased riverine flux of light ⁷Li into the oceans. Sun et al. point out the strong correlation of Li with Al, implying that Li is controlled by detrital input, but do not discuss the potential impact of diagenesis on either Li abundance or δ⁷Li. In Mg isotopic data (Chen et al., 2020), there is a signal of increased weathering starting in the mid-Dienerian, but mineralogical controls on Mg distribution obscure the weathering signal in the earliest Triassic, including the section covered by our sampling.

We have compared the Meishan section to other sedimentary rocks by calculating elemental enrichment factors (ef) with respect to PAAS (Taylor and McLennan, 1985). We use a clastic reference material for dominantly carbonate rocks because many trace elements in the Meishan section are controlled by the detrital, clastic fraction, as is evidenced by trace element correlations with Al. The ef is calculated as (element/Al)_sample/(element/Al)_PAAS. The most striking feature of ef values as a function of stratigraphy is that most elements have ef ∼1 above bed 27 (0 cm from PTB; Figure 8): the concentration of most elements above bed 27 is comparable to that in average shale. This underscores the observation that the detrital siliciclastic component controls trace element distributions in the upper part of the section.

Because of low Al content in carbonate-rich strata, ef values in the carbonate beds are systematically high (c.f., Xiang et al., 2020), and ef calculated with respect to PAAS are potentially misleading. Of particular concern are ef for redox-sensitive trace elements like Ni, As, and Mo, and for Zn, none of which are correlated with Al. In our analyses, these elements have relatively high DL (20, 5, 2, and 30 ppm, respectively). Using the DL values to calculate ef (i.e., for samples with values below DL, substituting DL for the element value in the ef expression) shows interesting patterns that reveal the weakness of our data and of PAAS-based ef in carbonate strata. Above bed 24e, in the siliciclastic section, median ef calculated with DL values are 0.9, 8.6, 5.2, and 1.0 for Ni, As, Mo, and Zn, respectively. In bed 24e, the median ef are 12.5, 125, 75, and 14, respectively, for Ni, As, Mo, and Zn. In the carbonate-dominated beds below 24e, the median ef are 4.2, 42, 25, and 4.7. Because Fe is correlated with Al in beds below 24e, low Fe samples (Fe₂O_3T < 0.5%) below bed 24e (red points in Figure 8) show this tendency toward higher overall ef.

Sc, Ba, and Th show little variation of ef, regardless of stratigraphic position, suggesting that these elements, throughout the section, are controlled almost exclusively by the siliciclastic component. The apparent insensitivity of Ba to changing biological activity and redox conditions is unexpected. By contrast, the ef_Zr is highly variable below bed 27. Though Zr is usually part of the detrital fraction, and is considered to be immobile in low-temperature environments, ef_Zr variability implies that there has been post-depositional mobility. Zr can be mobilized as hydroxyl and fluoride complexes (Censi et al., 2015), can interact with phosphate binding sites on microbial cell surfaces (Censi et al., 2015), and potentially can also be mobilized as carbonate complexes (Kobayashi and Sasaki, 2017). Mobilization of Zr as hydroxyl or carbonate complexes would require neutral to alkaline pH in the pore water environment.

Other elements that show little variation of ef include Fe and Ti (Figure 8). There is a clear perturbation of ef_Fe in the extinction horizon (24f), and a suggestion of increased ef_Fe at low Fe concentrations in the beds below 24e. Variability of ef_Fe potentially supports the argument of Fe depletion as the limitation on pyrite deposition at the top of 24f: the Fe-rich samples in beds 24d and 24f have ef_Fe > 4, whereas ef_Fe < 2 in beds 26 and above. There is a distinct break in ef_Ti at the extinction horizon, suggesting again that there are different detrital sources, different extents of chemical weathering, or different conditions of winnowing during transport contributing to the weathering flux above and below bed 24e.

The elements that have ef > 1 above bed 27 are Sb, W, As, and U; for these, ef is even higher in beds below 27. Among these elements, only U is measurable in the acetic acid extracts, so that only U enrichment is tied to the carbonate fraction. The controls on Sb, W, and As abundances are unclear.

Zn shows a dramatic change of ef from bed 24e to 24f; in 24e and below, ef_Zn ≥ 10, whereas in beds 24f and above, ef_Zn < 2. Zn is an essential micronutrient for marine phytoplankton (Conway and John, 2014). The change in ef_Zn is coincident with the extinction horizon, implying a collapse of phytoplankton-linked productivity (c.f., Algeo et al., 2012: this drop of productivity is unique to South China PTB sections). Elevated Zn concentration in bed 24 is interpreted, from Zn isotopic data, to be due to magmatic inputs (Liu et al., 2017); the change of ef_Zn then implies that these external inputs were less important in the beds above 24f. But elevated Zn in bed 24e is not accompanied by enrichments in other “magmatically-derived” elements, an observation that calls into question the magmatic provenance of Zn. V, Cr, Ni, and, to some extent, U show similar patterns. Because Zn, V, Cr, and Ni show ef ∼ 1 above bed 27, there is no continued input of an “excess” of these elements above the PTB. If input by volatilization during continuing volcanism were a primary control on their abundances, we would expect ef > 1 for these elements in the strata above bed 27. Co and Cu (not shown in Figure 8) also show ef ∼ 1 in bed 27 and above, but, unlike Zn, V, Cr, and Ni, do not show high ef in the section below 24f. Co, Cu, Ni, and Zn have low solubility products with sulfide, and are rapidly precipitated from sulfide-bearing (euxinic) waters, unlike V and Cr, which do not easily form sulfide compounds; a single mechanism involving euxinic fluids is unlikely to account for the behavior of all these trace elements. A large fraction of V and Cr is tied to the detrital fraction in most sedimentary systems (Cole et al., 2017).

The REE and Y show peaks in ef in beds 24e–24f and 27, with lower values in bed 26. Relative enrichment in the carbonates reflects low Al in these beds, and diagenetic increase of REE without an increase in Al.

Diagenesis is prevalent in carbonate rocks, making it difficult to extract information from them about primary depositional conditions. Bulk rock major and trace element data are insensitive to when changes in concentrations might have occurred: distinguishing stages or phases of diagenesis from bulk rock geochemical data alone is rarely possible.

Diagenesis in carbonates is usually evaluated by examining relations between CaO, MgO, MnO, and, to a lesser extent, FeO and Sr (e.g., Brand et al., 2012). At Meishan, the most sensitive indicator of diagenesis is ⁸⁷Sr/⁸⁶Sr (Dudás et al., 2017). The least altered samples are expected to have ⁸⁷Sr/⁸⁶Sr < 0.7071, based on the secular trend of seawater ⁸⁷Sr/⁸⁶Sr in the late Permian. These compositions are found primarily in bed 24e (Figure 9). Low values of MgO/CaO, MnO/CaO and FeO/CaO are not systematically linked with low ⁸⁷Sr/⁸⁶Sr, indicating that Sr is more susceptible to post-depositional modification than are Mg, Mn and Fe. In our data, MnO/CaO above ∼0.001 in carbonates is an indicator of diagenetic influence. In bed 26 and beds above 28, MnO/CaO is not an indicator of diagenesis.

The estimated abundances of calcite, dolomite, and rhodochrosite throughout the section (Figure 2) show that dolomitization was an important process in the beds above 24e. The MgO/CaO of most analyses below bed 24f is below 0.02 (Figure 9), an indication that low-Mg calcite is the dominant phase in these beds. That, in itself, is diagnostic of early diagenesis in which low-Mg calcite replaces likely primary carbonate phases (aragonite, high-Mg calcite, proto-dolomite; Morse, 2003). Above bed 24e, especially in bed 27, there are calcite-rich horizons, and there may well be low-Mg calcite, but the median whole-rock MgO/CaO is higher in bed 27 (0.237 ± 0.231) than in beds 24e and below (0.017 ± 0.015). Li et al. (2018) interpret the global pattern of dolomitization across the PTB to reflect the activity of sulfate-reducing bacteria in an anoxic environment.

Below bed 24e, the other sign of diagenesis is the formation of chert lenses. Silicification is reflected by variation in SiO₂/Al₂O₃ (Figure 2). The siliciclastic detrital component has SiO₂/Al₂O₃ below 8, but numerous samples below bed 24e have SiO₂/Al₂O₃ well above 10. From the geochemical data, it is not clear whether this increase in silica content is due to accumulation of sponge spicules (Zheng et al., 2013) or to addition of SiO₂ during diagenesis. The occurrence of chert lenses in the beds below 24e suggests that some variation in SiO₂ is due to diagenesis. Chert is underrepresented in our whole rock sampling because our primary target was to interpret the geochemistry of the carbonates. Additional insight into trace element distributions, particularly, could be gained by systematic sampling of cherts in the Meishan section.

Numerous studies have examined the character and distribution of pyrite in the Meishan section (Jiang et al., 2006; Shen et al., 2007; Chen Z. Q. et al., 2015; Li G. et al., 2016). Syngenetic, framboidal pyrite (diameter <5 µm), and presumably diagenetic, larger framboids and euhedral pyrite cubes occur at Meishan. The role of diagenesis in the cycling of S is clearly indicated by multiple S isotopic study (Shen Y. et al., 2011). The pyritization of carbonate fossils (Chen Z. Q. et al., 2015) also indicates diagenetic mobility of both S and Fe. Thus, the inference of redox conditions in the seawater column, based on the occurrence and character of pyrite, is confounded by the evidence that pyrite distribution is affected by diagenesis that proceeds under conditions in the sediment - pore water system.

Field observations indicate that there is a non-deposition surface (“hardground” or “firmground”) at the top of bed 24d (Cao and Zheng, 2009; Zheng et al., 2013). This is reflected in the geochemical data as a zone of silicification, with 5 samples from 24d5 to 24d6 containing 29–45 wt% SiO₂ and having SiO₂/Al₂O₃ between 10 and 20, well above the five to eight range that characterizes siliciclastic input. More importantly, these samples show an anomalous increase in V, Co, Ni, Cu, Zn and Pb in both the whole rock and the acetic acid extract data (Figures 3,6). In the whole rock data, Ba is not noticeably enriched in bed 24d, whereas the acetic acid extracts show elevated Ba, supporting a diagenetic process for a suite of elements that characterizes the carbonate-hosted “Mississippi Valley” type Pb-Zn deposits (Sverjensky, 1986; Paradis et al., 2007). Though the “hardground” surface at the top of bed 24d is a depositional feature reflecting sea level fluctuation, apparently diagenetic silicification produced a hydrological cap that trapped upwelling solutions carrying dissolved transition metals. These solutions are interpreted to be oxic because most transition metals cannot be transported in reduced solutions. Because the transition element enrichment in bed 24d does not continue upward into bed 24e, it is possible that diagenesis of 24d predated deposition of 24e, or that silicification at the top of 24d effectively restricted upward flow of diagenetic waters into 24e. We did not sample the “hardground” zones in bed 27 to determine whether similar features occur there.

The REE patterns of primary, marine carbonates are expected to show LREE depletion, with gradually increasing concentrations toward the HREE (Nothdurft et al., 2004). This pattern broadly reflects the abundances of REE in ocean water (Alibo and Nozaki, 1999). Particularly in the acetic acid extracts, the bulk of the samples from Meishan show a bell-shaped pattern with distinct enrichment of the MREE (Figure 7). This pattern is attributed to diagenetic redistribution of REE (Bright et al., 2009), and is particularly pronounced in marine phosphates, including conodonts (Chen J. et al., 2015).

As discussed above, some trace element data suggest that the factors controlling trace element abundances were different for beds above 24e and those below 24e. The simplest interpretation of these data is that, in the 24f–26 interval, there was significantly increased input from local, arc-related volcanism. Bed 25 is a volcanic ash that is linked to silicic, arc-related, volcanism (Gao et al., 2013; He et al., 2014). Silicic arc volcanism typically has relatively low Cr and V, but elevated Zr, Y, and REE compared with basaltic volcanism. The trace element patterns of beds 24f–26 reflect this local volcanic input. By contrast, continental flood basalt volcanism, though highly variable in trace element content in detail, would have Cr/Al and V/Al higher than those of silicic volcanics, and lower Zr, Y, and LREE—a pattern that is exactly opposite to that observed in our data. Similar changes in trace element abundances are not evident in the beds on either side of bed 28, which has a trace element profile that is very different from bed 25 (He et al., 2014). Hg abundances and isotopic data (Shen et al., 2021) provide a more complex picture. Of the four Hg abundance spikes in the South Chinese sections near the end-Permian extinction, the oldest three occur only in South Chinese PTB sections and can be linked to regional magmatism, whereas the fourth spike, associated with the extinction interval itself, is widely distributed and appears to include inputs both from regional sources and from the Siberian eruptions. If the trace element complement associated with Siberian magmatism derives primarily from burning of coal, and not from the magma itself, we cannot use the trace element systematics of igneous rocks as an interpretive framework for trace element distributions at Meishan.

Xiang et al. (2020) provide a framework, based on Fe speciation, for interpreting redox conditions within the Meishan section. It is not clear whether Fe speciation is a primary depositional marker or has been impacted by diagenesis. Xiang et al. (2020) argue that most of the section below bed 24e was oxic, with only brief episodes of anoxic conditions, as suggested by multiple S isotopic data (Shen Y. et al., 2011); that there is evidence of anoxia, with occasional euxinia, from bed 24e or 24f upward through bed 28; and that conditions were ferruginous (anoxic, but with Fe²⁺ in excess of available sulfide) in the beds above bed 28.

Our data are compatible with those of Xiang et al. (2020), and supplement them by adding information on the distributions of a broader range of trace elements. Our data, however, are insensitive markers of redox conditions, because they are limited by three factors: small sample size (∼10 mg), so that many elements have values below the limit of detection; relatively high DL (e.g., 20 ppm for Cr and Ni, 30 ppm for Zn) for critical elements; and sampling on a scale that may not capture high-frequency changes in redox conditions. The combination of these leads to missing values and an incomplete picture of elemental variation. Thus, out of 64 whole rock powders, only 9 were above DL for As, 17 for Mo, and 24 for Ni. Among the 9 values reported for As, four are in beds 24f and 26, but eight samples in 24f and 26 are BLOD; among the 17 values reported for Mo, 7 are in beds 24f and 26, but there are 5 additional samples in beds 24f and 26 that are BLOD. For both As and Mo, there are samples in bed 24d that are above the detection limit. By contrast, only 3 of 30 reported Zn values are in beds 24f and 26. Overall, our data are consistent with those reported by Zhou and Kyte (1988) for As, Sb, Zn, Th and U, and by Kaiho et al. (2001) for Ni and Mn.

Our data suggest that at least three factors affect abundances of redox-sensitive trace elements. The first is detrital input, signaled by increased Sc, V, Cr, Y, Zr, Nb and Ba in 24f3 and 26, where Sc, Y, Zr, and probably Nb, are not affected by redox conditions. The elevated V, Cr and Ba in beds 24f and 26 is thus partly due to detrital input. The second is redox state at the time of deposition, indicated by increased As, Mo, Sb and Tl in 24f7 and the black claystone of bed 26, which likely signal anoxic and possibly euxinic conditions. It is not clear whether these elements reflect redox conditions in the water column (primary) or in sediment pore waters (diagenetic). The third factor is diagenetic mobility of some of the redox-sensitive elements. We interpret transition metal enrichments in bed 24d to reflect diagenesis rather than primary depositional conditions, and suggest that trace metal distributions of at least some elements, including Ni, in the stratigraphy at Meishan cannot be used to infer redox conditions without assessing the extent of diagenetic mobility. Bennett and Canfield (2020) note that the behavior, during diagenesis, of some redox-sensitive trace metals, like V, is poorly understood.

Rothman et al. (2014) argued that the distribution of Ni in the Meishan section reflected trace element input from the eruptions of the Siberian Traps, and that Ni availability was a key prerequisite to a microbial methanogenesis event that could explain global carbon isotopic variations near the PTB. Two observations temper this interpretation. First, because individual horizons with elevated Ni content occur well below the extinction horizon (80–170 ppm in individual samples from beds 24a–24d, extending to ∼75 kyr prior to the extinction), the extinction in bed 24f, with Ni up to 170 ppm, is not uniquely linked with an increase in Ni supply. Our data do not constrain when and if there might have been an increase in Ni supply below bed 24. Our data show that Ni levels above bed 28, after the extinction interval, were lower than those in the section below the extinction horizon, consistently with Rothman et al.’s (2014) data. The increase in Ni is not consistently reflected in our data, in that not all samples in and immediately above the extinction horizon show increased Ni. Second, it is not possible to explain the increase of all of the trace elements in bed 24d—V, Co, Ni, Cu, Zn, Pb, and Ba—by appealing to aerosol input from the Siberian eruptions. Their erratic abundance distributions above bed 24d, especially their absence in bed 24e, is not consistent with a model of global aerosol input. A local diagenetic model for transition metal distributions is more consistent with our data.

In our data, there are no consistent correlations between the redox-sensitive trace elements and either Mn or Fe, contrary to expected patterns (Tribovillard et al., 2006; Lyons et al., 2009). In samples 24f7 and 24f7b, we sampled the pyrite layer with its admixed siliciclastic component (SiO₂ in both samples is ∼18 wt%); As and Mo are elevated in both samples, whereas V, Cr, Ni, and Zn are above detection limits in only one of these two samples—the trace element enrichment is not consistent at our level of sampling. Though U commonly accompanies elevated Mo (Algeo and Lyons, 2006), this association is only weakly reflected in our data. The range of Mo that we measured is below that usually encountered in euxinic strata (>30 ppm), but is within the range observed in suboxic sediments (Lyons et al., 2009).

The distribution of Th is controlled primarily by detrital input: Th is closely correlated with Al₂O₃ (Figure 4). The major change in the rate of siliciclastic input occurs at the top of bed 24e, concurrently with the development of euxinic conditions in beds 24f and 26 (c.f., Xiang et al., 2020). This change in redox conditions is reflected in an increase in the U content in beds 24f and 26. Thus, in our data, there are two populations of samples, based on Th/U: samples above bed 26, dominated by siliciclastic input, have Th/U 2.68 ± 0.74 (n = 20), and Th/U here is not indicative of redox conditions. Samples below bed 24f have Th/U 0.39 ± 0.25 (n = 30), reflecting both U enrichment and the lower input of Th in siliciclastic materials. This difference between the upper part of the stratigraphy and the lower part again suggests that there is either a change in the provenance of the siliciclastics, or a change in the proportion of mechanical to chemical weathering across the extinction horizon. The acetic acid extracts for samples above bed 26 consistently have lower U than those below bed 26. The carbonate fraction in beds above 26 is depleted in U, a pattern that is consistent with global U depletion in carbonates above the PTB (e.g., Ehrenberg et al., 2008), yet whole rock U enrichment factors compared to PAAS remain well above one in samples above bed 26. In our whole rock data, the median U content in beds above 26 is 2.4 ppm (±1.5, 1σ), whereas the median U content for beds 24e and below is 3.3 ppm (±2.8, 1σ), a difference that is not statistically significant.

The distributions of Sb, W, and Tl are difficult to interpret. All are redox-sensitive. All are above DL in most samples, with mild enrichment in beds 24d, 24e, 24f, and 26. With As, they are usually correlated in low-temperature hydrothermal systems (i.e., epithermal gold systems; Berger and Bagby, 1993). Little is known about their biological cycling. In the absence of systematic data about the distribution of Sb, W, and Tl in variously oxidized sedimentary systems, these data are suggestive but not definitive of suboxic to euxinic conditions. Zhou and Kyte (1988) also show elevated Sb in bed 24f2.

The low REE content of marine carbonates makes their REE patterns susceptible to large changes with even small admixture of REE from other sources (Nothdurft et al., 2004). Clean marine carbonates have REE patterns with low total REE content and low LREE (chondrite-normalized La/Yb < 1; Li and Jones, 2014). Admixture of REE from other sources typically increases both REE abundances and La/Yb so that REE patterns approach that of standard shales (e.g., Post-Archean Australian Shale, PAAS, Taylor and McLennan, 1985). In our data, REE profiles of the acetic acid extracts, i.e., carbonate-associated REE, mostly have La/Yb < 1 and elevated MREE, resulting in REE patterns that are bowed upward (Figure 7; Supplementary Figure 6) and typical of carbonates showing diagenetic enrichment of MREE (Bright et al., 2009). Whole rock REE concentrations are 100–1,000 times higher than in the acetic acid extracts, approach concentrations in average shale, and are relatively flat when normalized to PAAS (Figure 5): REE from detrital sources are controlling REE distributions in the whole rocks.

A suggested measure of the seawater component in carbonates is Y/Ho, for which the chondritic and crustal average value is 25–30, and the Y/Ho in oceanic carbonates is usually >45 (Song et al., 2012). Using this criterion, many of our samples have a detectable seawater component, but few have a dominantly seawater REE complement (Supplementary Figure 6). More significantly, the acetic acid extracts, nominally tracking the carbonate component, do not show Y/Ho much above the expected terrigenous average, and acetic acid extracts do not show a positive Y anomaly. Though Li and Jones (2014) do not provide Ho data, their REE patterns show no positive Y anomaly in marine aragonite (matrix and corals), but do show Y anomalies in calcitized - diagenetically modified - carbonate. Thus, Y/Ho itself is potentially tracking diagenetic reactions, not a primary seawater signal.

Within the range of oxidation conditions that occur near the earth's surface, Ce and Eu can change oxidation state. In seawater, changes in oxidation state affect the behavior and relative abundances of Ce and Eu with respect to other REE: chemical sediments formed in oxidizing marine environments typically show negative Ce and slight positive Eu anomalies. In our assessment of the REE data, we have adopted the computational formalisms of Bolhar et al. (2004) that account for possible positive La anomalies in calculating Ce/Ce* (the Ce anomaly), and for possible positive Gd anomalies in calculating Eu/Eu*.

The control of REE abundances by the siliciclastic fraction, and the diagenetic modification of carbonate REE, imply that variations in Ce and Eu are unlikely to be useful markers of the redox condition of the seawater column. Ce/Ce* < 1 indicates oxic conditions, and most of our samples show negative Ce anomalies (Figure 10). It is only in beds 29 and above, deposited under ferruginous conditions (Xiang et al., 2020), that Ce/Ce* is consistently above 1. This may reflect the association of Ce with Fe-bearing particulates (Alibo and Nozaki, 1999). In samples 24f7 and 24f7b, from the pyrite bed, Ce/Ce* is 1.07 and 0.91, with a value of 0.80 in the acetic acid extract from 24f7. Thus, even in the horizon in which euxinic conditions are indisputable, Ce/Ce* is recording a mixed signal. Zhao et al. (2013) have documented the same pattern in conodont albid crowns that are more resistant to diagenetic modification.

At Meishan, REE data do not distinguish between inputs from the continental tholeiites of the Siberian volcanic event and more local, arc-related volcanism. The mechanism of REE delivery would be different—aerosolized trace elements would be associated with the Siberian Traps, whereas trace elements in volcanic ash would be the mechanism of input from more local sources. Both would result in an overall increase in REE content, and in an increase in La/Yb, as observed upward in the stratigraphy at Meishan. Abundances of other trace elements like Cr, V, and Ti are higher in continental tholeiites, whereas those of Zr and Y are lower in continental tholeiites compared with silicic arc volcanics. These systematics suggest that input from a continental tholeiite source would result in elevated Cr, V, and Ti, with concomitant relative depletion of Zr and Y. This is opposite the observed changes in beds 24f–26, suggesting that the better option is input from local, arc-related volcanic sources. The interpretation is complicated by the contrasting trace element distributions measured in beds 25 and 28 at Meishan (He et al., 2014).