Mercury spikes as evidence of extended arc-volcanism around the Devonian–Carboniferous boundary in the South Tian Shan (southern Uzbekistan)

Recently, the end-Devonian mass extinction (Hangenberg Crisis, 359 Ma) was identified as a first-order mass extinction, albeit not one of the “Big Five” events. Many marine and terrestrial organisms were affected by this crisis. The cause of this mass extinction is still conjectural and widely discussed. Here we report anomalously high mercury (Hg) concentrations from the South Tian Shan (Uzbekistan), together with correlation using conodont biostratigraphic data. Hg enrichment (to 5825 ppb) was detected in marine deposits encompassing the Hangenberg Crisis. In the Novchomok section, the Hangenberg Crisis interval does not contain typical Hangenberg Black Shales; however, by means of inorganic geochemistry (enrichment of redox-sensitive elements such as Mo, V, and U) we detected an equivalent level despite the lack of marked facies changes. This is the first record of Hg and Hg/total organic carbon anomalies in marly shales, marls and carbonates that are totally independent of facies changes, implying that volcanism was the most probable cause of the Hangenberg Crisis. This conclusion is confirmed by the presence of a negative δ13C excursion, which may reflect massive release of isotopically light carbon from volcanogenic and thermogenic devolatilization likely combined with increased arc-volcanism activity worldwide at the end of the Devonian.


Figure 1. (A)
Late Devonian (360 Ma) palaeogeography (after 91 see also 12 ) with the locations of Uzbekistan, the Novchomok section and the National Kitab Geological Reserve marked. (B) Tectonic model (after 78 ) of the study area (shown by the red star), including an amalgamating cluster of island arcs and accretionary complexes (grey arrow indicate plate movements; for detail see Fig. 12 in 78 ).

Results
Conodont biostratigraphy. Two zonation schemes are in use for the interval across the Devonian-Carboniferous boundary: 4 and 26 suggested a scheme similar to the previous one of Ziegler & Sandberg 27 , whereas 28 and 29 introduced a wide Bi. ultimus Zone that combines the former upper expansa and early and middle praesulcata zones.
Recently, 30 proposed establishment of the Protognathodus meischneri Subzone for the upper part of the Bi. ultimus Zone (see 31 : this subzone is approximately equivalent to the Lower praesulcata Zone of Ziegler & Sandberg 27 and to the praesulcata Zone of 26 (Fig. 4). In the absence of the nominal species, the characteristic taxa are Siphonodella praesulcata Sandberg and Protognathodus collinsoni, the first occurrences of which are slightly above the first appearance datum (FAD) of Pr. meischneri 28 . As no Siphonodella representatives were found in the studied Novchomok section, the zonation of 29 was applied in the present study, including the new Pr. meischneri Subzone. However, because the index taxa are missing, the present biostratigraphy is based on compilation of the stratigraphic ranges of other taxa found in www.nature.com/scientificreports/ particular samples, using the data of 29 . Moreover, it should be stressed that scarcity and poor preservation state of conodonts have a negative effect on the precision of the biostratigraphic calibration. The age of the lowermost sample (Nov 22-019) is constrained by the total stratigraphic range of Pseudopolygnathus controversus Sandberg and Ziegler (Fig. 5: 4) to the interval from the upper part of the Palmatolepis gracilis manca Zone to the Bispathodus costatus Zone. If the occurrence of Bispathodus cf. aculeatus aculeatus (Branson and Mehl) ( Fig. 5: 8), which appears higher in the section in sample Nov 22-09A, corresponds to the FAD of this subspecies, it would constrain the age of sample Nov 22-019 to the interval of the Pa. gr. manca to Pa. gr. expansa zones. The low abundance of conodonts in the discussed samples, however, makes such age determination risky. The age of assemblages from samples Nov 22-09A and Nov 22-01 is estimated as the interval from the Bispathodus ac. aculeatus Zone to the Bispathodus ultimus Zone. The upper limit is set by the first appearance in the section, in sample Nov 22-0, of Protognathodus collinsoni ( Fig. 6: 1, 2).
The richest sample Nov 22-0 is attributed to the Pr. meischneri Subzone within the upper part of the Bispathodus ultimus Zone. This age assignment is based on the occurrence of Pr. collinsoni specimens ( Fig. 5: 1, 2) and a lack of typical representatives of the latter species. More precisely, the occurrence of a transitional element between Pr. collinsoni and Pr. kockeli (Fig. 5: 3) may suggest the upper part of the Pr. meischneri Subzone. In the same sample, species not previously reported at such a high stratigraphic level occur: Pseudopolygnathus cf. brevipennatus ( Fig. 5: 6, 7) and Palmatolepis perlobata ssp. (Fig. 5: 8). These taxa became extinct in the lower part of the Bi. ultimus Zone 29 . Their occurrence may be explained by reworking from older strata, as suggested also by other fragments of Palmatolepis (Fig. 5: 5 and Fig. 5: 9-from the sample Nov 22-1) that recall older forms. However, with the exception of the scarce reworked elements, the remaining assemblage fits well in this interval, as no younger species are present. The fact that there are no species which have their FAD in the Pr. kockeli Zone (Pr. kockeli, Polygnathus purus subplanus, Po. politus) strongly supports our attribution. Reworking of older conodonts is a common phenomenon associated with the Devonian-Carboniferous boundary 4 .
In summary, the investigated strata of the Novchomok section, spanning the interval between samples Nov 22-019 and Nov 22-0, certainly belong to the uppermost Famennian. It should be stressed, however, that because of the scarce biostratigraphic data the succession of biozones (SD. 1) is hard to define precisely. The key sample Nov 22-0 gives the most precise age constraint, i.e. the upper part of the Protognathodus meischneri Subzone within the Bi. ultimus Zone. In turn, the assemblage from the sample Nov 22-019 is certainly older than the Bi. ultimus Zone, as explained above. Thus, the interval between samples Nov 22-019 and Nov 22-0 most probably

Carbonate and organic carbon isotopes
The δ 13  . A less pronounced positive δ 13 C org excursion (to -23‰) parallels the δ 13 C carb excursion in the D-C boundary interval (Fig. 2). The negative δ 13 C excursion corresponds to the largest Hg peaks (see below; SD. 3). Carbon isotopes in marine carbonates reflect a real change in ocean chemistry but maybe also affected by diagenetic alteration. Diagenetic processes are particularly evident for negative carbon isotope shifts. The isotopic composition of bulk rock organic matter is influenced by variable sources of input and differential degradation of organic components 32 . An increase of thermal maturation is typically associated with 13 C enrichment 33 . Therefore, the higher δ 13 C org measured in the TOC-poor samples may reflect the thermally induced mobilization compounds that are enriched in 12 C relative to the bulk of the organic carbon ( 33,34 ; SD 3). Lower δ 13 C values within the HBS equivalent (see below) may be explained either by incorporation of 12 C derived from oxidized organic matter (δ 13 C carb ), a change in the TOC composition (δ 13 C org ) 35 , or liberation of massive amounts of 13 C-depleted CH 4 and CO 2 e.g. 36 . The latter were suggested as a major contributors to the negative excursion  www.nature.com/scientificreports/ of both carbonate and organic carbon isotopes during e.g. the Permian-Triassic boundary transition e.g. 37 and Early Toarcian 38 . A comparison of δ 13 C variation curves in the Novchomok and the other D/C sections from different facies of distant continents reveals similar variation patterns (see below), which may be explained by changes in the global carbon cycle.  www.nature.com/scientificreports/

Redox-sensitive trace elements and total organic carbon contents
Redox conditions can be deduced on the basis of indices such as the Th/U and C org/ P ratios [39][40][41] , as well as from enrichment of redox-sensitive trace metals such as Mo, U and V 42,43 . The Th/U ratio of anoxic siltstones or shales is less than 3, whereas in carbonates this ratio is typically below 1 (for details see references 39,40 ). The C org /P ratio in anoxic conditions is greater than 150, whereas for sediments formed under oxic conditions the value is below 30. Intermediate values are characteristic of dysoxic or high-productivity and periodically oxygen-restricted settings 41 .
In the studied section, the molybdenum values in many samples (especially in the upper part of the section) were very low (< 0.5 ppm, often below the detection limit of 0.1 ppm; Table 1). In contrast, in samples enriched in Hg, Mo levels were generally > 1 ppm with a maximum value of 11.1 ppm. Uranium levels in the samples were between 0.2 to 5.1 ppm. The Th/U ratios in samples enriched in Hg were low, < 1 in carbonates and < 3 in shales, indicating oxygen-depleted benthic redox conditions. However, these low values could in some horizons be an Table 1. Hg (ppb), TOC (%), Hg/TOC (ppb/%) ratio, CaCO 3 , Al 2 O 3 , Fe 2 O 3 , total sulphur (TS, %), Mo (ppm), As (ppm), V (ppm), U (ppm), Th (ppm), P (%), Th/U and C org/ P ratios, δ 13 C org (‰) and δ 13  www.nature.com/scientificreports/ artifact associated with strong depletion in thorium, as well as other detrital-fraction elements. Interestingly, total sulphur was below the detection limit (< 0.02%) in the entire section. The low C org /P ratios throughout the section are characteristic of low-productivity and oxic conditions, which is confirmed by the very low total organic carbon (TOC) values of less than 0.2%. Horizons with Hg anomalies are characterized by higher V values of 17-486 ppm ( Fig. 3; Table 1), whereas in almost all samples from other parts of the section the V content is below the detection limit, implying more oxic conditions. The level equivalent to HBS is also enriched in uranium and molybdenum (Fig. 3): even if the concentrations of the elements have been depleted by diagenesis, they still indicate more restricted conditions.

Mercury spikes around the Devonian-Carboniferous boundary
Hg enrichment was observed in several horizons in the Akbasay Formation (uppermost Famennian) in the carbonate-rich, organic-poor Novchomok section (Fig. 2) 3). An abrupt increase in Hg concentrations and the largest Hg/TOC peak coincide with the negative excursion in carbonate and organic δ 13 C (Fig. 2, SD. 3).

Discussion and conclusions
The Hangenberg Crisis is associated with a transgression pulse, development of anoxic conditions in the sea, and climate warming e.g. [4][5][6]12 . Several possible primary causes of the Hangenberg Crisis have been proposed: climate and glacioeustatic changes; global carbonate crisis resulting from oceanic acidification; salinity changes; phytoplankton blooms and expansion of land plants; volcanism; and extraterrestrial impact (e.g. 4,6,45 . There is as yet no clear consensus about the cause of the event, and the primary triggers are still a matter of vigorous debate. In the Novchomok section the HC interval does not contain typical HBS. The anoxic interval was detected in the studied section, however, by means of elevated Mo concentrations (Fig. 3, Table 1) in the dark shaly-marly www.nature.com/scientificreports/ package (Fig. 2), which may be regarded as the stratigraphic equivalent of the HBS horizon. The occurrence of the HBS equivalent implies that the base of the meischneri Subzone is located somewhere below the dark interval, i.e. below sample Nov 22-011. A similar late first occurrence of species of the genus Protognathodus has been documented in other localities and has been explained by ecological factors 46 . Submarine hydrothermal activity and volcanic eruptions are the main natural sources of mercury in recent and ancient environments, and are reflected by Hg spikes in sedimentary rocks [13][14][15] . Interestingly, all the "Big Five" mass extinctions are associated with mercury spikes, supporting extensive volcanism having been a primary cause of environmental changes during these events 7,17,47-51 . Recently, the volcanic "smoking gun" as a potential trigger of the Hangenberg Crisis has been questioned 8 . According to 8 , elevated UV-B radiation and a drastic drop in stratospheric ozone during global warming were responsible for this biotic overturn. Moreover, Fields et al. 52 , postulated supernova explosions as a trigger for the event. These hypotheses are very attractive, but there is a lack of hard evidence to support the extraterrestrial scenario, whereas the occurrence of extensive volcanic activity during that time in many parts of the world confirms an Earth-bound scenario (e.g. 6,7 ).
In many areas pyroclastic horizons occur in the uppermost Famennian (Fig. 7), such as in Poland 5,6,53 , Germany ( 7,54-56 ), the Iberian Peninsula 57 and China 58 . Recently, convincing evidence of increased volcanic and hydrothermal activity as given by mercury spikes has been detected in the HC interval in Vietnam 10 ; the Czech Republic 11 ; south China 7,11 ; the Holy Cross Mountains, Poland 7 ; Thuringia and Bavaria, Germany 6,7,9 ; and the Carnic Alps in Austria and Italy 6,12 . Menor-Salvan et al. 57 described a hydrothermal peak in the Iberian Peninsula, the last phase of which (at ~ 360 Ma) formed the world's largest cinnabar ore reservoir in Almadén.
Previously documented mercury spikes 9 in the Novchomok section, which is now dated by means of conodonts, partly predate the biocrisis; however, the main and upper Hg spikes encompass the HC interval (Fig. 2), confirming the volcanic scenario for this event. Our results from Novchomok include five Hg anomalies, of which the two lower anomalies (samples Nov 22-016 and Nov 22-013) predate the Hangenberg Crisis (Fig. 2), whereas the upper spikes (samples Nov 22-010 to Nov 22-05) could be related to the end-Devonian event.
The lower part of the Hangenberg Crisis interval (sensu 4 ) is characterized by a globally observed δ 13 C negative excursion in the carbonate 59-62 and organic-matter records 6 . The minima in the carbon isotope curves coincide with Hg enrichment and an Hg/TOC excursion in surprisingly many places around the world in the initial phase of the HC 64 . According to some authors 64 , Hg enrichment is connected to occurrences of Hg-enriched sulphides and cannot be interpreted as a volcanic proxy, especially for the Ordovician-Silurian, Frasnian-Famennian and Permian-Triassic boundaries. According Shen et al. 65 , a sulfide host phase for Hg occur in strongly euxinic environments with high TS contents (> 1.0%). Still, generally, Hg is most commonly associated with the organic fraction. However, other authors 43 have proposed that neither Hg/TOC nor Hg/S are significantly linked with the organic and sulphide fractions, and therefore are useful as a volcanic proxy (compare 66 ). These ratios are not significantly influenced by redox conditions 7,43 . This absence of linkage can be confirmed in our investigated section, because the observed spikes occur independently of lithology, in limestones and marls as well as in shales. The Hg vs. TOC correlation in the Novchomok section is very low (R 2 = 0.16; SD. 3), indicative of a volcanic Hg source, not a bioproductivity-controlled Hg oversupply 7 . The Hg vs. Al 2 O 3 correlation in the investigated successions is also very weak (R 2 = 0.24), indicating no correlation of Hg with the clay fraction and terrigenous input. In addition, the entire investigated section is strongly impoverished in total sulphur (trace amounts, Table 1), which excludes the sulphide fraction as a host of mercury.
Negative δ 13 C excursions can reflect massive release of isotopically light carbon from volcanogenic and thermogenic devolatilization in a giant volcanic system ( 67 ; see R&S hypothesis in 63 ). The killing effectiveness of a volcanic cataclysm depends on the geological setting of the host region, the size of the igneous province, the magma plumbing system and the eruption dynamics 68 . The form and eruption dynamics of the volcanic system control the magnitude and composition of the thermogenic outgassing that most probably causes the greatest disruption to the carbon cycle (e.g. [69][70][71] ). Moreover, a large igneous province (and possibly arc magmatism and arc-continent collisions) can drive anoxia via global warming 68,72 ; therefore, the global D-C biodiversity crisis, similarly to the Permian-Triassic extinction 73 , may have been mainly driven by volcanism-linked anoxia. As described by 12 , the presence of 55 pg/g dry weight of MeHg in the HBS interval (sample Nov 22-010) could indicate methylmercury poisoning, which may have been an additional driver of the end-Devonian Mass Extinction (for detail see 12 ).
One of the problems with the volcanism hypothesis is a lack of precisely dated Large Igneous Provinces (LIP) during this time interval ( 74 , for details see 7 ). Nevertheless, the activity of the continental silicic Maritimes (Magdalene) LIP in eastern Canada includes the 360-370 Ma interval, with a pulse at around 360 Ma 75,76 . Moreover, LIPs often precede the main extinction intervals, which can be explained by delayed ecosystem responses, because volcanism leads to climatic changes (warming) 7 (the press-pulse volcanic model; 7 ). According to Ernst et al. 76 , a second main pulse of the Kola-Dnieper and Yakutsk-Vilyui LIPs occurred around 360 Ma, which may be tentatively correlated with previous Famennian ocean anoxic episodes such as the Annulata or Dasberg Events and/or the Hangenberg Crisis.
Another possibility is that intraplate (?oceanic; 4 ) LIPs were consumed in subduction zones; in this case, Hg enrichment could be the only reliable evidence for increased volcanic activity at the planetary scale 7 . Many magmatic and volcanic rocks (Fig. 7) as well as ash layers and hydrothermal deposits occur near the D-C boundary, associated with observed mercury enrichments in different palaeogeographic regions 6,7 , confirming that the Late Devonian was a time of many active magmatic centres.
The high mercury concentration in the Novchomok section may also suggest an especially close location to a volcanic Hg source, which may have been submarine volcanic activity associated with hydrothermal pulses, see e.g. 6  www.nature.com/scientificreports/ Late Devonian and early Carboniferous 78 and the arc-volcanic Magnitogorsk zone (Fig. 1B). The zircon ages of this zone include the D-C boundary time 79,80 , and so the zone could potentially have been the source area of Hg. However, the main activity of the Devonian magmatic centre in Magnitogorsk Zone predated the Hangenberg Crisis. According to 81 , the Magnitogorsk island arc was active only until the end-Frasnian; if this is the case, another Hg source must be sought. In fact, volcanic activity appears to have been waning at the end of the Devonian Period in the east Magnitogorsk zone, but was still occurring 82 . Additionally, dating of sulphide mineralisation in the Urals volcanic-hosted massive sulphide deposits (362 ± 9 and 363 ± 1 Ma; 80 ) and intrusive magmatism in the Ural Platinum Belt, where continental-margin gabbro-tonalite-granodiorite magmatism was initiated at ~ 365-355 Ma 83,84 , include the time interval of the end-Devonian crisis. Many pyroclastic rocks related to volcanism within an intra-oceanic arc occur around the D-C boundary at Baruunhuurai Terrane, Mongolia 85 ; however, precise biostratigraphic data are not available for this interval. It is also possible that the Hg-source island arc was consumed in a subduction zone during closure of the Uralian or Turkestan Ocean.
Marshall et al. 8 described floral malformations in east Greenland and interpreted the observed floral mutagenesis to be a result of elevated UV-B radiation, reflecting ozone-layer reduction that was suggested to have been associated with global warming. Surprisingly, according to those authors, mercury data excluded planetary-scale volcanism as a potential trigger for the end-Devonian Hangenberg Crisis.
However, according to another study 86 , floral malformations during this time interval were associated with the mutagenic effect of regional acidification caused by explosive (arc-type?) volcanism, recorded in common pyroclastic horizons. Visscher et al. 87 and Foster & Afonin 88 argued that this type of floral mutagenesis observed at the Permian-Triassic boundary reflected the biotic response to environmental stress associated with increased volcanic activity of the Siberian Traps coupled with degradation of the ozone layer and increased UV radiation. According to 89 , volcanic activity of the Central Atlantic Magmatic Province and SO 2 emissions were responsible for malformation of plant cuticles. Thus, plant mutagenesis supports, rather than excludes, a volcanic scenario. LIP-related cooling connected to darkness and occurrence of SO 2 and its products in the atmosphere is rather short-term, in contrast to global warming induced by release of volcanogenic CO 2 , which could have caused ozone damage and a consequent increase in UV-B radiation 90 . Furthermore, as shown in the study by 77 for the Palaeocene-Eocene Thermal Maximum, the occurrence of Hg and Hg/TOC anomalies may be related to phreatomagmatic eruptions and submarine degassing from hydrothermal vent complexes leading to local deposition of Hg-enriched sediments.
The predicted causal relationship between large-scale volcanic activity, volcanically driven climatic and redox changes, and the response of the global carbon cycle is clearly visible in the end-Devonian (and also, for example, the end-Frasnian, P-T boundary and end-Triassic) stratigraphic (and biotic) record(s). In contrast, there is no hard evidence of an extraterrestrial cause of the Hangenberg Crisis, such as a supernova explosion.

Methods
Micropalaeontological preparation. Recovery of conodont elements was attempted from nine samples from the presumed D-C boundary interval, comprising the boundary between the Akbasay and Novchomok formations (SD. 1). Sample processing was carried out in the micropalaeontological laboratory of the Polish Geological Institute -National Research Institute, in Warsaw. The rock material-mostly dark grey, poorly metamorphosed limestones-was dissolved in 20% formic acid, and the insoluble residuum was enriched using heavy liquid (sodium metatungstate). No identifiable conodont elements were found in samples Nov 22-015, Nov 22-07 and Nov 22-3. All of the conodont elements are stored in the Polish Geological Institute-National Research Institute, in Warsaw.
In general, the obtained microfossils were difficult to identify because of their poor preservation. Much of the material consists of conodont fragments that are difficult to determine even at the genus level. The elements that were selected for taxonomic description are broken, fractured, twisted, compacted, plastically deformed and commonly covered with sediment and with authigenic mineral overgrowths. The high temperatures (300-480 °C) indicated by the Conodont Alteration Index values (see "Study area") may indicate the occurrence of regional heating associated with deep burial and/or weak metamorphism. Tectonic stress caused fracturing of the specimens, leading to their disintegration during sample processing.
The Isotope analysis. Bulk-rock samples for sedimentary organic carbon isotope analysis were pulverized and acidified with excess 10% HCl and held at 60 °C for at least 8 h to remove inorganic carbonate material. Samples were then rinsed with ultrapure (> 18 MΩ) deionized water to remove acid and oven-dried at 60 °C. Analyses of the isotope signature of organic carbon in sediment (δ 13 C org ) were performed using a Thermo Flash EA 1112HT elemental analyser combined with a Thermo Delta V Advantage isotope ratio mass spectrometer in continuousflow mode at the Institute of Geological Sciences, Polish Academy of Sciences (Warsaw). Isotope values for carbon are given in parts per thousand (‰) relative to the Vienna PeeDee Belemnite (VPDB) standard and calibrated according to certified international standards USGS 40, USGS 41 and IAEA 600. The measurement precision (1σ) was ± 0.15‰.
Samples for δ 13 C carbonate analysis were reacted with 100% H 3 PO 4 at 70 °C to produce CO 2 . The isotope measurements were carried out using a KIEL IV device connected online to a FinniganMAT Delta plus isotope mass spectrometer (IGS PAS, Warsaw). Results are expressed in δ notation relative to the VPDB and normalized to international standards NBS 18, NBS 19 and IAEA-CO-9. The measurement precision was better than ± 0.08‰. www.nature.com/scientificreports/