2 carbon sinks on geologic timescales. Yet, the fate of deeply subducted carbon remains an enigmatic aspect of the global carbon cycle.

The Makran subduction zone, with the highest carbon flux per unit length along trench of any modern subduction zone 9 , is uniquely well suited for investigating carbon recycling in a high-flux setting. The Makran continental margin is dominantly accretionary 10 and includes a Neogene-Quaternary volcanic arc that consists of three volcanic centers in Iran and Pakistan 11 . Here I present evidence that a carbonatite volcano in southern Afghanistan is a hitherto unrecognized eastern limb of the Makran volcanic arc. This volcano is proof that carbonatitic melts-theoretically generated in subduction zones at postarc depths (>140 km) 6,12 , but seldom attributed to volcanic arc processes 13 -can be voluminous enough to efficiently recycle subducted sedimentary carbon to the lithosphere. Isotope geochemistry and geochronology constrain the timing of sedimentary deposition and eruption, respectively, and imply that subducted carbon can return to shallow reservoirs on much shorter timescales than previously thought [14][15][16] .

As the Eurasian and Arabian plates converge, Arabian oceanic crust enters the Makran subduction zone at a rate of approximately 30 mm/yr 17 . The Arabian Plate carries Indus Fan sediments-the thickest sequence of sediments (5-7.5 km) entering any subduction zone-into the Makran trench, causing an unusually shallow grade of subduction and an exceptionally large accretionary prism 18 . The crustal wedge extends ~300 km north from Nature Geoscience Manuscript NGS-2020-12-02945 3 the trench, increasing in age from unconsolidated Quaternary sediments through imbricated Miocene ophiolites and metasediments 10 .

As previously defined, the Makran volcanic arc consists of three Miocene-Quaternary volcanic centers-Bazman, Taftan, and Koh-i-Sultan-along an east-northeast linear trend oblique to the trench (Fig. 1) 11 . Lavas erupted at these locations have basaltic to rhyolitic compositions and fall along calc-alkaline differentiation trends 19 . Subduction zone geometry varies from east to west: subduction in the east is shallower near the coast 20 and steeper at greater depths 21 compared to subduction in the west. This asymmetry might be due to oceanic lithosphere subduction in the east and continental lithosphere underthrusting in the west 22 .

The Khanneshin carbonatites (Extended Data Fig. 1) erupted through Neogene sedimentary rocks (Extended Data Fig. 2) of the Sistan Basin, southern Afghanistan, along intersecting regional faults 23 . The core of the 4-km-wide main vent consists of calcite-rich medium-to coarse-grained carbonatite (sövite) ringed by agglomeritic ankerite-barite carbonatite (Extended Data Figs. 345) 24 . Both units contain abundant mica-rich xenoliths of metasomatized wall rock (fenite) and are crosscut by fine-grained carbonatite (alvikite) dikes. A volcano-sedimentary apron extends radially 3-5 km from the main vent and is intruded by many late-stage dikes and volcanic plugs, the youngest of which are phonolitic. Aeromagnetic surveys suggest that as many as eight other minor alkaline igneous centers with unknown ages and compositions may be buried beneath Nature Geoscience Manuscript NGS-2020-12-02945 4 Holocene sands 23 . Unlike most volcano-forming carbonatites 25 , however, the Khanneshin volcano does not appear to be accessory to silicic volcanism.

Some geochemical data exist for the Khanneshin carbonatites 23,26,27 . This study adds two critical results: 40 Ar- 39 Ar geochronology that establishes the timing of Khanneshin volcanism and strontium isotopic constraints for Khanneshin rocks (see Supplementary Information for sample descriptions).

40 Ar- 39 Ar geochronology: Sövite sample RT-10K-09-inferred to be the least geochemically evolved because it has light carbon, oxygen, and thallium isotopic compositions 27 -was selected to represent the main stage of eruption. Unlike other micabearing Khanneshin samples, the coarse-grained phlogopite in RT-10K-09 is not associated with fenite xenoliths.

Step heating experiments on three phlogopite aliquots yielded 40 Ar/ 39 Ar plateau ages of 3.54, 3.74, and 3.83 Ma that consisted of 63, 45, and 35 percent of the total 39 Ar released, respectively (Extended Data Figure 6). Analytical uncertainty for each date is 0.02-0.04 Ma and the full external uncertainty for each analysis is 0.20 Ma. All three 40 Ar/ 39 Ar ages agree within uncertainty and could represent a single Ar closure age between 3.74 and 3.63 Ma. Alternatively, the oldest date (3.83 Ma) may represent the eruption age of the sövite if the younger dates record partial resetting caused by subsequent eruptive episodes that reheated the sample and caused Ar loss.

87 Sr/ 86 Sr, ranging from 0.707919 ± 7 (RT-11K-06) to 0.708061 ± 8 (KHAN-3) and with a mean of 0.708004 ± 36 (1s, n=20 rocks). This is comparable to previous results for Khanneshin rocks 26 and confirms that there is little Sr isotopic variability across the central vent of the volcano. The brecciated sandstone (represented by sample FH-10K-10) through which the carbonatitic lavas erupted has 87 Sr/ 86 Sr indistinguishable from the carbonatites; this value probably represents the isotopic composition of calcite veins (Extended Data Fig. 2) rather than the sedimentary protolith (see Supplementary Information).

Three lines of evidence indicate that the Khanneshin magmas were products of active subduction. First, the Khanneshin volcano is spatially and temporally associated with the Makran volcanic arc. Earthquake focal mechanisms within the subducting slab (Fig. 1) extend to 157 km depth 17 and the deepest earthquake occurred less than 70 km to the S-SE of the Khanneshin volcano 28 . This suggests that the slab passes beneath the volcano ~700 km north of the trench at a depth of ~180 km. The age of the Khanneshin volcano is bracketed by silicic volcanism at Bazman (4.6 to <0.6 Ma), Taftan (6.95 to <0.71 Ma), and Koh-i-Sultan (<2.5 Ma) volcanoes 29 , which are indisputably part of the Makran volcanic arc. Furthermore, the Khanneshin volcano is only slightly north of the linear trend defined by the other volcanoes (Fig. 1). These observations suggest that that the Khanneshin volcano is the easternmost manifestation of the Makran volcanic arc.

Nature Geoscience Manuscript NGS-2020-12-02945 6 Second, sediments on the downgoing slab beneath the Khanneshin volcano could be fertile sources of carbonatitic melt. In the Makran subduction zone, the slab-mantle interface may not heat to 450 °C until reaching a depth of 75 km 30 . Heating may accelerate as the slab passes through the lithosphere-asthenosphere boundary, located ~150 km beneath the Khanneshin volcano 31 . When heated above ~700 °C, sediment layers thicker than 1 km are prone to forming buoyant diapirs 2 , which undergo partial melting and efficient decarbonation as they ascend into the mantle wedge 32 . Sediments can also be convectively transferred from the slab into the mantle wedge 3 . During burial, subducted Indus Fan material (calcareous ooze and clay-rich terrigenous turbidites 33 ) would have consolidated and metamorphosed into carbonated pelites. At 950-1050 °C and 3-5 GPa 3,34 , these lithologies can produce carbonated alkali-rich melt. Such conditions are plausible near the top of the subducting slab and in the mantle wedge. Carbon-and alkalirich melt released from metasediments may separate into immiscible carbonatitic and silicate magmas during ascent. The buried igneous centers near the Khanneshin volcano 23 could represent conjugate silicate magmas. Thus, there are viable mechanisms by which carbonatitic melt might be generated beneath the Khanneshin volcano.

Exceptionally high rates of carbonatitic melt generation in the Makran subduction zone could be facilitated by (i) an abnormally high sediment flux, (ii) the abundant carbonate in the sediments (they can contain >50% CaCO3 33 ) and (iii) efficient carbon subduction past typical forearc depths 9 .

Third, the Khanneshin lavas contain geochemical evidence of crustal recycling. Their trace element patterns are consistent with the immiscible separation of carbonatitic magma from silicate magma derived from the melting of Indus Fan material (see Supplementary Information and Extended Data Fig. 7). Isotopically, Khanneshin, Bazman, and Taftan volcanic rocks fall along a common 143 Nd/ 144 Nd and 87 Sr/ 86 Sr trend (Fig. 2) that diverges from trends attributable to the magmatic fractionation of Rb/Sr and Sm/Nd. Such Sr-Nd decoupling has long been viewed as evidence of sedimentary recycling 35 . The Makran arc values (including the Khanneshin lavas) fall within the Enriched Mantle 2 array, which is likewise attributed to a recycled sediment component 36 .

Thallium isotope compositions of Khanneshin lavas also attest to ocean crust recycling, despite being strongly influenced by wall rock interactions 27 . Interestingly, the 206 Pb/ 204 Pb, 207 Pb/ 204 Pb, and 208 Pb/ 204 Pb (18.9, 15.6, and 40.0, respectively) of Khanneshin carbonatites 26 are very similar to average Indus River K-feldspars 37 ; Khanneshin lavas may have inherited this signature from continental detritus transported by the Indus River.

Elsewhere, carbonatites may contain ancient recycled material from lithospheric reservoirs reactivated by backarc extension, mantle wedge flow, or ascending magmas.

That is not a likely source of the recycled sedimentary component in the Khanneshin lavas. The Afghan block, through which Khanneshin magmas erupted, formed by the accretion of one or more Gondwana microcontinents to Asia by the Early Cretaceous 38 .

Southern Afghanistan, as part of the vast Alpine-Himalayan orogen, should have young lithospheric mantle (probably less than 50 Ma 39 ) and is far from cratons that could serve as long-lived geochemical reservoirs. Thus, and for the reasons outlined above-(i) spatiotemporal associations with arc volcanism and the Makran slab, (ii) viable mechanisms for voluminous carbonatitic melt production, and (iii) geochemical evidence for recycled sedimentary material-Khanneshin carbonatites can be viewed as products of Makran subduction.

Sediments in the Makran subduction zone are the most likely source of Khanneshin magmas. Because Khanneshin lavas have very high Sr concentrations (multiple wt% in most cases), substantial Sr isotopic contamination during magma ascent seems unlikely, even if Paleozoic or older rocks exist beneath the volcano. As noted above, the mantle lithosphere beneath the volcano is young, so remobilization of Sr implanted in the subcontinental lithosphere prior to the initiation of the Makran subduction zone also seems unlikely. Instead, the 87 Sr/ 86 Sr measured in Khanneshin rocks probably reflects the isotopic composition of subducted Indus Fan material. The Indus River initiated during the Eocene shortly after the India-Asia collision began and much of the fan appears to be Paleogene 40 . If the Khanneshin lavas inherited the 87 Sr/ 86 Sr of their sedimentary precursors, the monotonic increase in marine sedimentary carbonate 87 Sr/ 86 Sr since 40 Ma can be used as a chronometer (Fig. 3). Average marine carbonates 41 have 87 Sr/ 86 Sr overlapping with mean Khanneshin carbonatites (0.708004 ± 36) only once since 200 Ma.

The 87 Sr/ 86 Sr range in Khanneshin lavas suggests that their source rocks had a mean deposition age of 28.9 ± 1.4 Ma, which coincides with early Indus Fan growth. Assuming that 40 Ar/ 39 Ar plateau ages (3.83-3.54 Ma) record the main stage of volcanism, 24-27 Myr passed between deposition and eruption. The current length of the slab from the trench to beneath the volcano is ~700 km. However, the trench has migrated 140 km southward since the mid-Miocene 42 , so the slab segment currently beneath the volcano traveled only ~560 km since entering the trench. At the current rate of 32.6 mm/yr 17 , subduction to beneath the Khanneshin volcano would take 17 Myr. Thus, sediment Nature Geoscience Manuscript NGS-2020-12-02945 9 transfer to the trench and the ascent of recycled materials from the slab presumably lasted <10 Myr combined.

There is increasing isotopic evidence that some carbonatitic magmas worldwide contain recycled crustal material. Radiogenic isotope systematics suggest that carbonatites derive from either marine sedimentary carbon recycled through subcontinental lithosphere or deeply subducted carbonated oceanic crust that returns from either the transition zone or from the core-mantle boundary 14 . The role of recycling is confirmed by calcium isotopes 16 and boron isotopes 15 , but the recycling scenarios invoked by these studies require timescales of 100-1000 Myr. The Khanneshin volcano is evidence that sedimentary carbon recycles to and from typical postarc depths on short timescales (i.e., <100 Myr). Low-volume carbonatitic melts cannot survive metasomatic entrapment 43 and rapid devolatilization 44 during ascent, so the exceptionally high sedimentary carbon flux into the Makran trench may have been necessary to produce carbonatitic magmas voluminous enough to reach the surface. This might explain why carbonatite volcanism occurs in the Makran subduction zone, but not elsewhere on Earth 13 . Globally, subducting carbon inputs appear to be larger than volcanic arc outputs 6,8 . This imbalance is especially pronounced for the Makran and Andaman-Burma subduction zones, where the Himalaya-derived Indus and Bengal fans may transport >12 Mt/yr of carbon-roughly 7% of the global subducting carbon flux 1 -into the mantle wedge 9 .

Nature Geoscience Manuscript NGS-2020-12-02945 10 significant quantities of CO2 and SO2 (Barren Island, India 45 ), but there are several volcanic centers that have been active during the Quaternary (including Bazman, Taftan, and Koh-i-Sultan in the Makran arc) that are potential sources of diffuse volcanic CO2 and for which the modern and historical fluxes are unknown 46 . If Khanneshin volcanic CO2 emissions were comparable to those of the Oldoinyo Lengai carbonatite volcano in Tanzania 47 , it could have contributed 2.4 Mt/yr. Currently, most volcanic CO2 emissions in the region may come from the Tengchong volcanic field, China (4.5-7.1 Mt/yr 48 ), which is related to the subduction of the Burma slab 49 . Unless there are major CO2 sources undetected by satellite measurements 45 , the Makran and Andaman-Burma carbon inputs are probably not balanced by volcanic gas emissions, unlike in the Java and Sumatra subduction zones 50 . This suggests that the return of subducted carbon to the atmosphere is inefficient in subduction zones with the highest sedimentary carbon fluxes. In such settings, the carbon inputs must be balanced by carbon mixed into the convecting mantle 5 , the storage of carbon in lithospheric reservoirs 6 , or both. The Khanneshin volcano is evidence that subducted carbon can rapidly return to lithospheric reservoirs. It may be a rare surface expression of long-lived carbon reservoirs that form via carbonatitic magmatism in rear-arc lithosphere. The neodymium and strontium isotope compositions of Makran volcanic arc samples testify to sediment recycling. Khanneshin lavas fall along the same Enriched Mantle 2 trend as Bazman and Taftan samples 11,29 ; this trend cannot be explained by magmatic fractionation of Rb/Sr and Sm/Nd from mid-ocean ridge basalt values (hatched area) and indicates sediment recycling 35 . The Khanneshin data are from 26 and mantle geochemical trends are after 36 Khanneshin magmas probably derived from subducted sediments that melted near the top of the subducting slab or in buoyant diapirs that ascended into the mantle wedge.

Slab depths and the thermal structure are based on earthquake focal mechanisms 17 and thermal modeling 30 , respectively. The black dashed line along the top of the subducting Arabian Plate represents the trajectory of subducted Indus Fan material and the time elapsed since entering the trench. Note that only earthquakes with focal mechanisms >50 km deep (white) are plotted in Fig. 1.

Carbonatite samples were powdered by hand in an agate mortar and pestle. Between 5 mg and 10 mg of each sample was dissolved in a 3:1 mixture of HF and HNO3. Sr was separated and purified from the samples using Sr-Spec (Eichrom) resin. Sr isotopic measurements were performed on a Thermo-Finnigan Neptune ICP-MS at Woods Hole Oceanographic Institution. Isobaric interferences of 87 Rb on 87 Sr and 86 Kr on 86 Sr were corrected for by monitoring 82 Kr, 83 Kr, and 85 Rb and by applying a mass bias correction using an exponential relationship 51 . The internal precision for Sr isotopic measurements was 6-24 ppm. Raw Sr results are normalized using standard SRM987 ( 87 Sr/ 86 Sr = 0.7102140). Standard NBS987 was reproducible to within 25 ppm. See Extended Data Table 1 for strontium isotopic results.

Nature Geoscience Manuscript NGS-2020-12-02945 13 40

At the Oregon State University Argon Geochronology Lab, three phlogopite aliquots were cleaned for 30 min in 200 proof HPLC grade acetone and for 30 minutes in ethyl alcohol, and then were rinsed four times with triple-distilled water and dried at 55 °C for 12 hrs.

The samples were irradiated in the Oregon State University TRIGA reactor for six hours, along with Fish Canyon Tuff sanidines 52 that served as flux monitors. Following the procedures described in 53 , portions of each phlogopite aliquot (3.882, 9.351, and 4.251 mg for aliquots 1-3, respectively) were loaded in Cu-planchettes and step heated by rastering a 30 W Synrad CO2 laser beam across each sample under ultrahigh vacuum.

Reactive gases were cleaned with AP10 Zr-Al SAES getters at 450 °C and 21 °C before the argon was inlet into an ARGUS VI multicollector mass spectrometer for analysis.

Plateau ages (Extended Data Fig. 6)-based on contiguous extraction steps with apparent 40 Ar- 39 Ar dates that are indistinguishable at the 95% confidence interval-were calculated using ArArCALC v.2.6.2 software 54 using the decay constant 5.530 ± 0.097 x 10 -10 1/yr (2σ) from 55 and corrected by 56 . See Data S1, S2, and S3 for the complete results from the Argon Geochronology Lab.