The formation of new subduction zones, or Subduction Initiation (SI), is fundamental to the onset and stabilization of plate tectonics1, yet the processes and settings involved in the nucleation of a new convergent plate boundary remain challenging to reconstruct2. Critical to our understanding of SI are geological records of the incipient subduction plate contact, but these are inaccessible in modern subduction zones. Exposed relics of such interfaces have been recognized in the form of sub-ophiolitic metamorphic soles (soles) – thin slabs of metamorphic rocks found at the base of supra-subduction zone ophiolites and showing an inverted metamorphic gradient3,4,5,6. These metamorphic rocks are derived from upper oceanic crust and are interpreted as being accreted from the lower to the upper plate during SI5,7,8,9. The overlying supra-subduction zone ophiolites are interpreted as allochthonous relic forearc oceanic lithosphere of the upper plate formed during or shortly after SI10,11. Accordingly, the ophiolite-sole association is generally seen as diagnostic of intra-oceanic SI12,13. The high-temperature granulite facies peak metamorphic conditions reported in most soles (11–15 kbar, ≥800 °C7) are further interpreted as reflecting a high geothermal gradient at the site of SI, leading to the association of ophiolite-sole couples with SI at or near oceanic spreading centers3,12,13,14. However, the inference that ophiolite-sole couples are diagnostic of intra-oceanic SI may be challenged, especially where it can be demonstrated that SI was induced, whereby upper plate extension and ophiolitic crust generation post-dates initial underthrusting1,15.

Recent studies using garnet Lu-Hf geochronology of garnet-clinopyroxene amphibolite in soles revealed that initial underthrusting may predate upper plate extension and formation of the overlying supra-subduction zone ophiolitic crust by more than 8 Ma15,16,17. Determining the setting of SI therefore requires reconstructing the upper plate lithosphere at the time of metamorphic sole burial, rather than at the time of ophiolitic crust generation18,19. This is of particular importance in the case of ophiolite-sole pairs that formed next to continental margins, like the Xigaze Ophiolite20 and its sole21,22 in South Tibet.

The Xigaze Ophiolite was generated at the southern margin of the continental Lhasa Terrane of southern Tibet23,24 through forearc extension25,26,27 that generated a slow28 to ultra-slow29 spreading center above the north-directed subduction zone that eventually led to the India-Asia collision30. Estimates of timing of this SI vary: on the one hand, those using the age of ophiolite spreading and hyperextension as a proxy for the birth of the subduction zone place SI around ~130 Ma, in which case sole formation occurred below the forearc when ophiolites were spreading21,31,32. If, on the other hand, ophiolite spreading post-dates SI, then sole formation may have started below the lithosphere within which the ophiolites formed. Such an older SI age of even 170 Ma has been proposed to explain magmatism on the southern Lhasa Terrane30,33, although this magmatism may also be related to southward subduction along the northern Lhasa margin34,35,36. The current state of the art thus allows for multiple interpretations of SI timing, hampering interpreting the tectonic setting and conditions under which the Xigaze sole formed.

In this paper, we aim to determine the timing of initial lower plate underthrusting through Lu-Hf garnet geochronology, as a direct way to date whether the Xigaze Ophiolite metamorphic sole formed prior to or after forearc extension and ophiolite formation. We therefore investigate garnet-clinopyroxene amphibolite from the Xigaze sole and complement the Lu-Hf garnet isochron ages with Lu trace element maps to support age data interpretation. We discuss the implications of our results for the timing and setting of north-directed SI under the Lhasa margin and for the generic mechanism and setting of sole formation during SI in general.


Geological setting and sampling

The 2000 km long Indus Yarlung Zangbo Suture Zone, in South Tibet (Fig. 1A), exposes the remnants of the Neo-Tethys Ocean that once separated India from the Lhasa Terrane20. The central segment hosts the Xigaze Ophiolite (XO), which comprises several ophiolitic massifs (Fig. 1B) that are typically composed, from the base up, of a mantle tectonite section, rare cumulates and gabbro of the lower plutonic crust, a sill/dike complex and pillowed basalts20,37, and a Lower Cretaceous radiolarian chert38 sedimentary cover interfingered with arc-derived volcanic ash layers and Lhasa Terrane-derived turbiditic sandstones23,24. These Lower Cretaceous ages of the oldest sedimentary cover of the ophiolite are coincident with a ~ 130 Ma U/Pb zircon age of gabbro of the XO39, and with similar 130–120 Ma ages along-strike20,31. Paleomagnetic data and provenance of the XO sedimentary cover indicate formation at the immediate southern margin of the Lhasa Terrane as a forearc to the Gangdese arc23,24,26. The mantle section of the XO has a long history of interaction with subduction fluids and is not genetically related to the overlying crust40,41: instead, its characteristics are consistent with at least partial derivation of the XO mantle section from Lhasa Terrane subcontinental lithosphere25,26,42. To the South, the ophiolite is overthrusting a sheared serpentinite mélange that contains a dismembered sole21,22,43 and oceanic lithosphere-derived blocks44,45. Farther below is the Bainang subduction complex, a thrust-stack of trench-fill and abyssal radiolarian mudstone and chert offscraped from the Neo-Tethys seafloor during north-directed subduction46. This package was eventually thrust onto the continental margin-derived Tethyan Himalayan Sequence in Paleogene time47.

Fig. 1: Geology of the Xigaze ophiolite and its metamorphic sole.
figure 1

A Simplified tectonic map of the India-Asia collision zone with major faults and boundaries showing main ophiolite belts (in black) and study area (box B). B Simplified geological map of the central segment of the Yarlung Zangbo Suture Zone77; location of C is indicated. C Detailed geological map of the sampling area77. Red line is a major south-dipping thrust sense fault; location of D is indicated. D Field photo of sampling site with geological interpretation and sample location.

The XO sole consists of blocks of grt-cpx amphibolites, common amphibolite, and greenschist embedded in the serpentinite matrix of the ophiolitic mélange. The high-pressure granulite blocks show evidence of partial melting as mostly concordant leucosomes48. The maximum metamorphic conditions recorded in the grt-cpx amphibolite of the XO sole are in the high-pressure granulite facies ~14 kbar and over 850 °C22,32,43,48, in agreement with the reported leucosomes, and are consistent with the other soles worldwide7. They returned 133–119 Ma U-Pb zircon dates that are interpreted either as post-peak metamorphism or igneous protolith ages32,48, apatite U-Pb ages of 132–133 Ma48, and 40Ar/39Ar hornblende cooling ages of 130–119 Ma21,43, all overlapping with the age of overlying ophiolitic crust39,49. Such a dismembered sole sharing the same geochronological and petrological characteristics has also been described under the Saga ophiolite, farther west along the central segment of the Yarlung Zangbo Suture Zone43.

We analyzed four grt-cpx amphibolite specimens in this study, which were sampled from decametric blocks in the ophiolitic mélange of the valley east of Bainang (Fig. 1C, D). Field relationships, petrography, mineral chemistry, geochemistry, and Ar geochronology of these rocks were reported previously21,22. All four samples show a hornblende-dominated nematoblastic fabric with anhedral prehnitized plagioclase and slightly coarser-grained grt–cpx-rich horizons (Fig. 2A–D). Both domains show fine-grained ilmenite–titanite symplectites. Garnet of mostly almandine and grossular composition occurs as subhedral centimeter-scale pre- to syn-kinematic porphyroblasts with inclusions of clinopyroxene, plagioclase, hornblende, and titanite/rutile/ilmenite.

Fig. 2: Garnet petrochronology.
figure 2

AD µ-XRF element scans of thin sections for Fe, Ti and Mn. Pl plagioclase, Hbl hornblende, Ilm ilmenite, Ttn titanite, Ru rutile. EH Lutetium distribution maps of garnet porphyroblasts. IL Lu-Hf garnet isochrons.

Garnet chemistry

Garnet grains show remarkably diverse Lu zoning patterns between and within samples (Fig. 2E–H). Garnet from BAI01 exhibits a bell-shaped concentric zoning consistent with Rayleigh fractionation, with Lu content decreasing from core (~10 ppm) to rim (~3 ppm). Garnet from CG64 shows complex zoning with a patchy Lu-depleted core with inherited fabric, a slightly Lu-richer mantle (~6 ppm), and a variably Lu-poor rim (3–4 ppm). Garnet from LUS17 broadly shows homogeneous content, hinting to potential peak diffusion, with clear signs of resorption at some boundaries and associated Lu enrichment (up to 5 ppm). Some sub-grains still retain Rayleigh-type growth zoning (lower left corner of Fig. 2G). Garnet from LUS12 features sharp oscillatory zoning in a generally Lu-poor mantle overgrowing a patchy xenomorphic Lu-richer core (up to 20 ppm).

Lu-Hf garnet geochronology

The analyzed samples provided four Lu-Hf garnet ages that were identical within uncertainty (Table 1). Garnet from sample BAI01 yielded a Lu-Hf whole rock-garnet isochron age (Fig. 2I) of 143.7 ± 0.7 Ma (MWSD = 1.6; number of garnet analyses ngrt = 4), whereas that of sample CG64 (Fig. 2J) yielded a 144.6 ± 0. 9 Ma isochron (MSWD = 0.52; ngrt = 4). Garnet fractions were extremely difficult to extract for sample LUS17, mainly due to the fragile nature of grains. The sample yielded an Lu-Hf isochron age (Fig. 2K) of 140.1 ± 4.5 Ma (MSWD = 0.12; ngrt = 2). Garnet from sample LUS12 (Fig. 2L) yielded a 144.3 ± 3.5 Ma (MSWD = 0.29; ngrt = 3) and required omission of the fourth garnet fraction for which the isotopic ratio did not correspond to that of the other fractions. The weighted mean of these four isochron ages is 144.0 ± 1.2 Ma (MSWD = 1.8).

Table 1 Isotopic data.


Evolution of the south tibetan metamorphic sole

The robustness of the Lu–Hf geochronometer50,51 is largely governed by the diffusivity of Hf, which is demonstrably sluggish52. The peak temperatures that the Tibetan sole samples were subjected to do not exceed estimated closure temperatures of diffusive Hf loss, which is in excess of 950 °C for grains analyzed in this study50. Lutetium is more mobile, so the hypothesized effects of any diffusive net transfer of Lu between garnet and matrix52 are to be considered. The Lu element maps provide a useful means to do so. The Lu distribution and zoning in garnet from all four samples is different (Fig. 2E–H), both in terms of primary zoning and the degree to which resorption and possible Lu reuptake has affected grains. Yet, despite these differences, the Lu–Hf dates obtained for all samples are identical within uncertainty. The only apparently different age component present among the samples is observed for coarse fragments derived from particularly large grains in LUS12a; this material (fraction Grt4) appears older than most of the garnet in that sample, which is inconsistent with any age bias by Lu redistribution in mafic rocks52 and instead may indicate inherited cores (Fig. 2H). Regardless, the prevalence of c. 144-Ma Lu-Hf ages for garnet of different average grain size, composition and zoning indicates that actual age bias, either due to differences in zoning or possible diffusive Lu reuptake (e.g., LUS17), did not significantly influence the age data; the Lu-Hf age data thus reliably represents the timing of garnet growth at ~144 Ma. As prograde garnet growth in supracrustal mafic protoliths requires an increase in both P and T, we interpret those growth ages as a minimum age for formation of the XO metamorphic sole at an incipient plate contact, as is the case in other soles15,16,17. Other studies have reported zircon ages of 133–119 Ma for the same locality48 and an adjacent one32. Apatite ages of 132–133 Ma have also been reported from the Bainang valley48. These authors interpreted some of the zircon and all apatite grains as igneous, implying that the older range of the 133–119 Ma dates would be protolith ages. However, the authors could not rule out a metamorphic origin. We note that the zircon ages overlap with 130–121 Ma Ar–Ar on hornblende cooling ages for the same locality, as also reported from the well-studied Semail ophiolite-sole couple of Oman. Zircon of the Semail sole was demonstrated to have formed during the crystallization of partial melts generated during granulite metamorphism of the sole15,53,54. In addition, apatite is highly unstable in the presence of leucosome and will likely be consumed by melt producing reaction before crystallizing upon cooling55. The garnet Lu-Hf and hornblende Ar-Ar geochronology of the Bainang samples constrain prograde metamorphism near 144 Ma and cooling between 133 and 121 Ma; we therefore interpret the reported zircon and apatite ages from granulite of the XO metamorphic sole as dating the crystallization of a melt fraction following peak metamorphism, consistent with the leucosomes reported in the field and on samples48. Metamorphism of the XO sole involved a long prograde stage to granulite-facies peak metamorphism followed by rapid cooling from peak conditions, a similar scenario as reported for the Semail sole15,56. Definitive evidence for the age of the protolith in metamorphic soles remains elusive, or controversial at best, as reviewed elsewhere13.

Our new data thus show that subduction of Neotethyan seafloor was underway by 144 Ma, providing the earliest direct evidence for northward subduction of Neo-Tethys seafloor south of the Lhasa Terrane. As the XO formed above this north-directed subduction directly at or close to the margin of the Lhasa Terrane23,25,26,42, the underlying sole must have been buried following SI around the ocean-continent boundary (Fig. 3A, B). After nucleation of the subduction zone, the sole was transferred from the lower to the upper plate, accreting to Lhasa Terrane continental mantle (Fig. 3B, C).

Fig. 3: Tectonic model for sole formation during subduction initiation under a continental margin.
figure 3

A lithospheric cross-section of the pre-subduction Tibetan margin, (B) forced subduction initiation at the continental margin causes metamorphism in the lower plate and garnet growth, (C) increasing slab pull force during convergence causes slab roll-back, (D) slab-pull force reaches a critical point triggering viscous necking of the subduction interface, trench retreat, upper plate extension, and exhumation of the metamorphic sole. (E) Continued northward subduction of Neotethyan oceanic lithosphere under the Lhasa Terrane margin with the Xigaze ophiolite and sole forming the basement of the clastic forearc basin.

Numerical models of this setting predicts that viscous necking of the subduction interface (Fig. 3D), trench retreat, and continental forearc hyperextension (Fig. 3E) followed as the increasing slab pull force eventually overcomes coupling at the subduction interface27. The sole was then exhumed between 130 and 120 Ma, as indicated by sole cooling ages, synchronous with limited mantle melting and generation of the XO crust. The forearc extension model3,11,26,27 following induced subduction initiation therefore provides a comprehensive explanation for the lag time between burial and exhumation of soles as well as a mechanism for sole exhumation synchronous with upper plate extension and supra-subduction zone ophiolite spreading. Ensuing north-directed subduction of Neotethyan oceanic lithosphere resulted in the formation of the Gangdese magmatic arc at the southern margin of the Lhasa Terrane34 and deposition of the Xigaze basin clastic sediments on the XO24, until obduction over the Tethyan Himalayan continental margin30,47.

Soles and subduction initiation at continental margins

Our new results demonstrate that garnet growth in the Tibetan sole was already underway 14 Ma before forearc extension and formation of the Xigaze Ophiolite. This has several first-order implications for the regional tectonic evolution, and for the formation of soles in general.

First, northward underthrusting of oceanic lithosphere below the Lhasa block largely predating ophiolite formation implies that SI was induced by far-field stresses15, ruling out spontaneous subduction initiation. The 14 Ma lag time between lower plate burial (Lu-Hf garnet ages of the sole) and upper plate extension (U-Pb zircon ages of ophiolitic crust) and associated, synchronous, rapid sole exhumation and cooling (U-Pb zircon and apatite and Ar-Ar hornblende ages of the sole) is longer than in Oman (8 Ma15), California (~10 Ma16), and Turkey (~12 Ma17,57). This longer delay in Tibet may possibly reflect a stronger coupling at the subduction interface, consistent with a thicker continental upper plate and thus a longer plate interface. However, some of the timing estimates could also be minimum estimates. Still, the similarity and differences in timing is an important point and an avenue of future research that needs to be explored.

Second, the trigger of SI should be sought at or before 144 Ma. The identification of this trigger is beyond the scope of this paper, but we note that SI occurred as the Lhasa Terrane was approaching the Eurasian margin58, if not colliding with it35. Subduction may thus have been initiated by transference, perhaps diachronously: the forces required to induce SI at a continental margin are significantly lowered in the case of lateral propagation along inherited rift structures59.

Third, the 144 Ma garnet growth age for the Xigaze ophiolite sole directly demonstrates that soles also form during subduction below continental lithosphere, as proposed before19 arguing against the inference that ophiolite-sole couples are diagnostic of intra-oceanic SI. Moreover, the P-T conditions, lithologies, mineralogy, and chemistry of the XO sole rocks are typical of soles worldwide, including those that unequivocally formed in intra-oceanic settings (e.g., in the Mirdita Ophiolite60). This is surprising, as peak temperatures in excess of 850 °C in soles have long been inferred to be indicative of subduction initiation under a hot mantle hanging wall3,5,8,9. Here, we demonstrate that SI under a continental mantle resulted in a metamorphic sole that recorded the same P-T conditions22,32. Perhaps this demonstrates that the mantle below the southern Lhasa margin was anomalously hot, e.g. related to the arc magmatism that affected this region already since Jurassic times or before30,34. On the other hand, metamorphic soles of Oman and Turkey formed by subduction initiation in ancient lithosphere, devoid of pre-existing magmatism that could have caused elevated temperatures18,19, and these soles recorded near-identical conditions as the XO sole7,15,17,61. So regardless of dynamic interpretations and speculations on the causes of these high-temperature conditions, our analysis shows that the metamorphic conditions recorded are the same for soles that form by SI near a mid-ocean ridge62,63,64, along ancient transform faults in >100 Ma old oceanic lithosphere18,19, or even along a continental margin as we show here, spanning a remarkable variety of settings with contrasting thermal gradients. Moreover, these conditions are consistently found for soles that date back at least 480 Ma8,65. We thus conclude that the sole formation process is irrespective of the setting of SI but may rather relate to the subduction initiation process and/or the formation of the overlying ophiolite during forearc extension, processes implicitly common to all soles. Constraints from metamorphic soles may accordingly be considered generic and provide key insight in the processes that governed the initiation of Phanerozoic subduction zones in general, regardless of the setting in which they formed.


Samples were collected from four adjacent outcrops (see Fig. 1D) around location CG-64 at a longitude of 89.33311°E and a latitude of 29.14045°N. Thin sections were mapped using a μ-X-ray fluorescence spectroscopy (μ-XRF) Tornado M4 instrument at Université Laval. Analytical conditions were a 20 µm step size with an acquisition time of 3 ms per pixel. The X-ray tube was set at 50 keV and 300 mA for a total acquisition time of c. 2.5 h per sample. Garnet grains showing the strongest major element zoning from core to rim, mainly observed as a bell-shaped spessartine zoning, were selected for LA-ICP-MS Lutetium mapping.

Trace element analyses of garnet were performed by LA-ICP-MS at LabMaTer (Université du Québec à Chicoutimi), using a Australian Scientific Instruments RESOlution 193 nm excimer laser and an S155 Laurin Technic ablation cell system coupled to an Agilent 7900 quadrupole ICP-MS. High-resolution mapping was performed in-situ on garnet to document trace element zoning. Analytical conditions were a 20 μm beam moving at a speed of 80 μm s−1 and pulsing of 30 Hz at 5 J cm−2 in a 4 ms per mass cycle. The synthetic reference glass GSE-1G66 was used for calibration. Data reduction was processed with the Iolite freeware67 using mean EPMA 29Si data as an internal standard.

Lutetium-hafnium chronology was performed in two separate laboratories. Samples were crushed and garnet fractions were hand-picked under a binocular microscope and contained all sizes of grain and random core/rim proportions. Garnet and whole rock (WR) splits from samples LUS12 and LUS17 were analyzed at the Department of Earth Sciences, University of California, Santa Barbara. Additional garnet materials from these samples, as well as all analyses for samples BAI01 and CG64 were done at the Pacific Centre for Isotopic and Geochemical Research, University of British Columbia, Vancouver. Both methods utilize mixed 176Lu-180Hf isotope tracers made from different base metals and oxides, and independently developed and calibrated68,69. For Lu-Hf analysis, garnet fractions and WR powder were transferred to screw-top PFA vials. Garnet grains were washed using de-ionized water and ethanol, dried, transferred to PFA vials, and bathed in 1 N HCl at room temperature for 1 h. After removing the HCl, samples were mixed with a 176Lu-180Hf isotope tracer with matrix-equivalent Lu/Hf ratio, and digested through repeated addition of HF:HNO3:HClO4 and 6 N HCl, each step followed by evaporation to dryness. After tracer admixing, the WR powders were digested in stainless-steel Parr vessels at 180 °C for 7 days using HF:HNO3. After digestion, all samples were dried, re-dissolved in 6 N HCl, diluted to 3 N HCl using de-ionized H2O, and centrifuged. The solution containing the garnet elemental solute was then loaded onto polypropylene columns containing a 1-ml Ln-Spec resin bed before being analyzed through REE-HFSE chromatography70.

Hafnium and Lutetium isotope analyses were done with a Nu Instruments Plasma/multi-collector inductively coupled plasma mass spectrometry (MC-ICPMS) instrument. For Lu analyses, isobaric interference of 176Yb on 176Lu was corrected using an exponential correlation of 176Yb/171Yb and 174Yb/171Yb, calibrated through replicate analyses of NIST Yb solution standards at different concentrations 10-100 ppb71; For Hf isotope analyses, 180Ta and 180W interferences were corrected by analyzing 181Ta and 183W. Mass bias was assumed to follow an exponential law and was corrected applying 179Hf/177Hf = 0.7325 and 173Yb/171Yb = 1.129672. Drift was corrected by assuming linear time dependence. Hafnium isotope values are reported relative to those of ATI-475, an in-house-developed Hf isotope reference material made from the original Hf metal ingots from which the international reference solution JMC-475 was made, with 176Hf/177Hf = 0.2821673. Replicate analyses of ATI-475 done at concentrations bracketing that of samples helped estimate external reproducibility of 176Hf/177Hf and was 38 ppm during the course of our analytical session74. The Lu-Hf isochrons were established using a λ176Lu decay constant of 1.867 × 10−11 yr−1,75,76. All uncertainties are cited at the 2-s.d. level.