The monazite record of (ultra)high-temperature metamorphism in southern Madagascar: Gondwana terrane correlations and structural trapping of hot middle crust

The Precambrian bedrock exposed in southern Madagascar (Fig. 1) has been subdivided into lithotectonic domains bounded by faults and shear zones (GAF-BGR 2008a–e; Boger et al. 2012). Table 1 summarizes these domains, a few of their defining characteristics and putatively correlative units in Africa and India.

In the east, the Antananarivo domain (GAF-BGR 2008c) consists of Archean to Paleoproterozoic felsic orthogneisses thought to be part of the Greater Dharwar Craton (Tucker et al. 2011a, 2014) or the Proterozoic Azania microcontinent (Collins and Pisarevsky 2005), which are overlain by the Paleoproterozoic to Neoproterozoic sedimentary rocks, including the Itremo Group or domain (GAF-BGR 2008d; Archibald et al. 2015; Boger et al. 2019). The Antananarivo domain and its supracrustal sequences were intruded by the c. 800 Ma plutons of the Imorona–Itsindro suite (Handke et al. 1999; Moine et al. 2014; Nédélec et al. 2016; Zhou et al. 2018). West of the Antananarivo domain is the narrow NNW-trending Ikalamavony domain, composed of metavolcanic and metasedimentary rocks including ocean-floor basalts, c. 1000 Ma calc-alkaline mafic to felsic intrusions of the Dabolava Suite and the c. 800 Ma Imorona–Itsindro suite (GAF-BGR 2008d; Tucker et al. 2014; Boger et al. 2019). To the SW of the Ikalamavony domain, the Anosyen domain consists of intercalated Neoproterozoic cordierite–garnet–sillimanite pelitic gneisses, clinopyroxene–scapolite calc-silicate gneisses and leucocratic garnet–sillimanite–mesoperthite gneisses referred to as leptynites (Windley et al. 1994; GAF-BGR 2008b; Boger et al. 2014, 2019), the protoliths of which appear to have been deposited on c. 2 Ga para- and orthogneiss basement (Paquette et al. 1994; Tucker et al. 2014). The vertical, north-striking Beraketa shear zone (Martelat et al. 1999, 2000) separates the Anosyen domain from the Androyen domain. The Androyen domain has a similar variety of rock types to the Anosyen domain, but with less pelitic material in the metasedimentary gneisses (biotite-, hornblende- and/or garnet-bearing gneisses), more abundant quartzofeldspathic gneiss and more common hornblende–pyroxene metabasic rocks (Windley et al. 1994; GAF-BGR 2008a; Boger et al. 2019). The protoliths of the Androyen gneisses appear to have ages of c. 2 Ga (Tucker et al. 2014), with Nd model ages of 3.3–2.6 Ga (Boger et al. 2019). Based on differences in whole-rock geochemistry and general lithology, Boger et al. (2019) proposed a Graphite subdomain in the southwestern Androyen domain. The Androyen domain (specifically the Graphite subdomain) was intruded at c. 900 Ma by the bimodal Ankiliabo suite, composed of anorthosite and alkali-granite. The Androyen domain is separated from the Vohibory domain, largely composed of Neoproterozoic metabasites and metasediments (Emmel et al. 2008; GAF-BGR 2008e; Jöns and Schenk 2008; Tucker et al. 2014; Boger et al. 2019), by the vertical north-striking Ampanihy shear zone (Martelat et al. 1999, 2000).

The monazite-bearing rocks of this study include pelitic, semi-pelitic and quartzofeldspathic migmatites (of sedimentary or granitoid protolith) (sample coordinates and mineral assemblages are listed in Supplementary material Table S1). Monazite grains were dated in thin section by laser ablation split-stream (LASS) inductively coupled plasma mass spectrometry (ICP-MS) at the University of California, Santa Barbara. Analytical details are provided in Supplementary material Document 1.

Monazite composition varies orders of magnitude within and among samples. To assess this variability and compare samples, logarithms of four elemental ratios were plotted as functions of U–Pb date: La_n/Sm_n, Eu_n/Eu* (Eu anomaly: Eu* = Eu_n/(Sm_nGd_n)^0.5), Gd_n/Yb_n and Th/U (subscript n denotes chondrite-normalized concentration; McDonough and Sun 1995). These were then fitted with smoothing curves (moving average and cubic spline), with 95% confidence intervals calculated by Monte Carlo analysis (resampling n = 1000). For samples and elemental ratios that showed statistically significant change, defined as variation of the smoothing curve greater than its 95% confidence bound, those plots were centred and scaled for ease of comparison (analogous to taking a z-score). Each spline fit was scaled such that the standard deviation of its distribution of values became unity. The splines were then centred such that their value at 545 Ma was zero; 545 Ma was chosen for centring because it represents the approximate middle of the main stage of orogenesis and is a date shared by all samples. This centring and scaling allow for direct visual comparison of the direction and magnitude of compositional change between many samples, even if the absolute values differ substantially (for example, owing to bulk-rock composition and different mineral modes). Additional details of the fitting are provided in Supplementary material Document 1.

Sample F4824B from the southern Androyen domain (24.6568°S, 45.55638°E; Fig. 1) was selected for phase-equilibrium modelling. This sample is a garnet–cordierite–sillimanite–biotite gneiss (Fig. 2) with minor ilmenite, zircon and monazite. Two thin sections were made from different textural domains for petrographic analysis. The first thin section (F4824B1), a cordierite-rich gneiss with minor garnet, is most representative of the gneiss. The second thin section (F4824B2) was made from a c. 1 cm thick garnet- and sillimanite-rich band. Both record the same mineral relationships but with different mineral modes. A bulk-rock composition was obtained from the same portion of the sample as the first thin section (F4824B1) because it is more texturally homogeneous than the thin sillimanite–garnet-rich layer (F4824B2). Analytical details are provided in Supplementary material Document 1.

A P–T pseudosection was calculated using the internally consistent mineral endmember thermodynamic database of Holland and Powell (2011) and the activity–composition relationships for solid solutions of White et al. (2014). Calculations were conducted using a version of Theriak–Domino (de Capitani and Petrakakis 2010) compiled by D. K. Tinkham (downloaded from https://www.dtinkham.net/peq.html on 6 December 2021). The H₂O content was estimated from the proportion of biotite in thin section F4824B1 (0.35 wt%). The sensitivity of the results to assumed Fe³⁺/Fe^total was tested by repeating the calculations with Fe³⁺/Fe^total = 0, 0.06 and 0.12. Higher proportions of ferric iron were not considered reasonable owing to the absence of magnetite–spinel solid solutions in this rock (Diener and Powell 2010; Boger et al. 2012). To assess plausible prograde phase equilibria, we followed the melt-reintegration method of Korhonen et al. (2013) (see also White et al. 2004). We report two steps of melt reintegration following this approach (a total reintegration of c. 12 vol% melt). Beyond this amount of melt reintegration, solidus temperatures are too low to compare the results with the observed mineral assemblages and textural relationships. The whole-rock and reintegrated melt compositions used for the phase-equilibrium calculations are given in Supplementary material Table S2. Additional description of the calculations is provided in Supplementary material Document 1.

We compiled mineral compositions published by previous metamorphic studies (Martelat et al. 1997; Jöns and Schenk 2008, 2011) (Supplementary material Table S3) and one new sample from the Vohibory domain (F4827A; garnet–biotite–sillimanite gneiss: 23.8974°S, 44.86678°E). These were used as inputs for the THERMOCALC AvPT method of thermobarometry (Powell and Holland 1994), which calculates a set of equilibrium P–T conditions based on the intersection of independent mineral reactions determined from mineral compositions. For mineral compositions taken from previous studies, published P–T results are available. The P–T conditions recalculated from those samples do not differ significantly from those in the original studies; however, the recalculation by a single method, and with the same internally consistent thermodynamic data, minimizes possible systematic uncertainties that might exist. Calculations were conducted using the programs THERMOCALC (Powell and Holland 1988) version 3.40 and AX2 by T. Holland (downloaded from <https://filedn.com/lU1GlyFhv3UuXg5E9dbnWFF/TJBHpages/index.html> on 25 April 2016).

All measurements and calculations are tabulated in the Supplementary material.

Sample F4824B1 from the southern Androyen domain contains garnet, cordierite, sillimanite, alkali feldspar, plagioclase, quartz and biotite with minor ilmenite, zircon and monazite. Garnet and cordierite contain inclusions of sillimanite and biotite. Alkali feldspar contains inclusions of sillimanite. Representative photomicrographs are shown in Figure 2. Biotite occurs primarily as xenoblastic grains in the interstices of matrix cordierite and feldspar or as coronae around ilmenite. Broadly, the petrographic observations from this location are similar to those reported for many rocks of the Anosyen domain by Boger et al. (2012).

The observed mineral assemblage (garnet–cordierite–sillimanite–biotite–quartz–alkali-feldspar–plagioclase–ilmenite) was calculated to be in equilibrium just below the solidus: ≤800°C and c. 0.4–0.5 GPa (Fig. 3). Isopleths of garnet core compositions also intersect in this field at 700–775°C and c. 0.5 GPa (Fig. 3 and Supplementary material Fig. S1).

The main differences in the topology of the phase equilibria after melt reintegration are related to changes in the stability of garnet and melt (Fig. 3). Using the measured rock composition, garnet is stable at >0.4 GPa for the entire temperature range investigated (700–950°C), both above and below the solidus. The solidus is the biotite-breakdown melting reaction at 750–800°C. After melt reintegration, garnet stability is restricted to >750°C and >0.4 GPa, forming only as a peritectic mineral during biotite-breakdown melting. The solidus after melt reintegration is the fluid-saturated melting reaction at low pressure (<0.4 GPa) and the muscovite-breakdown reaction at higher pressure (>0.4 GPa). The solidus temperature is 50–150°C lower after melt reintegration.

The thermobaric ratio (T/P) of peak metamorphism varies across southern Madagascar (Fig. 4). It is lowest in the Vohibory domain, with an inverse-variance weighted mean AvPT result of 766 ± 40°C, 0.90 ± 0.06 GPa and 846 ± 34°C GPa^–1 (±1 SE of the mean; n = 7). The weighted mean AvPT result for the Androyen domain is 726 ± 39°C, 0.70 ± 0.05 GPa and 1032 ± 44°C GPa^–1 (1 SE of the mean; n = 7). Metamorphic temperature does not vary significantly across the Ampanihy shear zone, but mean metamorphic pressure and thermobaric ratio are statistically significant at 98.96 and 99.9% confidence intervals, respectively.

Fewer AvPT results are available for the easternmost Androyen domain and the Anosyen domain, but available data (Jöns and Schenk 2011) indicate an increase in thermobaric ratio eastward into the Beraketa shear zone and Anosyen domain (to as much as c. 1700°C GPa^–1). Phase-equilibrium modelling (Boger et al. 2012; Holder et al. 2018a; Horton et al. 2022), which has provided more precise P–T constraints than conventional thermobarometry for the eastern portion of the study area (Frost and Chacko 1989), indicates higher metamorphic temperature (≥900°C) and thermobaric ratio (>1300°C GPa^–1) in the Beraketa shear zone and Anosyen domain compared with the Androyen domain (<850°C, c. 1000°C GPa^–1). No discernible difference in metamorphic pressure is recognized between the Anosyen and Androyen domains.

Concordant monazite U–Pb dates range from c. 630 to 500 Ma (Fig. 5 and Supplementary material Fig. S2), comparable with previous results of U/Th–Pb geochronology in southern Madagascar (Paquette et al. 1994; Ashwal et al. 1999; Kröner et al. 1999; Markl et al. 2000; Martelat et al. 2000; de Wit et al. 2001; Emmel et al. 2008; Jöns and Schenk 2008, 2011; Giese et al. 2011; Tucker et al. 2011b, 2014; Boger et al. 2014, 2019; Horton et al. 2016, 2022; Holder et al. 2018a; Holder and Hacker 2019). Samples from the Ikalamavony, Anosyen and Androyen domains yielded discordant dates that define a broad discordia with an upper intercept of c. 2 Ga, a recognized basement age for the Androyen and Anosyen domains (Fig. 5) (Tucker et al. 2014). Concordant c. 630–600 Ma dates are observed most prominently in the Vohibory and Androyen domains, but they are also clearly present in multiple samples of the Anosyen domain (Fig. 5). Six samples in the Anosyen domain yielded multiple analyses with concordia dates of 630–600 Ma that do not fall along mixing lines with a c. 2 Ga component. Another sample gave a single concordant date that fits these criteria. One sample also yielded numerous concordant 630–600 Ma dates, but with high ²⁰⁷Pb/²⁰⁶Pb uncertainty such that they might simply be mixing with a c. 2 Ga component.

Southern Madagascar monazites have both lobate and oscillatory chemical zoning. Sample F4818D1 monazites (Fig. 6) from near Tôlanaro (Fig. 1) have chemical zoning representative of Anosyen domain samples: irregular lobate zoning exists in the cores and oscillatory zoning, often interpreted as growth in the presence of melt (Lederer et al. 2013), is observed in the outer portions of some grains. Monazites in sample F4824B2 (the sillimanite-rich domain of the sample used for phase-equilibrium modelling) have representative Androyen zoning: discrete lobate domains of variable composition that cross-cut each other (Fig. 7). The irregular, cross-cutting, lobate zoning observed in both domains is commonly interpreted to indicate recrystallization of monazite by interface-coupled dissolution precipitation (Putnis 2009) owing to evolving fluid and/or melt composition during metamorphism (Seydoux-Guillaume et al. 2002, 2012, 2019; Hetherington et al. 2010; Budzyń et al. 2011; Harlov et al. 2011; Williams et al. 2011; Grand'Homme et al. 2016; Varga et al. 2020).

Monazite composition (La/Sm, Eu/Eu*, Gd/Yb and Th/U) changes systematically as a function of U–Pb date in many of the studied samples. Moving-average and cubic-spline smoothing curves for all samples are presented in Supplementary material Figure S3. Representative Anosyen (F4818D1; Fig. 6) and Androyen (F4824B2; Fig. 7) samples have common features: the oldest dates are c. 620 Ma and there are inflections at c. 580 Ma. In F4818D1, additional inflections occur at c. 540 Ma. For both samples, Eu/Eu* decreases approximately one order of magnitude between c. 580 and c. 540 Ma, whereas Gd/Yb and Th/U increase approximately an order of magnitude over the same interval. In F4818D1, there is also a decrease in La/Sm during this interval.

Of the 77 samples dated, seven had fewer than five analyses, owing to either analytical issues or scarcity of monazite. Smoothing curves were not fitted to these samples. Among the remaining 69 samples, 54 samples showed statistically significant variation in the cubic-spline fits of Eu/Eu* (variation greater the 95% confidence interval of the spline), 45 samples showed change in Gd/Yb, 34 in La/Sm and 51 in Th/U. For most samples that did not yield statistically significant change in Eu/Eu*, Gd/Yb or Th/U, the spread in U–Pb dates was relatively low (<40 Ma), such that trends in composition would be difficult to resolve with the typical 15–20 Ma uncertainty (±7–10 Ma 2 SE) of individual spot analyses. The lower number of samples that yielded statistically significant change in La/Sm can probably be attributed to the smaller variability in this ratio (typically less than an order of magnitude) compared with Eu/Eu*, Gd/Yb or Th/U.

The scaled and stacked spline fits reveal similar and coincident changes in monazite composition in the Anosyen domain, Androyen domain and the Beraketa shear zone that divides them (Fig. 8). The ratio La/Sm shows the most consistent pattern: decreasing prior to 540 Ma, then increasing in some samples after 540 Ma. The ratio Eu/Eu* also decreases consistently before 540 Ma and increases again afterward in some samples; however, three of 54 samples anomalously show the opposite behaviour (two Anosyen, one Androyen). The ratios Gd/Yb and Th/U are less consistent in their behaviour among the studied samples. Some samples show increases prior to 540 Ma, whereas others show decreases; regardless of the sign of change, samples often show inflection or reversal c. 540 Ma. The largest compositional changes occur between 580 and 540 Ma; compositional changes prior to 580 Ma are smaller in magnitude.

First, we discuss the regional P–T–t history of southern Madagascar, advocating for a shared metamorphic and structural history of the Anosyen and Androyen domains. We then use these inferences as a basis for reassessing the geochronological and structural correlations of the Madagascar terranes with the adjacent crustal provinces of Gondwana (southern India and eastern Africa). We propose an updated reconstruction of the orogen in which the Anosyen–Androyen domain of Madagascar was part of an upright, transpressional structural zone between divergent thrust systems. We then present a hypothesis for sustaining long-lived (ultra)high-temperature metamorphism in southern Madagascar in the structural and temporal context of the orogen.

The rocks of southern Madagascar experienced two metamorphic events, at 630–600 and 580–500 Ma. The c. 630–600 Ma event has long been recognized in the Vohibory and Androyen domains and corresponds to the main stage of regional metamorphism in much of Eastern Africa, from Tanzania (Fritz and Hauzenberger 2021) to Israel and Jordan (Elisha et al. 2017, 2019). This period of orogenesis corresponds to the Eastern African Orogeny, whereas the young phase of orogenesis at 580–500 Ma is the Kuunga Orogeny (Meert 2003). A few sparse dates have been used to speculate that the 630–600 Ma metamorphic event might also have affected the Anosyen domain (Jöns and Schenk 2011; Horton et al. 2016). However, because these data were sparse (a handful of in situ U–Pb spot analyses across a couple of samples), only the 580–500 Ma event has been widely recognized in the Anosyen domain (Boger et al. 2015, 2019; Holder et al. 2018a).

From our survey of metamorphic monazite dates, we identify eight samples in the Anoysen domain with concordant dates in the 630–600 Ma range (Figs 1 and 5). These include granulites as far SE as Tôlanaro and as far north as Ihosy. We therefore advocate that the entirety of the Anosyen domain, like the Androyen, experienced the earlier regional metamorphic event. The weaker signal of this event in the Anosyen domain than in the Androyen domain may be due to (1) the earlier metamorphism that was not as high grade to the east or (2) the higher peak metamorphic temperatures (900–1000°C) and the greater abundance of reactive and melt-fertile pelitic rocks in the Anosyen domain, which led to more complete recrystallization of pre-existing metamorphic monazite.

Similar monazite compositional trends and inflections are observed across the Anosyen domain, Androyen domain and Beraketa shear zone (Fig. 8). This further supports the interpretation of a shared metamorphic history through Ediacaran–Cambrian orogenesis. In the following paragraphs, we discuss each of the compositional ratios evaluated, their possible petrological interpretations and their implications for the timing of prograde, peak and retrograde metamorphism in this domain.

1. The light rare earth element (LREE) ‘slope’, La/Sm. The most consistent composition pattern observed among monazites from this study is a decline in the La/Sm ratio over time. In some samples, there is an inflection and reversal at c. 540 Ma. Controls on the La/Sm ratio in monazite have not been thoroughly investigated. Krenn et al. (2009) documented multiple generations of monazite within a high-pressure granulite with distinct LREE slopes. Based on textural observations and monazite Th content, they attributed this variation to reaction with apatite. Apatite generally has lower La/Sm and Th than coexisting monazite (e.g. Bea and Montero 1999), such that the release of LREE from apatite might produce monazite with distinctly lower La/Sm and Th than pre-existing monazite in that rock. In support of this hypothesis, small amounts of monazite often grow at the expense of apatite in high-grade rocks and in melting experiments (Wolf and London 1995; Harlov et al. 2007; García-Arias et al. 2012; Rocha et al. 2017). The decline of monazite La/Sm in the Madagascar monazite might therefore represent apatite breakdown during partial melting. As a caveat, however, the lower La/Sm ratios in the Madagascar monazite are not clearly correlated with lower Th concentrations, as one might predict if La/Sm decline was the result of apatite breakdown. An alternative hypothesis is that the decrease in La/Sm is due to more complex mass balance or fractionation of LREE among the LREE-rich phases (monazite, apatite, and melt) with progressive melting. In support of this hypothesis, fewer samples from the lower-temperature Androyen domain record resolvable decreases in La/Sm over time compared with the higher-temperature Anosyen domain.

2. The Eu anomaly, Eu/Eu*. Nearly all samples show statistically significant change in Eu/Eu*. In most samples, Eu/Eu* decreases, most notably between 580 and 540 Ma, with an inflection and sometimes reversal at c. 540 Ma. Decreases in monazite Eu/Eu* have most commonly been attributed to the growth of feldspar (e.g. Rubatto et al. 2006; Holder et al. 2015). Particularly for low-pressure rocks, the production of alkali-feldspar during mica-breakdown reactions may be important (Rubatto et al. 2006). However, Eu partitioning is very sensitive to fO₂ (Holder et al. 2018a, 2020; Yakymchuk et al. 2023), which may change substantially during metamorphism (e.g. Ague et al. 2001) owing to open-system fluid flow or closed-system unbuffered mineral reaction (e.g. Fe–Ti oxide reactions: Holder et al. 2018a, 2020). Given the differences in compatibility of Eu²⁺ and Eu³⁺, it is also feasible that melt extraction could influence this ratio (Holder et al. 2020), particularly in residual migmatites such as those of this study area. Whatever the mechanism (or combination of mechanisms), the vast majority of Madagascar samples show systematic decrease in this ratio associated with the high-grade metamorphic event. If this decrease corresponds to prograde metamorphism (mica breakdown, melt extraction, reduction or a combination of these processes), the reversal in behaviour at c. 540 Ma in some samples would be consistent with the interpreted age of peak metamorphism and subsequent retrogression (e.g. Taylor et al. 2015; Holder et al. 2018a; Horton et al. 2022).

3. The heavy rare earth element (HREE) ‘slope’, Gd/Yb. In most samples this ratio increases prior to c. 540 Ma, at which time it reverses in many samples. Changes to the HREE slope are most often attributed to garnet growth or breakdown. Prograde growth of garnet fractionates HREE (Bea and Montero 1999; Pyle and Spear 1999; Foster et al. 2000; Hermann and Rubatto 2003; Rubatto et al. 2006, 2013; Spear and Pyle 2010; Mottram et al. 2014), resulting in higher monazite Gd/Yb ratios. Subsequent breakdown of garnet during retrogression would release HREE, lowering Gd/Yb for retrograde monazite. Following this interpretation, we interpret the Gb/Yb inflection at c. 540 Ma as the timing of peak metamorphism, which is consistent with the interpretation of Eu/Eu* trends described above. A smaller number of samples show the opposite pattern: a decrease in Gd/Yb prior to c. 540 Ma and a subsequent increase. In the Anosyen domain, this is explained by high-temperature garnet breakdown (releasing HREE and lowering Gd/Yb) by the reaction garnet + sillimanite = Al-spinel + quartz (e.g. samples F4818A1, F4818B1, F4822H1: Holder et al. 2018a) and subsequent retrograde garnet growth (fractionating HREE, raising Gd/Yb) by reversal of this reaction during retrogression. Although the sign is opposite, the reversal at c. 540 Ma records the same transition from peak to retrograde metamorphism. In the Androyen domain, each of the samples that shows this opposing trend is anomalous. Sample F2914D1 is a quartz–plagioclase–alkali-feldspar gneiss with large (several millimetres) equant garnet. Garnet is mantled by biotite, which does not exist elsewhere in the rock. The majority of monazite dates correspond to the earlier c. 630–600 Ma metamorphic event and are associated with higher Gd/Yb ratios that are consistent with garnet growth. We therefore interpret the mineral assemblage of this sample to have been minimally affected by the main, second phase of regional metamorphism (c. 580–500 Ma); we attribute c. 540 Ma dates with lower Gd/Yb to minor fluid addition, which produced the biotite rims on the older garnet during the second metamorphic event. Sample F4825B1 contains abundant graphite, muscovite and barite, but no garnet; controls on the Gd/Yb ratio in this sample must be controlled by other mineral or fluid reactions. Sample F4825D2 contains an unusual skeletal network of garnet along the grain boundaries of quartz and feldspars, consistent with an interpretation of late garnet growth.

4. The Th/U ratio. Of the compositional variables investigated, this ratio is the least consistent among samples. In some samples, Th/U monotonically increases over time during the metamorphic event. In others, it decreases. And in many samples, Th/U changes reverse over time or exhibit more complex behaviour. Theoretically, Th/U ratio should increase through the granulite facies owing to greater compatibility of Th than U in monazite and therefore preferential retention of Th in monazite in residual gneisses (e.g. Alessio et al. 2018; Williams et al. 2018, 2022; Yakymchuk and Brown 2019). Many samples from both domains show a pattern of increasing Th/U, sometimes with a reversal at c. 540 Ma. These reversals might be interpreted as the timing of prograde-to-peak metamorphism and subsequent retrogression, as for Eu/Eu* and Gd/Yb ratios discussed above. However, many samples show the opposite or more complex trends. Given the sensitivity of Th/U ratios to fluid chemistry and redox state, we hypothesize that the inconsistency of this ratio among samples may be due to variable fO₂ of these gneisses (e.g. Boger et al. 2012) and the complex, high-temperature metasomatic history of this granulite terrane (Rakotondrazafy et al. 1996; Pili et al. 1997a, b; Boulvais et al. 1998, 2000; Moine et al. 1998; Ramambazafy et al. 1998; Martin et al. 2014). Whatever the controls on monazite Th/U, the most prominent inflections and reversals observed in individual samples occur at c. 540 Ma, coinciding with reversals of other compositional ratios.

The Vohibory domain experienced pressures 0.2–0.3 GPa higher than the Androyen domain (Figs 4 and 9). This is consistent with a significant dip-slip component in the Ampanihy shear zone that vertically displaced the Vohibory domain by c. 8–11 km relative to the Androyen domain (Martelat et al. 1997, 1999, 2000; de Wit et al. 2001). However, peak Vohibory metamorphism and development of the Ampanihy shear zone are linked to 630–600 Ma orogenesis (de Wit et al. 2001; Jöns and Schenk 2008), whereas peak Androyen metamorphism occurred at c. 540 Ma. Thus, the estimate of throw should be interpreted with caution because peak pressure east and west of the Ampanihy shear zone was not achieved at the same time. This diachronism supports the interpretation of this shear zone as an early suture that remained an important orogenic boundary.

We found no metamorphic evidence for preferential exhumation of the Anosyen or Androyen domain across the Beraketa shear zone (Figs 4 and 9). Both domains exhibit similar mineral reaction textures (e.g. this study; Jöns and Schenk 2011; Boger et al. 2012; Holder et al. 2018a) and P–T paths (Fig. 9). The main difference is higher peak temperature in the Anosyen domain (≥900°C) compared with the Androyen domain (c. 800–850°C). This is consistent with the interpretation of this shear zone as a strike-slip simple-shear-dominated structure (Martelat et al. 1997, 1999, 2000). Curiously, several studies that focused on samples in and near the Beraketa shear zone (Ackermand et al. 1989; Markl et al. 2000; Jöns and Schenk 2011; Horton et al. 2022) have reported metamorphic pressures of 0.8–1 GPa. These are 0.1–0.4 GPa higher than the preferred pressure estimates of 0.6–0.7 GPa for the Androyen and Anosyen domains (Fig. 9). This might represent systematic errors among the thermobarometric approaches employed or the different rock types studied, rather than actual differences in metamorphism. Alternatively, slightly higher-pressure rocks may have been exhumed within the Beraketa shear zone in response to rheological contrast between the lower crust and lithospheric mantle during shearing, a model that has recently been proposed based on gravity anomalies across the southern Madagascar shear zones (Martelat et al. 2020).

The Anosyen and Androyen domains have been combined and subdivided into several lithotectonic units over the last 30 years (Boger et al. 2019). The domains contain notable similarities: upright structural orientation (Martelat et al. 1999, 2000), high-grade metamorphism, c. 2 Ga continental basement ages (Tucker et al. 2014), possibly with an older Archean basement component (Cahen et al. 1984; Boger et al. 2019), and common lithologies (Windley et al. 1994; GAF-BGR 2008a, b; Boger et al. 2019). Based on these similarities, with emphasis on the shared c. 2 Ga basement ages, Tucker et al. (2011b, 2014) proposed that the Anosyen and Androyen domains were part of a single Paleoproterozoic microcontinent.

There are also prominent differences between the two domains. These include (1) more intense strain in the Androyen domain (Martelat et al. 1999, 2000), (2) a greater proportion of supracrustal rock in the Anosyen domain (Windley et al. 1994; GAF-BGR 2008a, b; Boger et al. 2015, 2019), (3) possibly younger supracrustal protoliths in the Anosyen domain (Boger et al. 2014, 2015, 2019) and (4) different suites of igneous rocks, such as the 580–520 Ma charnockite-bearing Anosyen Batholith in the Anosyen domain (e.g. Paquette et al. 1994; considered part of the Ambalavao suite by GAF-BGR 2008b) and the c. 900 Ma anorthosite-bearing Ankiliabo suite in the Androyen (Ashwal et al. 1998; Boger et al. 2015) or Graphite (Boger et al. 2019) domain. With particular emphasis on the apparent lack of 630–600 Ma metamorphism in the Anosyen domain, Boger et al. (2015) hypothesized that the Anosyen and Androyen domains were distinct terranes with unrelated Proterozoic histories that became juxtaposed during the 580–500 Ma metamorphic episode that affected both domains (GAF-BGR 2008b; Boger et al. 2014, 2015, 2019). It follows that the Beraketa shear zone would be a suture that separates rocks of eastern African affinity (Androyen) that experienced 630–600 Ma metamorphism from rocks of Indian affinity (Anosyen) that did not. However, this is inconsistent with the 630–600 Ma metamorphic dates reported here for the Anosyen domain.

We argue that the coincident 630–600 and 580–500 Ma metamorphism, similarity in monazite compositional trends and P–T paths, shared upright structure, lack of throw along the dividing Beraketa shear zone and prominent c. 2 Ga basement ages favour a hypothesis in which the Anosyen and Androyen domains comprise a single terrane with shared pre-orogenic history, as suggested by Tucker et al. (2011b, 2014). This hypothesis extends to the Graphite domain of Boger et al. (2019) because there are no discernible metamorphic or structural differences between it and the rest of the Androyen domain. From these structural and geochronological inferences, we present a revised hypothesis for terrane correlations, orogenic structure and palaeogeography of the southern Madagascar terranes with respect to correlative units of eastern Africa and southern India (Figs 10–12).

The c. 1.0 Ga calc-alkaline intrusions of the Dabolava suite, which are unique in Madagascar to the Ikalamavony domain (GAF-BGR 2008d; Boger et al. 2014; Tucker et al. 2014), have Nd model ages of c. 1.3 Ga (Boger et al. 2019) and indicate the generation of predominantly juvenile crust in the domain, rather than reworking of an older crustal component. From this and other lithological considerations, the currently favoured interpretation of the Ikalamavony domain is that it comprises, at least in part, a c. 1 Ga island arc (Tucker et al. 2014; Boger et al. 2019; Armistead et al. 2020; Fig. 12). By c. 820 Ma, the Ikalamavony domain was attached to the Antananarivo and Itremo domains, as evidenced by the intrusion of 820–760 Ma Imorona–Itsindro suite plutons across all three domains (Tucker et al. 2014; Fig. 12) during a period of continental extension (Nédélec et al. 2016; Zhou et al. 2018) or continental arc magmatism (Handke et al. 1999; Moine et al. 2014; Archibald et al. 2016, 2017).

In India, c. 1.0 Ga dates exist within the Achankovil zone (Taylor et al. 2015) and the central Madurai block at Sivagiri (Plavsa et al. 2012; George et al. 2015) and Kodaikanal (Bartlett et al. 1998). An Achankovil–Ikalamavony correlation (with possibly related c. 1 Ga rocks in the southern Madurai block; Plavsa et al. 2012) is consistent with the interpretation that the Anosyen domain (Madagascar) and Trivandrum block (India) are equivalent (e.g. Windley et al. 1994; Fitzsimons 2016); both are composed of near-identical metasedimentary rocks with c. 2 Ga basement ages that experienced coincident 580–520 Ma metamorphism culminating in ultrahigh-temperature conditions at c. 550–540 Ma (Taylor et al. 2014; Johnson et al. 2015; Blereau et al. 2016; Harley and Nandakumar 2016; Holder et al. 2018a; Horton et al. 2022). This correlation also aligns the Neoarchean Antananarivo domain (Madagascar) and Madurai block (India) with its associated Proterozoic marginal sedimentary sequences (Itremo domain) (Plavsa et al. 2012), the Mesoarchean Antongil–Masora domains (Madagascar) with the Dharwar craton (India) (Boger et al. 2014; Tucker et al. 2014) and the hypothesized Betsimisaraka suture (Madagascar) with the Palghat–Cauvery shear zone (India) (Collins et al. 2014). Although exact palaeogeographical positions and some interpretative details differ, the research community has generally reached a consensus about these correlations between India and Madagascar (Boger et al. 2014; Collins et al. 2014; Tucker et al. 2014; Fitzsimons 2016).

The most likely extension of the Ikalamavony domain in Africa is the Galana unit in southern Kenya, which also contains c. 1.0 Ga orthogneisses of island arc origins (Hauzenberger et al. 2007; Bauernhofer et al. 2009; Fritz and Hauzenberger 2021). Rocks to the east, or the Sobo unit (Fritz and Hauzenberger 2021; also referred to as Galana East by Hauzenberger et al. 2004, 2007; Bauernhofer et al. 2009), are dominantly metasedimentary and have model ages as old as 2.9 Ga. This same west-to-east sequence exists in Madagascar (e.g. Boger et al. 2015), where the metasedimentary Itremo group is east of the Ikalamavony arc, contains Archean detritus and was presumably deposited on the Antananarivo margin (Cox et al. 2004; Tucker et al. 2011b).

The Galana unit has also been hypothesized to be correlated with the Vohibory arc of southern Madagascar (Rakotovao et al. 2014; Fritz and Hauzenberger 2021). However, the Vohibory arc is distinctly younger and either formed at 850–700 Ma and was metamorphosed from 670 to 570 Ma (Emmel et al. 2008; Jöns and Schenk 2008) or formed at 670–630 Ma, immediately prior to collisional orogenesis that began at c. 620 Ma (GAF-BGR 2008e). As such, the Vohibory domain is more probably related to the arc systems of the Arabian–Nubian Shield and Eastern Granulites, west of the Galana Unit (Boyd et al. 2010; Mole et al. 2018; Boger et al. 2019; Mtabazi et al. 2019; Fritz and Hauzenberger 2021).

With the new geochronological dataset of this study, our Galana–Ikalamavony–Achankovil structure and terrane correlation (Figs 10 and 11) marks the eastern limit of >600 Ma metamorphism. In the Arabian–Nubian Shield (Meert 2003; Elisha et al. 2017, 2019), as well as across the Eastern and Western Granulites units of Tanzania, 630–600 Ma metamorphic dates are common (Möller et al. 2000; Muhongo et al. 2001; Sommer et al. 2003). In southern Kenya, only the East Galana–Sobo unit lacks such dates (Hauzenberger et al. 2007). In Mozambique, metamorphic dates of 630–600 Ma exist in the Eastern Granulite nappes (M'Sawize, Xixano, Meluco: Bingen et al. 2009; Boyd et al. 2010), as well as the Mugeba (Kroner et al. 1997) and Monapo (Grantham et al. 2013; Macey et al. 2013) klippen. Metamorphism at 630–600 Ma is largely absent from southern India (Ghosh et al. 2004) but scarce monazite dates compatible with c. 620–600 Ma metamorphism have been reported in the Trivandrum Block (Taylor et al. 2014; Blereau et al. 2016; Kadowaki et al. 2019), which we consider correlative to the Anosyen domain. Although, for brevity, we have limited our discussion to the crustal fragments immediately adjacent to Madagascar in Gondwana, we also note that metamorphism of similar age is documented elsewhere. In Sri Lanka, c. 620–600 Ma metamorphism affected the Highland complex (Sajeev et al. 2010; Osanai et al. 2016; He et al. 2018). In Antarctica, c. 600 Ma metamorphism has been identified in the Sør Rondane mountains of Central Dronning Maud Land (Shiraishi et al. 2008; Grantham et al. 2013; Higashino et al. 2013; Ishikawa et al. 2013).

Unlike the Ikalamavony and Vohibory domains, the Androyen and Anosyen domains have continental affinity, with a prominent basement crystallization age of 2.2–1.8 Ga (Tucker et al. 2014), Nd-isotope model ages that extend as far back as the Neoarchean Era (Boger et al. 2019) and Rb–Sr whole-rock isochrons are also consistent with the presence of Archean–Paleoproterozoic basement or detritus (Cahen et al. 1984). The c. 2 Ga inheritance is also seen clearly among inherited monazite from this study (Fig. 5). This older continental basement preserved between juvenile arc systems requires that the Androyen–Anosyen domains were a separate microcontinent in the early Neoproterozoic Era (e.g. Tucker et al. (2011b, 2014; Fig. 12).

Paleoproterozoic to Archean basement ages suggest that the microcontinent rifted from the margin of a craton. Prominent late Archean and Paleoproterozoic ages are common features of cratons, so this alone is not a strong palaeogeographical indicator, although the Archean Congo–Tanzania craton with its Paleoproterozoic fringing terranes, such as the Ubendian–Usagaran belts, are one such possibility (Fig. 12). Anorthosite plutons may provide a stronger link to eastern Africa. The Saririaky and possibly Ankafotia anorthosites in the Androyen domain have emplacement ages of c. 900 Ma (Ashwal et al. 1998; GAF-BGR 2008a; Boger et al. 2015). In the Eastern Granulites of NE Tanzania, the Uluguru and Mahenge mountains anorthosite complexes have concordant zircon U–Pb dates extending to c. 900 Ma (Tenczer et al. 2006). Given the rarity of anorthosites in the rock record it seems plausible that these domains were attached at the time of anorthosite emplacement. If so, the Anosyen–Androyen microcontinent may represent a fragment of the Eastern Granulites basement units that separated in the early Neoproterozoic Era. In terms of age and Nd isotopic composition, the Madagascar anorthosites are also similar to the Tete anorthosite complex in Mozambique (Evans et al. 1999), which is west of the nappe complexes and was therefore presumably contiguous with the Congo–Tanzania craton at the time of emplacement. A common origin for these anorthosites is consistent with a scenario in which several continental fragments, including some of the Eastern Granulites basement and the Androyen–Anosyen basement, separated from the Congo–Tanzania craton after 900 Ma.

Imorona–Itsindro suite magmatism provides another palaeogeographical link for the Androyen microcontinent. Although generally absent from the Androyen and Anosyen domains (Tucker et al. 2014), zircon dates (810–736 Ma) contemporaneous with Imorona–Itsindro plutonism were measured in Anosyen orthogneisses close to the Anosyen–Ikalamavony contact near Ihosy (Tucker et al. 2011b; Boger et al. 2014). These dates require that the Androyen microcontinent was attached to the Antananarivo domain by c. 800 Ma (Fig. 12), which is also consistent with greatest statistical similarity among detrital-zircon spectra of some Anosyen metasediments to other Neoproterozoic sedimentary units of the Antananarivo domain (Boger et al. 2014).

In summary, we posit that the Androyen microcontinent is part of the same system of rifted continental fragments as the Eastern Granulites basement. These continental fragments rifted from the Congo–Tanzania craton in the Neoproterozoic Era, some time after c. 900 Ma. However, whereas the Eastern Granulites basement remained as separate crustal fragments in the Mozambique Ocean until they accreted back to the Congo–Tanzania craton at 630–600 Ma (e.g. Fritz and Hauzenberger 2021), the Androyen microcontinent was accreted to the Antananarivo domain by c. 800 Ma.

The terrane correlations for which we advocate delineate three principal structural domains of Ediacaran–Cambrian orogenesis: divergent thrust systems separated by a central upright, transpressional structural zone (Figs 10 and 11). The west-vergent nappes of the Western and Eastern Granulites in Africa formed during east–west shortening associated with the 650–600 Ma collision (East African Orogeny) and were reactivated or further developed during the 580–500 Ma collision (Kuunga Orogeny) (Fritz et al. 2005, 2009; Rossetti et al. 2008; Tenczer et al. 2013). In central Madagascar (the eastern side of the orogen), early east–west shortening, possibly at 630–600 Ma, was accommodated by coaxial strain and steeply dipping north–south-trending folds (Armistead et al. 2020). Later east–west shortening, presumably at 580–500 Ma, caused non-coaxial, low-angle, top-to-the-east displacement of the Itremo, Ikalamavony and a small portion of the Anosyen domain above the Antananarivo domain along shallowly dipping shear zones (Armistead et al. 2020). This east-vergent structure is also seen in the correlative Sobo Unit of Kenya (Hauzenberger et al. 2007) and the southern Madurai block of India (Plavsa et al. 2012). Coincident with this later event, north–south-trending ductile shear zones formed in central and southern Madagascar (Martelat et al. 2000; Nédélec et al. 2000; de Wit et al. 2001). The kinematics of the Eastern Granulites in Tanzania is consistent with east–west compression (with respect to modern Africa) for the earlier phase of arc-accretions and perhaps the earliest collisional event, whereas the dominant central and southern Madagascar structure reflects a sinistral transpression regime owing to NW–SE principal compressive stress (Martelat et al. 1999). The latter is consistent with the overall kinematic interpretation of the orogenic belt (Jacobs and Thomas 2004) and with the dominant shear sense interpreted from the Galana shear zone of Kenya (Hauzenberger et al. 2004).

The central zone of upright transpressive structure comprises the 20 km wide Galana shear zone of Kenya, which bifurcates into the vertical shear systems of southern Madagascar (Ejeda, Ampanihy, Beraketa and Ranotsara; e.g. Martelat et al. 2000, 2020). These shear zones bound the upright Anosyen, Androyen and Vohibory domains. The eastern boundary of this upright structural zone in Madagascar is the sinistral Ranotsara zone (Schreurs et al. 2010). It defines the eastern limit of 630–600 Ma metamorphism and the structural transition from the central upright structural zone to the east-vergent nappe systems. Since the formation of the east-vergent nappe system, the Ranotsara zone accommodated only minor strain (c. 70 km sinistral offset). This is consistent with the majority of 580–500 Ma strain (or at least its latest component) being accommodated by the Galana shear zone in the north and being partitioned across the Androyen–Anosyen domains and their bounding shear systems in the south. Trapped within this central transpressive zone, the Androyen and Anosyen domains experienced both 630–600 and 580–500 Ma periods of orogenesis without extensional collapse (e.g. Martelat et al. 1997).

An outstanding question is whether the vertical shear zones of southern Madagascar continue south-southwestward into eastern Africa. Depending on the precise reconstructed position of Madagascar in Gondwana, one might expect these structures to extend into south Tanzania or north Mozambique, broadly in the transitional region between the Central and Southern Mozambique belts (Fig. 10). However, similar vertical shear systems are not observed, suggesting a change in strain accommodation, lack of recognition or disruption by younger rifting in this area. This region marks an important transition in the basement geology, tectonic transport directions and overall orogenic history of the region (e.g. Hauzenberger et al. 2014). North of this boundary, collision is associated with the Congo–Tanzania craton, whereas to the south it is associated with the Kalahari craton (e.g. Meert and Lieberman 2008). The fringing terranes around each craton differ and the cratons also experienced a contemporaneous orthogonal (north–south in present Africa directions) collision.

Another outstanding question is the precise craton collision responsible for the 580–500 Ma metamorphic event. One scenario, the Azania hypothesis, was summarized by Armistead et al. (2020): an Azania microcontinent consisting of most of the Madagascar domains (Antananarivo, Itremo, Ikalamavony and Anosyen–Androyen domains) together with their correlatives in southern India (North Madurai, South Madurai, Achankovil and Trivandrum, respectively) and Kenya (Galana and Sobo units) was accreted to the Congo–Tanzania craton along with the arc terranes of the Arabian–Nubian Shield and the Eastern Granulites of eastern Africa by 630–600 Ma (Fig. 12). Subsequently, 580–500 Ma metamorphism was caused by the terminal collision of East and West Gondwana along the Betsimisaraka suture between Azania and the Dharwar craton of India (Figs 10 and 12) with near-simultaneous oblique collision of the Australia–Mawson craton (East Antarctica) (e.g. Merdith et al. 2017a, b). There are two limitations of this hypothesis. First, there is no arc magmatism linked to the supposed closure of an Ediacaran ocean between Azania and the Dharwar craton (Imorona–Itsindro suite magmatism predated this event). Second, similar Archean basement ages (Tucker et al. 1999; Schofield et al. 2010), c. 800 Ma (Imorona–Itsindro) magmatism (Tucker et al. 2011a) and detrital zircon spectra in Neoproterozoic metasedimentary rock (Boger et al. 2014) on both sides of the proposed suture have led to the question of its existence as an Ediacaran–Cambrian structure.

In an alternative scenario, the 630–600 Ma metamorphic event marked the terminal suturing of the Congo–Tanzania and Dharwar cratons (e.g. Horton et al. 2016; Fig. 12), as well as the arc terranes of the Arabian–Nubian Shield further north (Elisha et al. 2019). The 580–500 Ma event was then caused by a later oblique collision of the Australia–Mawson (East Antarctica) craton (e.g. Meert and Lieberman 2008; Collins et al. 2014). In this case, no suture in Madagascar or its immediately correlative rocks of eastern Africa and southern India is associated with the 580–500 Ma collision.

Rather than being exhumed in either of the nappe systems or experiencing orogenic collapse (Dewey 1988), the Androyen and Anosyen domains remained laterally confined beneath the orogenic plateau by the bounding transpressional shear systems (Fig. 12). This is evidenced by only modest near-peak-T decompression of the Androyen and Anosyen domains, as well as the preservation of their dominantly upright structure, which indicates that metamorphism ended during continued transpression, without a prominent phase of extension (Martelat et al. 1999). Only a portion of the northernmost Anosyen domain shows a transition across the Ranotsara zone from an upright to an east-vergent structural orientation; there, metamorphic grade is generally thought to be lower (Raith et al. 2008; Jöns and Schenk 2011; Boger et al. 2012; Horton et al. 2016).

From these perspectives, the highest metamorphic grades in the Madagascar orogen (900–1000°C at c. 6 kbar in the southern Anosyen domain) appears to have occurred where thrusting and extensional collapse were both ineffective (Fig. 11). Two aspects of this orogenic model may help explain the occurrence of (ultra)high-temperature metamorphism (Androyen, up to 850°C; Anosyen, 900–1000°C) and exceptionally high thermobaric ratios (Androyen: c. 1000°C GPa^–1; Anosyen, >1300°C GPa^–1) in southern Madagascar. First, metamorphism during the 630–600 Ma event would have resulted in dehydration and some partial melting. Even if the rocks cooled substantially between orogenic episodes, this prior devolatilization would have substantially lowered the total heat needed to reach (ultra)high metamorphic temperature during the 580–520 Ma event (Schorn et al. 2018). Second, by remaining structurally bound within a central, thick portion of the orogen, the Androyen–Anosyen domain would have accumulated appreciably more radiogenic heat (Clark et al. 2015; Fitzsimons 2016; Horton et al. 2016) than if they had been tectonically extruded in the adjacent nappe systems or had undergone extensional collapse. Elevated mantle heat flow (e.g. Horton et al. 2016) or advective heating of the middle crust by mantle- or lower-crust-derived magmas (e.g. Kaare-Rasmussen et al. 2024) would have only contributed further to this already favourable tectonic setting for the generation and sustenance of long-lived, high-temperature metamorphism.

The major lithologies, structures and petrochronological boundaries in southern Madagascar can be traced into Africa and India. Our terrane correlations and orogenic reconstruction imply that the metamorphism in southern Madagascar is the result of two back-to-back collisions. Unlike their lithological equivalents that extruded eastward and westward in thrust systems, the Anosyen and Androyen domains remained structurally confined to a vertical, lithosphere-scale transpressional shear system during orogenesis. Centrally located in the orogen, these domains share a common P–T–t history since at least 630 Ma. Critically, within this structural setting, they experienced little exhumation for many tens of million years, creating a favourable orogenic setting for long-lasting high-temperature metamorphism.

S. Oriolo, M. Heilbron and an anonymous reviewer provided constructive feedback and discussion of this work. D. Hernández-Uribe provided feedback on an early version of the paper.

RMH: conceptualization (lead), data curation (lead), formal analysis (lead), funding acquisition (equal), investigation (lead), methodology (lead), project administration (equal), resources (equal), supervision (lead), validation (lead), visualization (equal), writing – original draft (lead), writing – review & editing (lead); FH: conceptualization (equal), data curation (supporting), funding acquisition (equal), investigation (equal), methodology (supporting), project administration (equal), resources (equal), validation (equal), visualization (equal), writing – original draft (equal), writing – review & editing (equal); CS: data curation (supporting), formal analysis (equal), visualization (supporting), writing – original draft (supporting); AFMR: project administration (supporting), resources (equal).

This work was supported by the National Science Foundation (grant numbers EAR-2022746, EAR-2022573 and EAR-11348003).

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

All data are provided as supplementary files linked to this publication. Readers should contact the corresponding author for additional questions or information.