The mollusc shells utilized in this work originate from the Samhan Formation in the Saiwan area of Oman in the Huqf desert (20°39^′ N, 57°31^′ E; see Fig. 1A). The biostromes in this locality were dated as late Campanian (∼ 75 Ma) based on ammonite biostratigraphy by Kennedy et al. (2000). The paleolatitude of the site at 75 Ma was 3° S according to reconstructions following http://paleolatitude.org (last access: 12 November 2025; van Hinsbergen et al., 2015) based on the paleomagnetic reference frame by Vaes et al. (2023). The locality was described by Schumann (1995) as exposing Vaccinites-dominated rudist biostromes in which rudist bivalves are preserved in their life position (see also Fig. 1B–C). The V. vesiculosus and T. sanchezi specimens used and reused in this study originate from Unit IV in profile 1 of Schumann (1995), which is equivalent to unit 2 in Philip and Platel (1995). The thick-shelled O. figari oysters were collected in the marly layer just above the biostrome (see also Fig. 1B–C and de Winter et al., 2017b).

All specimens described in this study were sampled in longitudinal cross-sections through the axis of maximum growth using a combination of hand drilling and computer-assisted microdrilling using slow-rotating tungsten carbide dental drills. Table 1 gives an overview of the data used in this study, its temporal resolution and the source of datasets in cases where they have been reused from the literature. V. vesiculosus specimen “B6”, T. sanchezi specimen “B10” and O. figari specimen “B11” were described in de Winter et al. (2017b) and stable isotope (δ18O and δ13C) and trace element (Mg, Sr, Ca, Mn, Fe) data presented in that study are used here. Specimen B10 was also subject of a study by de Winter et al. (2020a), in which more detailed, daily scale trace element measurements (Mg / Ca, Sr / Ca, Mg / Li and Sr / Li) were presented and discussed. Stable isotope (δ18O and δ13C) data from T. sanchezi specimens “H576”, “H579” and “H585” was previously reported in Steuber (1999). Of these, specimen H579 was sampled in 5 parallel sections (“H579A-E”).

For this study, we obtained newly measured stable (δ18O and δ13C) and clumped (Δ47) isotope analyses on one additional T. sanchezi specimen, called “HU-027”. This specimen was incrementally sampled (n = 135) at 50 µm resolution for δ18O and δ13C in a cross-section in the internal pillar of the rudist shell. A total of 96 clumped isotope measurements were carried out on incremental samples from two locations corresponding to the maximum and minimum δ18O and δ13C values in the above-mentioned profile (samples R_1–R_11) as well as larger bulk samples in two areas on the same cross-section (samples RB_1 and RB_2; Fig. 2A). Sample RB_2 was obtained from a different pillar in the same cross-section, whose timing of growth can be linked to the adjacent pillar by following the isochronous growth lines visible in the cross-section (Fig. 2B). The timing of deposition of the shell calcite sampled in RB_2 partly overlaps with incremental samples R_1–R_4, but this sample also contains a significant portion of shell material deposited earlier in the ontogeny.

Specimens H576, H579 and H585 were subject to detailed diagenetic screening in (Steuber, 1999) and the preservation of specimens B6, B10 and B11 was tested in de Winter et al. (2017b, 2020a) and de Winter and Claeys (2016). These previous studies concluded, based on a combination of scanning electron microscopy, cathodoluminescence microscopy and trace element analysis, that there was no detectable recrystallization in the areas of the shells sampled for geochemical analysis and that the low-magnesium calcite outer shell layer of these rudists preserves the original shell composition deposited during the lifetime of the animal. For the purpose of this study, additional screening based on high-resolution trace element analyses (Mg / Ca, Sr / Ca, Mn and Fe) will be discussed for specimens B6, B10 and B11, one specimen for each of the three studied species (see Sects. 3.2 and 4.1). The newly sampled specimen HU-027 was subject to detailed microscopic scrutiny using a combination of reflected light, cross-polarized light, scanning electron microscopy and energy dispersive X-ray spectroscopy (EDS) and micro-Raman spectroscopy to characterize original shell texture preservation and detect diagenetic alteration (Fig. 2C–F).

To contrast the paleoclimate reconstructions from the above-mentioned multi-proxy dataset, we extracted data on the occurrences of bivalves in modern oceans from the Ocean Biodiversity Information System database (OBIS, 2025). We accessed the OBIS database through the “occurrences” function of the “robis” package (Provoost et al., 2022) and processed the occurrences in the open-source computational software package R (R Core Team, 2023). A total of 2 199 523 occurrences of taxa in the class of Bivalvia were extracted from the database, and their localities were categorized in bins of 2° by 2° based on their latitude and longitude.

Modern seasonal SST ranges across the world oceans were extracted from the Extended Reconstructed Sea Surface Temperature (ERSST) dataset from the National Oceanic and Atmospheric Administration (NOAA; Huang et al., 2017). We extracted monthly data from the years 1981 until 2010 on a 2° by 2° latitude and longitude grid from the ERSST dataset. We extracted the warmest and coldest monthly temperatures from the grid cells that contained reported occurrences of bivalves in the OBIS dataset. This resulted in a distribution of the warmest and coldest monthly temperatures across the living environment of modern bivalves.

Our clumped isotope dataset on specimen HU-027 yielded information about the seasonal spread in temperature and the water oxygen isotopic value (δ18O_w) experienced by this specimen. We reconstructed these parameters from the Δ47 and δ18O_c values of the shell carbonate following the clumped isotope-temperature relationship by Daëron and Vermeesch (2023) and the calcite δ18O_c-δ18O_w-temperature relationship by Kim and O'Neil (1997). This dataset allowed us to characterize seasonal variability in climate in the Saiwan paleoenvironment. However, since the relatively short δ18O profile from HU-027 did not allow age modelling and the clumped isotope dataset only sampled one specimen, we augmented the dataset from HU-027 with δ18O, δ13C and trace element data from other specimens (see Sect. 2.2.3) in the same assemblage to study the relationship between temperature and growth rate in Saiwan. To achieve this, we carried out subsequent data processing steps outlined below.

To reconstruct shell growth rates of the molluscs and water temperatures in the Saiwan environment from chemical data measured in the shells, we carried out the following data processing steps (see also the flowchart in Fig. 3):

We used concentrations of Mn and Fe and Mg / Ca and Sr / Ca ratios to screen for diagenetic recrystallization of parts of the shells, and remove chemical data from suspicious shell sections for further analysis.

All specimens show distinct periodic patterns in δ18O_c and δ13C values when plotted in growth direction through the shells (Fig. 4). Stable oxygen isotope values (δ18O_c) in the shells vary between -7.1 ‰ VPDB and -1.5 ‰ VPDB with a mean δ18O_c value of -5.1 ± 0.9 ‰ VPDB (1σ) for the entire dataset. The lowest mean δ18O_c values are recorded in T. sanchezi shells (-5.5 ± 0.6 ‰ VPDB; 1σ), with higher values recorded in V. vesiculosus (-4.1 ± 0.6 ‰ VPDB; 1σ) and O. figari (-3.6 ± 0.4 ‰ VPDB; 1σ). Similarly, the lowest mean δ13C values are recorded in T. sanchezi (0.7 ± 0.9 ‰ VPDB; 1σ), followed by V. vesiculosus (0.8 ± 0.4 ‰ VPDB; 1σ) and O. figari (1.7 ± 0.6 ‰ VPDB; 1σ).

Trace element analyses highlight that shells of all three species are generally characterized by low concentrations of Mn and Fe (Fig. 5). T. sanchezi exhibits median Mn concentrations of 38 µg g⁻¹ (average: 50 ± 44 µg g⁻¹, 1σ) and median Fe concentrations of 56 µg g⁻¹ (average: 77 ± 116 µg g⁻¹, 1σ) with a few isolated locations in the shell with concentrations exceeding 300 and 1000 µg g⁻¹ for Mn and Fe, respectively. The carbonate in V. vesiculosus has somewhat higher median Mn concentrations of 176 µg g⁻¹ (average: 192 ± 94 µg g⁻¹, 1σ) and median Fe concentrations of 125 µg g⁻¹ (average: 173 ± 152 µg g⁻¹, 1σ) with some locations showing Mn and Fe concentrations exceeding 500 and 800 µg g⁻¹, respectively. A clear positive trend is observed towards higher Mn and Fe concentrations in V. vesiculosus samples (Fig. 5a). Finally, O. figari has median Mn concentrations of 63 µg g⁻¹ (average: 75 ± 33 µg g⁻¹, 1σ) and much lower median Fe concentrations of 4 µg g⁻¹ (average: 5.7 ± 5.8 µg g⁻¹, 1σ), with maximum Mn and Fe concentrations of 180 and 40 µg g⁻¹, respectively, in some locations.

Mg / Ca and Sr / Ca ratios are very similar between T. sanchezi and V. vesiculosus, with mean Mg / Ca ratios of 11.4 ± 2.4 mmol mol⁻¹ (1σ) for T. sanchezi and 11.5 ± 4.8 mmol mol⁻¹ (1σ) for V. vesiculosus and mean Sr / Ca ratios of 1.49 ± 0.21 mmol mol⁻¹ (1σ) for T. sanchezi and 1.09 ± 0.45 mmol mol⁻¹ (1σ) for V. vesiculosus. A subset of the samples from V. vesiculosus exhibit a clear trend towards lower Mg / Ca and Sr / Ca values. O. figari exhibits much lower Mg / Ca values (1.87 ± 1.15 mmol mol⁻¹; 1σ) and Sr / Ca values of 1.36 ± 0.18 mmol mol⁻¹ (1σ).

Applying ShellChron on Mg / Ca, Sr / Ca (only for specimen B10), δ18O_c and δ13C values through all specimens except HU-037 yielded information about the age-distance relationship in the direction of growth through the shells (Fig. 6). These distances in growth direction on cross-sections through the shells were interpreted as proxies for the growth rate of the individual during its life. ShellChron-based age models yield highly consistent age-distance relationships for different T. sanchezi specimens, regardless of whether they are based on Mg / Ca, Sr / Ca, δ18O_c or δ13C records (Fig. 6A). Contrarily, growth models based on δ18O_c and δ13C values in V. vesiculosus and O. figari differed (Fig. 6B–C).

Applying the Daydacna algorithm to trace element records through T. sanchezi specimen B10 (for which subdaily-resolved trace element data is available) yielded independent evidence for the age-depth relationship in shells of T. sanchezi (see Fig. 6D). Except for the Mg / Li record, all age-distance relationships obtained by applying Daydacna on trace element records through specimen B10 closely agree with the age-distance relationship obtained through the combined ShellChron growth models for this and other T. sanchezi specimens. (Fig. 6B). Of the Daydacna results, the model based on Mg / Li ratios deviates most strongly from the other Daydacna results (based on Mg / Ca, Sr / Ca and Sr / Li records) and the stable isotope-based ShellChron age models. The close agreement between these age models generated using different algorithms, based on different environmental cycles (daily vs. seasonal) on different geochemical records through the same specimen, highlights the reproducibility of the age-distance relationship found for our assemblage of T. sanchezi specimens from the Saiwan ecosystem. This lends confidence to the interpretation that the observed rhythms in Mg / Ca, Sr / Ca and Sr / Li records in specimen B10 represent daily cycles (de Winter et al., 2020a) and allows us to refine our age model for this species to quantify growth rates on a monthly scale.

Isotope profiles in Fig. 4 and data on the precision of the ShellChron model outcomes in Table 3 show that the seasonal pattern is in some specimens clearer in δ18O_c, while others show clearer seasonality in the δ13C records. This translates to a better precision of δ18O_c-based age models in some specimens, while others have better-defined age models based on δ13C. On average, the δ18O_c-based age models are more precise (23.1 d at 95 % confidence level) than the δ13C-based age models (27.6 d at 95 % confidence level; see Table 3). To account for inter-specimen differences, we decided to use the most precise growth model (either based on δ18O_c or δ13C) available per specimen for determining the age-distance relationship.

Clumped isotope analysis on T. sanchezi specimen HU-027 yielded a mean Δ47 value of 0.572 ± 0.047 ‰ I-CDES (1σ). The clusters created from the clumped isotope dataset of specimen HU-027 highlight the relationship between δ18O_c values in T. sanchezi and the temperature and δ18O_w values reconstructed from clumped isotope thermometry (Fig. 7). Clustering by location in the shell yields maximum reconstructed temperatures in HU-027 of 44.2 ± 4.0 °C and minimum temperatures of 19.2 ± 3.8 °C. These clusters in HU-027 sample a similar or larger spread in δ18O_c, δ13C and Δ47 compared to the statistical clustering approaches that do not take into account the sample location (maximum temperature range: 45.4 ± 17.1 to 24.7 ± 4.3 °C; see File S5 in the Zenodo repository; de Winter, 2025), demonstrating that the sampling strategy successfully resolves the seasonal variability recorded in T. sanchezi specimen HU-027.

Interestingly, while the lowest δ18O_c values in T. sanchezi are associated with the highest temperatures, as one would expect assuming a constant δ18O_w value throughout the year, the highest δ18O_c values do not represent the coldest season. This suggests that the Saiwan environment in the Late Campanian experienced significant seasonal variability in δ18O_w values. Combining clumped and oxygen isotope data on the clusters shows indeed that they record excursions towards very low δ18O_w values (-4.63 ± 0.86 ‰ VSMOW) in the coldest season, far off the δ18O_w value of -1 ‰ VSMOW commonly assumed for past greenhouse periods (Shackleton, 1986).

Ancient shell carbonates, such as the rudists studied here, are known to be susceptible to various forms of diagenetic alteration (Al-Aasm and Veizer, 1986b, a; Brand and Veizer, 1980, 1981; Ullmann and Korte, 2015). Common in carbonate systems, such as the carbonate platforms where the Saiwan rudists were growing, is open-system diagenesis, in which the diagenetic fluid with which the shell carbonate exchanges to alter its chemical and isotopic composition is continuously replaced (Al-Aasm and Veizer, 1986b; Brand and Veizer, 1981). In such systems, the chemical and isotopic composition of carbonates moves away from its original value in an approximately linear trend (mixing line), typically resulting in increased concentrations of trace elements such as Mn and Fe, reduced concentrations of Sr and trends towards lower δ18O_c values in the carbonate (Al-Aasm and Veizer, 1986b, a).

Our detailed geochemical investigation of specimens B6, B10 and B11 (Figs. 2, 5) highlights low Mn and Fe concentrations (typically < 300 and < 200 µg g⁻¹, respectively, below thresholds used by Schmitt et al., 2022; see Fig. 5) and high Sr concentrations (typically > 1.0 mmol mol⁻¹; see Sect. 3.2), which show a correlated, skewed distribution tailing towards a few locations on the shell where Mn and Fe concentrations are high and Sr concentrations are low, especially in V. vesiculosus specimen B6 (Fig. 5). In V. vesiculosus, the trend of covarying elevated Mn and Fe concentrations and coinciding reductions in Mg / Ca and Sr / Ca clearly shows the imprint of local open-system diagenetic remineralization. Open system diagenesis was observed in isolated localities in these fossil shells, which were avoided during sampling for stable and clumped isotope analysis (see Sect. 2.2.3). The chemical differences between well-preserved sections of the shells of these three taxa are likely to reflect taxon-specific variations in trace element concentrations, which are also common in modern molluscs and may reflect differences in shell microstructures and their associated formation pathways (e.g. Carré et al., 2006; Onuma et al., 1979).

Isotopic compositions of carbon and oxygen are often jointly depleted in diagenetically altered materials due to the exchange of shell carbonate with either isotopically depleted meteoric fluids during early diagenetic alteration (Allan and Matthews, 1990) or exchange with pore fluids under high temperatures (e.g. Brand and Veizer, 1981). However, a positive correlation between δ18O_c and δ18C values is not necessarily a reliable indicator of diagenetic alteration (Swart and Oehlert, 2018). Instead, a positive correlation between δ18O_c and δ18C values is often observed in modern (non-diagenetically altered) photosymbiotic species, such as tridacnids (Elliot et al., 2009; Killam et al., 2020). Such a correlation has been proposed to be caused by seasonal changes in the isotopic composition of the dissolved inorganic carbon pool due to variability in the activity of photosymbionts in phase with the seasonal effect of temperature on the oxygen isotope composition of the shell (Elliot et al., 2009; McConnaughey and Gillikin, 2008). The fact that T. sanchezi specimens in our dataset exhibit a strong positive correlation between δ18O_c and δ18C, while the other species do not, is corroborated by other evidence that T. sanchezi had photosymbionts such as the presence of specific adaptations in the shell thought to facilitate the hosting of photosymbiotic microorganisms in the mantle and the strong expression of diurnal cycles in shell structure and chemistry (see Skelton and Wright, 1987; Steuber, 1999; de Winter et al., 2020a). We therefore disregard this as evidence for open-system diagenesis.

Grain boundary diffusion is another potential diagenetic process that can influence the isotopic composition of biogenic carbonates. This process is rapid on geological timescales (< 100 years) and causes the exchange of oxygen isotopes with pore fluids (Adams et al., 2023; Cisneros-Lazaro et al., 2022; Nooitgedacht et al., 2021). In foraminifera, this process exchanges up to ∼ 3 % of the oxygen in the biomineral (Adams et al., 2023), meaning that even in the presence of strongly isotopically negative pore fluids this process can only change the δ18O_c value of the biomineral by a few tens of a permille, not enough to fully explain the high temperatures recorded in the fossils in this study.

Finally, our clumped isotope analysis results of specimen HU-027 may be susceptible to solid-state reordering of the clumped isotope signature, a form of closed-system diagenesis which occurs at elevated temperatures (> 100 °C; Chen et al., 2019; Looser et al., 2023; Passey and Henkes, 2012; Stolper and Eiler, 2015). This reordering effect requires large differences between the temperatures in which the carbonate was originally precipitated and the temperatures of the surrounding rocks and, when activated, is likely to affect the entire sample equally rapidly (on geological timescales; Henkes et al., 2014; Stolper and Eiler, 2015). Similarly, heating carbonate samples to 175 °C caused a resetting of the Δ47 value through exchange with internal waters without noticeable change to δ18O_c values (Nooitgedacht et al., 2021). Given the fact that significant temperature variability is recorded by the clumped isotope dataset from specimen HU-027, and that the recorded temperatures are far from the temperatures needed for the solid-state reordering process to significantly affect isotopic clumping (> 100 °C; Henkes et al., 2014), we feel confident in interpreting the recorded temperatures in terms of the paleoclimate and environment at the Saiwan site.

The Monte Carlo simulations of the seasonal δ18O_c-δ18O_w-temperature path based on the clumped isotope dataset from specimen HU-027 in Fig. 7 show a statistically significant difference between paleotemperatures and δ18O_w values between two parts of the annual cycle, especially for the middle range of the δ18O_c values of (δ18O_c between -5.0 ‰ and -6.0 ‰ VPDB). Shell increments deposited after the δ18O_c minimum record lower temperature and δ18O_w values, while parts of the shell before the δ18O_c minimum record high temperature and δ18O_w values. High temperatures are reconstructed for both the extreme ends of the δ18O_c variability (δ18O_c < -6.0 and δ18O_c > -5.0 ‰ VPDB). We interpret this as the signature of a shift in the temperature and δ18O_w maxima with respect to the δ18O_c cycle, with a high temperature extreme (44.2 ± 4.0 °C) at the low end of the δ18O_c cycle, which we interpret as a hot and dry summer season, and a milder temperature maximum (41.4 ± 4.8 °C) at the high end of the δ18O_c cycle, which we interpret as a warm and dry spring season (Fig. 7B, D). The coldest and wettest season (winter) has such a low δ18O_w value (-4.64 ± 0.86 ‰ VSMOW) that it is not represented by the highest δ18O_c values, as would be the case if δ18O_c would reflect a pure temperature signal (Fig. 7B, D). This interpretation of the seasonality in Saiwan is also consistent with the temporal relationship between the clumped isotope sampling locations in specimen HU-027: Sample RB_1 (winter) comes earliest in the chronology, closely followed by samples R_8–R_11 (spring), and RB_2 (spring) directly precedes R_1–R_7 (summer). The latter two partly overlap later in the chronology (Fig. 2). The result is a seasonality in which the temperature cycle and the hydrological cycle (reconstructed through the δ18O_w value) are out of phase. The δ18O_c value of carbonate precipitated under these conditions therefore exhibits hysteresis behaviour (see Fig. 7). Based on the clumped isotope dataset from specimen HU-027 alone, we reconstruct a seasonal sea surface temperature at Saiwan of 19.2 ± 3.8 to 44.2 ± 4.0 °C. The oxygen isotopic composition of the seawater varied from -4.62 ± 0.86 ‰ VSMOW in winter to +0.86 ± 1.6 ‰ VSMOW in summer during the lifetime of specimen HU-027.

While the strong seasonal variability in δ18O_w value of the seawater in Saiwan contradicts the typical explanation of δ18O_c fluctuations in mollusc shells, the assumptions under which the ShellChron model operates (see Sect. 2.4.1) are not violated by this observation. We believe our characterization of seasonal variability in temperature and δ18O_w value to be realistic for the following reasons:

Firstly, this temperature distribution over the year, which is offset in phase from the precipitation seasonality, is observed in modern tropical climates, especially those affected by monsoon-like precipitation seasonality.

Figure 8 shows the spread in monthly temperature, δ18O_w and growth rate averages. Table 4 also highlights the implications of the seasonal variability in δ18O_w throughout the year in the Saiwan environment (see Fig. 7C–D): Correcting δ18O_c values for δ18O_w variability in all specimens yields an average coldest month mean temperature (CMMT) of 18.7 ± 3.8 °C and a warmest month mean temperature (WMMT) of 42.6 ± 4.0 °C for the entire dataset. Mean annual average temperatures were 33.1 ± 4.6 °C. The uncertainties on these estimates are propagated from the uncertainties of the lowest and highest temperature clusters in the clumped isotope dataset (see Table 2 and Fig. 7).

Using the classic assumption of a seasonally constant δ18O_w value of -1 ‰ VSMOW yields a considerably narrower and warmer monthly temperature range (CMMT–WMMT) of 31.8–37.6 °C. Alternatively, when we exclude the coldest clumped isotope cluster in the HU-027 dataset, which has a very low δ18O_w value of -4.62 ± 0.86 ‰ VSMOW (Fig. 7), and might therefore be biased by a strong seasonal influx of meteoric water, and assume the mean of the other three clusters (-0.25 ‰ VSMOW) as a constant δ18O_w value, the CMMT-WMMT range becomes 35.7–41.8 °C. Recent studies demonstrated that the assumption of seasonally constant δ18O_w values in shallow marine environments often leads to an underestimation of the seasonal temperature range from δ18O_c measurements (e.g. de Winter et al., 2021a). We observe the same effect here and therefore use the seasonal temperature reconstructions that take into account seasonal δ18O_w variability from our clumped isotope T. sanchezi dataset throughout the remainder of the discussion.

There is a significant difference between the δ18O_c ranges recorded in V. vesiculosus, O. figari and T. sanchezi in our dataset (see Figs. 8 and 9). When considering the seasonal δ18O_w variability in Saiwan, the T. sanchezi specimens record significantly higher temperatures in their shells (19.2–45.1 °C range of monthly temperatures, with an average of 33.6 °C) than V. vesiculosus (12.2–36.2 °C; average: 31.6 °C) and O. figari (12.5–33.8 °C; average: 31.1 °C). While the age models allow growth stops, it is possible that some months were not recorded by one or more of these species, reducing the seasonal range. This is also evident from the minimum monthly mean growth rate modelled from some δ18O_c profiles (e.g. in specimen H579; see Table 4) being zero.

Nevertheless, the inter-species difference is surprising, given that these specimens originated from the same biostrome (or directly above, in the case of O. figari; see Fig. 1B–C). The lower temperatures (∼ 2 ‰ higher mean δ18O_c) recorded in O. figari may be attributed to its slightly higher stratigraphic position. Since the Saiwan site is embedded in a long stratigraphic record and is dated based on ammonite biozones, which yield a dating accuracy on the order of ∼ 100 kyr (Lehmann, 2015), there remains some uncertainty to the (internal) age variability between the units (Kennedy et al., 2000; Schumann, 1995). Therefore, it is possible that the O. figari specimen studied here sampled a different climate and paleoenvironment than the other specimens. This different paleoenvironment may be characterized by a different δ18O_w value or seasonal δ18O_w range, resulting in a mild underestimation of paleotemperature and its seasonal range, or a true difference in ambient temperature during the lifetime of the oyster compared to the rudists. Since the mean δ18O_c value of O. figari is almost 2 ‰ higher than that of T. sanchezi (Table 4), mean annual δ18O_w would have to have increased by 2 ‰ within a (geologically) very short time interval, given the close stratigraphic relationship between the species, to explain the difference. A plausible way to achieve such a change might be a shift in the seasonal δ18O_w regime: If the climate in which O. figari grew did not feature the highly δ18O_w-depleted winter season (see Fig. 7) and instead featured year-round δ18O_w values close to -0.25 ‰ VSMOW (the mean value of the summer and spring/autumn clusters in HU-027), the difference would explain ∼ 1.2 ‰ of the 2 ‰ difference in the δ18O_c value. The remaining 0.8 ‰ offset between O. figari and T. sanchezi could then be explained by a drop in mean annual temperature of ∼ 3.5 °C over time based on a typical δ18O_c-temperature sensitivity of 4.3–4.5 °C ‰⁻¹ (Epstein et al., 1953; Grossman and Ku, 1986; Marchitto et al., 2014). In this scenario, the seasonal temperature range experienced by O. figari was roughly 27.7–31.3 °C (average temperature of 29.8 °C), ∼ 1.3 °C lower than the average temperature recorded by O. figari when accounting for the δ18O_w seasonality reconstructed by clumped isotope analysis in specimen HU-027, but with a significantly reduced seasonal variability (Table 4). Note that this constant δ18O_w scenario would also significantly increase the estimated winter temperatures from O. figari, which are very low (12.5 °C; see Table 4) when applying the δ18O_w seasonality from our clumped isotope data. We therefore consider it likely that the strong drop in temperature and δ18O_w value that characterized the winter season in Saiwan may not be recorded in O. figari even if it occurred in the climate and environment of O. figari. The temperature distribution in Fig. 9D further confirms this by showing that these exceptionally low winter temperatures in O. figari are represented by a small fraction of the data from this specimen, while the majority of the samples record estimated temperatures between 25 and 35 °C.

This stratigraphic argument cannot explain the temperature differences between V. vesiculosus and T. sanchezi, since they were sampled in life position from the same biostrome. Given the rapid growth of these organisms (de Winter et al., 2017b, 2020b) and the rapid build-up and succession of the biostromes they are found in Gili et al. (1995), Gili and Skelton (2000), and Ross and Skelton (1993), it seems likely that these organisms lived within geologically short time periods from each other (conservatively less than 1000 years), and therefore sampled a similar climate in the Saiwan region. This seems plausible considering the comparatively large overlap in the frequency distribution of temperatures recorded by both species (Table 4; Fig. 9D). The growth rate of V. vesiculosus was on average similar to that of T. sanchezi (see Table 4 and Fig. 8E) and the spatial sampling resolution in V. vesiculosus (250 µm) was often higher than that in T. sanchezi shells (∼ 100–500 µm (de Winter et al., 2017b). Thus, it seems unlikely that the differences between the seasonal ranges obtained from the shells of both species can be attributed to undersampling of the full recorded temperature of V. vesiculosus, a source of variability discussed in Goodwin et al. (2003), Judd et al. (2017), and de Winter et al. (2021b). Applying the high seasonal range in δ18O_w observed in HU-027, the maximum monthly temperature recorded by V. vesiculosus (36.2 °C) is at least 5° lower than the maximum monthly temperatures recorded in the T. sanchezi specimens (> 43 °C in all specimens, average of 44 °C; Table 4). At the same time, mean annual temperatures in V. vesiculosus (31.6 °C) and T. sanchezi (33.0 °C) differ by less than 1.5°. As in O. figari, winter temperatures in V. vesiculosus are exceptionally low compared to temperature ranges in other specimens in the assemblage (Table 4) and the temperature distribution in this specimen (Fig. 9D). This is a consequence of the fact that winter datapoints in the V. vesiculosus dataset are associated with the very low δ18O_w values reconstructed from the winter datapoints in the clumped isotope dataset, resulting in low temperature estimates when applied on the δ18O_c values measured in V. vesiculosus. Since all T. sanchezi specimens yield consistent winter temperatures of ∼ 20 °C, we consider it plausible that V. vesiculosus did not record the seasonally low δ18O_w conditions and conclude that winter SSTs in Saiwan are more accurately estimated by the T. sanchezi data (19–20 °C).

The growth rate of T. sanchezi specimens is diminished when the range of maximum tolerable temperatures (upper thermal limits) for modern bivalves occurring in hot, shallow marine settings (Roebuck Bay, Australia, tolerable thermal range: 32.7–41.8 °C; Compton et al., 2007) is reached or exceeded (see File S6 in the Zenodo repository; de Winter, 2025). While the frequency of temperatures recorded by V. vesiculosus and O. figari specimens in our dataset decreases when these maximum tolerable temperatures are reached (Fig. 9), temperatures recorded in T. sanchezi specimens regularly exceed modern temperature maxima. Higher temperatures recorded in T. sanchezi specimens suggest that the upper thermal tolerance of this species may exceed those recorded in modern bivalves and that T. sanchezi had a higher temperature tolerance than V. vesiculosus and O. figari.

This conclusion is further supported by the observation that almost all studied T. sanchezi specimens record a monthly temperature range that exceeds that of V. vesiculosus (and O. figari; Table 4) and that the monthly temperature range recorded by T. sanchezi specimens is consistent (see Fig. 8B and Table 4). The fact that all T. sanchezi specimens together record over 30 years of seasonality with consistently low summer δ18O_c values (monthly minima of -5.7 ‰ VPDB or lower, corresponding to maximum monthly temperatures > 40 °C; see Fig. 3 and Table 4) also shows that these conditions were not isolated events but a persistent feature of the local climate in Saiwan during the Campanian. Finally, daily rhythms in the shell chemistry of the large T. sanchezi specimen B11 have previously revealed that this specimen likely grew year-round, and thus recorded the full seasonal temperature cycle (de Winter et al., 2020a).

An important limitation to this inter-species paleoclimate comparison is that, while the δ18O_c profiles presented here record independent records of seasonal variability in different specimens, the interpretation of this δ18O_c variability in terms of paleotemperature relies on clumped isotope data collected in one specimen (HU-027). Our assumption that all rudist specimens occurred in the same biostrome and sampled a highly similar climate allows us to use the seasonal δ18O_w structure inferred from the clumped isotope dataset in HU-027 to obtain temperature seasonality from δ18O_c profiles. Therefore, the temperature reconstructions from different δ18O_c profiles are not fully independent.

Figure 9A–C presents an overview of the data on the occurrence of modern bivalves from the OBIS dataset (OBIS, 2025) cross-referenced with the monthly temperatures in these modern locations based on NOAA SST data (Huang et al., 2017) for the years 1981–2010. The lowest monthly temperature in environments containing modern bivalve recordings is -1.8 °C. The maximum SST of the warmest month recorded in the NOAA dataset at any location in which modern bivalves are reported is 34.1 °C. Based on the combination of these datasets, it becomes clear that the Saiwan environment was warmer than any shallow marine environment that sustains bivalves in the modern world (Fig. 9D). The range of temperatures experienced by bivalves in the warmest setting studied by Compton et al. (2007; 32.7–41.8 °C) is frequently exceeded in the Saiwan environment, according to the over 4 °C higher mean WMMT estimated from our data (42.6 ± 4.0 °C; Table 4; Fig. 8A). The warmest monthly temperatures in the Saiwan environment thus commonly exceed the maximum monthly temperatures in the living environments of modern bivalves, even when we consider the mean warmest monthly temperature for all specimens instead of the monthly extremes recorded by some T. sanchezi specimens in our dataset (44 °C). When we consider that bivalves are known to stop growing their shell during stressful periods in the year (e.g. winter in most modern bivalves; (Ivany, 2012) and summer in some hotter climates (Buick and Ivany, 2004)), it may be possible (though perhaps unlikely) that the seasonal ranges recorded in our specimens from Saiwan underestimate the true seasonal temperature range in this paleo-environment.

Our clumped isotope data corroborates evidence based on stable oxygen isotope profiles through low-latitude Tethyan rudists (with assumptions of constant seawater δ18O_w value; e.g. Steuber et al., 2005; de Winter et al., 2017b) that shallow seas in the Cretaceous Tethys Ocean margins warmed up to temperatures unseen in modern climates (Fig. 9), but are not unheard of in the fossil record (e.g. Paleocene-Eocene Thermal Maximum tropical mean annual SSTs > 40 °C; Aze et al., 2014). Clumped isotope data from T. sanchezi specimen HU-027 also reveals that assumptions of year-round constant δ18O_w value can be very far off from the true δ18O_w seasonality: in the Saiwan site, the coldest season recorded in HU-027 is characterized by a mean δ18O_w value of -4.62 ± 0.86 ‰ VSMOW, which is more than 5 ‰ lighter than the mean δ18O_w value during the warmest season (0.86 ± 1.61 ‰ VSMOW). If δ18O_w would be considered constant year-round (e.g. -1 ‰ VSMOW), δ18O_c-based temperature reconstructions would underestimate the true seasonality at this site during the Campanian by over 10 °C (see discussion in Sect. 4.3 and results in Table 4). Our data instead shows that the low-latitude Saiwan paleoenvironment experienced a higher seasonal temperature range than the roughly contemporary (78 Ma) higher mid-latitudes of the Campanian boreal chalk sea recorded in shells from the Kristianstad basin (18.7 ± 3.8–42.6 ± 4.0 °C, or 23.9 ± 6.4 °C seasonal range for Saiwan vs. 15.3 ± 4.8–26.6 ± 5.4 °C, or 11.2 ± 7.3 °C seasonal range for Kristianstad basin), for which seasonal clumped isotope reconstructions were previously presented (de Winter et al., 2021a).

Note that the absolute mean annual temperature for the Saiwan ecosystem (33.1 ± 4.6 °C) was significantly higher than that of the higher latitude Boreal Chalk sea (20.1 ± 1.3 °C; de Winter et al., 2021a), yielding a temperature gradient of ∼ 13 °C over the latitudes 3° S–50° N. The present-day mean annual sea surface temperatures at the closest weather stations to these localities are 8.2 °C (Kristianstad, Sweden, 56° N; Huang et al., 2017; de Winter et al., 2021a) and 27.9 °C (Muscat, Oman, 23° N; NOAA, 2024), yielding a present-day SST gradient of ∼ 19 °C, more than 1.5 × that for the same locations in the Campanian. This result corroborates previous evidence from data and models that the latitudinal temperature gradient was smaller compared to the present-day during periods of warmer climate, such as the Campanian (Amiot et al., 2004; Burgener et al., 2018). However, it must be noticed that the paleolatitudes of these sites are slightly different than their modern latitudes.

The Campanian climate in Saiwan is significantly different from that of present-day Oman, which experiences a sea surface temperature seasonality of 24.2 ± 1.6–31.5 ± 1.6 °C (NOAA, 2024) and a very limited seasonal precipitation range (0–11 mm per month; ECMWF, 2024). The seasonal pattern in temperature and seawater composition in the Campanian is likely caused by large seasonal variability in precipitation, which is common in present-day tropical climates. In this climate, seasons with hot and relatively dry conditions (summers) are interchanged with cooler seasons featuring an influx of isotopically light water, which diluted the shallow seawater at Saiwan or produced a layer of lower salinity waters close to the sea surface, which was recorded by these very shallow-dwelling photosymbiotic rudists.

The above-mentioned conditions make summers in the Saiwan paleoenvironment hot, even for the Campanian with its global mean annual temperature of 20–25 °C (O'Brien et al., 2017). T. sanchezi apparently thrived under these conditions while V. vesiculosus and O. figari stopped producing their shell at temperatures close to those that limit modern shallow marine bivalves (∼ 34 °C; Clarke, 2014; Compton et al., 2007) and perhaps also during the high-precipitation and lower-salinity phases of the winter season. This suggests that T. sanchezi may have been particularly well-adapted to the high seasonality in temperature and precipitation and hot summers in its environment. This hypothesis is further supported by the observation that the genus Torreites occurs exclusively in the late Cretaceous low latitudes (18–29° N) of the near East and middle America (Global Biodiversity Information Facility, 2024). The unexpectedly high seasonal temperature variability in the Saiwan paleoenvironment (18.7 ± 3.8–42.6 ± 4.0 °C; Table 4; Fig. 8) might have provided relief for the Saiwan molluscs, since they were not forced to complete their entire life cycle at exceptionally high temperatures and may have recovered from heat exposure during the cooler seasons.

While the conditions in Campanian Saiwan approach the limits of what has been observed for present-day eukaryotes (Clarke, 2014), recent growth experiments on modern gastropods show that temperatures up to 45 °C can be tolerated by molluscs for limited amounts of time (Prayudi et al., 2024). Organisms that need to withstand these stressful heat conditions often develop specialized proteins and enzymes to continue bodily functions while experiencing close to lethal temperatures (Tehei et al., 2005; Tehei and Zaccai, 2007). While the Campanian fossils studied here do not preserve remnants of these organic molecules, these molluscs probably employed similar strategies to survive through the extreme summer heat. We thus hypothesize that the members of the Saiwan ecosystem (especially T. sanchezi) must have been evolutionarily adapted to its seasonally hot environment. These observations raise the question whether, given enough time, multicellular organisms such as shallow marine bivalves may survive life in hotter climates, and to which degree the thermal tolerance of modern relatives can be used as a reference point for interpreting the fossil record. More research into extreme paleo-communities such as the Saiwan ecosystem is crucial for understanding the evolutionary limits to which metazoans can adapt to warmer climates, how much time is needed to make these adaptations, and whether such adaptation can come in time to save modern shallow marine communities from rapid climate change.