Corresponding author: Marcus P.S. Badger, School of Earth Sciences, Bristol University, Wills Memorial Building, Queens Rd, Bristol, BS8 1RJ, UK. (email@example.com)
 The development of a permanent, stable ice sheet in East Antarctica happened during the middle Miocene, about 14 million years (Myr) ago. The middle Miocene therefore represents one of the distinct phases of rapid change in the transition from the “greenhouse” of the early Eocene to the “icehouse” of the present day. Carbonate carbon isotope records of the period immediately following the main stage of ice sheet development reveal a major perturbation in the carbon system, represented by the positive δ13C excursion known as carbon maximum 6 (“CM6”), which has traditionally been interpreted as reflecting increased burial of organic matter and atmospheric pCO2 drawdown. More recently, it has been suggested that the δ13C excursion records a negative feedback resulting from the reduction of silicate weathering and an increase in atmospheric pCO2. Here we present high-resolution multi-proxy (alkenone carbon and foraminiferal boron isotope) records of atmospheric carbon dioxide and sea surface temperature across CM6. Similar to previously published records spanning this interval, our records document a world of generally low (~300 ppm) atmospheric pCO2 at a time generally accepted to be much warmer than today. Crucially, they also reveal a pCO2 decrease with associated cooling, which demonstrates that the carbon burial hypothesis for CM6 is feasible and could have acted as a positive feedback on global cooling.
 Accompanying the middle Miocene growth of the East Antarctic Ice Sheet (EAIS) are major perturbations in the global carbon system, represented by some of the largest fluctuations in marine carbonate δ13C values in the Cenozoic [Flower and Kennett, 1995; Zachos et al., 2001]. A broad positive carbon isotope excursion (the “Monterey Excursion” [Vincent and Berger, 1985]) begins in the early Miocene (approximately 16.9 Myr ago) and terminates in the middle Miocene (~13.6 Myr ago [Holbourn et al., 2007]). Within this broad δ13C excursion, higher frequency fluctuations have been recognised with at least seven carbon isotope maxima (CM) defined [Woodruff and Savin, 1991]. These positive carbon isotope excursions are traditionally interpreted as the result of increased burial of organic carbon leading to a drawdown of carbon dioxide from the atmosphere, and subsequent global cooling and ice build-up [Flower and Kennett, 1993a; Vincent and Berger, 1985]. The largest of the carbon isotope maxima (“CM6” [Woodruff and Savin, 1991]) immediately follows the major ice expansion event of the middle Miocene (“E3”; [Flower and Kennett, 1993b] or “Mi-3”; [Miller et al., 1991]) and therefore may represent an important positive feedback in the climate system.
 However, it has recently been suggested that carbonate carbon isotope maxima associated with glacial transitions may be evidence of a negative feedback in the climate system [Shevenell et al., 2008]. Under this alternative scenario, ice sheet expansion blankets an area of silicate basement that was previously a sink for atmospheric pCO2 via silicate weathering [Pagani et al., 1999; Lear et al., 2004; Shevenell et al., 2008; Tian et al., 2009]. Both scenarios could result in positive carbonate carbon isotope excursions: the former by removal of 13C-depleted carbon from the ocean-atmosphere reservoir as more organic matter is buried, and the latter by lowering buried organic matter δ13C values as a result of increased photosynthetic isotopic fractionation (εp) due to higher concentrations of dissolved carbon dioxide ([CO2(aq)]). Critically, these scenarios involve opposite changes in atmospheric carbon dioxide concentration. Therefore, to assess the likelihood of these mechanisms, we reconstruct atmospheric pCO2 following the middle Miocene ice sheet expansion using two independent proxies: alkenone and boron isotope paleobarometry.
2 Materials and Methods
 The Blue Clay Formation of Malta (at Ras il-Pellegrin (RIP); Figure 1; 35°54.93'N 14°20.06'E) is a continuous land-based section of Miocene pelagic clays and marls with high sedimentation rates (40 mMyr-1) [Abels et al., 2005] and estimated paleowater depth of 500 to 600 m [Bellanca et al., 2002; Bonaduce and Barra, 2002]. The high clay content of the Blue Clay Formation has resulted in excellent preservation of microfossils, and the simple tectonic history of Malta [Dart et al., 1993] has resulted in material that has never been exposed to high temperatures or pressures, suggesting that primary geochemical signals should be preserved. The transition to the Blue Clay Formation from the older Globigerina Limestone Formation below coincides with the ice volume build-up “E3” [Abels et al., 2005; Woodruff and Savin, 1991]. The associated change in preservation style precludes a continuous record across the entire climate transition. Nevertheless, we sampled the Blue Clay Formation at 35 cm (~9 kyrs) resolution enabling us to generate a stratigraphy of the 1.1 million years following ice volume build-up “E3”. In the field, samples were tied to the lithostratigraphic log of Abels et al. . Subsequent refinement was achieved by matching bulk isotope stratigraphies. The astronomical recalibration of the lower part of the Blue Clay Formation according to Mourik et al.  was used for the age model.
 We analysed bulk and fine fraction (<63 μm) carbonate for δ13C and δ18O, and the planktic foraminifera Globigerinoides trilobus for Mg/Ca, δ13C, δ18O, B/Ca, and δ11B. We also generated a compound specific (C37 alkenone) carbon isotope record.
2.1 Carbonate Stable Isotope Stratigraphy
 Bulk sediment and foraminiferal isotope analyses were performed on a ThermoFinnigan MAT 252 with online sample preparation using an automated Kiel III carbonate device at Cardiff University. Long term uncertainties based on repeat analysis of NBS-19 are ±0.08‰ and ±0.05‰ for δ18O and δ13C respectively (2σ). Fine fraction isotope samples (<63 μm) were prepared offline and resultant CO2 analysed using a VG Optima dual inlet gas source mass spectrometer at the NERC Isotope Geosciences Laboratory, analytical uncertainties are typically ±0.2‰ for both δ18O and δ13C (2σ). All isotope data are reported as per mil on the VPDB scale.
2.2 Planktic Foraminiferal Mg/Ca Analyses
 Between 19 and 30 individuals of G. trilobus were picked from the 250-355 μm size fraction and cleaned following the procedure of Barker et al. . In the modern ocean, G. trilobus is considered to be the same biological species as G. sacculifer, although G. sacculifer apparently did not evolve until the Pliocene [Hemleben et al., 1989]. Here we assume G. trilobus and G. sacculifer are the same species. Samples were dissolved in 0.065 M nitric acid and analysed by sector field inductively coupled plasma mass spectrometry Thermo Scientific ELEMENT-XR (ICP-MS) at Cardiff University. Each sample was calibrated against a standard with matched Ca concentration to reduce matrix effects. Long-term analytical precision is better than 2 % (r.s.d.).
2.3 Organic Geochemistry
 5 g powdered samples were saponified and ultrasonically extracted using sequentially, methanoic 0.1 M KOH (5 % H2O), methanol and a 3:1 (v/v) azeotrope of dichloromethane (DCM):methanol. Neutral fractions were obtained from the total extract by liquid-liquid separation with n-hexane:DCM (9:1 v/v) and silica gel column chromatography was used to divide samples into four fractions (F), eluted sequentially with n-hexane (4 ml, Fraction 1), n-hexane:DCM (2:1 v/v; 2 ml, F2), DCM (5 ml, F3) and DCM:Methanol (95:5 v/v; 5 ml, F4). Alkenones elute in F3, and their quantification for calculation was performed on a Carlo Erba Instruments HRGC5300 gas chromatogram (GC) fitted with a flame ionisation detector and a Chrompack fused silica capillary column (50 m × 0.32 mm internal diameter; CP Sil-5CB stationary phase, dimethylpolysiloxane equivalent, 0.12 µm film thickness). Compound specific carbon isotope analyses were performed on a Finnigan MAT Delta S coupled to a GC with on-column injector using a modified Finnigan MAT Type 1 GC combustion interface at the University of Bristol. A Zebron ZB-1 fused silica capillary column (50 m × 0.32 mm internal diameter; dimethylpolysiloxane equivalent stationary phase, 0.12 µm film thickness) was used. Samples were injected at 70°C, and the oven programmed to increase in temperature to 130°C at 20 °Cmin-1 then to 300°C at 4°Cmin-1, remaining isothermal at 300°C for 25 minutes. Long term uncertainties, on the basis of repeat standard measurements, are ±0.6‰ (2σ).
2.4 Foraminiferal δ11B and B/Ca Analyses
 Six samples of G. trilobus (250-355 µm) each consisting of around 100 tests (~2 mg of CaCO3) were analysed for their boron isotopic composition on a ThermoFinnigan NEPTUNE multicollector ICP-MS at the University of Bristol following the methodology described by Foster . Analytical precision of this approach is ± 0.23 ‰ at 95 % confidence as determined by the long term reproducibility of full procedural replicates of an in-house coral standard. Prior to isotope analysis, but following cleaning and dissolution, an aliquot of each sample was analysed for trace element content using a sector field ThermoFinnigan ELEMENT 2 ICP-MS [Foster, 2008; Ni et al., 2007] at the University of Bristol. This analysis ensured cleaning was successful and provided a measure of the B/Ca ratio for each sample. The long-term reproducibility of our in house consistency standards for B/Ca is ± 2% (95% confidence) and our accuracy is better than 5% [Ni et al., 2007].
2.5 Mg/Ca Paleotemperature Equation
 Critical to the accuracy of the alkenone and, to a lesser extent, the boron pCO2 reconstruction is an accurate record of sea surface temperature (SST). However, foraminiferal calcite Mg/Ca is controlled not only by temperature, but also the Mg/Ca ratio of the seawater in which they calcify. The application of this proxy beyond the oceanic residence time of Ca and Mg (1 and 22 Myrs respectively; Fantle and DePaolo [2005, 2006]) can therefore be complicated by changes in seawater Mg/Ca ratios (Mg/Casw) with time. This is often corrected for by assuming a linear relationship between seawater Mg/Ca and foraminiferal Mg/Ca [e.g., Lear et al., 2000] although it has been suggested that this dependency may be better described using a power function [Hasiuk and Lohmann, 2010] (equation (1)).
Where Mg/Caforam is the Mg/Ca ratio of foraminiferal calcite, Mg/Casw(0) and Mg/Casw(t) are the Mg/Ca ratios of seawater today and at time t respectively, A, B, and C are constants (‘exponential’, ‘pre-exponential’ and ‘power’ respectively) and T is temperature.
 Mg and Ca are both delivered to the oceans via rivers and removed into sediments including carbonates. Hydrothermal alteration of basalt at ocean ridge systems represents an important sink for magnesium and source for calcium. All of these fluxes have varied over time. The residence times of Ca and Mg mean that Mg/Casw is unlikely to have varied significantly over the duration of our record, but may have been considerably different in the Miocene compared to the present day. Attempts have been made to reconstruct Mg/Casw over time using geochemical models [Hardie, 1996; Wilkinson and Algeo, 1989, Fantle and DePaolo, 2005; 2006], echinoderm skeletons [Dickson, 2002, 2004], fluid inclusions in marine evaporites [Lowenstein et al., 2001, Horita et al., 2002] and carbonate veins precipitated in oceanic basalt [Coggon et al., 2010]. Although these various reconstructions agree that Mg/Casw was lower during the Miocene than today, its precise value remains unclear.
 Here we apply the fluid inclusion based Mg/Casw value of Horita et al.  (Mg/Casw = 3.43) to the generic planktic Mg/Ca paleotemperature equation of Anand et al.  (A = 0.09, B = 0.38 in equation (1)), and apply the power constant for G. sacculifer (C = 0.41 in equation (1)) as determined by Evans and Müller  based on the data of Delaney et al. . This results in the Mg/Casw corrected paleotemperature equation given in equation (2).
 The fluid inclusion data has the advantage over other methods of also providing absolute values for [Mg] and [Ca], as required to reconstruct carbonic acid dissociation constants K1 and K2 (see section 2.7).
2.6 Alkenone Paleobarometry
 Isotopic fractionation between dissolved inorganic carbon (DIC) and algal biomass (εp) occurs during photosynthesis and is strongly controlled by [CO2(aq)] [Laws et al., 1995], a relationship that serves as the basis of the phytoplankton-based pCO2 paleobarometer [Hollander and McKenzie, 1991; Pagani et al., 1999]. However, a range of other factors, including algal growth rate and cell geometry also govern εp values [Bidigare et al., 1997; Popp et al., 1998a]. In order to control source organism and restrict cell size effects, we measure the carbon isotopic composition of C37 alkenones (δ13C37). C37 alkenones are long chain methyl ketones produced by a restricted group of haptophyte algae [Marlowe et al., 1990] and have been used in several studies to reconstruct levels of atmospheric carbon dioxide over time [Jasper et al., 1994; Pagani et al., 2005; Seki et al., 2010]. Haptophyte biomass δ13C values can be determined by correcting δ13C37 for the biosynthetic offset of 4‰ [Popp et al., 1998a]; εp values are then determined by reconstructing the carbon isotopic composition of dissolved CO2 (δ13C[CO2(aq)]). Shallow dwelling foraminifera have been used previously to determine the DIC δ13C value used for alkenone paleobarometry. However, the low number of individuals available for analysis here results in a record with a low signal to noise ratio. We therefore estimate surface water DIC δ13C using fine fraction δ13Ccarb, with an isotopic offset of +2‰ calculated from the mean difference between G. trilobus δ13C and δ13Ccarb over CM6. Using this δ13C record we determine δ13C[CO2(aq)] with temperature dependant fractionation quantified using Mg/Ca paleothermometry as described above. Calculated εp values were converted to [CO2(aq)] using the relationship determined by Bidigare et al.  assuming a growth rate (phosphate) dependant b-value of 110.65 for Emiliania huxleyi, which is typical for oligotrophic regimes [e.g. Bidigare et al., 1997; Pagani et al., 2010]. Atmospheric pCO2 is then calculated by applying Henry's law, assuming equilibrium between ocean-atmosphere CO2 and again using reconstructed SSTs; where necessary, SST was estimated on the basis of linear interpolation in the time domain.
 Uncertainty propagation on our alkenone-derived pCO2 estimates was performed by Monte Carlo modeling (n = 25000). Salinity uncertainty was estimated by reconstructing δ18Osw using paired G. trilobus Mg/Ca SST and δ18O data, assuming all variation in δ18Osw was due to salinity variations using the sensitivity described in Maslin et al. . This approach produces a maximum uncertainty as most of the variability in δ18Osw will be due to changing ice volume. Uncertainties of 2°C and 0.2‰ were applied to temperature and δ13Ccarb, respectively (normal probability density function (pdf), 2σ error) and 2.5 mmol and 0.05 psu on salinity and [PO43-] respectively (uniform pdf). 2σ errors on δ13C37 were estimated from replicate runs. An 11% error on the slope of b = a[PO4] + c was assumed [Pagani et al., 1999].
2.7 Boron Based Paleobarometry
 There are two isotopes of boron, 10B and 11B, with natural abundances of ~20 % and ~80 %, respectively. Isotope variations are described in delta notation as follows:
where 11B/10BNIST951 is the 11B/10B ratio of NIST SRM 951 boric acid standard (11B/10B = 4.04367; Catanzaro et al. ).
 Boron exists as two species in aqueous solutions at typical ocean pH: boric acid (B(OH)3) and borate ion (B(OH)4-). The abundance of these species is pH dependent with ~80 % B(OH)3 at typical seawater pH. Due to structural differences between the two boron species in seawater there is a pronounced isotopic fractionation between them in seawater, with B(OH)3 being enriched in 11B. In order to maintain a constant δ11B of seawater, the isotopic composition of each species also varies according to pH, for instance, the isotopic composition of B(OH)4- is related to pH by the following:
where pK*B is the -1og10 of the stoichiometric equilibrium constant for boric acid [Dickson, 1990] at the in situ temperature, salinity and pressure, δ11Bsw is the isotopic composition of seawater (39.61‰; Foster et al. ), δ11BB(OH)4- is the isotopic composition of borate ion. The isotopic fractionation between the two aqueous species of boron in seawater (αB) has recently been determined as 1.0272 ± 0.0006‰ [Klochko et al., 2006].
 On the basis of isotopic measurements of marine carbonates, Hemming and Hanson  suggested that the borate ion species is preferentially incorporated into marine carbonate. However, NMR studies [Klochko et al., 2009] have shown that some trigonal BO3 is also present in CaCO3 which has been used by some to argue that boric acid may also be incorporated into CaCO3. Recent isotopic measurements of benthic foraminifera by MC-ICP-MS however confirm that only very minor amounts of boric acid can be incorporated in foraminifera (<1 %; Rae et al. ). Consequently, for epifunal benthic foraminifera measured by MC-ICP-MS δ11Bforam = δ11BB(OH)4-, and equation ((4)) can be used to calculate pH.
 For the planktic foraminifera G. trilobus used here (assuming G. trilobus is the same as G. sacculifer), it has been shown [Sanyal et al., 2001], using negative ion thermal ionisation mass spectrometry (NTIMS), that there was a strong pH dependency of δ11B but δ11Bforam ≠ δ11BB(OH)4-. Here we follow Foster et al.  in order to correct for these “vital effects” and calculate δ11BB(OH)4 from the δ11B of G. trilobus (300-355 µm) by:
 The δ11BB(OH)4- can then be inserted into equation ((4)) to calculate pH. Since G. trilobus is a predominantly mixed layer dweller, we assume that the calculated pH is that of surface water.
 The stoichometric dissociation constant of boric acid (K*B) is temperature, salinity and pressure dependent [Dickson, 1990]. These variables do not have a large impact on the calculated pH and pCO2 (e.g. ~10 ppm/°C; ~2 ppm/psu) and here salinity is assumed to be 35 psu throughout and SST is given by G. trilobus Mg/Ca as detailed above.
 From equation ((4)) it can be seen that use of δ11B in G. trilobus to reconstruct pH and hence pCO2 requires an estimate of the isotopic composition of seawater in the past. The oceanic residence time of boron is ~14 Myrs [Lemarchand et al., 2002], and modeling approaches suggest that the seawater boron isotope ratio was lower by up to a few per mil during the Miocene, but the absolute magnitude of the change in seawater δ11B is largely uncertain [Lemarchand et al., 2002; Simon et al., 2006; Pearson and Palmer, 2000; Paris et al., 2010]. Using a modification of the depth profile approach of Pearson and Palmer , Foster et al.  calculate δ11Bsw was 37.8‰ for the middle Miocene, a value that is in good agreement with the modeling of Lemarchand et al.  and the reconstruction of Pearson and Palmer . Given the long residence time of boron in seawater (~14 Myrs; [Lemarchand et al., 2002]), that value is likely appropriate for our entire record.
 To fully reconstruct changes in the ocean carbonate system, two of the six co-varying parameters (pH, [CO2(aq)],[HCO3-], [CO32-], Total Alkalinity and DIC) must be known. Foster et al.  calculate surface water alkalinity for the middle Miocene using a modeling approach that closely follows Tyrrell and Zeebe  supplemented with new δ11B measurements of benthic foraminifera to estimate deep water pH. By assuming a similar to modern surface to deep alkalinity gradient, Foster et al.  estimate surface water total alkalinity in the middle Miocene to be 1293 ± 200 µmol/kg. The uncertainty in this estimate accounts for uncertainties in the surface-deep gradient, the depth of carbonate compensation depth, and any variations in total alkalinity during the course of our record (see Foster et al.  for details). The uncertainties in pCO2 we calculate from the boron based-pH are dominated by uncertainties in this second carbonate system parameter (± 40 ppm). A quadratic addition of other likely uncertainties, e.g., temperature (± 1 °C) and δ11B (around ± 20 ppm), gives a total uncertainty of approximately ± 50 ppm. It is important to note that following Tyrrell and Zeebe  we also account in all carbonate system calculations for the effect of changing [Mg] and [Ca] of seawater on the carbonic acid dissociation constants K1 and K2 and the solubility product of CaCO3 (Ksp), by using the [Ca], [Mg], and Mg/Casw taken from the fluid inclusion data of Horita et al. .
 The pH dependent speciation of boron in seawater also forms the basis for the B/Ca proxy of the carbonate system. It is thought that the incorporation of boron into calcium carbonate can be described by the following equilibria [Hemming and Hanson, 1992]:
 Following Zeebe and Wolf-Gladrow  the following exchange equilibria (KD) can be defined:
 It therefore follows that, since the B(OH)4- concentration is pH dependent, if KD can be calibrated and the pH determined (e.g. using boron isotopes), equation ((4)) can be solved for HCO3- providing a second variable of the carbonate system which would allow the whole system to be resolved. However, the KD for planktic species such as G. sacculifer has been shown to be species specific and dependent on test size [Ni et al., 2007] as well as one or several environmental variables (e.g. temperature and/or [CO32-]; [Yu et al., 2007; Foster, 2008]). A recent culture study has shown that for G. sacculifer, temperature had a negligible effect on boron incorporation but KD was inversely proportional to [CO32-], perhaps as a consequence of [CO32-] competing with B(OH)4 for the same lattice site [Foster et al., 2008; Allen et al., 2011, 2012; Allen and Hönisch, 2012]. Although Foster  presents a core top calibration for G. sacculifer (500-600 µm) the relationships between KD and temperature and [CO32-] remain rather poorly defined [Allen and Hönisch, 2012]. Furthermore, as the evolution of seawater [B] over these timescales is also unknown, we consider that a more quantitative treatment of our B/Ca data would be somewhat premature.
2.8 Carbon Cycle Modeling
 We use the mass balance model of Kump and Arthur  to calculate changes in the isotopic composition of inorganic and organic carbon in the ocean/atmosphere system. Changes in the modeled carbon content of the system are driven by changing inputs from weathering and metamorphism/volcanism and outputs from the burial of carbon as carbonate minerals and organic matter. The model accounts for changing εp values due to varying pCO2 and we also include a negative weathering feedback on changing pCO2 as detailed by Kump and Arthur :
where Fwsil is the global silicate weathering flux. In order to simulate Miocene climate we set initial pCO2 at 330 ppm (derived from our alkenone paleobarometry results) and ran the models for 1 Myrs to reach steady state before starting perturbations. Models were solved numerically using MATLAB (Simulink) with variable step solver ode45.
3 Results and Discussion
 Our bulk carbonate oxygen isotope record (δ18Ocarb) shows the final step (“E3”) of the globally recognised increase in δ18O at the MMCT, and our fine fraction (<63 µm) carbonate carbon isotope record (δ13Ccarb) shows the distinctive double peak of CM6, confirming that the RIP section faithfully records global changes (Figure 2). Alkenone δ13C (δ13C37) values are somewhat variable, but display an overall increase across CM6 (Figure 3). The boron-based proxy data are limited to the older portion of CM6 due to sample availability, but nevertheless also display a consistent increase (Figure 3).
G. trilobus Mg/Ca ratios decrease from 6.6 mmol/mol at the start of the record at 13.799 Ma, to 3.9 mmol/mol by 13.021 Ma (Table 1). For much of the early part of the record only the di-unsaturated alkenone is present suggesting SSTs exceeded 29°C (Table 1; [Müller et al., 1998]); this could reflect enhanced degradation of the tri-unsaturated alkenone, but such differential degradation, even under highly oxidising conditions, appears to increase by only ~0.1 (equivalent to <0.5°C; [Huguet et al., 2009]). Moreover, the tri-unsaturated alkenone is present in younger sediments, and indices reach a minimum value of 0.91 at 13.709 Ma. The high Mg/Ca ratios and indices suggest high SSTs (> 29°C) at this point in the Miocene; the SST record determined using our Mg/Ca calibration (equation ((2))) indicates a decrease of about 4°C following the ice sheet expansion, independent of the value of Mg/Casw (Figure 4). We note that even given the large uncertainties surrounding Mg/Casw and the response of foraminifera to changing Mg/Casw, our Mg/Ca SST estimates are predominantly within uncertainty of our based estimates (Figures 4 and 5), providing support for the power law approach of Evans and Müller .
Table 1. Temperatures and Planktic Foraminiferal (G. trilobus) Mg/Ca and Mg/Ca-Temperatures From the Blue Clay Formation at Ras il-Pellegrin, Malta
aNote that the paleothermometer cannot be used for temperatures greater than 29°C, so temperatures marked with a * should be considered minimum estimates.
 Our calculated εp values are almost identical to published Miocene values [Pagani et al., 1999] determined from similar aged oligotrophic regions, giving us confidence that growth rate effects were minimal in the RIP section, and that the reconstructed pCO2 is a global, rather than local signature [Pagani et al., 1999] (Figure 5). Our calculated pCO2 levels of 260 to 350 ppm are slightly higher than those reported previously (~240 ppm) [Pagani et al., 1999, 2005]. This difference is due to our higher SST estimates, which affect air-sea CO2 equilibria. Post-depositional preservation effects at the deep-sea carbonate-rich sites used by Pagani et al.  almost certainly resulted in artificially cool temperature estimates from oxygen isotope paleothermometry [Pearson et al., 2001], and recently revised Miocene pCO2 estimates are similar to those observed here [Pagani et al., 2010]. The clay-rich lithology at RIP has resulted in excellent preservation of foraminifera enabling more accurate estimates of past SST using Mg/Ca paleothermometry. These differences result in alkenone pCO2 estimates that are somewhat higher than previously reported by Pagani et al. [1999, 2005] but are still generally low (~300 ppm) and comparable to those measured in other Neogene studies [Kürschner et al., 2008; Foster et al., 2012] (Figure 5).
G. trilobus δ11B values increase across CM6 (Figure 3), which implies an increasing pH and decreasing pCO2 regardless of our choice of seawater δ11B (Table 2) or second carbonate system parameter. In support of this conclusion, we note that our G. trilobus B/Ca record also increases across CM6 (Figure 3), which, in the absence of other complicating factors [Yu et al., 2007; Foster, 2008; Allen et al., 2011], also implies an increase in surface water pH and hence decrease in pCO2 [Foster, 2008].
Table 2. Sea surface pH and atmospheric pCO2 calculated from the maximum and minimum δ11Bforam values from our record for a range of assumed δ11BSW calculated assuming total alkalinity =1300
Assumed Seawater δ11B (‰)
Calculated surface pH for δ11Bforam= 16.66
Calculated surface pH for δ11Bforam= 15.86
Calculated pCO2 for δ11Bforam= 16.66
Calculated pCO2 for δ11Bforam= 15.86
Calculated CO2 change between maximum and minimum δ11Bforam
 Using a δ11Bsw = 37.8‰ [Foster et al., 2012], δ11B-based pH changes from 7.91 ± 0.03 to a maximum of 8.02 ± 0.03. Using this pH record, along with a total alkalinity of 1293 ± 200 µmol, yields pCO2 estimates that drop from 311 ± 55 ppm to a minimum of 230 ± 43 ppm (ΔpCO2 of ~80 ppm). Importantly, these new pCO2 estimates using boron isotopes are in very good agreement with our alkenone-based estimates (Figure 4), which provide confidence in the validity of the δ11Bsw and total alkalinity reconstructions of Foster et al. . Boron isotope compositions of G. trilobus from the RIP section are also very similar to middle Miocene aged G. trilobus measured by MC-ICP-MS by Foster et al.  from ODP Site 761 and Site 926.
4 The Miocene Carbon Cycle and Climate
 Our δ13C37, B/Ca and δ11B records show positive trends over CM6 (Figure 3); for the former, this yields a decrease in εp values. Therefore, regardless of the absolute values or treatment of Mg/Casw, the alkenone and boron based-CO2 proxies all indicate CM6 is associated with a pCO2 decrease (Figure 4). In order to evaluate potential causes of CM6, we ran a simple ocean carbonate system model [Kump and Arthur, 1999] to assess the feasibility of the organic matter burial hypothesis. We inverse model the δ13Ccarb increase by assuming it is driven by two discrete pulses of increased organic carbon burial. Under this scenario, the magnitude of the δ13Ccarb increase across CM6a and CM6b requires a 70% and a 50% increase in organic carbon burial respectively, with each perturbation lasting 41 kyrs (inset; Figure 6a). This results in a 60 ppm CO2 drawdown, which is in agreement with both our alkenone and boron pCO2 records (Figure 6). This is in contrast to the silicate-weathering hypothesis which would have led to an increase in pCO2.
 The model and the records cannot be used to identify a cause of the suggested increase in organic matter burial (i.e., production versus preservation), which prevents a useful feasibility assessment of this modeled scenario. Nevertheless, we note that the magnitude of the organic carbon burial increase required to produce the observed carbon isotope excursion is substantial, and suggests that other factors may also have been involved. Although our preferred causal mechanism for CM6 involves at least some component of increased organic carbon burial, we cannot rule out other influences on the size and carbon isotopic composition of the ocean-atmosphere reservoir. For example, cooler bottom water temperatures could have facilitated increased sequestration of methane in hydrates, which would represent a reduced flux of 13C-depleted carbon to the atmosphere from the marine realm. Moreover, our model and records do not distinguish between increased organic matter sequestration in marine vs terrestrial realms or, in the case of the former, between burial at continental margins or in the deep sea. However, export productivity records from the Atlantic do show a pronounced increase over CM6 [Diester-Haass et al., 2009], and there is correlation between the presence of organic-rich sediments and carbon isotope maxima within the Monterey formation [Flower and Kennett, 1993a], suggesting that increased burial of organic matter in the marine realm was likely responsible for a major component of the global δ13Ccarb excursion.
 Our records reveal that organic carbon burial during CM6 could have acted as an important positive feedback on ice sheet growth. Global cooling and sea-level fall associated with the initial growth of the ice sheet would have increased meridional thermal gradients and may have increased ocean ventilation [Flower and Kennett, 1993a]. This may have led to enhanced biological productivity and CO2 drawdown as a result of a stronger biological pump, and consequently further cooling [Flower and Kennett, 1993a; Vincent and Berger, 1985]. Thus, CM6 likely illustrates an important positive feedback in the global climate system as opposed to a negative feedback [Shevenell et al., 2008] or a primary forcing mechanism [Vincent and Berger, 1985] as has been previously suggested.
5 Summary and Conclusions
 The largest of the middle Miocene “carbon maxima” events was associated with a pCO2 decrease of 59 ± 63 ppm (from δ13C37) or 82 ± 72ppm (from δ11Btrilobus) (e.g. ~ 20%). Both the magnitude and direction of the isotopic shift and observed pCO2 change are consistent with an increase in organic carbon burial, perhaps fueled by increasing oceanic temperature gradients and overturning following the expansion of the Antarctic ice sheet. At this time, we estimate atmospheric pCO2 was near 300 ppm, somewhat higher than previous alkenone- [Pagani et al., 1999] and boron-based [Pearson and Palmer, 2000] techniques but in agreement with more recently published long term records [Kürschner et al., 2008; Foster et al., 2012]. We attribute these differences to the use of higher sea surface temperatures in the case of the alkenone estimates and more accurate δ11B determinations. Crucially, our estimates suggest an emerging consensus for Miocene pCO2 between the alkenone, boron isotope and leaf stomatal approaches (Figure 5). Moreover, these values are only slightly higher than modern pre-industrial values and overlap with the threshold values thought to be required for bipolar glaciation [~280 ppm; DeConto et al., 2008]. Recent modeling studies suggest that changes in North American topography and other boundary conditions in the Miocene prevented glaciation and may have caused a lower pCO2 threshold for full blown northern hemisphere glaciation to exist during the Miocene [Foster et al., 2010]. Nonetheless, since sea surface temperatures estimated from both alkenones and G. trilobus Mg/Ca ratios at this Mediterranean site indicate temperatures >29°C, our results reaffirm the enigma of a warmer Miocene-world with, at times, less ice than today but only slightly higher levels of atmospheric pCO2. One potential resolution of these data is that climate sensitivity to pCO2 is greater than previously thought [Pagani et al., 2010]. The impact of high latitude vegetation on Earth's albedo may have also played an important role in Earth's energy budget in the Miocene [Knorr et al., 2011]. However, further global temperature records are required to solve the Miocene paleoclimate enigma.
 This manuscript is dedicated to the memory of Ben Flower, an inspirational paleoceanographer who will be missed by many. We thank F. J. Hilgen for introducing us to the section, A. A. Mourik for assistance in the field, G. Debono for arranging permissions to work on the section, F. Gill (OGU), J. Becker and J. Green for analytical assistance, P. Pearson for taxonomic assistance and S. Barker for modeling assistance. We also thank Ian D. Bull of the NERC Life Sciences Mass Spectrometry Facility for assistance with alkenone δ13C determinations. This manuscript was improved by the careful comments of two anonymous reviewers and the editor. This work was supported by NERC and the National Museum Wales in the form of a CASE studentship (M.P.S.B.), NERC grants NE/D008654/1 and NE/D010241/1 (C.H.L. and R.D.P.), a NERC Advanced Fellowship NERC NE/D00876X/2 (G.L.F.) and NIGL award IP/920/1106 (C.H.L. and R.D.P.).