1. Introduction
The detection and characterization of habitable-zone exoplanets represent a significant frontier in modern astronomy. Traditionally, the search for exoplanetary habitability and biosignatures has focused on rocky exoplanets, guided by Earth’s example as a life-bearing world (Kasting et al. 1993; Meadows & Barnes 2018). However, the extraordinary diversity of exoplanetary systems discovered over the past three decades has prompted the exploration of alternative habitable environments. This broader perspective could reveal planetary conditions that are both more prevalent and more conducive to atmospheric characterization. One such promising category is Hycean worlds, a recently proposed class of water-rich exoplanets that are observable with current astronomical facilities (Madhusudhan et al. 2021).
Hycean worlds are characterized by their extensive global oceans beneath hydrogen-rich atmospheres, providing a significantly wider habitable zone than terrestrial planets. These planets typically have radii between 1 and 2.6 R⊕ and masses ranging from 1 to 10 M⊕. As a subset of temperate sub-Neptunes, they exhibit diverse atmospheric and internal structures (Madhusudhan et al. 2020, 2021; Nixon & Madhusudhan 2021; Piette & Madhusudhan 2020). Given the prevalence of sub-Neptune-sized exoplanets (Fulton & Petigura 2018), Hycean planets are likely abundant in the exoplanet population. Their volatile-rich interiors lead to lower densities, larger radii, and lower surface gravity than rocky planets of comparable mass. These characteristics, combined with their hydrogen-rich, low mean molecular weight (MMW) atmospheres, result in large atmospheric scale heights, making them highly accessible for atmospheric characterization. This accessibility extends to potential biomarker detection using relatively modest observational time with the James Webb Space Telescope (JWST) (Madhusudhan et al. 2021; Phillips et al. 2021, 2022; Leung et al. 2022).
The concept of Hycean planets was largely inspired by the characterization of the habitable-zone sub-Neptune K2-18 b (Montet et al. 2015; Cloutier et al. 2017; Benneke et al. 2019b; Cloutier et al. 2019). Its bulk properties suggest the possibility of a water-rich interior and a liquid-water ocean beneath a hydrogen-dominated atmosphere (Madhusudhan et al. 2020). With a mass of 8.63 ± 1.35 M⊕ and a radius of 2.61 ± 0.09 R⊕, the planet has an equilibrium temperature of approximately 250–300 K for an albedo between 0 and 0.3 (Benneke et al. 2019b; Cloutier et al. 2019). While K2-18 b is a plausible Hycean candidate, alternative internal structures and non-habitable surface conditions remain viable interpretations (Madhusudhan et al. 2020; Nixon & Madhusudhan 2021; Piette & Madhusudhan 2020), especially in the absence of cloud/haze-free atmospheres (e.g., Scheucher et al. 2020; Innes et al. 2023; Pierrehumbert 2023). Initial near-infrared transmission spectroscopy (1.1–1.7 μm) from the Hubble Space Telescope (HST) WFC3 suggested a hydrogen-rich atmosphere with strong H2O absorption (Benneke et al. 2019b; Tsiaras et al. 2019; Madhusudhan et al. 2020). However, subsequent analyses identified degeneracies between H2O and CH4 in the HST spectrum (Bézard et al. 2022; Blain et al. 2021) and potential stellar contamination effects (Barclay et al. 2021), rendering the earlier H2O inference inconclusive.
Observations with JWST offer an opportunity to refine our understanding of K2-18 b’s atmospheric, surface, and interior conditions. Theoretical studies suggest that even in the presence of high-altitude clouds, JWST can detect key carbon, nitrogen, and oxygen (CNO) molecules such as H2O, CH4, and NH3, alongside potential biosignatures like dimethyl sulfide (DMS), methyl chloride (CH3Cl), and carbonyl sulfide (OCS) (Madhusudhan et al. 2021; Constantinou & Madhusudhan 2022). Additionally, the atmospheric abundances of prominent CNO molecules may reveal the presence of underlying surfaces beneath hydrogen-rich atmospheres (Hu et al. 2021; Tsai et al. 2021; Yu et al. 2021). For instance, an ocean beneath a shallow hydrogen atmosphere, a hallmark of a Hycean world, may manifest as an enhanced abundance of CO2, H2O, and CH4, along with a depletion of NH3 (Hu et al. 2021; Tsai et al. 2021; Madhusudhan et al. 2023).
In this work, we present the first JWST transmission spectrum of K2-18 b, obtained using the NIRISS SOSS and NIRSpec G395H instruments across the 0.9–5.2 μm wavelength range. These observations enable constraints on the planet’s atmospheric composition and surface conditions, marking the beginning of a new era in the atmospheric characterization of low-mass exoplanets with JWST. The remainder of this paper is structured as follows: Section 2 details our JWST observations and data reduction, Section 3 describes our atmospheric retrieval analysis, and Section 4 summarizes our findings and their implications.
2. Observations and Data Reduction
We obtained transmission spectra of K2-18 b using JWST’s NIRSpec (Ferruit et al. 2012; Birkmann et al. 2014) and NIRISS (Doyon et al. 2012, 2023) instruments as part of the JWST GO Program 2722 (PI: N. Madhusudhan). Two primary transits of the planet were observed: one with each instrument.
2.1 NIRSpec
The first transit was observed with the NIRSpec G395H grating on 2023 January 20-21, spanning 5.3 hours—approximately twice the expected transit duration. The observation utilized the Bright Object Time Series (BOTS) mode with the F290LP filter and SUB2048 subarray. Spectra were dispersed across two detectors (NRS1: 2.73–3.72 μm; NRS2: 3.82–5.17 μm) with a gap between 3.72 and 3.82 μm. The spectral resolution was R ∼ 2700. Given the host star K2-18’s brightness, target acquisition (TA) was conducted using a nearby reference star (2MASS J11301306+0735116).
The data reduction followed standard JWST Science Calibration Pipeline procedures (Bushouse 2020) with custom enhancements for spectral extraction. Stage 1 included saturation flagging, bias subtraction, linearity correction, dark current subtraction, jump detection (5σ threshold), and ramp fitting. Background subtraction was performed to mitigate 1/f noise (Alderson et al. 2023; Rustamkulov et al. 2023). Stage 2 applied wavelength calibration, omitting flat-field correction for differential transit measurements (Alderson et al. 2023). Stage 3 involved custom spectral extraction using an optimal extraction algorithm (Horne 1986), incorporating principal component analysis for point-spread function (PSF) estimation.
2.2 NIRISS
The second transit was observed with the NIRISS Single Object Slitless Spectroscopy (SOSS) mode on 2023 June 1, spanning 4.9 hours. The GR700XD grism (R ∼ 700) and CLEAR filter provided a 0.85–2.85 μm wavelength range. Data reduction utilized the JWST Science Calibration Pipeline (Bushouse 2020) and JExoRES pipeline (Holmberg & Madhusudhan 2023). A multi-step background subtraction technique mitigated 1/f noise (Radica et al. 2023; Albert et al. 2023). The first spectral order was used for analysis due to the second order’s lower flux and higher contamination sensitivity.
Further refinements were made through iterative reductions, improving background modeling and noise corrections, ensuring optimal spectral extraction for scientific analysis.
2.3. Starspot Occultations
Starspot and faculae crossings during exoplanet transits can significantly affect the apparent transit depth (e.g., Pont et al. 2008; Czesla et al. 2009). If uncorrected, starspot occultations reduce the observed transit depth, while faculae occultations increase it. This effect is wavelength-dependent and must be accounted for to ensure accurate transmission spectra. Our observations provide strong evidence of starspot occultations, particularly in the NIRISS transit. Fortunately, transmission spectrum corrections are possible by measuring the intensity ratio and size of the occulted features from transit light curves.
We perform a joint inference of transit parameters and active region properties using the semianalytical spot modeling code SPOTROD (Béky et al. 2014), allowing us to retain affected data. SPOTROD computes transit light curves with arbitrary limb darkening (equally affecting the stellar photosphere and spots) and models homogeneous circular starspots or faculae. The spot model consists of four parameters: the spot-to-star radius ratio (Rspot/R*), the spot-to-unspotted stellar surface intensity ratio (f), and the spot center’s projected coordinates (\u03b8, r2). Intensity ratios below and above 1 correspond to starspots and faculae, respectively. Using nested sampling (Skilling 2004) via MultiNest (Feroz et al. 2009), we perform Bayesian model comparisons between transit models with and without stellar spots, applying uniform priors to ensure comprehensive sampling.
For the NIRISS white light curve, we find a single starspot model is strongly preferred over a spot-free model, with a Bayes factor of ln B = 21.6 (6.9\u03c3 significance). The best-fit parameters yield an intensity ratio of f = 0.9329_{-0.0091}^{+0.0082} and a spot-to-star radius ratio of Rspot/R* = 0.254_{-0.044}^{+0.033}. The NIRSpec white light curve (NRS1 and NRS2 combined) marginally favors a spot model, with ln B = 1.15 (2.1\u03c3 significance), and best-fit parameters of f = 0.82_{-0.39}^{+0.60} and Rspot/R* = 0.113_{-0.077}^{+0.094}. Given these results, we incorporate spot modeling for both observations.
2.4. Light-Curve Analysis
The 1D spectral time series from both observations are analyzed to derive transit depths in three stages. First, we use white light curves to determine wavelength-independent system parameters. Second, we bin light curves to R \u223c 20 and fit wavelength-dependent limb-darkening coefficients (LDCs). Finally, we fix these LDCs and fit transit depths at native resolution to extract the final transmission spectrum.
We model transit light curves with SPOTROD, assuming a circular orbit and using system parameters from Benneke et al. (2019b). We adopt a quadratic limb-darkening law (consistent with previous JWST studies of M-dwarf systems; e.g., Lustig-Yaeger et al. 2023; Moran et al. 2023) and use Kipping’s (2013) parameterization for LDC priors. Baseline flux is modeled with both linear and quadratic trends, finding a preference for a linear trend in NIRISS data and a weak-to-moderate preference for a quadratic trend in NIRSpec data. For white light curves, we use a linear trend for both instruments, as system parameters remain consistent across trend choices. Additionally, we discard the first five minutes of observations due to a settling ramp. Apart from starspot occultations, no other systematics are identified, underscoring JWST’s exceptional data quality.
For each observation’s white light curve, we fit for midtransit time (T0), normalized semimajor axis (a/R*), orbital inclination (i), planet-to-star radius ratio (Rp/R*), quadratic LDCs (u1, u2), baseline flux parameters, and spot parameters. An uncertainty scaling parameter is included to match photometric residual scatter. Noise levels are measured at 1.2\u00d7 and 2.2\u00d7 expected photon and read noise levels for NIRSpec and NIRISS, respectively. MultiNest sampling yields system parameters consistent within 1\u03c3 regardless of trend choice or starspot modeling.
We bin spectroscopic light curves to R \u223c 20 to fit wavelength-dependent LDCs, ensuring a balance between precision and expected variability. Empirical LDCs are used instead of stellar atmospheric models to maximize accuracy (Csizmadia et al. 2013; Espinoza & Jord\u00e1n 2015). Fixing system and spot parameters to white light-curve values, we fit high-resolution transit depths while allowing Rp/R*, f, trend parameters, and uncertainty scaling to vary. NIRSpec light curves are fit at the pixel level, and NIRISS data are binned at two pixels per bin. The Levenberg\u2013Marquardt algorithm is used for high-resolution fitting (Alderson et al. 2023; Moran et al. 2023). The resulting high-resolution transmission spectra contain 1010 and 3401 data points for NIRISS and NIRSpec, respectively, covering 0.9\u20135.2 \u03bcm.
For NIRSpec, we derive two spectra: one with a linear trend and another with a quadratic trend. A quadratic trend is separately fitted for NRS1 and NRS2, fixing its components when fitting spectroscopic light curves (Moran et al. 2023). NRS1 spectra show negligible trend dependence (~10 ppm difference), while NRS2 spectra exhibit a ~60 ppm offset. Given a single transit observation for NIRSpec, an offset parameter is included in the atmospheric retrieval to account for baseline shifts.
3. Atmospheric Retrieval
We retrieve atmospheric properties of K2-18 b’s day-night terminator region using the AURA retrieval code (Pinhas et al. 2018), following methods from previous studies (e.g., Welbanks et al. 2019; Madhusudhan et al. 2020, 2021). The atmosphere is modeled as a plane-parallel structure in hydrostatic equilibrium with uniform chemical composition. Chemical abundances and the pressure-temperature (P-T) profile are free parameters. The retrieval employs a free chemistry approach, treating individual molecular mixing ratios as free parameters. The temperature structure is modeled with a parametric P-T profile (Madhusudhan & Seager 2009). Cloud and haze properties are also included (MacDonald & Madhusudhan 2017; Pinhas et al. 2018).
Molecular opacity contributions from H2O, CH4, NH3, HCN, CO, and CO2 are included based on previous studies (Pinhas et al. 2018; Welbanks et al. 2019). Absorption cross sections are derived from recent datasets (e.g., Polyansky et al. 2018; Hargreaves et al. 2020). Additional molecules associated with biomarkers in habitable exoplanets (e.g., CH3Cl, OCS, N2O, CS2, and DMS) are considered, with absorption cross sections sourced from HITRAN and other databases (e.g., Gordon et al. 2017; Sharpe et al. 2004). Collision-induced absorption from H2-H2 and H2-He is also accounted for (Borysow et al. 1988; Richard et al. 2012).
3.1. Retrieval Setup
Our canonical model includes 22 free parameters: 11 for the individual mixing ratios of chemical species, 6 for the P–T profile, 4 for clouds/hazes, and 1 for the reference pressure (Pref), set at a planetary radius of 2.61 R⊕. Bayesian inference and parameter estimation are conducted using the MultiNest nested sampling algorithm (Feroz et al. 2009) via PyMultiNest (Buchner et al. 2014). The retrieval setup and priors follow recent AURA retrieval framework implementations (Welbanks et al. 2019; Madhusudhan et al. 2020; Constantinou & Madhusudhan 2022), as detailed in Appendix C. Model variations include a cloud/haze-free atmosphere, Mie scattering by hazes, and stellar heterogeneities affecting the spectrum.
We analyze JWST NIRISS and NIRSpec transmission spectra for K2-18 b, covering 0.9–5.2 μm at native resolution. NIRSpec spectra are derived using two baseline flux models: linear and quadratic trends. Due to potential offsets between NIRSpec G395H grating detectors (NRS1 and NRS2) (Moran et al. 2023), retrievals are conducted under different offset scenarios.
Two broad data combinations are considered:
- NIRISS with linear-trend NIRSpec
- NIRISS with quadratic-trend NIRSpec
For each, four offset scenarios are examined:
- No offset
- One combined NIRSpec offset
- One NIRISS offset
- Separate offsets for NRS1 and NRS2
The last case, motivated by a reported transit depth offset (Moran et al. 2023), is the most conservative. Bayesian evidence suggests some offset is preferred. For linear-trend data, a single offset (on either NIRISS or NIRSpec) is favored, with a slight preference for NIRSpec. For quadratic-trend data, two separate offsets for NRS1 and NRS2 are favored, implying a quadratic trend introduces an offset. Based on these findings, we select three nominal cases:
(a) Linear trend with no offsets
(b) Linear trend with one offset
(c) Quadratic trend with two offsets
These cases inform our subsequent atmospheric analyses (Tables 2 and 3).
3.2. Prominent CNO Molecules
Our retrieval identifies CH4 and CO2 as dominant CNO molecules in the H2-rich atmosphere. The spectral fit is shown in Figure 3, and posterior distributions in Figure 4. For the no-offset case, log volume mixing ratios are:
CH4: log(X${CH4}$) = $-2.04{-0.72}^{+0.61}$
CO2: log(X${CO2}$) = $-1.75{-1.03}^{+0.45}$
For the one-offset case:
CH4: log(X${CH4}$) = $-1.74{-0.69}^{+0.59}$
CO2: log(X${CO2}$) = $-2.09{-0.94}^{+0.51}$
These estimates are consistent across retrieval cases, with median abundances around 1% and uncertainties below 1 dex. This marks the first high-precision detection of CH4 and CO2 in a sub-Neptune exoplanet.
H2O and NH3 are not detected, with 95% upper limits of log(X${H2O}$) = -3.21 and log(X${NH3}$) = -4.46. This contrasts with previous HST WFC3 inferences of H2O (Benneke et al. 2019b; Tsiaras et al. 2019), which were potentially affected by CH4-H2O degeneracy (Blain et al. 2021; Bézard et al. 2022). Our results resolve this degeneracy, showing CH4 as the dominant absorber.
While NIRISS spectra align with previous HST WFC3 data (Benneke et al. 2019b) in the 1.1–1.7 μm range, discrepancies arise at the blue end. Multiple CH4 features reinforce its detection (Figure 6). The upper limit on H2O suggests condensation in the upper troposphere, leading to a dry stratosphere.
CO and HCN are also undetected, with 95% upper limits of log(X${CO}$) = -3.00 and log(X${HCN}$) = -2.41 (Table 2). The absence of CO aligns with equilibrium chemistry predictions in H2-rich, low-temperature atmospheres (Moses et al. 2013), though disequilibrium processes may contribute trace CO (Hu et al. 2021; Tsai et al. 2021). The high CO2/CO ratio is consistent with an ocean-covered planet under a thin H2-rich atmosphere (Hu et al. 2021; Madhusudhan et al. 2023).
3.3. Biosignature Molecules
Retrievals constrain two potential biosignatures: dimethyl sulfide (DMS) and methyl chloride (CH3Cl), both predicted to be detectable in Hycean atmospheres (Madhusudhan et al. 2021). DMS abundances vary across retrieval cases:
No-offset case: log(X${DMS}$) = $-4.46{-0.88}^{+0.77}$
One-offset case: log(X${DMS}$) = $-6.35{-3.60}^{+1.59}$
Two-offset case: log(X${DMS}$) = $-6.87{-3.25}^{+1.87}$
Constraints weaken with increasing offsets, as DMS spectral features span multiple detectors (Figure 5). The potential detection of DMS is significant, given its role as a robust biomarker on Earth (Seager et al. 2013b; Catling et al. 2018; Madhusudhan et al. 2021).
CH3Cl exhibits a nominal peak in posterior distributions, more significant than other nondetections (e.g., H2O, NH3). In the no-offset case, log(X${CH3Cl}$) = $-6.62{-3.40}^{+3.08}$, with a 95% upper limit of -2.50. These constraints are consistent across retrieval cases.
CH4, DMS, and CH3Cl exhibit strong spectral features in the 3–3.5 μm range, leading to some degeneracy (Figure 5). However, strong CH4 features in the NIRISS band reduce degeneracies between CH4 and DMS, whereas CH3Cl remains relatively unconstrained.
3.4. Molecular Detection Significance
Beyond the abundance constraints discussed earlier, we assess the detection significance of key molecules using Bayesian model comparisons (Benneke & Seager 2013; Pinhas et al. 2018; Trotta 2008). Specifically, we evaluate the significance of a molecule’s detection by comparing a model that includes the molecule against the same model without it (Benneke & Seager 2013; Pinhas et al. 2018). This comparison is influenced by the chosen model parameters and dataset combinations, as detailed in Section 3.1. Table 2 presents the detection significances for prominent molecules across three retrieval scenarios. We note that the Bayesian evidence obtained via the nested sampling algorithm carries an intrinsic statistical uncertainty of approximately 0.1σ.
Among the major CNO molecules, CH4 exhibits the highest detection significance, ranging from 4.7σ to 5.0σ across all retrieval cases. This robust detection is attributed to multiple CH4 spectral features appearing consistently across the 1–5 μm wavelength range (Figure 3). CO2 is also detected with a confidence of approximately 3σ in all cases, facilitated by its strong spectral feature near 4.3 μm, which falls within the same NIRSpec detector (NRS2) and maintains a consistent spectral baseline. In contrast, we find no significant evidence for NH3, H2O, CO, or HCN.
Regarding potential biomarkers, dimethyl sulfide (DMS) shows some evidence depending on the retrieval scenario. Its detection significance varies with the offsets considered, as its primary spectral feature at 3.3 μm spans multiple detectors: NIRSpec NRS1 for the feature peak, NIRISS at shorter wavelengths, and NIRSpec NRS2 at longer wavelengths. Consequently, the estimated significance and abundance of DMS depend on detector offset assumptions. The no-offset case yields a 2.4σ detection, which reduces to ~1σ for the one-offset case and becomes statistically insignificant for the two-offset case. Despite these variations, all three retrieval cases exhibit notable DMS posterior distribution peaks within 1 dex of each other (Figure 4). Even in offset cases where the posterior distributions include long low-abundance tails due to baseline degeneracies, the results differ from nondetections of other molecules like H2O, NH3, and HCN, suggesting a potential presence of DMS. Further observations will help clarify this finding, as discussed below and in Section 4.
There is also a possible contribution from CH3Cl, though without significant detection confidence, as shown in Figure 4. Since CH3Cl shares spectral features with DMS (Figure 5), its presence marginally increases the detection significance of the DMS+CH3Cl combination to 2.7σ, compared to 2.4σ for DMS alone in the no-offset case. However, we find no significant evidence for CH3Cl independently or for any other biomarkers analyzed.
Overall, our most confident detections are CH4 and CO2, followed by tentative evidence for DMS. These initial insights into the chemical composition of K2-18 b will be further tested by upcoming JWST observations, including transmission spectroscopy with JWST MIRI (5–10 μm, JWST Program GO 2722, PI: N. Madhusudhan) and additional JWST NIRSpec G395H and G235H data (JWST Program GO 2372, PI: R. Hu).
3.5. Clouds, Hazes, and Photospheric Temperature
The observed transmission spectrum provides limited constraints on clouds and hazes in K2-18 b’s atmosphere. Table 3 summarizes these constraints. The inferred cloud-top pressures for gray clouds are generally below the observable photosphere, typically >100 mbar. Although the scattering slope (γ) is poorly constrained, the enhancement factors (a) are generally higher than the expected H2 Rayleigh scattering value (a = 1), albeit still consistent within 3σ uncertainties for the no-offset case. The haze coverage fraction at the planet’s day-night terminator is estimated at ~0.6, with an uncertainty of ~0.2. Bayesian model comparisons indicate a preference for models including clouds and hazes over cloud-free models, with a confidence of 2.8σ–3.2σ across retrieval cases. However, additional optical and near-infrared observations are required for more precise characterization (Appendix B). The molecular abundance estimates remain consistent within 1σ uncertainties between retrievals with and without clouds/hazes.
We also explored alternative haze models, replacing parametric clouds/hazes with Mie scattering hazes based on the methods of Pinhas & Madhusudhan (2017) and Constantinou et al. (2023). Using optical constants from Khare et al. (1984) and He et al. (2023), we found no preference for these models over our canonical cloud/haze model. Despite unconstrained haze properties in these retrievals, the abundance estimates for gas species remain consistent with previous retrieval cases.
Temperature constraints for K2-18 b’s photosphere are relatively weak. The temperature at 10 mbar is estimated between 2355678 K and 25774127 K across the three retrieval scenarios (Table 3). Transmission spectroscopy generally provides weaker constraints on atmospheric temperature structures compared to emission spectroscopy (Madhusudhan et al. 2016). Nevertheless, the retrieved temperature range, coupled with the nondetection of H2O, allows for the possibility of H2O clouds deeper in the atmosphere. The pressure levels probed by spectral features suggest that the planetary photosphere, where optical depth τ = 1, lies between ~0.1 and 100 mbar.
The presence of a hydrogen-rich atmosphere could lead to significant greenhouse warming, potentially raising ocean surface temperatures. However, clouds and hazes play a critical role in cooling the atmosphere and reducing the vertical temperature gradient (Madhusudhan et al. 2020, 2021, 2023; Piette & Madhusudhan 2020). Cloud layers could facilitate temperate ocean surface conditions compared to predictions from cloud-free models (Innes et al. 2023). Future observations will help refine these atmospheric models further.
3.6. Stellar Heterogeneities
We also conducted retrievals to assess the impact of unocculted stellar heterogeneities on the transmission spectrum using our AURA retrieval framework (Pinhas et al. 2018). Across all three retrieval cases, we found no significant evidence for such effects. The retrieved spot covering fraction remained consistent with zero within 2σ uncertainties, and models incorporating stellar heterogeneities were not favored over models without them. Furthermore, the abundance constraints for detected molecules remained largely unaffected by the inclusion of stellar heterogeneities. The nondetection of H2O further supports the absence of unocculted stellar heterogeneities, as cool unocculted starspots could introduce spectral contamination with H2O features, as previously reported (Barclay et al. 2021; Moran et al. 2023).
4. Summary and Discussion
We present a transmission spectrum of the candidate Hycean exoplanet K2-18 b, observed using JWST. The spectrum, obtained with the JWST NIRISS and NIRSpec instruments, spans the 0.9–5.2 μm range and contains absorption features of key CNO molecules and biomarkers predicted for Hycean worlds. We report strong detections of CH4 and CO2 in a H2-rich atmosphere at 5σ and 3σ confidence levels, respectively, with high volume mixing ratios (~1%) for both species. However, we do not detect H2O, NH3, CO, or HCN, and the derived upper limits on their abundances align with expectations for an ocean beneath a cold, thin H2-rich atmosphere (Hu et al. 2021; Madhusudhan et al. 2023). Additionally, we find potential evidence for DMS, a molecule predicted to be a robust biomarker in both terrestrial and Hycean worlds, further supporting the Hycean nature of K2-18 b and the possibility of biological activity.
The observed mass, radius, and equilibrium temperature of K2-18 b suggest a range of possible internal structures (Madhusudhan et al. 2020, 2023), including: (a) a Hycean world with a thin H2-rich atmosphere over a water-rich interior, (b) a mini-Neptune with a deep H2-rich atmosphere, or (c) a predominantly rocky super-Earth with a deep H2-rich atmosphere. Our retrieved atmospheric composition helps distinguish between these scenarios, which we discuss below along with future research directions.
4.1. A Potential Hycean World
K2-18 b was originally proposed as the prototype of a Hycean world (Madhusudhan et al. 2021), featuring habitable oceans beneath a H2-rich atmosphere. Our retrieved atmospheric composition aligns with theoretical predictions for such a scenario (Hu et al. 2021; Madhusudhan et al. 2023). Specifically, Hu et al. (2021) predicted CO2 abundances between 4 × 10−4 and 10−1 and CH4 abundances between 1.5 × 10−2 and 5.3 × 10−2, values consistent with our findings. Additionally, their models anticipated lower abundances of CO, NH3, and stratospheric H2O, which aligns with our nondetections. A low gas-phase H2O mixing ratio at pressures below ~100 mbar is consistent with condensation due to a tropospheric cold trap (Madhusudhan et al. 2023), similar to Earth’s stratosphere. This suggests that while H2O could be abundant at deeper atmospheric levels, the observed transmission spectrum does not probe these depths.
A primary challenge to K2-18 b’s classification as a Hycean world stems from its climate conditions. In the absence of substantial cloud or haze cover, a greenhouse effect within a thick H2-rich atmosphere would likely elevate temperatures at pressures exceeding ~10 bar, potentially driving an ocean into a steam-dominated phase and ultimately leading to a supercritical state at depth (Piette & Madhusudhan 2020; Scheucher et al. 2020; Innes et al. 2023; Pierrehumbert 2023). For an ocean to exist, its surface pressure must remain below ~10 bar. However, an excessively thin H2 atmosphere may be vulnerable to atmospheric escape over time (Kubyshkina et al. 2018a, 2018b; Hu et al. 2023), suggesting a narrow parameter space for sustaining a habitable ocean without substantial cloud or haze cover. High-albedo tropospheric clouds or scattering hazes could mitigate the greenhouse effect by reducing the planet’s absorbed stellar energy (Piette & Madhusudhan 2020; Madhusudhan et al. 2021).
Our retrieved atmospheric temperatures and upper limits on H2O support the possibility that H2O is condensing into clouds beneath the photosphere, indicating a cold upper troposphere (Benneke et al. 2019b; Madhusudhan et al. 2023). As discussed in Section 3.5, our retrievals provide some evidence for scattering hazes at the day–night terminator, although additional optical observations are needed for confirmation. If clouds or hazes are widespread, particularly on the dayside, they could increase planetary albedo, potentially allowing for habitable conditions in a surface ocean. However, given that clouds can also contribute to greenhouse warming, maintaining a sufficiently low surface temperature would likely still require a relatively shallow H2 atmosphere.
4.2. Is a Deep Atmosphere a Possibility?
Our analysis of K2-18 b’s atmospheric composition, combined with its level of irradiation, suggests that the planet does not conform to the characteristics of a mini-Neptune with a deep atmosphere. In such a scenario, photochemically produced carbon and nitrogen species would be recycled into their thermodynamically stable forms, CH4 and NH3, in the deep atmosphere before being transported back to the observable upper layers (Tsai et al. 2021; Yu et al. 2021; Madhusudhan et al. 2023). While CO2 can reach significant concentrations in a high-metallicity, low-C/O H2-rich atmosphere (Moses et al. 2013), our findings indicate that maintaining both ~1% CO2 and CH4 requires a high C/H metallicity, an extremely low C/O ratio (~0.02), and efficient vertical mixing. For instance, a combination of C/H = 30× solar, O/H = 690× solar, and Kzz ≥ 10^7 cm^2 s^−1, alongside a temperature profile with Tint = 60 K, yields CH4 and CO2 mixing ratios of the expected magnitude. However, this scenario raises questions about the origin of such an oxygen-enriched atmosphere and implies a high H2O abundance below a few hundred mbar, leading to an increased mean molecular weight. Additionally, NH3 and CO would be present in significant quantities, which contradicts our observations.
Similarly, our data do not support the alternative scenario of a shallow, H2-rich atmosphere overlaying a solid surface at a few-bar pressure level. While CH4 and CO2 abundances align with such models (Tsai et al. 2021; Yu et al. 2021; Madhusudhan et al. 2023), K2-18 b’s bulk density is inconsistent with a thin H2 atmosphere atop a silicate mantle (Madhusudhan et al. 2020). Even a purely silicate interior would necessitate a thick (≥10^3 bar) H2-rich envelope to explain the planet’s observed mass and radius. Although high-pressure interactions between a thick H2 atmosphere and a deep silicate mantle might generate atmospheric CO2 (Kite et al. 2020; Kite & Barnett 2020; Schlichting & Young 2022; Tian & Heng 2023), such models generally do not predict a 1% CO2 mixing ratio or the lack of N2 recycling to NH3. Ultimately, the planet’s bulk density and atmospheric composition strongly support the classification of K2-18 b as a Hycean world rather than a mini-Neptune or a rocky planet with a thin or thick H2 atmosphere.
A caveat to this discussion is that existing photochemical models for mini-Neptune atmospheres assume ideal-gas behavior and do not account for potential chemical interactions between a primordial H2 atmosphere and a supercritical water layer or silicate magma at depth. Additionally, our understanding of super-Earth and mini-Neptune atmospheric chemistry remains incomplete, and alternative explanations for the depletion of NH3 and CO may yet emerge. Furthermore, our retrieved molecular abundances represent averages in the observable photosphere (~0.1–100 mbar) at the day-night terminator. While CH4 and CO2 are expected to be relatively uniform in this pressure range (Hu et al. 2021; Yu et al. 2021; Madhusudhan et al. 2023), future high-precision observations may reveal nonuniform distributions.
4.3. Possible Evidence of Life
The potential detection of dimethyl sulfide (DMS) on K2-18 b raises intriguing questions about biological activity on the planet. While the current evidence for DMS is not as strong as that for CH4 or CO2, upcoming JWST observations will provide further constraints on its presence and abundance (Madhusudhan et al. 2021).
On Earth, DMS is predominantly produced by marine phytoplankton and is considered a potential biosignature due to its exclusive biological origin (Charlson et al. 1987; Barnes et al. 2006; Catling et al. 2018). Though less abundant than other biosignatures like O2, CH4, and N2O, DMS and CH3Cl may be significant in H2-rich environments with large biomass (Seager et al. 2013b; Madhusudhan et al. 2021). Our inferred DMS abundance spans a broad range, from log(XDMS) = -4.46 (+0.77, -0.88) in the no-offset case to log(XDMS) = -6.87 (+1.87, -3.25) in the two-offset case. Earth’s ocean surface DMS mixing ratios typically reach a few hundred parts per trillion (Hopkins et al. 2023), with rapid photochemical depletion at higher altitudes due to reactions with OH and other radicals, leading to SO2 production. While the upper end of our inferred DMS abundance is much higher than Earth’s, the lower end remains plausible (Seager et al. 2013b). However, limited infrared absorption cross-section data for DMS (Sharpe et al. 2004; Gordon et al. 2017; Kochanov et al. 2019) could impact our estimates, and future observations may reveal additional sulfur-bearing species.
DMS survival also depends on stellar activity. A quiescent M-dwarf, with lower UV flux than a Sun-like star, could extend DMS’s atmospheric lifetime (Domagal-Goldman et al. 2011). Predictions for Earth-like planets and super-Earths orbiting low-activity M dwarfs suggest DMS mixing ratios of ~10^−7 to 10^−6 (Domagal-Goldman et al. 2011; Seager et al. 2013b). However, K2-18 b’s host star is a moderately active M3 dwarf (Benneke et al. 2019a) and may have high extreme-UV flux (dos Santos et al. 2020), potentially depleting DMS via interactions with atomic oxygen from CO2 photolysis (Seager et al. 2013b).
If future observations confirm a DMS abundance above ~10^−6, it would suggest high biological production rates or require new theoretical insights into DMS chemistry, including possible abiotic pathways. Additionally, our retrievals considered a range of molecules with strong spectral signatures, and expanded future searches may reveal other relevant species. Besides DMS, the detected chemical disequilibrium in K2-18 b’s atmosphere may hint at biotic CH4 production, similar to Earth’s methanogenic bacteria.
Our findings demonstrate that JWST can detect biosignature molecules in the atmosphere of a habitable-zone sub-Neptune, providing a framework for biosignature assessment (Catling et al. 2018; Meadows et al. 2022). If verified, the potential presence of DMS on K2-18 b could represent a breakthrough in the search for extraterrestrial life.
4.4. Resolving the Missing Methane Problem
Our 5σ detection of CH4 resolves the long-standing “Missing Methane Problem” in exoplanetary science (Stevenson et al. 2010; Madhusudhan & Seager 2011). CH4 and NH3, common in solar system giant planets, are expected in H2-rich atmospheres below ~600 K (Burrows & Sharp 1999; Lodders & Fegley 2002). However, pre-JWST observations failed to detect CH4 in exoplanets below ~800 K (Stevenson et al. 2010; Knutson et al. 2014; Benneke et al. 2019a, 2019b).
Disequilibrium processes like photochemistry, vertical mixing, and outgassing affect sub-Neptune atmospheres (Yung & DeMore 1999), but none adequately explain the missing CH4 in previous studies (Line et al. 2011; Moses et al. 2013). Our CH4 detection on K2-18 b confirms JWST’s ability to characterize temperate exoplanet atmospheres, suggesting that other sub-Neptunes and gas giants in this temperature range may also host detectable CH4, enabling comparative studies of carbon chemistry across planetary systems.
4.5. Future Directions
Our findings highlight the potential of candidate Hycean worlds as prime targets in the search for life beyond Earth. These results encourage further observational and theoretical efforts to better understand the atmospheric and possible surface conditions of K2-18 b and other similar exoplanets (Madhusudhan et al. 2021). Upcoming JWST observations, particularly those using NIRSpec G395H (JWST GO 2372) and MIRI LRS (5–10 μm) (JWST GO 2722), will help validate our findings. While NIRSpec G395H will refine current measurements with higher precision, MIRI LRS will be instrumental in confirming the presence of dimethyl sulfide (DMS), which exhibits a distinct spectral feature around 7 μm (e.g., Figure 7 of Seager et al. 2013b). Similar observations are warranted for other promising Hycean candidates orbiting nearby M-dwarf stars, some of which may be even more suitable for study than K2-18 b (Madhusudhan et al. 2021).
Overall, our results mark the beginning of a new era in atmospheric characterization of potentially habitable exoplanets and the search for biosignatures with JWST. They also inspire further theoretical investigations into the physical, chemical, and biological processes governing Hycean worlds. This study represents an initial step toward spectroscopic identification of extraterrestrial life and a deeper understanding of our place in the cosmos.
Acknowledgments
This research is based on observations from the NASA/ESA/CSA James Webb Space Telescope, conducted as part of Cycle 1 GO Program 2722 (PI: N. Madhusudhan). We extend our gratitude to NASA, ESA, CSA, STScI, and the broader exoplanet science community for their contributions to the JWST mission. This work was supported by research grants awarded to N.M. by UK Research and Innovation (UKRI) Frontier Grant (EP/X025179/1), the MERAC Foundation (Switzerland), and the UK Science and Technology Facilities Council (STFC) Center for Doctoral Training in Data Intensive Science at the University of Cambridge (STFC grant No. ST/P006787/1).
N.M. and M.H. acknowledge support from STFC and the MERAC Foundation for M.H.’s doctoral studies. N.M. also thanks Tony Roman, Elena Manjavacas, Nestor Espinoza, and Sara Kendrew at STScI for their assistance in planning the JWST observations. J.M. acknowledges funding from JWST-GO-02722, provided by NASA through a grant from the Space Telescope Science Institute (STScI), which operates under NASA contract NAS 5-03127. We appreciate the valuable feedback from the anonymous reviewers.
This research utilized the NASA Exoplanet Archive, operated by the California Institute of Technology under NASA’s Exoplanet Exploration Program, as well as the NASA Astrophysics Data System. The analysis was conducted using Python libraries NUMPY, SCIPY, and MATPLOTLIB.
Computational resources were provided by the Cambridge Service for Data Driven Discovery at the University of Cambridge (www.csd3.cam.ac.uk), supported by Dell EMC and Intel through Engineering and Physical Sciences Research Council (EPSRC) Tier-2 funding (capital grant EP/P020259/1) and DiRAC funding from STFC (www.dirac.ac.uk).
Author Contributions
N.M. conceived, planned, and led the project. The JWST proposal was led by N.M. with contributions from S.C., S.S., A.P., and J.M. N.M. and S.S. designed the JWST observations. Data reduction and analysis were carried out by N.M., M.H., and S.S., while atmospheric retrievals were performed by N.M. and S.C. Theoretical interpretation was conducted by N.M., J.M., and A.P. N.M. authored the manuscript, with input and revisions from all co-authors.
Facility
JWST (NIRISS and NIRSpec)
Data Availability
The data analyzed in this study were obtained from the Mikulski Archive for Space Telescopes (MAST) at the Space Telescope Science Institute and can be accessed via DOI: 10.17909/3ds1-8z15. The transmission spectra of K2-18 b reported here are available on the Open Science Framework at DOI: 10.17605/OSF.IO/36DJH.
Appendix A: Comparison with HST WFC3 Observations
K2-18 b was previously observed in the 1.1–1.7 μm range using HST WFC3 (Benneke et al. 2019b; Tsiaras et al. 2019). These observations indicated the presence of H₂O in the planet’s terminator atmosphere (Benneke et al. 2019b; Tsiaras et al. 2019; Madhusudhan et al. 2020), though they also placed a high upper limit on CH₄ (Madhusudhan et al. 2020). The spectrum has been interpreted as either CH₄-dominated or as a mix of CH₄ and H₂O due to degeneracy between these molecules in the WFC3 band (Blain et al. 2021).
Figure 6 overlays the previous HST WFC3 observations (Benneke et al. 2019b) with our new JWST spectrum. Overall, the new NIRISS data align well with the prior WFC3 measurements, except for two data points at the blue end of the WFC3 band, which show a 2σ–3σ deviation from the NIRISS observations and the corresponding spectral fit. These discrepancies suggest that earlier interpretations favoring H₂O over CH₄ may have been influenced by these two points, as they are inconsistent with the CH₄ absorption peak in our retrieved spectral fit.
