Main

The disk around TWA 7 is one of the youngest (6.4 ± 1 Myr old; ref. 12) debris disks known to date. TWA 7 is a close (about 34 pc; ref. 13), low-mass (0.46 M☉; ref. 14) member of the young TW Hydra association, sometimes classified as a weak-line, non-accreting T-Tauri star15. The disk, resolved by the Near Infrared Camera and Multi-Object Spectrometer (NICMOS) on the Hubble Space Telescope (HST)16, is one of the rare ones resolved around M stars. It is seen almost pole-on17,18, a most favourable configuration to precisely estimate its radial distribution and to look for planets. The most recent modelling of the disk surface density deduced from polarimetric data from the Spectro-Polarimetic High Contrast Imager for Exoplanets Research (SPHERE) on the Very Large Telescope (VLT)18 includes a ring peaking at 28 au and extending out to more than 100 au (R1), a narrow (less than 7 au full-width at half-maximum) ring at 52 au (R2) and a broader (more than 40 au full-width at half-maximum) structure (93 au; R3; Figs. 1 and 4 from ref. 18). No planet has been detected so far, with detection limits roughly estimated to 0.5–1 MJ beyond 50 au (Extended Data Fig. 6 and Supplementary Information).

The coronagraphic images of TWA 7 taken with the F1140C filter (central wavelength = 11.3 μm, bandwidth = 0.8 μm) of the Mid-Infrared Instrument (MIRI) on the James Webb Space Telescope (JWST)19 were obtained on 21 June 2024 during cycle 2 (ID 3662; principal investigator, A.-M.L.; Extended Data Table 1). The details on the data reduction procedure are described in the Methods. The main critical step is the subtraction of the residual diffracted light leaking from the MIRI coronagraph using a reference star observed with the same set-up. This process is necessary to bring the contrast ratio with respect to the star to a level of 10−5–10−4 beyond angular separations of about 0.5 arcsec. The final image, presented in Fig. 1, reveals three sources within 10 arcsec from TWA 7, the properties of which are listed in Extended Data Table 2. One source at about 4.7 arcsec from TWA 7 (position angle 107°) was classified as a stellar background source, already detected in ancillary optical data from the Space Telescope Imaging Spectrograph (STIS) on the HST as well as near-infrared data from the NICMOS on the HST and the SPHERE on the VLT. The second one, located about 6.7 arcsec east of TWA 7, and spatially resolved, has no counterpart in the data from the SPHERE on the VLT, or in the data from the NICMOS or the STIS on the HST (taking into account TWA 7 proper motion (−118.751 ± 0.023 mas yr−1, −19.648 ± 0.026 mas yr−1; ref. 20)). Its location in the MIRI image is consistent with that of a bright source in Atacama Large Millimeter Array (ALMA) band 7 (346 GHz) data from 2016 (ref. 21), given the proper motion of the TWA 7 system between the ALMA and MIRI observations. This object therefore has the characteristics of a highly reddened background source. It is reminiscent of the JWST observations of the HR 8799 multi-planetary system, for which a z ≈ 1 galaxy was detected in the MIRI data taken at both 10 and 15 μm, and in ALMA band 7 data as well, but never identified in the near-infrared22. The third source, located about 1.5 arcsec northwest of TWA 7 (about 51 au, projected separation) is unique to these MIRI observations. Hence, this source (hereafter, CC#1) is extremely red, and is not compatible with any background or foreground star.

Fig. 1: JWST MIRI image of TWA 7 in the F1140C filter.
Fig. 1: JWST MIRI image of TWA 7 in the F1140C filter.
Full size image

North is up, and east is left. The status of three identified sources is indicated. Note that the faint signal north of the background galaxy is an artefact. bgd, background.

No data are available to test whether this third source shares a common proper motion with the star. In this context, in the following, we discuss the possible nature of this object. The first origin that one can consider is a Solar System object. Yet, most Solar System objects have proper motion between 5 and 40 arcsec h−1 (ref. 23). Even very remote, low-proper-motion small Solar System objects such as the dwarf planets Eris (semi-major axis of about 68 au) and Sedna (semi-major axis of about 510 au) showed proper motion of 1.4 arcsec h−1 (ref. 24) and 1.7  arcsec h−1 (ref. 25), respectively, at the time of their discovery. No trail was observed during the 2-h-long exposures, and no apparent motion was observed between the two images recorded during the sequences taken 2 h apart, indicating that CC#1 has a proper motion of less of than about 0.05  arcsec h−1. A Solar System object with such a low proper motion would be located at more than 200 au. For such a cold object, reflected light would dominate in the MIRI F1140C filter and would require a Neptune-like size to fit the flux measured for CC#1 (for a geometric albedo of 0.1–0.3). It was checked whether such a scenario could be compatible with the hypothetical Planet Nine26. On the basis of the constraints from the planetary ephemeris27 and the predictions of the orbit of Planet Nine28, a Solar System origin for CC#1 can definitely be excluded.

The second possible origin is a background galaxy. Like the background galaxy seen east of TWA 7, CC#1 has no reported counterpart at optical and near-infrared wavelengths. However, in contrast to this galaxy, it has no detected counterpart in the ALMA band 7 data (see details in Supplementary Information). The detection of this unresolved source at 11.3 μm, its non-detection in ALMA band 7 and the measured upper limits at 1.6 μm in the data from the SPHERE on VLT (Supplementary Information) could still be compatible with intermediate-redshift star-forming galaxies. Using such galaxy templates at various redshifts and published galaxy counts in JWST fields of view, we estimated that the probability of finding one such galaxy in a region of 1.5 arcsec radius centred on TWA 7 is about 0.34% (Methods). This probability is low, albeit non-zero. The location of the source with respect to the disk structure, right in a circumstellar ring gap (see below), makes the galaxy hypothesis even more unlikely.

The third and last possible origin is a planet. A forward modelling approach is used to constrain the properties of this planet. Using the HADES model29, it is possible to find fits for the JWST photometric data point while accounting for the 5σ upper limits provided by the high-contrast images at 1.59 μm (H2) and 1.67 μm (H3) from the SPHERE on the VLT (Fig. 2). HADES considers the thermal evolution of the planet; it assumes that the planet and the stars are coeval. Atmospheric fits incorporating water clouds indicate a narrow range, regardless of other parameters, for the effective temperature between 305 and 335 K (Extended Data Fig. 3 and Extended Data Table 3), and a mass of about 0.3 MJ. A metallicity range above solar is required. Additional data will be necessary to further constrain this parameter.

Fig. 2: Fitting of the candidate companion’s available data using the HADES model.
Fig. 2: Fitting of the candidate companion’s available data using the HADES model.
Full size image

Modelled photometry is shown with crosses. The 1–15-μm spectra correspond to representative solutions that fit the observed flux of TWA 7b in the JWST F1140C filter (blue), with respect to the 5σ upper limits from the VLT SPHERE data in the H2 and H3 filters (red), and are consistent with an age of 6.4 ± 1 Myr. The bandwidths of H2 and H3 are 0.052 μm and 0.054 μm, respectively. A zoomed-in view around H2 and H3 highlights their bandwidths and the integrated model spectrum points below the upper limit. met, metallicity; Teff, effective temperature; fsed, sedimentation rate.

As a comparison and extra mass estimation, evolutionary models of cold, low-mass planets30 combined with our estimation of effective temperature are used; they lead to a comparable mass of close to 0.3 MJ (Extended Data Fig. 4) for metallicities less than 2.5, as available in their framework. These two consistent results indicate that the planet mass is substantially below 1 MJ. The current best estimate is around 0.3 MJ, and is only weakly dependent on the underlying details of the two models used. It, however, depends on the age of the planet, assumed here to be coeval with its parent star. A younger planet would lead to a smaller mass.

The observed source is located right on the R2 narrow ring, and, moreover, within a region identified by ref. 18 as underdense compared to the rest of the ring (Fig. 3a). This is very reminiscent of simulations of resonant rings predicted by early works31,32 for closer and less massive planets, which led to the proposal of such a possible situation for the TWA 7 system18.

Fig. 3: Image of the TWA disk and candidate companion and simulations.
Fig. 3: Image of the TWA disk and candidate companion and simulations.
Full size image

a, Polarimetric image (in log scale) of the disk composed of the sum of three epochs (26 April 2016 presented in ref. 17, 20 March 2017 presented in ref. 18, and a new epoch, 8 February 2022, reduced as in ref. 18) from the SPHERE Infrared Dual-Band Imager and Spectrograph (IRDIS), with the MIRI image (resampled to the SPHERE pixel size) as an overlay with contours. The log of these data is provided in Supplementary Table 2. The peak densities of the rings are also indicated. The central hatched disk is a numerical mask to hide the stellar residuals. b, Disk simulations. Top view of a disk of massless planetesimals perturbed by a 0.34-MJ planet at 52 au, on a circular orbit, after 6 Myr (see details in text). The orbit of the perturbing planet is sketched in green and the location of the planet on its orbit is shown in red.

Dedicated N-body simulations were conducted for a planet with a mass of 0.34 MJ, located at 52 au around the 0.46 M☉ central star. This value is consistent with the measured projected separation, assuming that the planet and the ≈13°-inclined disk are coplanar. The simulation also included a disk of 200,000 planetesimals, distributed between 20 and 130 au. These parameters were selected to roughly match the boundaries of the observed disk. The planetesimal disk was assumed to be coplanar with the planet’s orbit, with an initial surface density proportional to 1/r. The eccentricities of the planetesimals were chosen to range between 0 and 0.01, and the planet was assumed to evolve on a circular orbit. The simulations, which spanned 10 Myr, were performed using the symplectic N-body code swift_rmv3 (ref. 33), which provides a first-order treatment of close encounters.

A top view of the resulting distribution of planetesimals after 6 Myr is shown in Fig. 3b. Planetary perturbations efficiently carve the disk over about 30 au, but leave a narrow ring at 52 au, as well as a relative void (underdensity) around the planet. The latter structure is characteristic of a ring of co-orbital planetesimals with the planets, trapped in a 1:1 resonance with it. The similarity between the TWA 7 disk image and the simulation (Fig. 3a,b) is remarkable. In this context, the observed R2 ring would correspond to the ring of co-orbiters with the planet. We nevertheless expect some differences between both distributions owing to the radiation pressure acting on the grains. Additional information on the grain size distribution is needed to compute the effects of radiation pressure and refine the dynamical modelling.

The low likelihood of a background galaxy, the successful fit of the MIRI flux and SPHERE upper limits by a 0.3-MJ planet spectrum and the fact that an approximately 0.3-MJ planet at the observed position would naturally explain the structure of the R2 ring, its underdensity at the planet’s position and the gaps provide compelling evidence supporting a planetary origin for the observed source. Like the planet β Pictoris b, which is responsible for an inner warp in a well-resolved—from optical to millimetre wavelengths—debris disk34, TWA 7b is very well suited for further detailed dynamical modelling of disk–planet interactions. To do so, deep disk images at short and millimetre wavelengths are needed to constrain the disk properties (grain sizes and so on). Refining the planet mass determination can be carried out with additional JWST photometry and possibly spectroscopy. Measuring the orbital parameters (eccentricity, in particular) is more challenging given the long orbital period (about 550 yr) of the planet. Yet, one notes that a planet on an eccentric orbit would rapidly destroy R2.

As it is angularly well resolved from the star, TWA 7b is suited for direct spectroscopic investigations, providing the opportunity to study the interior and the atmosphere of a non-irradiated sub-Jupiter-mass, cold (about 320 K) exoplanet, and start comparative studies with our much older and cooler Solar System giants, as well as with the recently imaged cold (about 275 K) but more massive (6 MJ) planet eps Ind Ab35. Improved estimations of its metallicity and temperature will further constrain its mass.

The present results show that the JWST MIRI has opened up a new window in the study of sub-Jupiter-mass planets using direct imaging. Indeed, TWA 7b (about 100 M⊕) is at least ten times lighter than the exoplanets directly imaged so far, and planets as light as 25–30 M⊕ could have been detected if present at 1.5 arcsec from the star or beyond.

Methods

JWST observations and data reduction and analysis

Data

Coronagraphic observations were performed with the 4QPM_1140 coronagraph paired with the F1140C filter. The details of the observations are given in Extended Data Table 1. We obtained two roll angles (difference of 7.835°) to mitigate the attenuation of the coronagraph in the field of view, in case an object falls close to one phase transition of the 4QPM. Each coronagraphic observation was 2 h long, hence a total of 4 h on the science target. Background observations were observed immediately after the science exposures in a two-point dithering mode, for a total of 4 h.

A reference star, CD-23-9765, was observed back-to-back with the target in the same configuration with the aim to subtract the starlight diffraction after the coronagraph. The reference shares similar brightness and spectral type with the target, and is angularly close. It was observed with nine-point small-grid dithering (SGD) to apply post-processing algorithms such as principal component analysis (PCA)36. In total, the reference star was observed for about 1 h and comes with dedicated background observations.

Using comparison with simulated coronagraphic images, as in ref. 37, we were able to estimate a pointing accuracy on the 4QPM of about 2 mas per axis, significantly lower than the 10-mas step of the SGD. We also confirmed the detector coordinates of the 4QPM_1140 mask (119.758, 112.158 as provided in the JWST Calibration Reference Data System).

Data analysis

The data reduction follows the steps described in refs. 21,38. Level 1 data are retrieved from the Mikulski Archive for Space Telescopes (MAST), processed with v1.14.0 of the pipeline together with Calibration Reference Data System file 1241. Images are registered to the coronagraph centre. Calibrated files (‘cal’ files) are produced in-house with the JWST pipeline for each roll, by subtracting the background and converting the photometric units (data numbers per second to mega-janskys per steradian). The background is built from the minimum per pixel of the two dithers. We skipped the flat-field correction, which is not appropriate for the MIRI coronagraph22.

We took advantage of the diversity brought by the SGD mode to build a reference frame to subtract the stellar diffraction. We tested various algorithms and retained a linear combination (which uses the downhill simplex minimization) of the nine SGD reference star images, as well as the PCA, as the two algorithms providing the best detection of CC#1. To mitigate the over-subtraction effect, a numerical masking is implemented to ignore some parts of the image. We obtained the best compromise by selecting the annular region between 0.5 arcsec and 3 arcsec, and the three sources were masked with a 1 λ/D patch, with λ representing the observing wavelength and D the telescope diameter (we checked that the CC#1 flux measurements are similar with a different region: 2–3 arcsec). We proceeded similarly for the PCA, using eight components to build the final image that is subtracted to the data, in an annular region from 0.5 arcsec to 5 arcsec (point sources not masked). Despite the bad-pixel correction applied on the raw coronagraphic images, the subtracted images with the reference star are still affected by a few bad pixels, both with the linear combination and PCA. We further apply a σ-clipping function to correct for these remaining bad pixels. Finally, the images are rotated to align north up, considering the aperture position angles: 121.45° for roll 1 and 129.27° for roll 2, as well as the V3 axis orientation on the detector (4.835°).

Next, extracting the flux and position of the CC#1 requires modelling its point spread function (PSF), which can vary spatially and nonlinearly owing to the attenuation of the coronagraph for which the phase transitions extend across the whole field of view. We used both the diffraction code developed in refs. 39,40 and WebbPSF41 to simulate the MIRI PSF, taking into account the coronagraph, considering the configuration of mask, stop and filter (‘FQPM1140’, ‘MASKFQPM’ and ‘F1140C’, according to the WebbPSF terminology). The position of CC#1 is approximated with a Gaussian fit, passing the sky coordinates to the former diffraction code and detector coordinates to WebbPSF to calculate the PSF of CC#1 accounting for the coronagraph attenuation. We measured the coronagraph transmission with both PSF estimates. The flux of the PSF model at the position of CC#1 is integrated in a 1.5 λ/D aperture (to match the aperture used for photometric measurements), and compared to that at 10 arcsec (far away from the coronagraph influence). The two approaches give similar transmissions: 0.66 and 0.62 in roll 1 and 0.31 and 0.28 in roll 2. Therefore, CC#1 is significantly closer to one 4QPM transition in roll 2 than in roll 1, so both its astrometry and photometry can be affected (Extended Data Fig. 1). We measured a signal-to-noise ratio of 30 and 18, respectively, for roll 1 and roll 2, using the linear combination method. In the roll 2 image, the PSF is more asymmetrical as the planet is closer to the quadrant edge in comparison to roll 1, so we decided to consider only roll 1 data for the photometric analysis.

On the basis of these CC#1 PSF models, we extracted the flux and the photometry of the object by minimizing the residuals between the reduced JWST data and a PSF model in a 1.5 λ/D area with three free parameters (positions and flux) and using either a downhill simplex algorithm or the Nelder–Mead algorithm42. For comparison, we also used aperture photometry, but this required implementation of an aperture correction based on simulated PSF (ratio of the total flux in the PSF to the flux integrated in the 1.5 λ/D aperture).

The flux extraction is applied both on the photometrically calibrated files (.cal), which directly provides CC#1 flux in mega-janskys per steradian, and on the uncalibrated files (.rate), which requires to measure a contrast with respect to the non-coronagraphic image of the star. As detailed previously22, the contrast measurement relies on commissioning data either on target acquisition images that come with the telescope pointing procedure but are obtained with a neutral density filter, or from images obtained on and off the coronagraph on another star. The method using target acquisition shows some net discrepancies, probably because the target acquisition filter is very broad (about 8–18 μm) and the targets have different spectral types (M3 for TWA 7 and K0 or K5 for the commissioning targets). Besides, the emission of the TWA 7 disk is expected to become significant beyond 15 μm. As a result, we did not use the target acquisition method in the following. The final photometric values are based on the linear combination and PCA technique to suppress the starlight. We averaged the values of the different methods (calibrated files and contrast, aperture and PSF model for the photometric extraction) and the error bar is built from the extreme values ((max − min)/2) for being conservative. We measured a flux density of 5.60 ± 0.97 × 10−19 W m−2 μm−1 for CC#1. The fluxes of the other sources are given in Extended Data Table 2. Note that we also tested the typical ‘injection–recovery’ method directly in the raw data, but did not notice any significant differences for the extracted flux of the planet.

Estimation of the probability for an intermediate-redshift star-forming galaxy

To estimate the probability that the source labelled CC#1 is a galaxy, we have taken into account the three constraints on the fluxes or upper limits at 1.6, 11 and 870 μm. The source has a flux of 22 μJy at 11 μm, and is not detected with ALMA at 870 μm, but with a tapered resolution of 2 arcsec. The measured 3σ upper limit for an unresolved source at the position of the MIRI source is 96 μJy (ref. 21). Combining all ALMA observations (see Extended Data Fig. 5), the 3σ upper limit is 76 μJy, in a beam of 0.29 arcsec × 0.24 arcsec (see a detailed analysis and estimation in the Supplementary Information). The third constraint is from the non-detection at 1.6 μm with VLT SPHERE, with an upper limit of approximately 0.6 μJy for a point source (about 60 mas in size). However, we have to take into account the fact that a low-z galaxy could be extended in the calculation of its maximum flux at 2 μm. Indeed, the size of a galaxy is expected to vary from one wavelength to the other: at 1.6 μm, the disk of old stars dominates, so the source is more extended, of the order of 10 kpc, whereas at 11 μm and 870 μm, the nuclear star-forming region dominates, and will appear more concentrated, of the order of 100–500 pc. We therefore considered the flux limit of an extended (0.6 arcsec) source in the SPHERE data (that is, 60 μJy).

Two types of galaxy might comply with these three constraints: star-forming galaxies or active nucleus (AGN) or a combination of both, with a redshift between z = 0.1 and 1. For such galaxies, at lower redshifts, where 1 arcsec is smaller than 2 kpc, the source would probably look extended (up to 5 arcsec) for wavelengths between 1 and 4 μm (ref. 43). At higher redshifts, the peak of the emission usually located at 100 μm will enter the ALMA domain at about 1 mm, and it should have been detected by ALMA44.

As the density of the cosmic star formation rate increases considerably with redshift between z = 0 and 1, starbursts at z = 0 are good templates for star-forming galaxies at z = 0.1 to 1. The starburst is in general nuclear. It can be highly peaked in the centre, or distributed in a ring of about 100-pc radius (like in M82), but the emitting size at 100-μm rest frame will be no more than typically 1 kpc. This corresponds to an angular size of about 0.55 arcsec at z = 0.1, about 0.16 arcsec at z = 0.5, and about 0.12 arcsec at z = 1. We also considered AGN templates. The nucleus is then a point source at long wavelengths, while the galaxy host is extended (about 10 kpc) at the shortest wavelengths.

The probability to find a z = 0.1 to 1 star-forming galaxy with a flux at 11 μm of 22 ± 3.8 μJy (hence, within a range of 7.6 μJy) in a field of view of 10 × 10 arcsec can be estimated readily through the source counts, from refs. 45,46,47,48. A probability of 50% is found in the redshift range z = 0.1 to 1.

To take into account the other constraints at 1.6 and 870 µm, we used the Spitzer Wide-Area Infrared Extragalactic Survey (SWIRE) templates, from ref. 49, http://www.iasf-milano.inaf.it/~polletta/templates/swire_templates.html which include 14 templates of starbursts and AGN, representative of star-forming and active intermediate-redshift galaxies. These templates were computed for a half-dozen redshifts, and calibrated to have a flux of 22 μJy at 11 μm. An illustrative plot of their spectral energy distribution (SED) is shown in Extended Data Fig. 2, in which the flux constraints at 1.6 and 870 µm are also indicated by black symbols (square and triangle).

We used the rest-frame 8-μm luminosity function of galaxies at redshifts between z = 0.1 and z = 1 in the Great Observatories Origins Deep Survey (GOODS) fields50 to estimate the abundance of these star-forming galaxies. The 8-μm luminosity was deduced from the templates. This led to a probability of 12% to find such a galaxy in the right redshift range (z = 0.1 to 1) in the 10 × 10 arcsec field of view. We estimated that there are about 80% star-forming galaxies and 20% active nuclei, including low-luminosity ones such as Seyfert galaxies51,52. As can be seen in Extended Data Fig. 2, all templates comply with the three constraints below z = 0.5, and some starbursts are found brighter at 870 μm above this redshift. We therefore reduced the probability accordingly for z > 0.5, attributing each star-forming template equal probability, as shown by the observations53. The final probability to find such a galaxy in the 10 × 10 arcsec field of view is about 5%, with significant uncertainties: +3/−2%. Hence, in a radius of 1.5 arcsec around TWA 7, the probability to have such a galaxy is about 0.34% (+0.22/−0.14%). We note that the distribution of galaxies on the sky might be clustered in some rare places, and our error bars should be increased, but no more than 30% given the typical errors on luminosity function of galaxies45,47,50. Altogether, the probability of having a galaxy satisfying the various constraints within 1.5 arcsec would be 0.34% (+0.29/−0.18%). Note that the current calculation applies to these specific observational constraints. The probability for CC#1 to be a rare type of background galaxy compatible with the JWST, ALMA and SPHERE observational constraints is therefore low. Such a low-probability event can admittedly arise would a large number of independent datasets be analysed. However, this is not the case for this study. Indeed, only two datasets among about 30 JWST MIRI datasets dedicated to exoplanet searches are analysed. Moreover, most of the other datasets are not as deep as the present ones, and there are no corresponding 1.6- and 870-μm data available. Hence, the likelihood that any research group would have found such a peculiar background galaxy contaminant within the whole corpus of relevant JWST data is much lower than the simple 10.2% (30 × 0.34%) one could derive.