© The Authors 2025

1. Introduction

Ultra-diffuse galaxies (UDGs) continue to challenge our understanding of galaxy formation and evolution. Defined by their low surface brightnesses (μ_g, 0 > 24 mag arcsec⁻²) and large effective radii (R_e > 1.5 kpc), UDGs are found in diverse environments, from dense galaxy clusters to the low-density field. Their origins remain debated, with two primary formation pathway hypotheses emerging: the ‘failed galaxy’ scenario, where early quenching prevents the typically extended star formation (van Dokkum et al. 2015; Danieli et al. 2022), and the ‘puffed-up dwarf’ scenario, in which low-mass galaxies are expanded via internal and external processes (Amorisco & Loeb 2016; Di Cintio et al. 2017; Carleton et al. 2019). However, recent discoveries suggest that some UDGs form through alternative mechanisms, including high-speed interactions or collisions (Silk 2019; van Dokkum et al. 2022; Shin et al. 2020; Lee et al. 2021, 2024).

One such case is GAMA 526784, an isolated UDG with a striking two-component structure: a compact, quiescent inner stellar body and an extended, actively star-forming outer region (Buzzo et al. 2025, hereafter Paper I). In Paper I we analysed deep imaging and integral field spectroscopy of the system and determined a mass-weighted stellar age of ∼9.9 Gyr and a metallicity of [M/H] ∼ −1.0 in the central body; we also found a significantly younger (∼0.9 Gyr), more metal-poor ([M/H] ∼ −1.2) population in the outskirts. Star formation is ongoing and highly localised, with a total star formation rate (SFR) of ∼0.01 M_⊙ yr⁻¹ concentrated in compact clumps aligned in a ∼5 kpc string-like structure. These regions show elevated electron densities (inferred from [SII]λ6716/λ6731 ∼ 1.1), significant dust attenuation (A_V up to 2.2 mag), and gas velocity dispersions reaching 50 km s⁻¹, all indicative of dense, possibly shock-compressed star-forming regions. The ionised gas is kinematically misaligned from the stellar body by ∼20°, in line with the idea of a recent external perturbation. In total, 29 star cluster candidates were identified, including a population of young clusters with ages of 8–11 Myr and stellar masses of log(M_⋆/M_⊙)∼5.0, and a more metal-poor, old globular cluster (GC) population with ages of ∼9 Gyr and masses of log(M_⋆/M_⊙)∼5.5. This bimodal star cluster population, along with its gas and stellar properties, raises the possibility that GAMA 526784 represents a transitional phase in the formation of GC-rich UDGs.

While the optical data provide strong evidence for a recent interaction, the nature of this event remains unclear. The ‘bullet dwarf’ scenario, in which a high-speed collision separates baryonic and dark matter and triggers a burst of star cluster formation (Silk 2019; van Dokkum et al. 2022), remains a compelling possibility. However, alternative explanations, such as a minor merger with a gas-rich dwarf, could also account for the observed properties (Fensch et al. 2019). To distinguish between these scenarios, a detailed analysis of the galaxy’s molecular and neutral gas content, as well as its large-scale environment, is essential.

We used Atacama Large Millimeter/submillimeter Array (ALMA) and Green Bank Telescope (GBT) observations of GAMA 526784 to probe its molecular and neutral gas reservoirs. These data allowed us to assess the availability of gas for future star formation, determine whether the observed shocks are associated with a gas-rich merger, and evaluate whether the system has sufficient gas to sustain massive star cluster formation. If GAMA 526784 is indeed forming young star clusters that will evolve into GCs, some of these young clusters (with ages < 3–5 Myr) might still be embedded in giant molecular clouds (GMCs; Lada & Lada 2003; Kim et al. 2021). If the clusters have available gas to become more massive, this would provide direct support that the galaxy is becoming GC-rich and offer new insights into how UDGs evolve over time.

By combining ionised, molecular, and neutral gas data, this study aims to clarify the evolutionary pathway of GAMA 526784 and its implications for UDG formation models. More broadly, understanding the gas content and star formation potential of this system contributes to the ongoing effort to map the diversity of UDG evolutionary pathways and to determine the role of high-speed interactions in shaping these enigmatic galaxies.

This paper is structured as follows: In Sect. 2 we describe the ALMA and GBT observations and data reduction processes. In Sect. 3 we present the methodology used to derive molecular and neutral gas measurements. Section 4 discusses the results and their implications for the formation of GAMA 526784, and we compare our findings to other UDGs and galaxy types to place GAMA 526784 in context. Finally, Sect. 5 summarises our conclusions.

2. Data

To investigate the properties of GAMA 526784, we compiled a comprehensive multi-wavelength dataset. Our analysis builds upon previous optical observations (Paper I), which provided insights into the star clusters, stellar populations, and ionised gas properties of GAMA 526784. In this work, we extend our study to the cold gas component by utilising observations from ALMA and GBT. The ALMA data allow us to trace the molecular gas through CO(1 − 0) emission, probing the star-forming potential of the system, while the GBT data provide a measurement of its neutral hydrogen (HI) reservoir, essential for understanding the galaxy’s overall gas content and potential for continued star formation. Below, we describe the details of these observations and data reduction procedures.

2.1. ALMA observations

We carried out the first ever ALMA observations of an UDG using ALMA’s Band 3 during Cycle 8 (PI: A. Zanella, Project ID: 2021.1.00898.S) with the goal of detecting the CO(1 − 0) emission line at a rest-frame frequency of ν_rf = 115.271202 GHz and the underlying continuum, redshifted to an observed frequency of ν_obs = 114.325475 GHz, assuming the observed radial velocity of the galaxy of V = 2758 km s⁻¹ (Paper I). The diffuse body of GAMA 526784 (hereafter ‘low-resolution’) was observed for 3 hours with the 7m array, while the higher-resolution dataset focused on the star clusters and their parent GMCs (hereafter ‘high-resolution’) was observed for nearly 6.5 hours with the 12m array, reaching sensitivities of 1.0 mJy beam⁻¹ and 9 μJy beam⁻¹, respectively.

Observations were conducted between November and December 2021 for the low-resolution data and between November 2021 and July 2022 for the high-resolution data. The native spectral resolutions of the observations are 10.255 km s⁻¹ for the low-resolution data and 1.282 km s⁻¹ for the high-resolution data. The full width at half maximum (FWHM) of the imaged beam sizes for the low-resolution data are 15.8″ × 9.9″, and the observations were carried out with the Atacama Compact Array (ACA). For the high-resolution data, the beam sizes are FWHM = 0.74″ × 0.56″. Table 1 summarises the ALMA observations.

Fig. 1. Comparison of the Hubble Space Telescope optical image (left panel) with the ALMA high-resolution cube of GAMA 526784 (middle). The star clusters (which are expected to be embedded in GMCs) are marked with labels corresponding to their numbers, following the naming convention used in Paper I. The beam size and orientation are indicated by the white ellipse in the bottom-left corner of the middle panel. Right: Spectra of the four brightest clusters, with the expected CO(1 − 0) emission wavelength marked in red. No detectable CO(1 − 0) emission is observed in any of the spectra.

The data were reduced using the standard ALMA pipeline with the CASA software (McMullin et al. 2007). The calibrated data cubes were converted to uv fits format and analysed in the uv-plane using GILDAS (Guilloteau & Lucas 2000), which is typical for faint sources. The analysis in the uv-plane offers several advantages over the image plane: it uses raw interferometric data (visibilities) and avoids image reconstruction and deconvolution, eliminating biases and artefacts from cleaning and providing a model-independent view. It also accounts for the actual spatial frequencies sampled by the array, respecting the instrument’s resolution and sensitivity limits, which is particularly important for low-S/N data like ours. Additionally, it allows a fine control over how different baselines are weighted or tapered, helping isolate spatial scales or emphasise specific structures. For compact, unresolved, low-S/N sources such as ours, uv-plane analysis is therefore preferable, while image-plane analysis remains valuable for spatially visualising complex or extended sources.

2.2. GBT observations

GBT observations of GAMA 526784 were conducted on September 29, 2024, under project ID GBT24B-319. The total integration time was 160 minutes, achieved through a sequence of 600-second ON/OFF pairs (300 s ON, 300 s OFF).

We utilised the L-band receiver coupled with the Versatile GBT Astronomical Spectrometer (VEGAS) operating in Mode 10, providing a bandwidth of 23.44 MHz and a spectral resolution of 0.7 kHz. The spectral setup was centred at the HI 21 cm line (1.42041 GHz) redshifted to correspond to a velocity of 2743 km s⁻¹ for GAMA 526784.

The L-band receiver provides a beam size of approximately 9 arcmin and a native velocity resolution of 0.15 km s⁻¹. The data were reduced using the standard GBTIDL routine getps, followed by baseline subtraction and spectral smoothing to enhance the signal-to-noise ratio.

3. Analysis

In what follows, we analyse the ALMA and GBT observations of GAMA 526784 to constrain its molecular and neutral gas content, assess its star formation efficiency, and investigate the role of its environment.

3.1. CO(1 − 0) emission

To investigate the molecular gas content of GAMA 526784, we first constructed the velocity-integrated CO(1–0) line maps. We adopted the method used by Zanella et al. (2023) and Zanella et al. (2018). The first step is to determine the optimal spectral range for integration. Since the measurement is performed directly in the uv-plane, we extracted the CO spectrum by applying the Fourier transform of a point spread function (PSF) model and looked for a detection around the expected frequency of the CO emission based on the redshift of the galaxy. We adopted a PSF model because, at the resolution of the ACA observations (15.8″ × 9.9″), the galaxy is unresolved with an effective radius of 13.2 arcsec (Paper I).

For the PSF modelling, we fixed the central coordinates based on the optical images (see Paper I), allowing the flux to vary across the channels. We then analysed the resulting spectrum to detect any positive emission line signals. However, no CO emission was detected in the spectra, suggesting that the diffuse body of GAMA 526784 does not host a significant amount of molecular gas that is traced by CO(1 − 0) emission.

We then averaged the data over the channels where the CO line was expected, based on the galaxy’s redshift as determined by Paper I (i.e. 2758 km s⁻¹). Since the CO(1–0) line is at ν = 115.271202 GHz in the rest frame, when we applied the redshift of the galaxy, we expected to find the CO(1–0) line at a frequency of ν = 114.2316936 GHz. We found the channel corresponding to this frequency and averaged the neighbouring channels, allowing for a velocity width of 30 km s⁻¹ up and down, i.e. three channels at each side. No CO(1 − 0) signal was detected in the averaged map of the low-resolution data targeting the diffuse body of the galaxy. Although this whole analysis was done in the uv-plane, we also inspected the imaging and arrived at the same conclusions, i.e. no CO detection. The final averaged CO map of the low-resolution data is shown in Appendix A.

We applied the same method to the high-resolution data cube, using the star cluster positions identified by Paper I, assuming that star clusters and their parent GMCs are co-spatial or are offset by less than the beam size (0.74″ × 0.56″) and that GMCs are unresolved at the resolution of our observations and fitting them as point sources. For the average map, we assumed a velocity width of 10 km s⁻¹ around the expected CO line. Again, no CO(1 − 0) emission was detected at the location of any star clusters. All data have been corrected for primary beam attenuation.

The resulting averaged map and spectra extracted at the location of the brightest star clusters in GAMA 526784 are presented in Fig. 1. The spectra of the fainter clusters is of similar quality and also do not show any clear sign of CO emission.

Although no CO emission was detected, we estimated upper limits on the molecular gas mass using the 5σ uncertainties in the flux measurements derived from the point-source fits.

3.2. Continuum emission

In a similar approach to the CO emission, we generated continuum maps by integrating over the entire spectral range, excluding the channels where CO(1 − 0) emission was expected based on the galaxy’s redshift. The continuum was not detected in the low- nor the high-resolution data, indicating a lack of significant emission in this spectral range as well.

3.3. Molecular gas mass

To estimate molecular gas mass constraints from the upper limits of our CO flux densities, we primarily adopted a metallicity-dependent CO-to-H₂ conversion factor, which is more appropriate for the low-metallicity, low-mass environment of GAMA 526784 (Accurso et al. 2017; Ramambason et al. 2024; Cormier et al. 2014; Shi et al. 2014). For comparison, we also calculated masses using the standard Milky Way conversion factor, which is a commonly used benchmark in extragalactic studies.

The metallicity-dependent conversion, introduced by Accurso et al. (2017), accounts for both the gas-phase metallicity and the galaxy’s offset from the star formation main sequence (SFR–M_⋆ relation) of Whitaker et al. (2012), and is given by

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill log{\alpha }_{\mathrm{CO}}(\pm 0.165\mathrm{dex})& =15.623-1.732[12+log(\mathrm{O}/\mathrm{H})]\hfill \\ \hfill & +0.051log\mathrm{\Delta }(\mathrm{MS}),\hfill \end{array}\end{array}$$(1)

where 12 + log(O/H) represents the gas-phase metallicity, and Δ(MS) is the distance of the galaxy from the star formation main sequence, defined as

$$\begin{array}{c}\hfill \mathrm{\Delta }(\mathrm{MS})=\frac{{\mathrm{sSFR}}_{\mathrm{measured}}}{{\mathrm{sSFR}}_{\mathrm{ms}}(\mathrm{z},{\mathrm{M}}_{\star })},\end{array}$$(2)

where sSFR_measured is the measured specific SFR, and sSFR_ms is the sSFR predicted for a galaxy on the star formation main sequence. The latter is given by

$$\begin{array}{c}\hfill \begin{array}{cc}\hfill log({\mathrm{sSFR}}_{\mathit{ms}}(z,M_{\star }))& =-1.12+1.14z-0.19z^2-(0.3+0.13z)\hfill \\ \hfill & \times (log{M}_{\star }-10.5)[{\mathrm{Gyr}}^{-1}],\hfill \end{array}\end{array}$$(3)

where z is the galaxy redshift, and M_⋆ is its stellar mass. Using this approach, the velocity-integrated CO line luminosity can be converted into molecular gas mass using the standard methods thoroughly explained in Solomon & Vanden Bout (2005). Assuming a global gas-phase metallicity of 12 + log(O/H) = 8.3 (Paper I; derived using the O3N2 calibration from Marino et al. 2013), and the recovered α_CO, met = 17.01 M_⊙ (K km s⁻¹ pc²)⁻¹ to derive 5σ upper limits on the molecular gas content of GAMA 526784, which leads to log(M_H2/M_⊙)≲7.6 M_⊙.

For comparison, we also computed the molecular gas mass using the traditional Milky Way conversion factor, α_CO, MW = 4.36 M_⊙ (K km s⁻¹ pc²)⁻¹ (Strong & Mattox 1996; Abdo et al. 2010), which is frequently applied to galaxies with solar-like metallicities and has been used in studies such as the PHANGS survey (e.g. Leroy et al. 2021). This yields a molecular gas mass estimate approximately four times smaller than those obtained using the metallicity-dependent method, highlighting the importance of applying conversion factors tailored to low-metallicity systems (e.g. Accurso et al. 2017; Ramambason et al. 2024; Cormier et al. 2014; Shi et al. 2014).

We extended the same metallicity-based method to individual GMCs using the SFRs and metallicities measured within the apertures defined for their flux extraction. While CO-to-H₂ conversion in solar-metallicity GMCs often assume the Milky Way value (Bolatto et al. 2013), systems like GAMA 526784 ([M/H] ≃ −1.5 dex) are expected to require significantly larger α_CO values due to reduced dust shielding and enhanced photodissociation of CO (Schruba et al. 2012). For comparison, resolved studies of GMCs in the Large Magellanic Cloud ([M/H] ∼ −0.3 dex) yield α_CO = 6.9 ± 2.3 (Hughes et al. 2010), and the Small Magellanic Cloud ([M/H] ∼ −0.7 dex) shows conversion factors (α_CO) exceeding 50 (Bolatto et al. 2008). In our case, the metallicity-dependent conversion results in α_CO values ranging from 6 to 25 across different GMCs, consistent with expectations for both more metal-rich and metal-poor cloud populations. The recovered physical properties of GAMA 526784 and the GMCs are shown in Table B.1.

The resulting upper limits on individual GMC masses (M_H2 < 10^6.0 − 10^7.0 M_⊙) fall within the range of the lowest-mass GMCs detected in PHANGS galaxies (Sun et al. 2020). These values are also comparable to those of well-known local clouds, including the most massive Milky Way GMC (6 × 10⁶ M_⊙), the largest M33 GMC (7 × 10⁵ M_⊙), and the Orion complex (5 × 10⁵ M_⊙; Engargiola et al. 2003; Wilson et al. 2005).

3.4. HI mass calculation

Fig. 2. HI spectrum of GAMA 526784, smoothed by a factor of 3 (resulting in a spectral resolution of 0.45 km s−1). The profile exhibits a clear double-horned shape, characteristic of a rotating disc, with a velocity width at 50% of the peak flux (W50) of 105 km s−1. Integrating the emission yields a total HI flux of 1.26 Jy km s−1.

In addition to the molecular gas, we estimated the HI mass of GAMA 526784 using the GBT data. The HI spectrum (Fig. 2) shows a double-horned profile characteristic of a rotating disc centred at V = 2758 km s⁻¹ with a velocity width of the HI line profile, at 50% of the peak intensity, of W₅₀ = 105 km s⁻¹. The integrated HI flux density is S_HI = 1.26 ± 0.13 Jy km s⁻¹ at the native resolution of 0.15 km s⁻¹.

The HI mass was calculated using the standard relation,

$$\begin{array}{c}\hfill M_{\mathrm{HI}}=2.36\times {10}^5\times D_L^2\times S_{\mathrm{HI}}{M}_{\odot },\end{array}$$(4)

where D_L = 40 Mpc is the luminosity distance. This yields log(M_HI/M_⊙) = 8.7 ± 0.9. With the recovered W₅₀, one can estimate the rotation velocity of the galaxy and its dynamical mass, using the approach of Spekkens & Karunakaran (2018). We assumed that W₅₀ = 2 V_rotsin i, where $sini=\sqrt{\frac{1-{(b/a)}^2}{1-q_0^2}}$, q₀ = 0.2 (Hubble 1926; Haynes et al. 1984; Bradford et al. 2016) and b/a = 0.52 (Paper I). This renders a rotation velocity for GAMA 526784 of V_rot = 59.3 ± 5.9 km s⁻¹, after correcting for inclination. This value is consistent with the ionised gas rotation amplitude (Paper I ). To calculate the dynamical mass of the galaxy, we assumed that the HI gas distribution extends out to three effective radii (Broeils & Rhee 1997) and used

$$\begin{array}{c}\hfill M_{\mathrm{dyn}}^{3Re}=3.5\times {10}^5{R}_e{W}_{50}^2{M}_{\odot },\end{array}$$(5)

with R_e = 2.5 kpc (Paper I). We estimate a dynamical mass within 3R_e of log(M_dyn^3R_e/M_⊙) = 10.0 ± 1.0. The baryonic mass, derived from the stellar mass (log(M_⋆/M_⊙)∼8.3) and HI mass (log(M_HI/M_⊙)∼8.7) totals approximately log(M_baryonic/M_⊙)∼8.9. This implies that the galaxy is likely dark matter-dominated within 3 R_e.

Fig. 3. Scaling relations for the molecular and neutral gas content of disc-dominated galaxies as a function of stellar mass. Grey scatter points are literature values from Karunakaran et al. (2024), Leisman et al. (2017), Papastergis et al. (2017), Wang et al. (2020), and Cao et al. (2024). Points with arrows pointing downwards are 5σ upper limits. GAMA 526784 is shown in all panels with the red star symbol. Black edges indicate measurements made using a metallicity-dependent CO-to-H2 conversion (Accurso et al. 2017); otherwise, the measurements were made using the standard Milky Way αCO conversion. GMCs are shown with plus signs; their masses were also calculated using the metallicity-dependent conversion.

4. Results and discussion

GAMA 526784 provides a valuable case study of gas-rich, low-mass galaxies. In what follows we discuss its significant neutral hydrogen reservoir and the non-detection of molecular gas traced by CO(1 − 0) emission, as well as its environment. With this information, we place constraints on its formation history.

4.1. Molecular and neutral gas content

Figure 3 compares GAMA 526784 with literature data for HI and H₂ masses as a function of stellar mass. The top panel shows that the HI properties of GAMA 526784 are typical for a star-forming galaxy of its stellar mass. Its M_HI/M_* ratio of ∼2.9 is consistent with other gas-rich UDGs and low-surface brightness (LSB) galaxies from Karunakaran et al. (2024). In contrast, quiescent field UDGs like those studied by Papastergis et al. (2017, M_HI/M_* ∼ 0.5) have a much lower M_HI/M_⋆ ratio, likely due to environmental gas stripping or suppressed star formation. The galaxy’s HI content is also less extreme than that of HI-selected systems like those in Leisman et al. (2017), which are biased towards higher gas fractions due to their selection criteria.

The middle panel of Fig. 3 depicts the molecular-to-stellar mass ratio, highlighting the stringent upper limits on the CO-derived H₂ mass. Recent studies of CO in UDGs and LSBs are shown for comparison in the figure (e.g. Wang et al. 2020; Cao et al. 2024). These works have found similarly low CO-detected H₂ masses. It is important to note that for a correct comparison between the studies of Wang et al. (2020), Cao et al. (2024) and ours, we converted their 3σ upper limits in the H₂ mass estimates into 5σ upper limits assuming Gaussian statistics.

The galaxy’s HI reservoir (M_HI = (4.7 ± 0.5)×10⁸ M_⊙) and CO non-detection can be explained by several mechanisms: (1) the predominance of CO-dark H₂, which remains invisible to CO observations but contributes to star formation; (2) a time delay in HI-to-H₂ conversion following a recent interaction (Cullen et al. 2007; Renaud et al. 2019); or (3) elevated turbulence inhibiting gas collapse (Krumholz & Dekel 2012). The non-detection of CO despite active star formation is consistent with trends observed in other low-metallicity dwarfs (Leroy et al. 2008; Cormier et al. 2014; Ramambason et al. 2024), reinforcing that GAMA 526784 is not an outlier but rather a representative example of its class.

Fig. 4. Possible interacting system of GAMA 526784 and a putative companion, separated by 48 kpc in projection. Young star clusters in GAMA 526784 align along the axis possibly connecting the galaxies, with their ages and the star formation history of the galaxy suggesting formation during a close encounter ∼100 Myr ago. This spatial and possibly kinematic configuration provides compelling evidence of a physical association between the galaxies. The image is 12 arcmin on the side. North is up, and east is to the left. The only massive galaxy close in projection to GAMA 526784 is PGC 025165 (highlighted by the dotted rectangle), and it lies at a redshift of z = 0.0391, i.e. far in the background.

4.2. Environment

GAMA 526784 is an isolated field UDG, with no obvious massive companion galaxies in its vicinity. In projection, the closest galaxy is PGC 025162 (highlighted in Fig. 4), located approximately 3 arcmin away (∼34 kpc at the redshift of GAMA 526784). However, its redshift from the HyperLEDA survey (z = 0.0391) places it well in the background compared to GAMA 526784 (z = 0.0091), confirming that the two are not physically associated. A comprehensive search of the surrounding region using photometric and spectroscopic redshifts from the GAMA catalogue reveals no massive galaxies within several hundred kiloparsecs. A larger-scale view of the environment, presented in Fig. 4, spans a 12 × 12 arcmin² field-of-view and clearly illustrates the lack of nearby companions, reinforcing the exceptional isolation of GAMA 526784.

Despite its isolation, the gas and star formation properties of GAMA 526784 suggest a complex evolutionary history. The spatial distribution of young clusters, extending over ∼5 kpc along a preferential axis, motivates the search for a companion. We hypothesise that a small galaxy, located approximately 4 arcmin (48.6 kpc) away from GAMA 526784 (Fig. 4), is a promising candidate based on its remarkably similar g − r = 0.4 colour to GAMA 526784 and the presence of similar star clusters.

The putative companion lies within the Subaru Hyper Suprime-Cam (HSC) field of view centred on GAMA 526784, allowing for a detailed analysis of its morphology and physical properties via galaxy fitting, and stellar mass estimation through spectral energy distribution (SED) fitting using the five grizy HSC filters. The data reduction, as well as the galaxy and SED fitting procedures applied to the companion, are identical to those used for GAMA 526784; we refer the reader to Paper I for further methodological details. The analysis yields an effective radius of 0.64 kpc (assuming the same distance as GAMA 526784), a central surface brightness of 23.2 mag arcsec⁻² in the g-band, and a stellar mass of log(M_⋆/M_⊙) = 7.3 ± 0.2 – roughly an order of magnitude less massive than GAMA 526784.

GBT observations offer critical constraints on this scenario. While the putative companion falls within the GBT beam, raising the expectation of a second line in the HI spectrum in Fig. 2 if it were gas-rich, no such line is detected. However, the HI signal at the companion’s location would be significantly attenuated by the beam response, reduced to ∼40% of peak sensitivity at a 4.5 arcmin offset. The absence of a second line could indicate one of several possibilities: (1) the companion is gas-poor, (2) its HI line is perfectly blended with that of GAMA 526784, which would require a coincidental velocity alignment, (3) it lies in the background, or (4) the signal is simply too attenuated and faint to detect. The undisturbed HI profile of GAMA 526784 makes blending less likely. The companion’s blue colours suggest recent star formation, making a gas-poor scenario less probable, though it is possible the gas reservoir is small and confined to the immediate vicinity of the star clusters, rather than spread throughout the galaxy. If the companion lies in the background, only deep spectroscopy will be able to resolve this. Still, the presence of extremely similar blue star clusters in the two systems, along with the alignment of the companion along the star cluster axis of GAMA 526784, strengthens the case for a physical association, though a confirmation would require deeper observations.

If physically associated, the system’s properties may point to either a high-speed encounter or an ongoing minor merger. The relatively undisturbed morphology of the lower-mass companion is puzzling, as simulations and observations generally predict significant disruption for the smaller galaxy in such interactions. However, this may favour a high-speed flyby scenario, in which the gravitational interaction is brief and insufficient to destroy the companion. The ∼100 Myr star formation burst in GAMA 526784 (Paper I) implies a relative velocity of ∼475 km s⁻¹ to account for the current projected separation of over 4 arcmin, i.e. consistent with a high-speed collision (Lee et al. 2024) capable of triggering widespread disc star formation. Alternatively, a minor merger unrelated to the companion could explain the presence of a two-component stellar population in GAMA526784 (Paper I). However, this scenario struggles to account for the formation of massive young clusters, as these clusters are very young, whereas any such merger would have occurred some time ago. Whether gas collapse leading to the formation of massive clusters can be sustained for many millions of years after a minor merger remains an open question, yet to be tested by simulations.

Current data cannot definitively distinguish between these scenarios. The similar optical properties strengthen the case for association, while the GBT non-detection and companion’s morphology present challenges. Future observations could resolve this through: (1) deeper HI mapping to detect low-surface-brightness features or confirm the companion’s distance, (2) targeting lines that are sensitive to CO-dark molecular gas, and (3) deep spectroscopy of both systems.

GAMA 526784 serves as an intriguing case for studying dwarf galaxy transformations. As shown in Fig. 4, the spatial coherence between the galaxy, its hypothesised companion, and star cluster string offers a rare snapshot of a possible interaction, highlighting how even isolated systems can bear traces of their dynamical history.

5. Conclusions

Our multi-wavelength study of GAMA 526784 has revealed a remarkable UDG that is possibly in the midst of transformation, providing new insights into UDG formation and the origins of unusual GC populations in low-mass systems. The galaxy’s HI reservoir (M_HI = (4.7 ± 0.5)×10⁸ M_⊙) contrasts with the mis-detection of CO. This apparent contradiction can be explained by considering both the prevalence of CO-dark molecular gas expected at its low metallicity (12 + log(O/H) = 8.3) and the effects of a possible ongoing interaction with a putative companion approximately 4 arcmin away.

Compelling evidence of this interaction includes the companion’s projected separation of 48 kpc, the presence and distribution of shocked gas along the string of young star clusters in GAMA 526784 (Paper I), and the alignment of these clusters along the same axis and in the direction of the companion. Their spatial arrangement and ages are consistent with having been triggered during a close encounter approximately 100 Myr ago; this would mean that the galaxies are travelling at ∼475 km s⁻¹ with respect to each other and that therefore this is a high-speed encounter. Moreover, the system’s two-component structure, with an old core surrounded by a younger, star-forming envelope, suggests GAMA 526784 represents a transitional phase between a gas-rich dwarf and a GC-rich quiescent UDG.

This system challenges simple UDG classification schemes, demonstrating how dwarf–dwarf interactions can produce extended morphologies while simultaneously seeding GC populations. The combination of high-velocity dispersion in the gas from the interaction (possibly suppressing global star formation) and local gas compression (enabling star cluster formation) helps explain the apparent paradox of a gas-rich yet inefficiently star-forming galaxy. GAMA 526784 may represent a progenitor phase for some GC-rich UDGs, with its young massive clusters potentially evolving into a GC system over time.

Future investigations should focus on high-resolution HI mapping to reveal tidal structures, lines that are sensitive to CO-dark molecular gas, and deeper spectroscopy of the putative companion to confirm their association. Wider-field searches for similar interacting dwarf pairs could establish how common this formation pathway is. As one of the clearest examples of an isolated UDG caught in the act of transformation, GAMA 526784 provides unique insights into the environmental processes that shape these enigmatic galaxies and their star cluster systems.

Acknowledgments

AZ acknowledges support from the European Union – NextGeneration EU within PRIN 2022 project n.20229YBSAN – Globular clusters in cosmological simulations and in lensed fields: from their birth to the present epoch and from the INAF Minigrant “Clumps at cosmological distance: revealing their formation, nature, and evolution” (Ob. Fu. 1.05.23.04.01). MH acknowledges financial support from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy – EXC-2094 – 390783311. KS acknowledges support from the Natural Sciences and Engineering Research Council of Canada.

References

Appendix A: Diffuse body (low-resolution data) CO map

In this appendix we present the averaged CO map of the low-resolution data targeting the diffuse body to show the lack of any clear CO emission throughout the galaxy, shown in Fig. A.1.

Fig. A.1. Comparison of Hα and CO maps, highlighting where the molecular gas would be expected in the diffuse body if it were co-spatial with the ionised gas. The right panel shows the averaged CO map of the low-resolution ALMA data, focused on the diffuse body. In both panels, Hα contours are shown to facilitate visualisation.

Appendix B: Table with star cluster properties

In this appendix we show the physical properties of GAMA 526784 and its star clusters measured in Paper I and in this work.

All Tables

All Figures

Fig. 1. Comparison of the Hubble Space Telescope optical image (left panel) with the ALMA high-resolution cube of GAMA 526784 (middle). The star clusters (which are expected to be embedded in GMCs) are marked with labels corresponding to their numbers, following the naming convention used in Paper I. The beam size and orientation are indicated by the white ellipse in the bottom-left corner of the middle panel. Right: Spectra of the four brightest clusters, with the expected CO(1 − 0) emission wavelength marked in red. No detectable CO(1 − 0) emission is observed in any of the spectra.

In the text

	Fig. 2. HI spectrum of GAMA 526784, smoothed by a factor of 3 (resulting in a spectral resolution of 0.45 km s−1). The profile exhibits a clear double-horned shape, characteristic of a rotating disc, with a velocity width at 50% of the peak flux (W50) of 105 km s−1. Integrating the emission yields a total HI flux of 1.26 Jy km s−1.
In the text

Fig. 3. Scaling relations for the molecular and neutral gas content of disc-dominated galaxies as a function of stellar mass. Grey scatter points are literature values from Karunakaran et al. (2024), Leisman et al. (2017), Papastergis et al. (2017), Wang et al. (2020), and Cao et al. (2024). Points with arrows pointing downwards are 5σ upper limits. GAMA 526784 is shown in all panels with the red star symbol. Black edges indicate measurements made using a metallicity-dependent CO-to-H2 conversion (Accurso et al. 2017); otherwise, the measurements were made using the standard Milky Way αCO conversion. GMCs are shown with plus signs; their masses were also calculated using the metallicity-dependent conversion.

In the text

Fig. 4. Possible interacting system of GAMA 526784 and a putative companion, separated by 48 kpc in projection. Young star clusters in GAMA 526784 align along the axis possibly connecting the galaxies, with their ages and the star formation history of the galaxy suggesting formation during a close encounter ∼100 Myr ago. This spatial and possibly kinematic configuration provides compelling evidence of a physical association between the galaxies. The image is 12 arcmin on the side. North is up, and east is to the left. The only massive galaxy close in projection to GAMA 526784 is PGC 025165 (highlighted by the dotted rectangle), and it lies at a redshift of z = 0.0391, i.e. far in the background.

In the text

	Fig. A.1. Comparison of Hα and CO maps, highlighting where the molecular gas would be expected in the diffuse body if it were co-spatial with the ionised gas. The right panel shows the averaged CO map of the low-resolution ALMA data, focused on the diffuse body. In both panels, Hα contours are shown to facilitate visualisation.
In the text