T he formation of realistic populations of galaxies requires one or more forms of self-regulating feedback to suppress the conversion of gas into stars. Cosmological simulations invoke various mechanisms to regulate star formation, in the form of energydepositionandwindlaunchinglinkedto supermassive black hole activity, supernovae, and/or radiation pressure from massive stars. The strength and scalings of these processes play critical roles in the evolution of galaxies by regulating the growth of stellar mass relative to the dark-matter halo, connecting the properties of central black holes to their host galaxies, and enriching the circumgalactic medium with heavy elements (1)(2)(3)(4).

O u t f l o w i n gw i n d so fg a sa r eu b i q u i t o u si n nearby galaxies. The gas in outflows spans many orders of magnitude in temperature and density (5)(6)(7), and different components of the winds are observable from x-ray to radio wavelengths. Observing winds in the distant Universe is difficult: Not only are the spectral features faint, but outflow tracers observed in emission may be less reliable because of the ongoing processes of galaxy assembly (8). The constrained geometry of absorption lines provides signatures of inflowing and outflowing material, but these lines have thus far eluded detection.

Passive galaxies with stellar masses of ~10 11 M ⊙ (where M ⊙ is the solar mass), low star formation rates (SFRs) (≲ 10M ⊙ yr -1 ), and stellar ages of ~0.8 billion years were already in place when the Universe was 2 billion years old, implying that these galaxies formed stars at rates of hundreds of M ⊙ per year before z = 5 (where z is the redshift) (9). Galaxies with such high SFRs are extremely rare in rest-ultraviolet surveys, implying that such galaxies are incapable of becoming sufficiently massive by z ~4t o reproduce the observed passive population. This suggests a connection between early passive galaxies and high-redshift dusty star-forming galaxies (DSFGs) observed at far-infrared wavelengths (10)(11)(12). With a redshift distribution that includes a substantial number of objects at z >4 ( 13,14), DSFGs represent a plausible progenitor population of the earliest passive galaxies. If this evolutionary connection is correct, many DSFGs should show signs of the feedback process(es) acting to suppress their rapid star formation.

We present observational evidence of a massive molecular wind being launched from SPT-SJ231921-5557.9 (hereinafter referred to as SPT2319-55), a DSFG observed when the Universe was only 1 billion years old. SPT2319-55 was discovered in the 2500-square-degree South Pole Telescope survey (15)onthebasisofits thermal dust emission. Earlier observations of this source from the Atacama Large Millimeter/ submillimeter Array (ALMA) determined its redshift to be z source =5.293 (14)andshowed that it is gravitationally lensed by an intervening foreground galaxy (16). As is typical for these objects, SPT2319-55 is gas rich, containing ~1.2 × 10 10 M ⊙ of molecular gas and ~1.2 × 10 8 M ⊙ of dust, and is forming stars very rapidly, with an SFR of ~790 M ⊙ yr -1 (17) (these values account for the lensing magnification).

We used ALMA to observe the rest-frame 119-mm ground-state doublet transition of the hydroxyl molecule, OH, and the thermal dust emission at this wavelength (17). This transition is a good tracer of gas flows in nearby galaxies (18,19). The ALMA observations reach a spatial resolution of 0.25 by 0.4 arc sec and resolve the lensed images of SPT2319-55 (Fig. 1). We detected a molecular outflow from SPT2319-55, seen in blueshifted absorption against the bright dust continuum emission, a signature of outflowing molecular material (Fig. 1).

We fit the spectrum in Fig. 1 with the sum of two velocity components, each consisting of two equal-amplitude Gaussian profiles separated by 520 km s -1 , the separation of the two components of the OH doublet (17). As is common practice for low-redshift observations of OH, we assigned the higher-frequency component of the doublet to the systemic velocity of the galaxy. Because the redshift of this source is known to better than 50 km s -1 from other observations ( 14), we fixed the velocity offset of one pair of Gaussian profiles to the systemic velocity. This component represents absorption due to gas from within the galaxy, with a fitted full-width-at-half-maximum (FWHM) linewidth of 330 ± 80 km s -1 .Weallowed a velocity offset for the second pair of Gaussian profiles; this component represents the blueshifted molecular outflow. We found that this second component is blueshifted relative to the galaxy by 440 ± 50 km s -1 and derived a maximum velocity of ~800 km s -1 (17).

The ALMA observations spatially resolve and clearly detect the dust continuum even in relatively narrow velocity channels. Because SPT2319-55 is gravitationally lensed by a foreground galaxy, we determine its intrinsic structure by using a lens modeling technique that represents the galaxy as an array of pixels (17,20). In addition to the line-free continuum emission, we also reconstructed the OH absorption components by using velocity ranges relative to the higherfrequency OH transition of -700 to -200 km s -1 for the wind component and +300 to +700 km s -1 for the internal component; the latter velocity range corresponds to velocities of -220 to +180 km s -1 relative to the lower-frequency transition and traces gas within SPT2319-55. These v e l o c i t yr a n g e sf a i r l yc l e a n l ys e p a r a t et h ew i n d and internal absorption (17).

The doubly-imaged continuum emission in Fig. 1 is consistent with the lensing of a single background galaxy, with a well-determined extent, by a single foreground lens. The reconstructions of the dust continuum and each absorption component are shown in Fig. 2. The continuum is dominated by a single bright region ~1.2 kpc in diameter (17), with flux density S 400GHz = 9.0 mJy (1 Jy = 10 -26 Wm -2 Hz -1 ) magnified by a factor m = 5.8. The internal absorption, arising from gas within the galaxy, is concentrated toward the center of the object. This is to be expected, as the continuum is brightest and gas column densities are highest toward the nuclear region, making the detection of absorption easier.

The geometry of the molecular outflow is more complex and is not confined to the center of the galaxy. Instead, it is clustered into multiple clumps, separated from each other by a few hundred parsecs, corresponding to ≈2.5 FWHM resolution elements (17). Because absorption is easier to detect against a strong continuum, we would expect a geometry similar to that of the internal absorption if the continuum were the limiting factor in the reconstruction. Tests using mock data also show that absorbing components weaker than the outflow are well recovered by our reconstructions (17).

The overall covering fraction of the wind is high-80% of pixels with a continuum signal-tonoise ratio of >8 have significant wind absorption (although these pixels are not all independent) (17). If the covering fraction along the line of sight we observe is the same as that for the rest of the source, this implies a total wind opening angle of ~0.8 × 4p sr. On the other hand, we can set a rough lower limit on the opening angle if we assume that we have observed the entirety of the outflow and that the molecular material is destroyed by the time it reaches a few galaxy radii away, as in local starburst galaxies (6). If the maximum radius of the outflowing material is 2 galaxy radii, for example, then the solid angle of a sphere of this radius subtended by the source RESEARCH | REPORT on October 9, 2018 http://science.sciencemag.org/ Downloaded from

winds in low-redshift objects studied in OH (19,22). If the OH absorption in SPT2319-55 is also optically thick, we expect that the wind contains substantial substructure on scales below our effective resolution limit, with most of the absorption arising from small, highly optically thick clumps. Alternatively, because the outflow reconstruction spans a wide range of velocities, it is possible that the covering fraction is lower in narrower velocity ranges. In this case, we also expect substantial substructure on smaller scales, as the lower covering fractions in narrow velocity bins must still sum to the high overall covering fraction we have observed. On the other hand, if the OH excitation temperature is high, the OH molecules can substantially fill in the absorption profile with re-emitted 119-mm photons, reducing the absorption strength while maintaining a high covering fraction without small-scale structure in the wind, although this implies high densities in the absorbing gas.

We estimated the mass outflow rate by using a simple model for the outflow geometry (17). The primary uncertainties in this estimate are the unknown optical depth of the OH 119-mmdoublet and the detailed geometry of the wind. In the limiting case of optically thin absorption, we find a minimum mass outflow rate M out ≳ 60 M ⊙ yr À1 . A more likely outflow rate, on the basis of the empirical correlation between the OH equivalent width and M out in low-redshift dusty galaxies (17), is M out ∼ 510 M ⊙ yr À1 .T h o u g hh i g h l yu ncertain (17), this value is within a factor of 2 of the SFR of SPT2319-55, indicating that the wind is capable of depleting the molecular gas reservoir on a time scale similar to that of star formation itself. We compare the SFR and molecular outflow rates we derived for SPT2319-55 under both assumptions in Fig. 3, along with molecular outflow rates for low-redshift objects (21,23).

The mass loading factor of the wind, M out = SFR ∼ 0:7, is high, even accounting for molecular material alone; the inclusion of unobserved wind phases, namely, neutral atomic and ionized gases, would increase this value still further. High mass loading factors are expected from simulations of galaxies (24,25), which require strong feedback to prevent the overproduction of stars. On the other hand, both the origin of the molecules in the SPT2319-55 wind (whether entrained in hotter material or formed in situ) and the driving source are unclear. Current data place an upper limit of 30% on the contribution of an active galactic nucleus to the total luminosity of SPT2319-55 but cannot rule out nuclear activity at a lower level (17).

Despite the high overall outflow rate, it is likely that only a small fraction of the wind mass is sufficiently fast-moving to escape the gravitational potential of the galaxy. Given the molecular gas mass of SPT2319-55, assuming a typical gas fraction for DSFGs at this redshift (26) and using the size from the source reconstruction yield an escape velocity from the galaxy of ~650 km s -1 . The fraction of the absorption at speeds greater than this value indicates that ~10% of the outflowing material will escape SPT2319-55, under the same assumptions as those for the outflow as a whole. This escape fraction is within the range observed in local galaxies (27). The remainder of the material, therefore, will stay within the galaxy'sd a r kmatter halo.

Our results show that self-regulating feedback is acting to disrupt and remove the molecular gas in SPT2319-55 and will likely suppress the rapid star formation in this galaxy in ≲ 100 million years. Whether this is sufficient to quench the star formation on a more permanent basis is less clear. A large fraction of the wind material is likely to remain within the galaxy, capable of fueling future star formation at later times. On the other hand, SPT2319-55 likely resides in a dark-matter halo of mass ~10 12 M ⊙ (28), sufficiently massive to shock infalling gas to high temperatures and prevent its accretion (29). Other forms of feedback, such as from an active galactic nucleus, may also act to maintain the low SFR after the initial suppression (30). M ol e cu l a rou t f l o wsl ik et h eo n ew eh a v eo bserved are likely an important step in the suppression of star formation and the emergence of early passive galaxies.