Context :

Aims : This paper analyses the behaviour of the gas-to-dust mass ratio ( G/D ) of local Universe galaxies over a large metallicity range .
We especially focus on the low-metallicity part of the G/D vs metallicity relation and investigate explanations for the observed relation and scatter .

Methods : We combine three samples : the Dwarf Galaxy Survey , the KINGFISH survey and a subsample from Galametz et al .
( 2011 ) totalling 126 galaxies , covering a 2 dex metallicity range , with 30 % of the sample with 12+log ( O/H ) \leq 8.0 .
The dust masses are homogeneously determined with a semi-empirical dust model , including submm constraints .
The atomic and molecular gas masses are compiled from the literature .
Two X _ { { CO } } Â are used to estimate molecular gas masses : the Galactic X _ { { CO } } and a X _ { { CO } } Â depending on the metallicity ( \propto Z ^ { -2 } ) .
Correlations with morphological types , stellar masses , star formation rates and specific star formation rates are discussed .
The trend between G/D and metallicity is empirically modelled using power-laws ( slope of -1 and free ) and a broken power-law .
We then compare the evolution of the G/D with predictions from chemical evolution models .

Results : We find that out of the five tested galactic parameters , metallicity is the galactic property driving the observed G/D .
The observed G/D versus metallicity relation can not be represented by a power-law with a slope of -1 over the whole metallicity range .
The observed trend is steeper for metallicities lower than \sim 8.0 .
A large scatter is observed in the G/D for a given metallicity , with a dispersion of 0.37 dex on average in metallicity bins of \sim 0.1 dex , for both X _ { { CO } } Â values .
The broken power-law reproduces best the observed G/D compared to the single power laws and provides estimates of the G/D that are accurate to a factor of 1.6 .
The good agreement of the G/D and its scatter with respect to metallicity with the predictions from the three tested chemical evolution models allows us to infer that the scatter is intrinsic to galactic properties , reflecting the different star formation histories , dust destruction efficiencies , dust grain size distributions and chemical compositions across the sample .

Conclusions : Our results show that the chemical evolution of low-metallicity galaxies , traced by their G/D , depends strongly on their local internal conditions and individual histories .
The large scatter in the observed G/D at a given metallicity reflects the impact of various processes occurring during the evolution of a galaxy .
Disentangling between these various processes , despite the numerous degeneracies affecting them , is now the next step .