Variability versus infrared color diagrams. Upper left: MAD_WISE1/WISE2 versus IRAC1–IRAC4. The MLR is calculated by using the formula from Groenewegen & Sloan (2018), which is increasing along with the variability and probably approaching an upper limit of ∼-6:1 M⊙ yr^-1 as shown by the dotted lines. The limits of MLR= -8.0M⊙ yr^-1 and -7.0M⊙ yr^-1 are also shown by the dashed lines. Upper right: a linear relation between IRAC1–IRAC4 andK_S –WISE3/WISE1–WISE3 which can be used to derive MLR in K_S –WISE3 and WISE1–WISE3 colors with slightly larger error. For comparison, the MLR calculated by using K–[12] color based on the MIR data from IRAS (Josselin et al. 2000) has been added in the K_S –WISE3/IRAC1–IRAC4 diagram as shown by the open triangles (MLR_K−[12] = − 8.0; −7.5M_⊙ yr^-1). It can be seen that the MLR calculated by using K–[12] is largely underestimated compared with the one calculated by using IRAC1–IRAC4. Bottom: Previous known “blue” RSGs(WISE1–WISE2/IRAC1–IRAC2 < 0) in WISE1–WISE2/IRAC1–IRAC2 are turning redder (WISE1–WISE2/IRAC1–IRAC2 > 0) along with the increasing of variability.