by cannot cope with pdata.frame converted to data.table

I found by= returns an error when performed on a former pdata.frame (from plm package). Here is an example:

library(data.table)
set.seed(45L)
DT1 <- data.table(V1=c(1L,2L),
                 V2=LETTERS[1:3],
                 V3=round(rnorm(4),4),
                 V4=1:12)


DT1
   V1 V2      V3 V4
 1:  1  A  0.3408  1
 2:  2  B -0.7033  2
 3:  1  C -0.3795  3
 4:  2  A -0.7460  4
 5:  1  B  0.3408  5
 6:  2  C -0.7033  6
 7:  1  A -0.3795  7
 8:  2  B -0.7460  8
 9:  1  C  0.3408  9
10:  2  A -0.7033 10
11:  1  B -0.3795 11
12:  2  C -0.7460 12

DT1[ , sum(V3), by=c("V2", "V1")]
   V2 V1      V1
1:  A  1 -0.0387
2:  B  2 -1.4493
3:  C  1 -0.0387
4:  A  2 -1.4493
5:  B  1 -0.0387
6:  C  2 -1.449

library(plm)
pDT1 <- pdata.frame(DT1, index=c("V2", "V4"), drop.index=FALSE, row.names=TRUE) # convert to a pdata.frame


DT2 <- as.data.table(pDT1) # convert back to a data.table
is.data.table(DT2)
TRUE

DT2
V1 V2      V3 V4
1   1  A  0.3408  1
2   2  A -0.7460  4
3   1  A -0.3795  7
4   2  A -0.7033 10
5   2  B -0.7033  2
6   1  B  0.3408  5
7   2  B -0.7460  8
8   1  B -0.3795 11
9   1  C -0.3795  3
10  2  C -0.7033  6
11  1  C  0.3408  9
12  2  C -0.7460 12

DT2[ , sum(V3), by=c("V2", "V1")]
Error in `[.pdata.frame`(DT2, , sum(V3), by = c("V2", "V1")) : 
  unused argument (by = c("V2", "V1"))


# Delete attributes introduced by converting to pdata.frame
DT2 <- as.data.table(lapply(DT2, function(x){ attr(x, "index") <- NULL; x}))
DT2 <- as.data.table(lapply(DT2, function(x){ attr(x, "class") <- NULL; x}))

DT2[ , sum(V3), by=c("V2", "V1")] # works now
   V2 V1      V1
1:  1  1 -0.0387
2:  1  2 -1.4493
3:  2  2 -1.4493
4:  2  1 -0.0387
5:  3  1 -0.0387
6:  3  2 -1.4493

plm_1.4-0
data.table_1.9.4