I have 4 RNA-seq samples A,B,C,D in triplicate. They're aligned with STAR and counted with summarizeOverlaps, following the Bioconductor RNA-seq workflow. I'm comparing A vs B and C vs D (other comparisons later but these for now). I'm getting different and odd patterns in the MA-plot depending on whether I first subset the summarizedExperiment (se) into 2 sets (A+B, C+D) or keep everything together:

se1 <- se[,1:6]
se2 <- se[,7:12]

dds <- DESeqDataSet(se, design = ~sample)
dds1 <- DESeqDataSet(se1, design = ~sample)
dds2 <- DESeqDataSet(se2, design = ~sample)

dds <- DESeq(dds)
dds1 <- DESeq(dds1)
dds2 <- DESeq(dds2)

resultsa <- results(dds, contrast=c("sample","B","A"))
results1a <- results(dds1, contrast=c("sample","B","A"))

resultsb <- results(dds, contrast=c("sample","D","C"))
results2b <- results(dds2, contrast=c("sample","D","C"))

plotMA(resultsa, ylim = c(-5, 5))
plotMA(results1a, ylim = c(-5, 5))

plotMA(resultsb, ylim = c(-5, 5))
plotMA(results2b, ylim = c(-5, 5))

I'm wondering about the patterns on the left side of each image, I think the separate ones look normal but the together ones don't, is that correct? What could be causing the difference?

The difference you see is because the X axis is always the average abundance across all samples, not just the samples involved in the specific contrast you tested. Hence, on the right-hand plots, the X axis is the average of 2 groups, while on the left-hand plots, it is the average of all 4 groups. Since adding more samples into an average makes it more stable, analyzing all groups together mitigates some of the discreteness artifacts observed at low abundances. Note that this entire section of the plot is grey, indicating that none of these low-abundance genes have attained statistical significance. You may want to play around with DESeq2's fold change shrinking features, which should also somewhat smooth out these artifacts.