Base composition vs cycle, MiSeq vs HiSeq X

[JBKri]

I have noticed that in our MiSeq runs we seems to always have an abrupt change in base composition in the last one or two cycles. The proportion of A goes to zero. Also I noticed we have a cyclic pattern in base proportions in the rest of the cycles, except the first 10 or so where there seems to be some other bias going on.
On the other hand, the data from a HiSeq X run shows no such cyclic variation, nor the sudden drop in As at the end.
See this image for base composition comparisons

Does anyone have an idea of what is causing these patterns in the MiSeq, and why they are not seen in the HiSeq run?
Note that these were two different libraries from different organisms. Both were TruSeq DNA PCR-free with Covaris ultrasonic shearing.
Jon

[JBKri]

On further investigation I see that the pattern varies depending on the number of cycles in the sequencing kit, and also some variation among runs. It seems 2*150 cycles causes little or no sudden change in base proportion in the last cycle. A few 2*250 runs showed some change in bases proportions, while the 2*300 cycle runs typically show the sharp change in the last cycle. Now the quetion is, should I trim off that last base? I have noticed the pattern can remain also after doing quality trimming.

[GenoMax]

You could run "n+1" cycles and then trim the last base, if you need that last base. You may want to chat with Illumina support to see if they have an explanation.

[moatman]

Does this relate to the first base(s) of the adapter in your sample sheet? I've run into that problem before, where adapter trimming will eat the last little bit, even if it happens to just be the first base of the adapter, in this case maybe your missing A's.