In a multiple sequence alignment, the algorithms will attempt to align the sequences along their length (global alignment). What you need is a local alignment approach with a high penalty for base mismatch.
The E-INS-i algorithm in MAFFT might provide the desired functionality. Select the Advanced Settings and there:
Strategy -> E-INS-i algorithm
What you want to perform is commonly called a multiple sequence alignment. As @Wayne_Yux said, the first step is to put all of your protein sequences in a single fasta file. You can then use one of several online tools to apply different alignment algorithms to your protein sequence set.
A popular sequence alignment algorithm is Clustal, which ...
You can store the 40 sequences in a fasta file and then use blastp to align them all at once to your reference sequence.
After that, you can inspect the alignment hits and see wether they fit your quality expectations
I imagine that for case 1 you have data of the type:
Plant1, $t_f$, $t_l$;
Plant2, $t_f$, $t_l$;
If you are interested in comparing the mean temperatures between flowers and leaves, your data would consist of n measurements coming from flowers and n measurements from leaves. You may surely try a t-test for independent samples, especially if you have ...