Firefly Luciferase Gene: Structure and Expression in Mammalian Cells

Jeffrey R. De Wet, Keith V. Wood, Marlene DeLuca, Donald R. Helinski, Suresh Subramani

Mol. Cell. Biol., Vol. 7, No. 2 (Feb. 1987), pp. 725–737

Continuous production beats flash kinetics. With excess exogenous luciferin you get a bright flash that decays to 10% within a minute. But your endogenous pathway would supply luciferin at a steady trickle.
Single-copy gene. Luciferase is one gene in the P. pyralis genome, not a family.
Peroxisomal targeting was already visible. They saw punctate staining in mammalian cells and didn't fully understand it yet. The luciferase has the C-terminal -SKL signal and will go to peroxisomes in plant cells too. Each cell has dozens to hundreds of peroxisomes, so you get diffuse glow from distributed reaction sites rather than single bright spots. That's what you want.
Use the cDNA, not the genomic gene. Firefly luciferase introns are tiny (<60 bp each) and don't splice in heterologous systems. Your Addgene plasmid already has the intronless cDNA version. If you ever needed to clone another firefly gene yourself (e.g. directly from genomic DNA rather than from an existing plasmid), you'd want to either use cDNA or codon-optimize a synthetic version rather than using the raw genomic sequence. Firefly introns are small but they won't splice properly in plant cells.
Intracellular ATP is sufficient (re-confirmed). Adding exogenous ATP and Mg²⁺ to intact cells didn't increase signal. The cell's own pools are enough. Don't worry about ATP availability in peroxisomes — they import it for β-oxidation anyway.
The core argument for your system choice is right here. Bacterial luciferase needs two subunits from two genes plus three more genes for substrate regeneration. Firefly luciferase is one polypeptide. The tradeoff: the bacterial system is genetically complex but self-contained for substrate. The firefly system is enzymatically simple but requires you to solve substrate supply separately.
Everything traces back to this sequence. Their original 1985 cDNA clone was incomplete (missing the first six N-terminal codons). This paper fixed it with the full genomic sequence. Every downstream variant (Luc2, codon-optimized versions, your Addgene plasmid) descends from the complete sequence established here.