(A) The CRISPR-Cas12a operon consists of Cas genes and a CRISPR array. (B) Each gRNA consists of a repeat and a spacer. Pre-processing repeats contain a ~16–18 nt fragment, here denoted CRISPR separator, which gets excised by Cas12a and an unknown enzyme. (C) The separator has previously been omitted when expressing Cas12a arrays in mammalian cells. We asked if the separator serves to insulate gRNAs from the negative influence of secondary structure in spacers. (D) We designed CRISPR arrays consisting of two gRNAs, the first with a non-targeting dummy spacer, and the second targeting the promoter of GFP, genomically integrated in HEK293T cells. (E) Experimental setup; Lb-dCas12a-miniVPR was used to activate GFP, and GFP fluorescence was analyzed as a measure of array performance. (F) CRISPR arrays can display hypersensitivity to the composition of the dummy spacer. In extreme cases, replacing the last nucleotide from T to G can lead to almost complete abrogation of GFP activation. (G) A library of 51 CRISPR arrays, each with a dummy spacer of different GC content. A strong negative correlation is seen between the GC content of the dummy spacer and GFP fluorescence. Each dot represents one of the 51 CRISPR arrays (average of three replicates). Arrays were divided into three groups based on the level of GFP fluorescence they enabled. Boxes indicate two groups that were analyzed in (I and J). (H–J) For each group, the average GC content of a sliding 5-nt window was calculated. The best-performing arrays were the ones where the dummy spacer happened to have low GC content at its 3’ end (H). Some arrays showed unexpectedly high or low GFP activity for the GC content of their dummy spacers (G). These arrays contain low (I) or high (J) GC content at the very 3’ end of their dummy spacers, suggesting that the GC content of the last few bases is an important predictor of array performance. Shaded regions in (H–J) represent standard error. (K) The predictive power of knowing the GC content of 3-nt regions in the dummy gRNA (Materials and methods). Merely knowing the GC content of the las three bases is more predictive than knowing the overall GC content. (L) A plot showing the relationship between GC content of the 51 dummy spacers and the secondary structures they are predicted to form with the GFP-targeting gRNA (the larger the value on the y-axis, the more stable the predicted secondary structure). (M) This predicted secondary structure formation is anticorrelated with performance of the GFP-targeting spacer, suggesting that strong secondary structures is what impedes array performance.