Crossing enhanced and high fidelity SpCas9 nucleases to optimize specificity and cleavage

Extending an sgRNA spacer to 21 nucleotides with a 5′ G interferes with the activity of increased-fidelity SpCas9 variants, and counterintuitively a matching G extension is far more detrimental than a mismatching one. Measured across 100 sgRNAs by EGFP disruption in N2a.EGFP cells, this effect applies to eSpCas9 and SpCas9-HF1 and explains their apparently poor performance whenever guides carry the 5′ G extension commonly added to satisfy the human U6 promoter's preference for a G transcription initiation site.

None of the routine workarounds for the U6 promoter's 5′ G requirement are compatible with increased-fidelity SpCas9 variants: changing a non-G 5′ nucleotide to G, using the spacer unaltered with its non-G 5′ end, and truncating the spacer back to the next G (yielding 17 to 19 nucleotides) all reduce activity, though to different extents, with eSpCas9 less sensitive than SpCas9-HF1. eSpCas9, SpCas9-HF1 and HeFSpCas9 can therefore routinely be used only with perfectly matching 20-nucleotide spacers.

Increased-fidelity SpCas9 variants trade breadth of usable targets for specificity, in a consistent order. With perfectly matching 20-nucleotide spacers across 24 targets in an EGFP disruption assay, eSpCas9 retains more than 80% of wild-type SpCas9 activity at 83% of targets and SpCas9-HF1 at 41%, while HeFSpCas9 is active on only a subset. Target selectivity ranks eSpCas9 as least discriminative, then SpCas9-HF1, then HeFSpCas9 as most discriminative — yet each variant can match wild-type activity undiminished on particular individual targets.

SpCas9-HF1 has higher fidelity than eSpCas9 against single-base mismatches. Testing 16 targets with 144 mismatching sgRNAs carrying mismatches at PAM-distal positions 19 to 14, where SpCas9 tolerates mismatches most, wild-type SpCas9 barely distinguishes perfect matches from mismatches at most targets; eSpCas9 shows no detectable cleavage (below 3%) at 20 of 48 mismatch positions while SpCas9-HF1 shows none at 28 of 47. Counting spacers whose mismatched guides still give more than 80% of matching-guide disruption gives 42 for wild-type SpCas9, 18 for eSpCas9 and only three for SpCas9-HF1, and deep sequencing of 67 samples confirms SpCas9-HF1's advantage with more than twofold lower off-target cleavage at 7 of 14 mismatch-position guides.

HeFSpCas9, carrying seven mutations drawn from both eSpCas9 and SpCas9-HF1, is most specific precisely on the targets where the two parent variants fail. On targets for which even SpCas9-HF1 shows considerable off-target cleavage, HeFSpCas9 gives detectable off-target activity at only 4 of 12 mismatching positions by deep sequencing, with the rest indistinguishable from background — a roughly 50- to 400-fold specificity gain over either parent. Overall fidelity ranks eSpCas9 below SpCas9-HF1 below HeFSpCas9.

Targets themselves can be ranked by which nuclease they require, not just nucleases by fidelity. A target cleaved efficiently by one nuclease with both matching and mismatching guides tends to be cleaved efficiently by a higher-fidelity nuclease with matching guides but much less so with mismatching ones. On the hardest targets, eSpCas9 and SpCas9-HF1 retain on average 93% and 60% of their on-target activity when given single-mismatch guides, against only 26% and 6% for the remaining 12 of 16 targets — so no single variant is generally superior and each target must be matched to an appropriate nuclease.

The nuclease and target rankings are intrinsic properties rather than artefacts of one cell line: repeating the experiment with nine targets of varying rank and 81 single-base mismatching guides in HEK-293.EGFP cells, which carry a single integrated EGFP copy, gives patterns almost identical to those in N2a.EGFP cells.

Differences in on-target activity and fidelity between eSpCas9 and SpCas9-HF1 are primarily intrinsic to the proteins rather than consequences of expression level, shown by comparing activities as a function of matched expression levels — though somewhat lower SpCas9-HF1 expression from these vectors cannot be excluded as a contributor to its lower on-target activity and apparently higher fidelity.

How a given target responds to one kind of imperfect guide predicts how it responds to others, implying a shared mechanism. Disruption activities of eSpCas9 and SpCas9-HF1 with single-mismatch sgRNAs correlate positively and significantly with their activities using 5′ mismatching-G-extended spacers at the same sites (Pearson 0.83 and 0.78), and activities at different mismatch positions within the same target also correlate (0.73 to 0.93 for eSpCas9 and 0.73 to 0.97 for SpCas9-HF1).

A nuclease with fidelity between SpCas9-HF1 and HeFSpCas9 can be built by reverting a single mutation out of HeFSpCas9's seven, which fills the gap for targets too off-target-prone for SpCas9-HF1 but not cleaved by HeFSpCas9. Reverting the eSpCas9 mutation K1003A gives HeFm2SpCas9, with higher on-target activity than HeFSpCas9 at 11 of 17 such targets and only slightly higher off-target activity at 2 of 5, and higher specificity than SpCas9-HF1 on the five targets where it reaches more than 60% of wild-type activity. Reverting the SpCas9-HF1 mutation N497A instead gives HeFm1SpCas9, which behaves much like HeFSpCas9.

At FANCF site 2 off-target site 1 — an off-target bearing a PAM-proximal mismatch that SpCas9-HF1 cleaves efficiently — all HeF variants produce only background-level cleavage measured by TIDE, while differing in on-target activity. HeFm2SpCas9 gives the highest on-target activity of the HeF variants, comparable to wild-type SpCas9, making it the practical choice for this class of target.

The fidelity mutations in these SpCas9 variants impair cleavage without impairing DNA binding. Although increased-fidelity nucleases show no cleavage with sgRNAs truncated by more than two nucleotides, eSpCas9, SpCas9-HF1 and HeFSpCas9 charged with 14-nucleotide spacers plus a non-matching 5′ G at the Prnp promoter driving an EGFP cassette give 15- to 20-fold transcriptional activation, matching their catalytically dead counterparts. Electrophoretic mobility shift assays confirm this directly: HeF nucleases retain most DNA-binding ability at three targets they cannot cleave, and eSpCas9 and SpCas9-HF1 charged with 21-nucleotide sgRNAs still bind targets whose cleavage the matching G extension abolishes.