Defining the subset of mutations in polymerase epsilon (POLE) associated with loss-of-proofreading (LOP) functionality

Mutation in the gene polymerase epsilon (POLE) has been associated with a hypermutant phenotype and improved prognosis, and more recently, investigated as a predictive biomarker for immunotherapy response.¹ However, there is currently ongoing confusion in the oncology community regarding which specific POLE mutations drive the hypermutant phenotype and immune response, with some authors arguing that this applies to all nonsynonymous mutations in the coding regions of POLE.² POLE is a unique DNA polymerase that in addition to its primary function as a polymerase also has the ability to proofread and remove mismatched base pairs via an exonuclease domain. Selective loss-of-proofreading (LOP) function leads to error-prone synthesis which drives elevated tumor mutational burden (TMB) with a characteristic mutational signature.³ Here we sought to end the controversy regarding which specific POLE-mutant alleles predict response to immune therapy by generating for the first time a comprehensive list of POLE-mutant alleles causing selective LOP function.

LOP POLE mutations are known to cause base substitution errors, resulting in a unique trinucleotide mutation signature which has been previously reported and experimentally confirmed.⁴ Using a manually curated list of 19 known LOP POLE mutations, we identified 66 tumors from The Cancer Genome Atlas and used the whole-exome sequencing (WES) to generate a gold standard LOP POLE mutational signature (Supplementary Figures S1 and S2A and Methods, available at https://doi.org/10.1016/j.annonc.2024.04.009). Screening a total of 1358 unique POLE-mutant alleles from a pan-cancer cohort of 46 332 patients profiled with WES, only 24 were identified as LOP, 13 alleles included in the positive LOP control signature and 11 additional LOP alleles (Supplementary Figure S2B, available at https://doi.org/10.1016/j.annonc.2024.04.009). Adding these 11 identified LOP alleles to 19 known LOP alleles, we create a comprehensive list of 30 unique POLE variants causing LOP, involving 17 unique codons (Table 1). As expected, most LOP mutations were in the exonuclease domain; however, there was also one LOP allele in the polymerase domain (S964Y) and one in a domain of unknown function (F1907L, Supplementary Figure S2C, available at https://doi.org/10.1016/j.annonc.2024.04.009). This was in sharp contrast to non-LOP POLE mutations which were distributed randomly throughout the POLE coding sequence (Figure S2D, Supplementary Table S1, available at https://doi.org/10.1016/j.annonc.2024.04.009). The most frequent LOP mutations were P286R (n = 47, 1.8% overall incidence, 30.9% of LOP POLE-mutant tumors) and V411L (n = 41, 1.5%, 27.0% of LOP POLE-mutant tumors). Tumors with LOP POLE mutation had remarkably high TMB with an average of 134.8 mut/Mb; this was similar between newly and previously identified variants (127.1 mut/Mb versus 164.4 mut/Mb, unpaired t-test P = 0.43) and significantly higher compared to non-LOP POLE mutations (134.8 mut/Mb versus 23.21 mut/Mb, P < 0.0001, Supplementary Figure S2E, available at https://doi.org/10.1016/j.annonc.2024.04.009).

Table 1.

Comprehensive list of LOP POLE mutations

Allele			Source	Incidence in this study	Pearson correlation with positive control	Average TMB (mut/Mb)	OncoKB annotation
D	275	H	Identified in this manuscript	1	0.87	15.3	Not reported
D	275	N	Identified in this manuscript	1	0.98	27.5	Not reported
P	286	H	Identified in this manuscript	1	0.96	69.6	Likely oncogenic
P	286	L	OncoKb/control signature	0	—	—	Likely oncogenic
P	286	R	OncoKb/control signature	47	0.95	201.2	Likely oncogenic
P	286	S	OncoKb/control signature	1	0.21a	76.8	Likely oncogenic
M	295	R	OncoKb/control signature	0	—	—	Likely oncogenic
S	297	F	OncoKb/control signature	8	0.97	169.6	Likely oncogenic
S	297	Y	OncoKb/control signature	1	0.94	62.8	Likely oncogenic
F	332	C	Identified in this manuscript	1	0.93	320.8	Not reported
F	367	C	Identified in this manuscript	2	0.95	159.8	Not reported
F	367	S	OncoKb/control signature	1	0.97	82.2	Likely oncogenic
F	367	V	OncoKb/control signature	0	—	—	Likely oncogenic
D	368	Y	OncoKb/control signature	0	—	—	Likely oncogenic
V	411	L	OncoKb/control signature	41	0.96	201.2	Likely oncogenic
V	411	M	OncoKb/control signature	0	—	—	Likely oncogenic
L	424	I	OncoKb/control signature	1	0.94	41.8	Likely oncogenic
L	424	V	OncoKb/control signature	3	0.51a	18.1	Likely oncogenic
P	436	S	Identified in this manuscript	6	0.79	86.6	Likely oncogenic
P	436	R	OncoKb/control signature	4	0.98	144.4	Likely oncogenic
M	444	K	OncoKb/control signature	3	0.96	262.8	Likely oncogenic
A	456	P	OncoKb/control signature	14	0.94	231.1	Likely oncogenic
Y	458	C	Identified in this manuscript	1	0.92	11.3	Not reported
S	459	del	OncoKb/control signature	0	—	—	Likely oncogenic
S	459	F	OncoKb/control signature	13	0.96	165.2	Likely oncogenic
S	459	Y	OncoKb/control signature	1	0.99	111.7	Likely oncogenic
S	461	P	Identified in this manuscript	2	0.98	160.8	Not reported
S	461	L	Identified in this manuscript	1	0.99	46.8	Not reported
S	964	Y	Identified in this manuscript	1	0.89	378.8	Not reported
F	1907	L	Identified in this manuscript	1	0.99	120.8	Not reported

Starting from the 19 literature-reported LOP POLE mutations that were used in the control signature, here are listed all POLE variants identified as causing LOP and their incidence, correlation with positive control signature, and average TMB in our cohort.

LOP, loss-of-proofreading; TMB, tumor mutational burden; VAF, variant allele frequency.

^aL424V and P286S had a VAF <0.16%, representing in our cohort a subclonal mutation.

Identifying the small subset of POLE mutations that cause LOP is critical to identify patients likely to respond to immunotherapy or appropriate for de-escalation of chemotherapy as has been recently proposed in endometrial cancer.⁵ The comprehensive list of 30 LOP and 1328 non-LOP POLE-mutant alleles provided here will allow for appropriate patient selection without needing to determine TMB, which requires sequencing a significant fraction of the genome and is not often possible in resource-limited settings.