We analyzed all cases that underwent CGP testing at Foundation Medicine, Inc between January 2014 and August 2020. Formalin-fixed, paraffin-embedded (FFPE) tissue of either whole section samples, biopsies, or cytology specimens were received as paraffin blocks or unstained slides from outside institutions during routine clinical care. A board-certified pathologist assigned a diagnosis for each specimen based on microscopic examination of a H&E stained slide from the FFPE tissue, the accompanying pathology report, and additional information provided by the ordering physician.

CGP was performed on hybridization-captured, adaptor ligation-based libraries using DNA and/or RNA extracted from FFPE tumor in a Clinical Laboratory Improvement Amendments (CLIA)-certified and College of American Pathologists (CAP)-accredited laboratory (Foundation Medicine, Inc, Cambridge, Massachusetts, USA). The samples were sequenced for up to 406 cancer-related genes and select gene rearrangements.15 CD274 non-amplification SV mutations were defined as missense mutations, truncations, splice site mutations, and insertion/deletions, as previously described.15 CD274 amplification was defined as ploidy +4. Tumor mutational burden (TMB) was determined on up to 1.24 Mb of sequenced DNA and TMB ≥10 mutations/Mb (mut/Mb) was considered TMB-High per CDx approval.16 17 Microsatellite instability (MSI) was performed from DNA sequencing up to 114 loci and MSI-High (MSI-H) was considered positive per CDx approval.18 19 In addition, as research use only (RUO), ultraviolet mutational signatures were called as described by Zehir et al.20

For the purposes of this study, a subclonal SV mutation was defined as a sample where <50% of tumor cells were predicted to harbor the variant based on both the variant allele fraction and the pathologic and/or computational tumor cell purity estimates. A RUO somatic germline zygosity (SGZ) bioinformatics algorithm was used to determine whether mutations were likely somatic or germline, as previously described.21 We assessed CD274 missense mutation’s functionality with several in silico methods including SIFT, MutationTaster, fathmm-MKL, and MetaSVM and recalibrated the scores to a rankscore so they can be compared with each other.22–25 The rankscore was on a scale of 0 to 1 with 0 being predicted to be a non-functional protein and 1 being predicted to be a functional protein.

For a subset of cases, the PD-L1 DAKO 22C3 assay was run according to manufacturer instructions in a CLIA-certified and CAP-accredited laboratory (Foundation Medicine, Inc, Morrisville, North Carolina, USA).26 The IHC cases were interpreted by board-certified pathologists specifically trained on the DAKO tumor proportion score (TPS) method, where tumor cell expression of PD-L1 was quantified. The DAKO TPS scoring method was defined as TPS=# PD-L1 positive tumor cells/(total # of PD-L1 positive + PD-L1 negative tumor cells).27

Overall, the frequency of CD274 SV mutations was low (0.3%, 1081/314,631) in our cohort of 314,631 samples. A total of 577 unique variants were discovered; some mutations were recurrent, while others occurred only once in the entire cohort (table 1). Of the 1081 mutations, 49.9% (539/1081) were from metastatic specimens and 42.3% (457/1081) were from primary specimens. In 7.8% (85/1081) of cases it was unknown whether it was a primary or metastatic specimen. Of the 1081 samples with CD274 SV mutations, only 1.4% (15/1081) had cooccurring CD274 amplification.

The most common CD274 mutations were R260H (n=51), R260C (n=18), R125Q (n=12), C272fs*13 (n=11), R86W (n=10), and R113H (n=10) (table 1, figure 1). R260C/H was the most frequent recurrent missense mutation; both substitutions have been observed in the germline of healthy subjects (gnomAD). Based on the SGZ algorithm, we found that 29.0% (20/69) of the codon R260 mutations were likely somatic and 55.1% (38/69) were likely germline. The algorithm could not predict whether the variant was germline or somatic in 16.0% (11/69) of the samples. In addition, when we examined all the missense mutations (n=974), we saw that 51.0% (497/974) were predicted to be somatic, 24.7% (241/974) were predicted to be germline, and the algorithm could not predict whether the variant was germline or somatic in 24.2% (236/974) of the cases (online supplemental table 1). In addition, C272fs*13 is at an indel at a poly-A homopolymer, a sequence context that is highly mutable in the setting of MSI-H status. This mutation was significantly enriched in the MSI-H group (0.01%, 5/5139) when compared with the non-MSI-H group (0.002%, 6/309,492) (Fisher’s exact test, p<0.0001), suggesting that the variant is often a result of mismatch repair protein deficiency. This finding is reflected in the high CD274 SV mutation frequency in non-serous endometrial adenocarcinomas in this study.

The types of mutations in this cohort also varied, with missense mutations being the most common (83.8%, 906/1081) and insertion/deletions being less common (0.8%, 9/1081) (table 2). Multiple samples had complex CD274 mutations, defined as more than one CD274 genomic alteration observed in the sample. The most common type of complex CD274 mutation was alterations with two missense mutations (1.9%, 21/1081), while other complex mutations consisted of a CD274 mutation with concurrent CD274 amplification (1.4%, 15/1081) and/or rearrangement (0.3%, 3/1081).

The prevalence of CD274 mutations also varied depending on tumor type. The top five tumor types (minimum 800 total samples) with the highest rates of CD274 mutations in descending order were: diffuse large B-cell lymphoma (1.9%, 19/997), cutaneous squamous cell carcinoma (1.6%, 14/868), endometrial adenocarcinoma (1.0%, 36/3740), unknown primary melanoma (0.9%, 33/3679), and cutaneous melanoma (0.8%, 32/3874) (figure 2, (online supplemental table 2). Interestingly, three of the five tumor types with the highest prevalence of CD274 mutations usually occur on the skin. When we examined the mean TMB for these cases, we saw very high mean TMB and median TMB (cutaneous squamous cell carcinoma (151 mut/Mb, 100 mut/Mb), cutaneous melanoma (133 mut/Mb, 126 mut/Mb), unknown primary melanoma (125 mut/Mb, 92 mut/Mb), respectively), suggesting that these were likely caused by ultraviolet exposure-induced hypermutation. This was further supported by the high prevalence of ultraviolet mutational signature in these tumor types (cutaneous squamous cell carcinoma (84.6%, 11/13), cutaneous melanoma (93.8%, 30/32), and unknown primary melanoma (100%, 32/32). Whether these CD274 mutations are random ‘passenger’ mutations due to the high mutation rate should be further investigated.

Of the 1081 cases with CD274 mutations, 19.7% (213/1081) cases had PD-L1 IHC data; and of the 313,550 cases without CD274 mutations, 18.6% (58,218/313,550) cases had PD-L1 IHC data.

Most of the CD274 non-truncating mutations had low to no tumor-cell expression of PD-L1 (figure 3A, online supplemental table 3). Of the 11 R260H cases concurrently tested with PD-L1 IHC, 81.8% (9/11) had no PD-L1 expression and 2 (18.2%) cases had low PD-L1 expression. Of the 5 E237K cases, 20% (1/5) had no PD-L1 expression, 40% (2/5) had low PD-L1 expression, and 40% (2/5) had high PD-L1 expression (figure 3B). This difference in protein expression of the two mutations was significantly different (Fisher’s exact test, p=0.036). When we compared the PD-L1 protein expression of cases with CD274 missense mutations (n=153) and cases without CD274 mutations (n=58 218), we saw a lower level of PD-L1 IHC expression in the cases with CD274 mutations, though the difference was not statistically significant (mean: TPS=11 vs 13, respectively; analysis of variance (ANOVA), p=0.404). When examining the clonal (n=153) versus subclonal (n=28) missense mutations, we saw significantly lower PD-L1 IHC expression in the clonal missense mutation (mean: TPS=11 vs 22, respectively; ANOVA, p=0.049) (figure 3C).

In addition, we examined the predicted functionality of each CD274 missense mutation with multiple functionality prediction models including SIFT, MutationTaster, fathmm-MKL, and MetaSVM (online supplemental table 4). However, the predicted functionality did not have any significant correlation with PD-L1 IHC expression (online supplemental figure 1).

Thirty-nine putative truncating variants with concurrent PD-L1 IHC testing were identified, including 12 nonsense mutations, 10 frameshift indels, and 7 canonical splice variants (figure 4A, online supplemental table 2). When we compared the PD-L1 protein expression of cases with CD274 clonal truncating variant (nonsense or frameshift indel) (n=18) and cases without CD274 mutations (n=58,218), we saw a lower level of PD-L1 expression in the cases with CD274 mutations (mean: TPS=1 vs 13, respectively; ANOVA, p=0.069). In addition, we saw a significantly lower level of PD-L1 expression in samples with a clonal truncating variant (nonsense or frameshift indel) when compared with samples with subclonal truncating variants (mean: TPS=1 vs TPS=38; ANOVA, p<0.001) (figure 4B).