Assessing the Genomic Landscape of Cervical Cancers: Clinical Opportunities and Therapeutic Targets

Cervical cancer remains a prevalent cancer in women in the United States, with an estimated 13,960 newly diagnosed cases and 4,310 deaths expected in 2023 (1). There are limited treatment options for patients with locally advanced and metastatic cervical cancer, with platinum-based combination chemotherapy the mainstay of systemic treatment (2). In patients with previously treated, programmed death-ligand 1 (PD-L1)-positive disease, the objective response rate (ORR) with pembrolizumab monotherapy is only 14.6% (3). The addition of pembrolizumab to frontline chemotherapy, with or without bevacizumab, results in a modest improvement in median progression-free survival (PFS; 10.4 months compared with 8.2 months for patients treated with chemotherapy alone; ref. 4). Novel treatment approaches are urgently needed for patients with recurrent and metastatic cervical cancer.

The World Health Organization classification system recognizes several histologic subtypes of cervical cancer (5). Squamous cell carcinoma (SCC) is the predominant histologic subtype, accounting for approximately 75% of all cervical cancers. Endocervical adenocarcinoma and adenosquamous cell carcinoma account for 10%–15%, and the remaining 10%–15% are of “other” or unspecified histology (6, 7). On the basis of clinical data from the Surveillance, Epidemiology, and End Results (SEER) database, small cell carcinoma and adenocarcinoma are the histologic subtypes associated with the poorest survival (8). Compared with other common solid tumors, the predictive and prognostic value of tumor genomic data in cervical cancer remains poorly defined. The Cancer Genome Atlas (TCGA) identified 14 genes significantly mutated in cervical cancer, including PIK3CA, ARID1A, and KRAS, as well as an APOBEC (apolipoprotein B mRNA-editing enzyme, catalytic polypeptide) mutagenesis pattern (9). Although an important milestone, tumor samples included in TCGA analyses were predominantly collected from patients with early-stage SCC (n = 144), with few tumors exhibiting adenosquamous (n = 3) or adenocarcinoma (n = 31) histology. Other previously analyzed cohorts included 115 women from Norway and Mexico (10), as well as 118 women from Uganda (11); these cohorts also predominantly had SCC. Consequently, there are limited data on the genomic landscape of non-squamous cervical carcinomas, including gastric-type endocervical adenocarcinoma (GEA) and small cell carcinomas of the cervix. Moreover, no study has evaluated the feasibility of prospective molecular characterization of tumors and matched normal specimens in patients with advanced cervical cancer and the potential utility of tumor genomic profiling to guide clinical care.

Here, we sought to leverage a prospectively generated, real-world dataset of genomically profiled cervical cancers to define the frequency of clinical actionable genomic alterations and to determine whether prospective molecular characterization of primary and recurrent cervical cancers might reveal histology-specific differences and guide enrollment of patients onto therapeutic clinical trials.

Patients provided written informed consent to a prospective tumor sequencing study under an Institutional Review Board–approved protocol (ClinicalTrials.gov, NCT01775072) from February 2014 through May 2019. Patients were initially eligible for sequencing if they had metastatic or recurrent disease; eligibility was expanded in October 2017, through a philanthropic grant, to include all patients with cervical cancer, regardless of stage or prior treatment. Detailed patient demographic and disease-specific clinical data, including clinical trial enrollment and treatment, were collected. All cases underwent pathologic review at our institution by an expert gynecologic pathologist. This study was conducted in accordance with the Declaration of Helsinki.

Next-generation sequencing (NGS) was performed in the Clinical Laboratory Improvement Amendments–certified Memorial Sloan Kettering Cancer Center (MSK) Molecular Diagnostics Service Laboratory using DNA extracted from tumor and matched normal DNA from blood. Tumor samples could either be primary or metastatic, depending on tissue availability. Tumor and paired germline DNA were analyzed using the MSK-IMPACT (MSK – Integrated Mutation Profiling of Actionable Cancer Targets) assay, an exon capture assay targeting all coding exons of 341 (n = 14), 410 (n = 33), or 468 (n = 130) cancer-associated genes, as described previously (12). DNA was sequenced to an average of 648× (range, 52×–1,191×) sequence coverage. All variants were reviewed by a board-certified molecular pathologist prior to signing out the results into the medical record. All patient-level clinical and genomic data are available via cBioPortal for Cancer Genomics (www.cbioportal.org); genomic data are also publicly accessible prepublication via AACR GENIE (American Association for Cancer Research Genomics Evidence Neoplasia Information Exchange; ref. 13). Signature proportions were calculated using maximum likelihood method employed by tempoSig (https://github.com/mskcc/temposig). OncoKB (www.oncokb.org), a FDA-recognized tumor mutation database and treatment classification system, was used to determine the clinical actionability of individual genomic variants (14). In brief, OncoKB annotates the level of evidence supporting use of a drug based on the therapeutic level of evidence (Supplementary Fig. S1). OncoKB annotations were applied using version 3.1.4 on July 22, 2022.

The presence of microsatellite instability (MSI) was assessed using MSISensor, version 0.2 (15). MSISensor assigns a numeric score based on the percentage of unstable microsatellite sites divided by the total number of microsatellite sites tested from aligned sequencing data. On the basis of prior clinical validation of MSISensor with MSK-IMPACT, MSI status is defined on the basis of the following scores: <3%: microsatellite stable (MSS); |$ \ge $|3 and <10%: microsatellite (MS)-indeterminate; and |$ \ge $|10%: MSI-high (MSI-H; ref. 16).

High-risk human papillomavirus (HR-HPV) status for each tumor was determined on the basis of either positivity of p16 IHC as diffuse, strong and continuous, nuclear and cytoplasmic staining (17); and/or HR-HPV RNA ISH based on the presence of dark-brown, dot-like cytoplasmic and/or nuclear positivity (18); or by detection via NGS, as described below. Patients with a history of HR-HPV documented on a prior Papanicolaou test were also classified as HPV positive for the purposes of this study, as were all SCCs (19).

Sequencing reads from MSK-IMPACT testing were filtered to analyze only reads that did not align with the human genome. The non-human reads were then reanalyzed with the “blastn” algorithm utilizing the National Cancer for Biotechnology Information Nucleotide (NCBI NT) Database. Paired-end reads that mapped with greater than 90% identity to an entry in the database were labeled as a single read present for the database entry. The database entry was cross-walked to the NCBI Taxonomy Database, and the read quantity was calculated as the number of reads that mapped to the HPV subtype (20, 21).

We assessed the enrichment of genomic alterations across histology using the Fisher exact or χ² test (where appropriate), and nominal P values are specified. Comparisons of gene mutation prevalence across cohorts (this study compared with TCGA) and across histologic subtypes were performed. For patients enrolled on clinical trials, ORR was assessed using RECIST v1.1 (22), and PFS was defined as the period from the start date of treatment until investigator-determined date of progression (by RECIST V.1.1) or death, whichever occurred first. Overall survival (OS) analysis was performed on the subset of patients who presented to our center at the time of initial diagnosis (MSK survival cohort). OS was defined as the period from the date of consent for MSK-IMPACT analysis until the date of death. To account for the method of sample selection (selection bias), that is, because patients must have survived long enough to sign consent for MSK-IMPACT testing, the Kaplan–Meier method with left truncation was used to estimate the median survival for the survival cohort (n = 97). The Cox proportional hazards model with left truncation at the date of MSK-IMPACT consent was applied to obtain P values (23). PFS for patients enrolled in a clinical trial was analyzed using the Kaplan–Meier method to estimate values for medians and two-sided 95% confidence interval (CI). Cox proportional hazards analyses were calculated using the R survival package.

Publicly available data generated by others were used by the authors. The clinical and genomic data for TCGA cohort were obtained from https://www.cbioportal.org/study/summary?id=cesc_tcga_pan_can_atlas_2018. Somatic mutational data and associated clinical data are available via the cBioPortal for Cancer Genomics (https://www.cbioportal.org/study/summary?id=cervix_msk_2023).

To define the clinical utility of prospective tumor genomic profiling, we integrated clinical data from 177 patients with cervical cancer treated at our institution with tumor genomic data generated within the context of a prospective tumor sequencing initiative (MSK Cervical Cancer Cohort). Patient demographic data are summarized in Table 1. Eighty percent of the tumors (n = 142) in the MSK Cervical Cancer Cohort were positive for high-risk HPV. The most common histologic subtypes were SCC (n = 69, 39%), usual endocervical adenocarcinoma (UEA; n = 50, 28%), and cervical adenosquamous carcinoma (CAS; n = 21, 12%), all of which are classically HPV associated (24). Rarer histologic subtypes, including GEA, accounted for 21% (n = 37) of the cohort (Fig. 1). Of note, these less-prevalent histologic subtypes were not studied in the cervical cancer–specific TCGA analysis (Fig. 1A).

Thirty-five percent (n = 63) of the patients in the MSK cohort were initially diagnosed with stage III (n = 29), IVA (n = 11) or IVB (n = 23) disease, compared with 19% (n = 63) in TCGA cohort (P = 0.008; Supplementary Fig. S2). In addition, rather than only focusing on primary disease, as TCGA did (9), 40% (n = 70) of the tumor samples subjected to MSK-IMPACT sequencing in our study were obtained from distant sites, most commonly the lung (n = 14) and lymph nodes (n = 12; Fig. 1A and B). MSK-IMPACT sequencing could be performed on either primary or metastatic tumor samples, depending on tissue availability. Of the 108 patients who had MSK-IMPACT sequencing performed on their primary tumor, at the time of database lock, 40 (37%) were alive and without evidence of disease, 32 (30%) were alive with disease, and 34 (31%) were dead. When comparing sequencing results obtained from primary versus distant sites, no significant differences in genomic drivers were identified.

To minimize the impact of referral bias, OS was calculated for the subset of patients who presented to our center at the time of initial diagnosis (MSK survival cohort). The median OS for this cohort (n = 97) was 59.2 months (95% CI: 46.7–89.4) at a median follow-up of 50.9 months (range, 3.9–176.4 months; Fig. 1C). There was no significant difference among histologic subtypes when comparing SCC versus UEA versus other subtypes (P = 0.099). There was, however, a significant OS difference when comparing stage at diagnosis [stage I, 106.8 months [95% CI: 52.3 not estimable (NE)]; stage II, 59.2 months (95% CI: 23–89.4); stage III, 57.5 months (95% CI: 10.1–NE); and stage IV, 18 months (95% CI: 7.5–66.7); P = 0.027].

We then explored whether clinical differences in the MSK Cervical Cancer Cohort compared with TCGA cohort were accompanied by differences in genomic profiles. Overall, the distribution of somatic alterations affecting cancer-related genes between the two cohorts was similar, with a few notable differences. More specifically, pathogenic somatic mutations in TP53, KRAS, and ERBB2 were enriched in the MSK Cervical Cancer Cohort compared with TCGA cohort, although only KRAS reached statistical significance [TP53: 11% (19/177) vs. 8% (25/297), P = 0.51; KRAS: 12% (21/177) vs. 5% (15/297), P = 0.019; ERBB2: 12% (22/177) vs. 8% (24/297), P = 0.2; PIK3CA: 25% (44/177) vs. 37% (111/298), P = 0.052; and PTEN: 7% (12/177) vs. 11% (34/297), P = 0.14, respectively]. These differences were largely attributable to the variability in histologic subtypes within these cohorts, including the inclusion of GEA tumors, as opposed to the narrower focus of TCGA cohort. Compared with the MSK Pan-Cancer Cohort, recurrent hotspot mutations were observed in the MSK Cervical Cancer Cohort (Supplementary Fig. S3). No novel mutations associated with cervical cancer were detected (Supplementary Fig. S4).

The most prevalent genomic alterations were mutations or amplifications in PIK3CA (44/177, 25%), ERBB2 (22/177, 12%), KMT2C (17/177, 10%), KMT2D (16/177, 9%), and KRAS (21/177, 12%; Fig. 2A). Consistent with the published literature, an APOBEC mutational signature was detected in 46% of the SCCs, 30% of the UEAs, 9% of the GEAs, and 33% of the CASs (Fig. 2A). There were notable differences in the prevalence of genomic drivers among histologic subtypes. In the adenosquamous carcinoma samples, mutations in TP53, ARD1A, and ERBB2 were completely absent, in contrast to STK11 alterations, which were more prevalent in CAS (n = 7/21, 33%) compared with SCC (n = 3/69, 4%) and UEA (n = 5/50, 10%) tumors. ERBB2 amplification and oncogenic missense mutations were common in both UEA (amps = 2/50, muts = 12/50, 28%) and SCC (amps = 4/69, muts = 2/69, 9%). Notably, ERBB2 and KRAS alterations were mutually exclusive in these histologic subtypes. MYC amplifications (n = 2/50, 4%) were seen only in UEA (Fig. 2A and B).

There have been previous reports of mutual exclusivity between oncogenic TP53 variants and HPV positivity (25, 26); this was also observed in our cohort. Notwithstanding the high rates of oncogenic TP53 alterations in the in non–HPV-associated GEA, oncogenic TP53/HPV-positive mutual exclusivity was also observed in other subtypes with only 3.5% (5/142) HPV-positive cases harboring oncogenic TP53 alterations (four SCC and one small cell carcinoma of the cervix). This contrasts with 53.8% (7/13) of non-GEA, HPV-negative cases with concurrent oncogenic mutations in TP53 (P = 0.00007).

When comparing the sequencing results from primary (P) (n = 107) versus metastatic (M) (n = 70) samples in our cohort, we observed enrichment of oncogenic mutations in metastatic tumors in genes including PIK3CA (P 23% vs. M 27%), ERBB2 (P 12% vs. M 13%), TP53 (P 8% vs. M 14%), KRAS (P 8% vs. M 17%), KMT2C (P 7% vs. M 13%), and KMT2D (P 7% vs. M 14%; Supplementary Fig. S5A). Despite the observed differences, none reached statistical significance. We then compared the sequencing results from patients diagnosed with stage I–IVA (n = 154) versus stage IVB (n = 23) cancers. Oncogenic mutations in genes including PIK3CA (I–IVA 22% vs. IVB 43%) and KMT2C (1 = I–IVA 8% vs. IVB 22%) were enriched in stage IVB patients. In contrast, the rate of oncogenic mutations in genes including KRAS (I–IVA 12% vs. IVB 9%) and TP53 (I–IVA 11% vs. IVB 9%) were consistent across stages (Supplementary Fig. S5B). Despite these observed trends, none of the trends reach statistical significance and may have been influenced by the prevalence of histologic subtypes by stage.

Multiple diagnostic methods were used to characterize HPV positivity. HPV ISH was performed on 58 tumor samples, 53 of which were positive, including 23 SCCs, 17 UEAs, and 10 CASs. NGS data were also queried for the presence of HPV DNA. To understand the relevance of the NGS HPV calls, we calculated the precision and recall (i.e., positive predictive value and sensitivity) of the NGS calls for those samples with HPV ISH, across all subtypes. In total, 68% (36/53) of ISH-positive cases had HPV reads positively identified by MSK-IMPACT off target reads. Of the five negative cases by ISH, all were negative for HPV reads, resulting in 100% precision (Supplementary Table S1A). Furthermore, 66 SCC cases (all considered HPV positive clinically) were tested for the presence of HPV reads via NGS, with a recall rate of 71% (47/66). CASs and UEAs had recall rates of 57% (8/14) and 72% (26/36), respectively, when compared with the clinical (non-NGS) HPV testing and 100% precision (Supplementary Table S1B–S1D). When looking at the most common HPV-associated histologic subtypes (SCC, UEA, CAS) that were HPV positive by NGS, including samples without orthogonal clinical testing, HPV-16 and -18 were identified in 52% (51/98) and 32% (32/98) of samples, respectively. Other high-risk HPV subtypes were identified in an additional 15% of tumors (15/98); one tumor was positive for HPV-53, which is considered intermediate risk (Supplementary Table S1E).

In contrast to TCGA, the MSK Cervical Cancer Cohort included 22 patients with GEA, a rare histologic subtype of cervical adenocarcinoma that is not HPV associated. Among the 22 GEAs, 12 (55%) harbored somatic mutations in TP53, 6 (27%) in KRAS, 3 (14%) alterations in ERBB2 (amps = 1, muts = 2), and 3 (14%) in STK11. The RAF/RAS and PI3K/AKT pathways were altered in 10 (45%) and four (18%) GEA tumors, respectively. All the GEA cases were negative for high-risk HPV by MSK-IMPACT or HPV-ISH (if tested). Forty-one percent were stage I at diagnosis, 23% were stage II, 23% were stage III, and 14% were stage IV (IVA 5%, IVB 9%). The MSK Cervical Cancer Cohort also included 15 patients with other rare histologic subtypes, including mesonephric carcinoma (n = 3), cervical clear cell carcinoma (n = 3), small cell carcinoma of the cervix (n = 5), mixed neuroendocrine histology (n = 3; 2 mixed adenocarcinoma and neuroendocrine, 1 with mixed mesonephric and small cell carcinoma), and endometrioid cervical carcinoma (n = 1). Of the 3 patients with mesonephric carcinoma, 2 had KRAS alterations and 2 had ARID1A mutations. All 3 were diagnosed with stage IBI disease, all were HPV negative, and 2 of the 3 are alive with disease. In patients with pure small cell carcinoma, 2 of 5 had concurrent TP53 mutations and MYC amplifications (Supplementary Fig. S6). Hotspot alterations in ERBB2 and STK11, which were common in the other subtypes, were absent in this rare tumor subgroup. All patients with either pure small cell or mixed neuroendocrine histology were either alive with disease or dead of disease at the time of database lock; 5 out of the 8 patients (63%) were positive for HPV by IMPACT or HPV-ISH.

There are currently no FDA-approved targeted therapies specifically for cervical cancer. Kinase inhibitors selective for NTRK and RET fusions and BRAF V600E mutations have received tumor-agnostic FDA approvals, but neither oncogenic NTRK or RET fusions nor BRAF V600E mutations were identified in our tumor samples. In the MSK Cervical Cancer Cohort, 23 (13%) of 177 patients had tumors that were MSI-H (n = 3) or were MSS but had a TMB ≥10 mut/Mb (n = 20), both of which are FDA-recognized biomarkers predictive of response to pembrolizumab (Fig. 2C; Supplementary Table S2; refs. 27–29). Notably, all three MSI-H tumors were SCCs and all were TMB-high (TMB-H). One patient with an MSI-H cervical cancer was treated with pembrolizumab and derived no clinical benefit; interestingly, although the tumor was also TMB-H (19.3 mut/Mb), MLH1, MSH2, MSH6, and PMS2 proteins were all retained by IHC; the tumor was also confirmed to be HPV positive by ISH. A second patient with an MSI-H tumor died of disease before the tumor-agnostic FDA approval of pembrolizumab for MSI-H tumors, and the third patient was cured with definitive upfront therapy. A total of 20 patients had tumors that were TMB-H and MSS; these tumors were mostly SCCs (14/20, 70%). Of these, 9 (45%) of 20 were treated with standard care immune checkpoint inhibition, with 2 (10%) exhibiting durable clinical benefit from pembrolizumab (defined as ≥1 year of therapy). None of the GEA or rare histology tumors were MSI-H or TMB-H.

Thirty-seven percent of patients (66/177) had at least one potentially actionable alteration designated as level 3B by OncoKB (Fig. 3A; Supplementary Table S3), most commonly PIK3CA (45, 25%) and ERBB2 (17, 9.6%). Of note, 3 patients had pathogenic somatic BRCA1 (n = 1, no LOH) or BRCA2 (n = 2, 1 LOH) alterations. Thirty patients (17%) had been enrolled on therapeutic clinical trials at the time of data lock, including 11 patients who participated in more than one study. Eighteen (10%) of 177 patients were matched with a clinical trial based on their MSK-IMPACT tumor genomic testing results, including 6 based on ERBB2 alterations, 6 based on a PIK3CA alteration (4 patients treated with GDC-0032), and 2 based on FGFR alterations, as well as an assortment of other targets, including ERK and TP53 (Fig. 3B; Table 2; Supplementary Table S4). Of note, an exceptional responder with a D769N ERBB2 mutation has been on a clinical trial of a HER kinase inhibitor trial for over 5 years (Fig. 3B).

Twenty patients in the MSK Cervical Cancer Cohort participated in a study of immune checkpoint inhibition, including 1 patient who enrolled on two such studies. Three patients achieved a response to immunotherapy based on RECIST, and 2 of these patients remain progression free with a minimum of 5 years of follow-up (Fig. 3B,–D). Of note, these 2 exceptional responders had tumors that were MSS and TMB-low (TMB-L), but PD-L1 positive by IHC and confirmed HPV positive (one case each with HPV-16 and HPV-18). Both patients received dual checkpoint blockade.

Oncologists are increasingly using tumor genomic profiling to guide selection of FDA-approved and investigational therapies for patients with advanced cancer. While few studies have explored the clinical utility of NGS in advanced cervical cancer, the recent FDA approvals of pembrolizumab (29) and dostarlimab (30) for TMB-H/mismatch repair-deficient advanced solid tumors provide justification for clinical genomic profiling for these patients, for whom treatment options in the advanced setting are lacking. NTRK and RET fusions and BRAF V600E mutations are now also recognized as tumor-agnostic biomarkers of response to NRTK and RAF inhibitors, respectively, but none of the patients in the MSK Cervical Cancer Cohort had tumors harboring these oncogenic kinase alterations. The median OS in the MSK survival cohort was 59 months, with significant differences in survival correlating with stage at diagnosis. In contrast to a prior SEER database study (8), we did not observe a significant difference in survival across histologic subtypes; however, this is likely due to small sample sizes and the limited absolute number of patients with small cell carcinoma and GEA. To facilitate future meta-analyses combining our data with that of other groups, all clinical and genomic data have been made available via the cBioPortal for Cancer Genomics and via AACR GENIE.

Given the poor response rates to chemotherapy in the recurrent setting of cervical cancer, it is notable that 13% of the patients in the MSK Cervical Cancer Cohort had OncoKB level 1 actionable alterations, defined as FDA-authorized biomarkers of treatment response; in this cohort, these alterations were all MSI-H or TMB-H. Unfortunately, ORRs with single-agent PD-1 blockade for advanced cervical cancer are modest, at approximately 10%–15% (3, 31, 32). This is consistent with our data, in which only 10% of patients had durable responses to immune checkpoint inhibition delivered as standard care or on a clinical trial. Of note, both patients with exceptional response to immune checkpoint inhibition on a trial had TMB-L/MSS tumors, raising questions about the utility of TMB as a biomarker of immunotherapy response in patients with advanced cervical cancer.

The sum of the data support the need for additional biomarker initiatives in cervical cancer to identify patients most likely to respond to single-agent PD-1 blockade, as well as the development of combinatorial strategies and novel immune-based approaches. These will likely include dual checkpoint blockade, such as an anti-CTLA-4 (CTL antigen-4) in combination with an anti-PD-1 (programmed cell death protein 1), for which ORRs have ranged from 22% to 46%, depending on the drugs, dosing, and schedule (33, 34). In studies by Naumann and colleagues (33, 34) and O'Malley and colleagues (33, 34), PD-L1 expression enriched for responders, but responses were also observed in PD-L1–negative patients. Promising novel immunotherapy approaches include bispecific antibodies such as AK104—an anti-PD-1 and anti-CTLA-4 bispecific antibody—which has demonstrated impressive clinical activity and reasonable safety when combined with standard chemotherapy in patients with cervical cancer in the upfront setting (35). Another bispecific, bintrafusp alfa, targets both TGFβ and PD-L1, demonstrating an ORR of 30.5% in HPV-associated malignancies (36).

Thirteen percent of the patients in the MSK Cervical Cancer Cohort had ERBB2 alterations (4% amplifications, 9% oncogenic mutations). ERBB2 mutations were enriched in UEAs (28%), but they were also seen in SCCs and GEAs. This finding suggests that HER2 is likely an important therapeutic target for this group of patients with a high unmet need. The pan-HER tyrosine kinase inhibitor neratinib has demonstrated clinical efficacy in patients with HER2-mutant cervical cancer, with an ORR of 25% and median PFS of 7.0 months (37). In the MSK Cervical Cancer Cohort, 1 patient had been on neratinib therapy for more than 5 years. More recently, the antibody–drug conjugate trastuzumab deruxtecan (T-DXd) demonstrated clinical activity in HER2-mutant non–small cell lung cancer, with an ORR of 51% (38), as well as in HER2-amplified breast (39) and gastric cancers (40). T-DXd is currently under investigation in HER2-overexpressed cervical cancer (Clinicaltrials.gov: NCT04482309), and our data suggest that clinicians should strongly consider testing for HER2 overexpression in advanced/recurrent cervical cancer, especially in UEAs and GEAs, to facilitate enrollment on HER2-targeting clinical trials.

The recent development of KRAS inhibitors also may have therapeutic implications in cervical cancer. Approximately 12% of our cohort had mutations in KRAS, including 1 patient with a G12C hotspot mutation. Adagrasib and sotorasib have recently received FDA approval for KRAS G12C-mutated non–small cell lung cancer (41), and there are other pan-KRAS G12X inhibitors in development that may be relevant for patients with cervical cancer.

One of the strengths of our study is the inclusion of rare histologic subtypes, including GEA, a rare and aggressive histologic subtype of cervical adenocarcinoma, whose pathogenesis is unrelated to prior HPV infection (42, 43). A proportion of GEAs had genomic drivers commonly observed in pancreatobiliary tumors, including alterations in KRAS, TP53, SMAD4, and CDKN2A. A subset of GEAs more closely resembled intestinal-type gastric adenocarcinoma, with shared gene alterations in KMT2D, ERBB3 and RNF43, which have been associated with a poor prognosis (44). Clinically actionable alterations were seen in ERBB2 and PIK3CA (45), offering potential targets for clinical trial enrollment.

NGS has been shown as a highly sensitive method for HPV detection due to its ability to detect HPV presence at a low copy number (46), which informs our ability to better attribute the role of HPV in the development of cervical cancer compared with traditional PCR-based methods (47). In the MSK Cervical Cancer Cohort, we were able to identify HPV in the majority of tumor samples expected to be HPV positive (24) with good concordance with HPV-ISH which is currently considered the gold standard for HPV identification. The ability to detect virus in a clinical, large-panel NGS platform (20) such as MSK-IMPACT opens up the potential to identify cancer-associated HPV subtypes that may be targeted using novel approaches, such as therapeutic vaccines (48) and T-cell receptor or chimeric antigen receptor T-cell therapies (49). Recently, the MSK-IMPACT assay was modified to include probes designed to capture DNA sequences found in high-risk HPV subtypes. This may result in improved detection sensitivity versus the off-target reads employed for HPV detection in the current study.

Our study also has its limitations. Although comparable with TCGA, our results are limited because of the overall small sample size and retrospective nature of the analysis. Furthermore, this study was conducted at a tertiary cancer center with a large referral base, resulting in overrepresentation of patients with rare histologies, as well as patients with advanced or recurrent disease. Finally, while we were able to enroll 18% of the patients in this cohort to a clinical trial, this was influenced not only by the sequencing results but also access to a diverse and active clinical trial portfolio, which is likely site dependent. Despite these limitations, our analysis of the MSK Cervical Cancer Cohort provides additional evidence of the genomic diversity of histologic subtypes in patients with cervical cancer and demonstrates that genomic profiling can inform clinical decision making and provide the rationale for novel treatment options for a subset of patients with cervical cancer.

C.F. Friedman reports grants from Puma Biotechnology during the conduct of the study; other support from Puma Biotechnology; personal fees from Aadi Biosciences/GOG Partners, GSK; grants from AstraZeneca, Mersana, Hotspot Therapeutics, Marengo, Daiichi, and Seagen; grants and other support from Merck, Genentech/Roche; grants and personal fees from Bristol Myers Squibb outside the submitted work. C. Vanderbilt reports other support from Paige.AI outside the submitted work. M.M. Leitao reports personal fees from Medtronic, Intuitive Surgical, JnJ/Ethicon, and Immunogen outside the submitted work. C. Kyi reports other support from Gritstone, Merus, and Acrivon outside the submitted work. D. Zamarin reports grants from NCI during the conduct of the study; grants and personal fees from Genentech, AstraZeneca, Synthekine; grants from Plexxikon and Merck; personal fees from Memgen, Xencor, Targovax, Tessa Therapeutics, Miltenyi Biotec, Celldex, Crown Biosciences, GSK, Astellas, Takeda; other support from Immunos, Calidi Biotherapeutics, Accurius outside the submitted work; in addition, D. Zamarin has a patent for Newcastle Disase Virus for cancer therapy licensed and with royalties paid from Merck. R.E. O'Cearbhaill reports grants from NCI/NIH during the conduct of the study; personal fees from Tesaro/GSK, Regeneron, Seattle Genetics/SeaGen, Fresenius Kabi, Gynecologic Oncology Foun, Curio/Onclive/PER/Aptitude/MJH, R-Pharm, Immunogen, SITC, Bayer, Miltenyi Biotec, and 2Seventy Bio; and non-financial support from Hitech Health, Gathering Around Cancer outside the submitted work. J. Konner reports personal fees from AstraZeneca and Clovis outside the submitted work. M.F. Berger reports personal fees from Eli Lilly, AstraZeneca, and Paige.AI outside the submitted work. B. Weigelt reports grants from Repare Therapeutics outside the submitted work. C. Aghajanian reports personal fees from Eisai/Merck, Roche/Genentech, AstraZeneca/Merck, Repare Therapeutics; as well as grants from Abbvie, AstraZeneca, Clovis, Genentech/Roche; other support from GOG Foundation, Board of Directors, and Blueprint Medicine outside the submitted work. D.B. Solit reports personal fees from Pfizer, Scorpion Therapeutics, FORE Therapeutics, Function Oncology, Elsie Biotechnologies, Inc, Paige.AI, Fog Pharma, Rain Therapeutics, and Pyramid Biosciences outside the submitted work. No disclosures were reported by the other authors.

C.F. Friedman: Conceptualization, data curation, formal analysis, writing–original draft, writing–review and editing. V. Ravichandran: Data curation, formal analysis, writing–original draft, writing–review and editing. K. Miller: Data curation, writing–original draft, writing–review and editing. C. Vanderbilt: Data curation, formal analysis, writing–review and editing. Q. Zhou: Data curation, formal analysis, writing–review and editing. A. Iasonos: Data curation, formal analysis, writing–review and editing. M. Vivek: Data curation, formal analysis, writing–review and editing. P. Mishra: Data curation, formal analysis, writing–review and editing. M.M. Leitao: Data curation, writing–review and editing. V. Broach: Data curation, writing–review and editing. Y. Sonoda: Data curation, writing–review and editing. C. Kyi: Data curation, writing–review and editing. D. Zamarin: Data curation, writing–review and editing. R.E. O'Cearbhaill: Data curation, writing–review and editing. J. Konner: Data curation, writing–review and editing. M.F. Berger: Data curation, writing–review and editing. B. Weigelt: Data curation, writing–review and editing. A. Momeni Boroujeni: Data curation, writing–review and editing. K.J. Park: Data curation, writing–review and editing. C. Aghajanian: Conceptualization, data curation, writing–review and editing. D.B. Solit: Conceptualization, data curation, writing–review and editing. M.T.A. Donoghue: Conceptualization, data curation, formal analysis, writing–original draft, writing–review and editing.

We thank members of the Kravis Center for Molecular Oncology, the Integrated Genomics Organization and Diagnostic Molecular Pathology. We thank Dr. Lisa Eli of Puma Biotechnology for her support and input. This work was supported in part by an NIH/NCI Cancer Center Support Grant (P30 CA008748), as well as Cycle for Survival, the Marie-Josée and Henry R. Kravis Center for Molecular Oncology, and a grant from Puma Biotechnology. C.F. Friedman is a member of the Parker Institute for Cancer Immunotherapy at MSK.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.