NEW YORK – Investigators from Harvard Medical School and partner institutions in the Boston area have developed a novel whole-genome liquid biopsy sequencing assay they call HPV-DeepSeek, which they believe can help improve the detection of viral cancers in the mouth and throat, describing the method and early validation results in a recent preprint.
Cancers that originate from human papilloma virus infection make up the majority of tumors in the oropharynx — the back of the mouth and the upper throat, and they offer a unique target for early detection and screening because they retain that originating viral DNA, which can be used as a biomarker.
A number of studies have shown that viral sequences can be detected in blood samples from patients with HPV-associated cancers with high sensitivity and specificity. But the Harvard team hoped to be able to push this detection as early as possible — not only into early-stage disease but months and even years before clinical signs of cancer emerge.
"You have to step back a bit and think about the natural history of what's going on in these cancers. We believe that infection with HPV in your oropharynx occurs about 30 years before someone develops a cancer. That's a very long time that the virus is there," said Daniel Faden, the pending publication's first author.
“But what happens in that time frame [is] a black box," he added. "We don't really know of a premalignant intermediary state, unlike in cervical cancer and anal cancer where there's a known transition from infection to low-grade dysplasia, then high-grade dysplasia, then invasive carcinoma. … People just kind of show up with their cancers."
Faden said that his team and others had seen signs that it could be possible to detect cancers years before they present clinically in previous work with droplet digital PCR. They just needed a more sensitive approach to make it work more reliably.
The result of their efforts to do that is HPV-DeepSeek, a viral whole-genome sequencing assay. Increasing the breadth of sequencing allows what Faden called "more shots on goal."
"With PCR, you're typically looking for, say, a single fragment of the HPV genome in the circulation … so you're ignoring a lot of the genome," he said. In addition, he explained, the DeepSeek assay allows analysis of other, complementary signals like viral integration events and PIK3CA mutations. "Those are other cancer signals you can see in the blood to help build confidence that you are detecting a cancer."
In their study, Faden and his colleagues first trained HPV-DeepSeek in a set of 123 samples from oropharyngeal cancer cases and another 123 age- and gender-matched controls.
Using the locked cutoff from that cohort, the team then tested 28 oropharyngeal cancer cases and 28 age- and gender-matched controls that had samples collected at various prior time points between about one and 10 years before diagnosis.
The assay detected cancer in 22 of the 28 pre-diagnostic samples with lead times up to almost eight years. Among samples taken within four years of diagnosis, the test detected cancer in 100 percent of cases. Using machine learning, the team was able to boost this to a pre-diagnostic detection of 27 of the 28 cases, with a maximum lead time of 10 years and 100 percent sensitivity up to nearly eight years before diagnosis.
Importantly, Faden said, all the control cases were negative. But because the control group was so small, confirming specificity will certainly be a concern moving forward.
"When we're talking population level, for a screening test you need very, very high specificity. You can't be telling patients that they have a cancer when they don't have one," he said. "Obviously, this is not a population-level study. However, if you look at the existing work from my group and lots of other groups looking at detectability of ct-HPV DNA in the general population, it’s not detectable. Specificity is extremely high.”
The hypothesized reason for this, Faden said, is that HPV should not be detectable in someone's blood unless they have an invasive cancer because while the virus is ubiquitous in mucosal surfaces, it is not known to be blood-borne. "Somewhere around 2 to 5 percent of the population has an HPV infection at any time, so if that were detectable in blood, we would see lots of controls in all the studies we are running have a positive test, and we don't see that."
Another crucial question for the group as it moves forward to clinically validate its assay is whether this ability to detect the presence of malignant transformation years before a clinical diagnosis actually improves patient care and outcomes.
"The answer right now is that we don't really know," Faden said. That said, there has been data coming in from other studies of ctDNA detection of viral cancers that can shed some light. Liquid biopsy pioneer Dennis Lo, for example, has developed a test for Epstein-Barr virus-associated nasopharyngeal cancers with prospective screening data published in the New England Journal of Medicine that found that screened individuals had tumors diagnosed at earlier stages than those followed with standard-of-care procedures.
"We know earlier-stage patients have improved survival," Faden said, "So we can extrapolate off of the existing literature in a similar type of cancer."
Apart from the detection results, which he said have been garnering a lot of attention as he has shared the study with colleagues and others in the field, Faden highlighted the group's attention to orthogonal validation, which helps boost confidence in the results.
In 68 percent of the 28 cancer cases, the team could see multiple other cancer signs, including gene mutations, viral genome integration events, and HPV antigens. The researchers also used their sequencing data to rule out cross-sample contamination by molecularly fingerprinting each patient's HPV genome, demonstrating that each viral genome was unique.
Glenn Hanna, a medical oncologist at the Center for Head & Neck Oncology at Dana-Farber Cancer Institute, wrote in an email that the case sample set of 28 is arguably too small to draw confident conclusions, but the study's accuracy metrics are "impressive nonetheless given the robust sensitivity."
"The biggest question is what the impact would be for early detection," he added. "Overall, HPV+ cancers have a favorable prognosis with modern therapy. It's not clear that detecting cases early with screening would change the impact of stage, prognosis, and cure," especially in cancer that is relatively uncommon and is likely to decline in event rate with modern vaccination efforts.
José Zevallos, chairman of otolaryngology at the University of Pittsburgh Medical Center, said in an email that Faden and colleagues' work shows "tremendous promise." However, he cautioned, to be most impactful, it would need to be coupled with advanced diagnostic capabilities for localizing the cancer or low morbidity therapeutic options that have yet to be defined.
"Advances in our ability to actually pinpoint an early-stage cancer deep in the tonsil or base of tongue (without a major surgical resection) have not kept pace with these remarkable liquid biopsy advances, leaving us with limited clinical actionability in the setting of a positive test."
Faden said he and his team are currently considering a variety of possible next steps. One way to make a prospective trial easier to power would be to use the test in high-risk patients, potentially individuals with HIV who are at a 60-or-so-fold increased risk of HPV cancers.
Another approach could be age- and gender-based. Faden said that there is some modeling data that suggests that a single time point screen in a population of men at age 50, for example, could lead to similar efficacy to existing cancer screening programs in other tumor types.
A third possibility would be thinking about building viral targets into existing multi-cancer early detection tests that are coming down the pipeline.
Faden said his group is also exploring utility for their approach outside of screening in diagnostic and monitoring applications.
"One of the things that we're really interested in, and what I think is sort of the next frontier of liquid biopsy, are really fully integrated diagnostics that take you from screening through diagnosis and treatment monitoring — a single test which can screen, confirm diagnosis, and then move through treatment monitoring on the same platform."