Neoadjuvant cemiplimab for resectable hepatocellular carcinoma: a single-arm, open-label, phase 2 trial

Summary

Background

Surgical resection of early stage hepatocellular carcinoma is standard clinical practice; however, most tumours recur despite surgery, and no perioperative intervention has shown a survival benefit. Neoadjuvant immunotherapy has induced pathological responses in multiple tumour types and might decrease the risk of postoperative recurrence in hepatocellular carcinoma. We aimed to evaluate the clinical activity of neoadjuvant cemiplimab (an anti-PD-1) in patients with resectable hepatocellular carcinoma.

Methods

For this single-arm, open-label, phase 2 trial, patients with resectable hepatocellular carcinoma (stage Ib, II, and IIIb) were enrolled and received two cycles of neoadjuvant cemiplimab 350 mg intravenously every 3 weeks followed by surgical resection. Eligible patients were aged 18 years or older, had confirmed resectable hepatocellular carcinoma, an Eastern Cooperative Oncology Group performance status of 0 or 1, and adequate liver function. Patients were excluded if they had metastatic disease, if the surgery was not expected to be curative, if they had a known additional malignancy requiring active treatment, or if they required systemic steroid treatment or any other immunosuppressive therapy. After resection, patients received an additional eight cycles of cemiplimab 350 mg intravenously every 3 weeks in the adjuvant setting. The primary endpoint was significant tumour necrosis on pathological examination (defined as >70% necrosis of the resected tumour). Secondary endpoints included delay of surgery, the proportion of patients with an overall response, change in CD8⁺ T-cell density, and adverse events. Tumour necrosis and response were analysed in all patients who received at least one dose of cemiplimab and completed surgical resection; safety and other endpoints were analysed in the intention-to-treat population. Patients underwent pre-treatment biopsies and blood collection throughout treatment. This trial is registered with ClinicalTrials.gov (NCT03916627, Cohort B) and is ongoing.

Findings

Between Aug 5, 2019, and Nov 25, 2020, 21 patients were enrolled. All patients received neoadjuvant cemiplimab, and 20 patients underwent successful resection. Of the 20 patients with resected tumours, four (20%) had significant tumour necrosis. Three (15%) of 20 patients had a partial response, and all other patients maintained stable disease. 20 (95%) patients had a treatment-emergent adverse event of any grade during the neoadjuvant treatment period. The most common adverse events of any grade were increased aspartate aminotransferase (in four patients), increased blood creatine phosphokinase (in three), constipation (in three), and fatigue (in three). Seven patients had grade 3 adverse events, including increased blood creatine phosphokinase (in two patients) and hypoalbuminaemia (in one). No grade 4 or 5 events were observed. One patient developed pneumonitis, which led to a delay in surgery by 2 weeks.

Interpretation

This report is, to our knowledge, the largest clinical trial of a neoadjuvant anti-PD-1 monotherapy reported to date in hepatocellular carcinoma. The observed pathological responses to cemiplimab in this cohort support the design of larger trials to identify the optimal treatment duration and definitively establish the clinical benefit of preoperative PD-1 blockade in patients with hepatocellular carcinoma.

Funding

Regeneron Pharmaceuticals.

Introduction

Hepatocellular carcinoma is the third leading cause of cancer-related mortality globally,¹ and although immunotherapy combinations have changed the prognosis of patients with advanced hepatocellular carcinoma, the majority of patients still die from this disease. The recommended first-line treatment for early stage hepatocellular carcinoma is surgery or thermal ablation in patients with preserved liver function. Improved hepatocellular carcinoma screening has increased the number of patients undergoing resection; however, the majority of tumours recur despite surgery.^2,3 Although negative margins are usually observed at the time of surgical resection, recurrence is typically within the liver, and thought to originate from residual micro metastatic disease or de novo tumourigenesis (depending on patients’ predisposing factors) following resection.^2,4 This pattern of recurrence highlights the potential benefit of perioperative therapy to improve outcomes in hepatocellular carcinoma. At present, however, there are no standard neoadjuvant or adjuvant therapies that have shown a durable survival benefit in patients with hepatocellular carcinoma, resulting in the absence of approved therapies in this setting.⁵

Immunotherapy has shown efficacy in unresectable hepatocellular carcinoma, with response rates to single-agent PD-1 blockade of around 20%, and the potential for higher responses to combination therapy.^6,7 Neoadjuvant immunotherapy incorporating anti-PD-1 antibodies, alone or in combination, has shown success in inducing pathological responses in multiple tumour types while also inducing expansion of tumour-specific T cells, which potentially imparts a vaccinal effect capable of systemic surveillance.^8,9 Although preoperative cytotoxic or targeted therapies, or both, have not shown benefit in hepatocellular carcinoma, there is a strong rationale for the use of immunotherapy in the perioperative setting. We postulate that the benefit of perioperative immunotherapy would be comparable or superior to its efficacy in the metastatic setting, based on evidence that the clinical benefit of immunotherapy is inversely correlated with tumour burden (in the adjuvant setting, tumour burden is typically microscopic if present) and preclinical and early clinical evidence specifically supporting the superiority of neoadjuvant immunotherapy over adjuvant therapy.^3,10 Pathological response is generally the endpoint assessed in this setting based on trials done in other tumour types, which have shown a disease-free survival benefit in patients who have achieved a pathological response following neoadjuvant immunotherapy.^11,12 However, pathological response to immunotherapy has not yet been shown to correlate with overall survival in these other tumour types, and this has yet to be explored specifically in hepatocellular carcinoma.

Cemiplimab is a high-affinity, fully human immune-globulin G4 monoclonal antibody to PD-1.¹³ We aimed to evaluate the clinical activity of neoadjuvant cemiplimab in patients with resectable hepatocellular carcinoma.

Methods

Study design and participants

This single-centre, open-label, single-arm, phase 2 trial of cemiplimab monotherapy administered before and after definitive surgery enrolled patients with early-stage hepatocellular carcinoma. Eligible patients were aged 18 years or older and had confirmed resectable hepatocellular carcinoma (Liver Imaging Reporting and Data System [LIRADS] score of 5 on imaging or biopsy-proven tumour, or both), an Eastern Cooperative Oncology Group performance status of 0 or 1, and adequate liver function. Patients were enrolled regardless of the underlying cause of hepatocellular carcinoma; patients with a history of hepatitis C virus (HCV) or hepatitis B virus (HBV) infection were eligible for enrolment if viral clearance had occurred or circulating virus was suppressed on HBV-directed therapies. Patients with HIV with an undetectable viral load by PCR and a CD4+ T-cell count higher than 350 cells per μL were also eligible for enrolment.

Patients were excluded from enrolment if they had metastatic disease, if the surgery was not expected to be curative, or if they had a known additional malignancy requiring active treatment. Patients could not be receiving chronic systemic immunosuppression or have active autoimmune disease requiring systemic treatment in the past year, except for patients with endocrinopathies on hormone replacement therapy. Pregnant women and patients who had undergone a transplant were excluded, as were any patients with a history of CNS or pulmonary inflammatory conditions. Full inclusion and exclusion criteria are in the study protocol available in the appendix (pp 39–42).

The study protocol and all amendments (available in the appendix pp 10–261) were approved by the institutional review board at Mount Sinai Hospital, and the study was done in accordance with the International Conference on Harmonisation of Technical Requirements for Pharmaceuticals for Human Use Good Clinical Practice guidelines. All patients provided written informed consent before enrolment.

Procedures

Patients deemed suitable candidates for surgical resection were enrolled, underwent a core needle biopsy of their tumour (per protocol) under CT guidance, and subsequently received two cycles of neoadjuvant cemiplimab 350 mg intravenously every 3 weeks. After the second dose of cemiplimab (which had a ±3-day window), patients underwent surgical resection. Gadoxetate-enhanced MRI was done before initiation of treatment and again within 10 days before surgical resection, unless contraindicated. Upon recovery from surgery, patients received an additional eight cycles of adjuvant cemiplimab intravenously 350 mg every 3 weeks; the adjuvant phase of the study remains ongoing and data from this phase are not presented here.

Tumour necrosis on pathological examination was assessed by consensus of two dedicated hepatopathologists (MIF and SCW) who in unison visually estimated and agreed upon the percentage of necrosis seen within the resected tumour bed, as defined by the region within the tumour capsule delineated from normal hepatocytes, and in the pre-treatment biopsy. Necrosis in the biopsy was estimated on the basis of the full core analysed; to measure the percentage of the tumour that was necrotic, the entire tumour bed was examined grossly for necrosis, then representative samples of the tumour (at least one section per cm of the largest dimension) were examined to confirm assessment. A complete pathological response was defined as an absence of viable tumour in all sections analysed. Tumour-infiltrating lymphocytes were also quantified in these pre-treatment and post-treatment specimens. Two abdominal radiologists (BT and MS) assessed the degree of tumour necrosis in consensus on preoperative MRI, defined as non-enhancing tissue on subtracted post-contrast T1-weighted images obtained during the portal venous phase; this necrosis quantification has previously been shown to correlate closely with the degree of tumour necrosis on histopathological assessment in hepatocellular carcinoma in patients who received locoregional therapy ahead of surgery.¹⁴

Blood collection for haematology and blood chemistry analyses was done at the screening visit; during the neoadjuvant treatment period on days 1, 8, and 22; on the day of surgery and every 4 weeks during the surgery follow-up visits; then during the adjuvant treatment period every 3 weeks for cycles 1–8, 30 days after the last dose of cemiplimab, and 90 days after the last dose of cemiplimab. Safety was monitored continuously throughout the trial: before surgery, postoperatively during adjuvant therapy, and for 90 days after cessation of cemiplimab (monitoring in the adjuvant phase is ongoing). Here, we report on the safety and tolerability of neoadjuvant treatment through to surgical resection. Adverse events were assessed with the Common Terminology Criteria for Adverse Events (CTCAE) version 5.0 and coded with the currently available version of the Medical Dictionary for Regulatory Activities (MedDRA).

Outcomes

The primary endpoint was significant tumour necrosis, defined as more than 70% necrosis of resected tumour (which was extrapolated from the limited literature available).¹⁵ Secondary endpoints included delay of surgery, defined as surgery more than 28 days following the second cycle of cemiplimab; overall response rate (according to Response Evaluation Criteria in Solid Tumors [RECIST] version 1.1 criteria), defined as the proportion of patients with a complete response (100% tumour shrinkage) or partial response (≥30% decrease in tumour size) as documented by the investigator; adverse events during the neoadjuvant period, defined as adverse events that were not present at baseline or represented the exacerbation of a pre-existing condition during the on-treatment period; immune-related adverse events, defined as adverse events that met protocol-defined immune-related criteria; and change in tumour-infiltrating CD8⁺ T-cell density, defined as the change from baseline to the time of surgery; the full list of secondary endpoints is provided in the appendix (pp 14–15). In an exploratory analysis, and to align with other trials done in this setting, we also recorded the proportion of patients with 50% or greater tumour necrosis on pathological examination of the resected tumour.

Tissue analysis

Pre-treatment biopsies were formalin fixed and paraffin embedded (FFPE) for multiplex immunohistochemistry (mIHC) and immunofluorescence analysis, and additional biopsies (up to four core needle biopsies were collected at the screening visit and every 12 weeks for up to 5 years after surgery) were preserved in RNAlater for bulk RNA sequencing (RNAseq) analysis.¹⁶ A fully automated mIHC assay was done on the Ventana Discovery ULTRA platform (Ventana Medical Systems, Tucson, AZ, USA); additional details are provided in the appendix (p 2).¹⁷ Surgical tumour resections obtained after treatment were similarly preserved for analysis. The degree of tumour necrosis and the presence of tumour-infiltrating lymphocytes were quantified on FFPE pre-treatment biopsies and post-treatment tumour resections. Following pathological assessment, resected tumours were sampled to allow for bulk tissue RNAseq. In a subset of patients from whom a substantial amount of tumour and adjacent tissue were available following standard pathological assessment, tissue was dissociated into a single-cell suspension, and was analysed by mass cytometry with previously used methods and established panels.

RNAseq was done on pre-treatment biopsies and resection samples that were stored in RNAlater. Immune scores were computed by summing the normalised unique molecular identifier counts of all genes comprising each signature and plotted on a log2 scale. Details of the sequencing and analysis methodology are provided in the appendix (p 2). To identify cell types within RNAseq data, previously described gene signatures for CD8⁺ T cells,¹⁸ naive, cytotoxic, or activated or dysfunctional lymphocytes,¹⁹ or B cells and T regulatory cells²⁰ were used to quantify lymphocyte populations within tumour specimens before and after treatment. The monocyte-derived macrophage population was defined with a gene signature that included CSF1R, CSF3R, CD163, CD68, C1QA, CD14, and TFEC. The scores were then generated by taking the log of the total transcript count of all genes comprising the signature.

Statistical analysis

Primary efficacy outcomes were measured in patients who completed surgery, secondary efficacy outcomes were measured in the full analysis set (ie, all patients who received any study drug), and safety outcomes were assessed in the safety analysis set (ie, patients who received any study drug; based on the treatment received). We calculated that 21 patients needed to be enrolled for a power of 80% to test a null hypothesis of a poor overall response of 5% or less, versus an alternative hypothesis of a promising response rate of 20% or greater, at a 10% one-sided significance level.

Rates of significant tumour necrosis were summarised with frequencies and percentages, and two-sided 95% CIs were calculated with the Wilson score method.

Correlation between radiographic and pathological estimations of necrosis and radiographic tumour shrinkage was evaluated by Spearman’s correlation, and nominal p values and correlation coefficients were reported. For cell subsets identified with gene signatures in RNAseq data, statistical significance was assessed by Wilcoxon signed-rank test for paired data (at baseline and after treatment) and the Wilcoxon rank-sum test for unpaired data (different individuals); p values are nominal significance values suggesting hypothesis-generating trends. The proportion of CD8⁺ T cells among CD4 cells in the tumour lesion and adjacent tissue was analysed by two-way analysis of variance (ANOVA); the two-way parameters being compared were the tissue (tumour lesion or normal tissue) and the proportion of necrosis groups (<50% or ≥50% necrosis). This study is cohort B of a larger, ongoing, multicohort, multihistology trial of perioperative cemiplimab, and is registered with ClinicalTrials.gov, number NCT03916627.

Role of the funding source

This study was sponsored by Regeneron Pharmaceuticals and was designed by investigators at Mount Sinai Hospital in close collaboration with investigators at Regeneron Pharmaceuticals. Data from the study were collected by investigators, analysed by statisticians employed by the sponsor (Regeneron Pharmaceuticals), and interpreted by the authors, including employees of the sponsor. Medical writing support was provided by Jenna Lee of Prime (Knutsford, UK), and funded by Regeneron Pharmaceuticals according to Good Publication Practice guidelines.

Results

Between Aug 5, 2019, and Nov 25, 2020, 21 patients were enrolled in the trial, underwent biopsies, and subsequently received two doses of cemiplimab; the final patient underwent surgical excision on Dec 28, 2020. Data reported here are from the neoadjuvant treatment period through to surgical resection in patients who were enrolled between Aug 5, 2019, and Nov 25, 2020, with a data cutoff date of Dec 31, 2020.The majority of patients were Asian (52%), and the most common underlying cause of hepatocellular carcinoma was HBV infection (table 1). 20 patients had stage Ib–II disease on the American Joint Committee on Cancer (AJCC) and American Joint Committee on Cancer (UICC) 8th edition, and one patient had stage IIIb radiographically confirmed disease (appendix p 2). The median time from initiation of cemiplimab to surgical excision was 29 days (IQR 27–35), with one patient undergoing surgery 22 days after initiation of immunotherapy; the longest time to surgery was 84 days. Upon surgical exploration, one patient was found to have significantly enlarged lymph nodes in the porta hepatis that were not fully visualised on imaging and were confirmed to represent metastatic disease on intra-operative frozen tissue analysis; resection was therefore aborted.

Table 1.

Patient demographics, baseline characteristics, and disposition

	Resectable hepatocellular carcinoma (n=21)
Age (years)	68 (63–73; 45–82)
Sex
Female	3 (14%)
Male	18 (86%)
Asian	11 (52%)
ECOG performance status
0	18 (86%)
1	3 (14%)
Cause of hepatocellular carcinoma
Hepatitis B virus	8 (38%)
Hepatitis C virus	5 (23%)
NASH or NAFLD	5 (23%)
Alcohol-related liver disease	1 (5%)
Liver fibrosis score
0	1 (5%)
1	7 (33%)
2	4 (19%)
3	5 (24%)
4	4 (19%)
Cancer stage at screening
Stage Ib	6 (29%)
Stage II	14 (67%)
Stage IIIb	1 (5%)
BCLC stage21*
A	13 (62%)
B	5 (24%)
C	3 (14%)

Data are n (%) or median (IQR; range). ECOG=Eastern Cooperative Oncology Group. NAFLD=non-alcoholic fatty liver disease. NASH=non-alcoholic steatohepatitis. BCLC=Barcelona Clinic Liver Cancer.

^*All patients were determined by a multidisciplinary tumour board to be good candidates for complete surgical resection before enrolment.

20 (95%) patients had adverse events of any grade during the neoadjuvant treatment period (appendix p 3). The most common adverse events of any grade were increased aspartate aminotransferase (in four patients), increased blood creatine phosphokinase (in three), constipation (in three), and fatigue (in three; appendix p 3). Seven (33%) patients had grade 3 adverse events; two had elevated blood creatine phosphokinase, which resolved without treatment and was of unclear origin, and one patient had hypoalbuminaemia, which resolved (appendix p 3). No grade 4 or 5 adverse events were observed. Treatment-related adverse events of any grade occurred in six (29%) patients, two (10%) of which were grade 3 (table 2). One patient had grade 3 maculopapular rash, and another had grade 3 pneumonitis during neoadjuvant therapy; this pneumonitis required treatment with steroids and resulted in a delay of surgery by 2 weeks according to protocol-defined criteria. Upon resolution of the event, successful surgical resection was done, although no adjuvant cemiplimab was given.

Table 2.

Summary of neoadjuvant treatment-related adverse events

	Any grade	Grade 1	Grade 2	Grade 3
Any	6 (29%)	1 (5%)	3 (14%)	2 (10%)
Fatigue	3 (14%)	2 (10%)	1 (5%)	0
Infusion-related reaction	2 (10%)	0	2 (10%)	0
Rash	1 (5%)	0	1 (5%)	0
Cough	1 (5%)	1 (5%)	0	0
Pneumonitis	1 (5%)	0	0	1 (5%)
Pruritus	1 (5%)	1 (5%)	0	0
Maculopapular rash	1 (5%)	0	0	1 (5%)

Data are n (%).

Of the 20 patients whose tumours were evaluable for the primary endpoint, four (20%) had significant tumour necrosis, including three (15%) who had complete tumour necrosis (100%) at histopathology. Seven (35%) of the 20 patients who underwent surgical resection had 50% or greater tumour necrosis (table 3), whereas the remaining 13 patients had 30% or less necrosis. Three of 20 patients had a partial response radiographically as per RECIST 1.1 (corresponding to an overall response of 15%), with all other patients maintaining stable disease (figure 1A). On treatment, imaging was done on 21 patients at a median of 24 days (IQR 23–29) following initiation of cemiplimab.

Table 3.

Pathological tumour necrosis at resection

	Number of patients (n=20)
Significant tumour necrosis (>70%)	4 (20%)
Complete tumour necrosis (100%)	3 (15%)
Tumour necrosis ≥50%	7 (35%)
Tumour necrosis <50%	13 (65%)

Data are n (%).

Figure 1: Response as assessed by standard imaging, and tumour necrosis assessed by pathological examination and imaging.

(A) Waterfall plot of responses in patients ordered by increasing response with standard RECIST measurements (green bars; dashed line correlates with 30% decrease in tumour size). Degree of necrosis as assessed pathologically by two expert hepatopathologists (light blue bars; based on absolute change in necrosis) and degree of necrosis on MRI done after treatment, before surgery (dark blue bars; dashed line correlates with 70% necrosis to achieve the primary endpoint of significant tumour necrosis). (B) Comparative analysis of necrosis measurements by MRI and as per pathological analyses. The blue dashed line is the regression line. *Imaging data not available for analysis of necrosis; denotes a patient for whom MRI was contraindicated, so MRI-based analysis of necrosis was not possible; pathological necrosis was 0%. †Patients with confirmed significant tumour necrosis. ‡Patient without MRI to quantify necrosis.

MRI identified patients with significant necrosis before resection, irrespective of radiographic tumour shrinkage (figure 1A). Estimation of necrosis defined by MRI was strongly correlated with pathological assessment of necrosis at surgery (figure 1B, r=0·62, p=0·0049). By contrast, there was only a moderate correlation between assessment of necrosis (either pathological or radiographic) and tumour response measured by standard RECIST 1.1, and this correlation was not significant (appendix p 5; p=0·065 for the correlation with pathological necrosis and p=0·19 for the correlation with radiographic necrosis). Focusing on patients with notable post-treatment necrosis, standard and pathological images from five of the seven patients who achieved 50% or greater necrosis are shown to highlight examples of radiographic and pathological necrosis (appendix p 6). Three patients had more than 70% necrosis on both MRI and pathological assessment (patients 16, 17, and 18); however, only one of these patients had a partial response (−30%) by RECIST 1.1 (patient 18), whereas the other patients’ lesions were considered stable disease as per the standard response.

In the two patients with the largest tumour shrinkage as per RECIST 1.1 (patients 19 and 20; figure 1A), the degree of post-treatment necrosis detected by imaging underestimated the necrosis subsequently assessed on pathological examination. This discrepancy highlights that each of the reported metrics have limitations in quantifying the decrease in viable tumour volume. For example, patient 19 had significant shrinkage of the tumour bed, but the imaging estimate of the viable tumour only quantifies the necrosis within this (now smaller) tumour bed (appendix p 6). The pathological and radiographic assessments of necrosis we report only quantified necrosis within the tumour bed at the time of resection and do not account for the change in size from baseline; only the standard RECIST 1.1 measure change in size from baseline, although the criteria do not include measurements of necrosis versus viable tissue based on post-contrast image subtraction within the defined tumour bed.

For the exploratory tissue analysis, we compared the seven patients with 50% or greater histopathological necrosis with the remaining 13 patients who had undergone resection and were found to have little to no necrosis in their resected tumours (≤30% necrosis for all 13 patients; appendix p 4). Six of the seven patients identified with this exploratory cutoff showed an increase in necrosis in the post-treatment versus pre-treatment sample, suggestive of a therapeutic effect (appendix p 4). One of these seven patients, patient 2, had a highly necrotic tumour at baseline and showed no appreciable change in necrosis following treatment on MRI or pathological examination, and the tumour size was slightly increased while on therapy. Of the seven patients with 50% or greater necrosis, three had a history of HBV, two had non-alcoholic steatohepatitis or non-alcoholic fatty liver disease (NASH or NAFLD), one had HCV-related cirrhosis, and one had alcoholic cirrhosis (appendix p 4).

Immunohistochemical analysis of post-treatment lesions showed increased density of immune infiltrates (appendix p 7) and greater numbers of tumour-infiltrating lymphocytes (appendix p 7) in patients with 50% or greater necrosis compared with patients with little or no necrosis. In eight patients with adequate tumour samples for analysis by mass cytometry, four patients with 50% or greater necrosis (three with 100% necrosis and one with 50% necrosis) had significantly higher CD8⁺ T-cell infiltration in the tumour than four patients with little to no necrosis (p=0·0010; two-way ANOVA followed by multiple-comparison Sidak test). This finding was unique to the tumour, as similar numbers of T cells were seen in the non-involved adjacent normal liver (appendix p 7).

Furthermore, quantifying the immune infiltrate in pre-treatment and post-treatment tissue specimens analysed by mIHC showed greater numbers of immune cells at baseline, which further increased after therapy in patients who had 50% or greater necrosis. These infiltrates were relatively unchanged in patients who had minimal to no necrosis following treatment, although given inter-patient variability the magnitude of the immune-cell increase from before treatment to after treatment did not differ significantly for the less than 50% and 50% or greater necrosis groups (figures 2A, B).

Figure 2: Tissue analysis by multiplex immunohistochemistry, mass cytometry, and RNA sequencing.

(A) Representative image of a multiplex immunohistochemistry panel (comprising CD3, CD8, FOXP3, CD68, and CD20) used to do quantitative image analysis. (B) Mean density of each immune subset at baseline and at resection in patients with 50% or more necrosis according to multiplex immunohistochemistry. Immune subsets are defined as T cells (CD3⁺, CD8⁺ T cells), conventional CD4 cells (CD3⁺, CD8⁻, FOXP3⁻ cells), regulatory T cells (CD3⁺, FOXP3⁺ cells), myeloid cells (CD68⁺), and B cells (CD20⁺). Error bars indicate one standard error of the mean. (C) Heat map representation of bulk RNA sequencing of biopsy cores and tumour resection from 11 patients (seven patients with little to no necrosis on resection [all <50% necrosis] and four patients with ≥50% necrosis). Genes used in previously published gene signatures correlating with immune lineage are shown at baseline and after treatment. (D) Comparison of immune lineage populations taken from RNA sequencing analysis with publicly available gene signatures associated with CD8 T cells, regulatory T cells, monocyte-derived macrophages, and B cells, as well as activated or dysfunctional, cytotoxic, and naive programmes. DAPI=4′,6-diamidino-2-phenylindole.

In RNAseq analyses of RNA from paired pre-treatment and post-treatment specimens, published signatures for CD8⁺ T cells, activated or dysfunctional cells, cytotoxic cells, monocyte-derived macrophages, and B cells were all enriched at baseline in patients who were subsequently found to have 50% or greater necrosis upon resection. Notably, all but the B-cell and naive T-cell signatures (neither of which was enriched at baseline) increased following therapy in these patients. By contrast, there was no change seen in the expression levels of any of these signatures in patients with little to no necrosis (<50%; figures 2C, D).

Discussion

Early detection of hepatocellular carcinoma is increasingly possible because of improved screening; however, more than 50% of patients who undergo resection will have recurrence, and no perioperative therapies have shown a survival benefit.^3,5 The application of cancer immunotherapy in neoadjuvant settings could have a substantial impact, by both driving immediate induction of tumour-cell killing and potentially inducing durable immune responses capable of eliminating residual micrometastatic disease from which early recurrence is thought to arise;⁴ however, recurrence might also be a metachronous (de novo) primary tumour resulting from the field effect of underlying liver disease. Neoadjuvant immunotherapy has proven to be effective in tumours classically responsive to immunotherapy (eg, melanoma and lung cancer)^9,22 as well as in tumour types with low response rates to immunotherapy, such as colorectal cancer.^12,23,24 Immunotherapy is also promising because, unlike thermal ablation, it could prime an immune response that eliminates micrometastatic disease outside the treatment field. To our knowledge, our study and the study by Kaseb and colleagues²⁵ (of preoperative nivolumab alone or in combination with ipilimumab in patients with resectable hepatocellular carcinoma) are the first reported clinical trials of neoadjuvant immunotherapy in patients with resectable hepatocellular carcinoma.

In the present trial, 20% of patients had significant tumour necrosis, with 35% having 50% or greater tumour necrosis at surgery; this finding is in line with the response rates of neoadjuvant PD-1 blockade observed in other trials of different tumour types,²⁶ and correlates well with the findings of Kaseb and colleagues’ study.²⁵ As described in the study by Kaseb and colleagues,²⁵ among patients treated with either preoperative nivolumab alone or nivolumab plus ipilimumab who subsequently underwent successful surgery, 25% had a pathological complete response, although as might be expected with the addition of CTLA-4 blockade, 43% of patients had grade 3 or worse toxicity before surgery compared with 23% of patients treated with nivolumab alone.²⁵ This is somewhat higher than the frequency of perioperative grade 3 treatment-related adverse events observed in our study (in 10% of patients); however, the study by Kaseb and colleagues²⁵ involved longer periods of induction therapy before surgery, which might increase the likelihood of perioperative toxicity. In this hepatocellular carcinoma population in whom surgery is a curative intent therapy, longer or more toxic therapies, or both, increase the risk of toxicities that might delay or preclude surgery. Furthermore, in Kaseb and colleagues’ study,²⁵ two patients in each group did not undergo surgical resection due to progression of disease, which is also a risk with longer periods of preoperative therapy, particularly if a patient does not respond to therapy.

Alongside optimal treatment duration, optimal endpoints in these novel trials have yet to be defined. In all open trials of neoadjuvant immunotherapy in hepatocellular carcinoma, the clinical endpoints vary because of the lack of validated surrogate endpoints. In neoadjuvant chemotherapy trials and meta-analyses in non-small-cell lung cancer and breast cancer, pathological responses—a major pathological response or pathological complete response, respectively—are validated endpoints that significantly correlate with survival, and as such these endpoints have been adopted as surrogate endpoints for neoadjuvant immunotherapy trials.²⁷ There is no validated pathological response surrogate in hepatocellular carcinoma or studies quantifying response to neoadjuvant therapy and correlating this with survival. Patients in the present study seemed to be divided into two populations, with seven showing a significant amount of necrosis (50–100%) compared with 13 patients who had minimal necrosis (≤30%). When designing the trial, given the absence of validated pathological cutoffs, we initially opted for the significant tumour necrosis cutoff of 70% based on an analysis extrapolated from a small retrospective study of transarterial chemoembolisation in patients with hepatocellular carcinoma, in whom this threshold correlated with overall survival and disease-free survival.²⁸ Conversely, Kaseb and colleagues²⁵ used an exploratory cutoff of 70% tumour necrosis. Given the differences in mechanisms of locoregional control and the potential for immune priming between transarterial chemoembolisation and cemiplimab, we acknowledge that this significant tumour necrosis cutoff is largely arbitrary, and larger, controlled neoadjuvant immunotherapy trials will be required to better define the implementation of a threshold that correlates with survival.

The response to immunotherapy might vary depending on the underlying cause of hepatocellular carcinoma, and surrogate endpoints might require evaluation within this context. Notably, in the seven patients who had 50% or more necrosis in this trial, all causes were represented; three had a history of HBV, two had NASH or NAFLD, one had HCV-related cirrhosis, and one had alcohol-related cirrhosis (appendix p 4). The pathological response seen in two patients with NASH or NAFLD is notable given that patients with NASH-related hepatocellular carcinoma fared significantly worse than patients with hepatocellular carcinoma from other causes in an analysis of large trials of anti-PD-L1 treatments in hepatocellular carcinoma.²⁹ In the present study, for one patient (patient 17) who had a confirmed diagnosis of NASH, the pathologists documented no tumour-infiltrating lymphocytes on pre-treatment biopsy, whereas a robust immune infiltrate was seen on the excised tumour, suggesting that NASH-related hepatocellular carcinoma might be responsive to immunotherapy, at least in the early stage setting. It is reassuring that a similar degree of pathological response was seen in Kaseb and colleagues’ study,²⁵ which had a somewhat different demographic in terms of fewer Asian patients, and nearly half of patients had hepatocellular carcinoma of non-viral origin, compared with a quarter of our cohort.

Although imaging is the standard method to assess response to therapy in the metastatic setting, we see substantial barriers to using either standard (RECIST 1.1) or novel (subtraction on MRI) exploratory surrogate imaging techniques to define response ahead of pathological assessment, which is the gold standard for induction therapy. Given the relatively short therapeutic intervention in this study, the extent of response was not fully captured radiographically in several patients via RECIST 1.1. By contrast, tumour viability assessed by MRI provided estimates of pre-treatment and on-treatment necrosis, which correlated with pathological assessment. Discordance between MRI-assessed necrosis and pathological assessment of necrosis, which was most notable in patients 9, 19, and 20, might have resulted from a perfusion flare in necrotic granulocytic regions being considered to represent residual viable tumour. As an example, patient 20 had significant tumour shrinkage (as per RECIST 1.1), as well as complete necrosis on pathological examination, but the inflamed tumour bed was still well perfused on MRI assessment, and therefore not classified as necrotic. The correlation between radiographic and pathological necrosis assessments supports the continued use of serial MRI in neoadjuvant clinical trials; however, the limitations observed when tumours are imaged after a brief therapeutic intervention highlights the need to refine imaging response criteria to account for both change in tumour size (RECIST 1.1) and enhancement on MRI, and to define other radiomarkers to differentiate necrotic tissue from viable tissue. Should these findings be recapitulated after adequate follow-up of a larger cohort, MRI could be further explored as a clinical decision-making tool.

As expected, the immune infiltrate in patients whose tumours had 50% or greater necrosis was more robust than in patients with little or no necrosis on surgical samples following treatment with cemiplimab, which correlates with the findings of Kaseb and colleagues.²⁵ Additionally, the density of the immune infiltrate in pre-treatment biopsy based on RNAseq data correlates with this higher necrosis post-treatment, complemented by trends observed from mIHC, suggesting that patients with an underlying immune recognition of their tumour are more likely to respond to anti-PD-1 monotherapy. Indeed, Kaseb and colleagues show that in patients without a significant baseline immune infiltrate, major pathological responses were primarily achieved in the context of combination therapy.²⁵ This suggests that patients without a significant immune infiltrate on pre-treatment biopsies might benefit from a combination approach, incorporating an agent to prime a de novo T-cell response, such as that observed with ipilimumab.²⁵ Other combinations, such as immunotherapies that have a proven safety record, or radiotherapy, would also be interesting to explore; agents that affect the vasculature (such as bevacizumab or tyrosine kinase inhibitors) might be less ideal in the neoadjuvant setting owing to potential perioperative toxicity.

To conclude, to our knowledge, this is the largest trial to date of perioperative PD-1-targeted monotherapy in hepatocellular carcinoma. Although this study is limited by the small number of patients, cemiplimab showed clinical activity in a patient population with an unmet clinical need. The results of this study are preliminary, and the small population might not account for covariance; however, we believe the patient population in this study is still representative of the global hepatocellular carcinoma population, and similar or comparable responses to immunotherapy in the advanced hepatocellular carcinoma setting have been observed between subgroups such as Asian versus non-Asian patients.^30,31 The present study, together with neoadjuvant immunotherapy trials in several other tumour types, supports continued evaluation of perioperative immunotherapy to decrease recurrence rates and the development of unresectable or metastatic disease. Larger trials will help to define the utility, safety, and overall survival benefit of perioperative PD-1 blockade.

Research in context.

Evidence before this study

The recommended first-line treatment for very-early-stage or early-stage hepatocellular carcinoma is surgery or thermal ablation in patients with preserved liver function, and patient outcomes have improved with advances in surgical techniques and perioperative care. However, there is a high incidence of postoperative recurrence and cancer-related deaths. PD-1 and its ligand, PD-L1, are generally overexpressed in hepatocellular carcinoma, and high PD-L1 expression by tumour cells has been associated with significantly poorer prognosis. In early reports, neoadjuvant immunotherapy (including anti-PD-1) showed some success in induction of pathological responses in hepatocellular carcinoma. We searched PubMed for papers published from May 1, 2014, to May 1, 2019 (the 5-year period before the start of the study), with the search terms “resectable hepatocellular carcinoma” AND “neoadjuvant”. No immunotherapy studies were found. The most relevant study was a single-centre study of combination neoadjuvant bevacizumab and chemotherapy in patients with hepatocellular carcinoma.

Added value of this study

This study shows that a short course of neoadjuvant cemiplimab resulted in pathological responses in patients with resectable hepatocellular carcinoma. The safety profile of cemiplimab was acceptable. In initial pathological assessments and based on the sequencing from pre-treatment and post-treatment tissue specimens, there was a positive correlation between molecular signatures of tumour immune activity and pathological necrosis, as well as a correlation between an increase in the immune infiltrate response from baseline and greater pathological necrosis. Additionally, although standard imaging response criteria (Response Evaluation Criteria in Solid Tumours [RECIST], version 1.1) do not identify pathological responses in most patients after a brief course of therapy, contrast-enhanced MRI was shown to be an accurate non-invasive method to assess tumour necrosis in response to therapy and could be used in conjunction with RECIST 1.1 to quantify the total change in viable tumour following therapy.

Implications of all the available evidence

Our findings support the design of larger trials to validate optimal clinical endpoints that correlate with improvement in survival, and to establish the utility and safety of pre-operative PD-1 blockade in patients with resectable hepatocellular carcinoma.

Data sharing

Qualified researchers can request access to study documents (including the clinical study report, study protocol with any amendments, blank case report form, and statistical analysis plan) that support the methods and findings reported in this manuscript. Individual anonymised participant data will be considered for sharing once the product and indication have been approved by major health authorities (eg, the US Food and Drug Administration, European Medicines Agency, and the Pharmaceuticals and Medical Devices Agency), if there is legal authority to share the data and there is no reasonable likelihood of participant re-identification. Requests for data should be submitted to https://vivli.org/.