Biomarker-Driven Personalization of Neoadjuvant Immunotherapy in Melanoma

The introduction of immune-checkpoint blockade (ICB) and targeted therapy has dramatically improved survival outcomes of patients with advanced melanoma (1–5). In stage III and II (anti–PD-1 only) melanoma, these agents have been shown to improve relapse-free survival (RFS) in large adjuvant phase III trials (6–11) and have been included in the standard of care for stage III melanoma as adjuvant therapy after surgical resection of the disease.

Adjuvant ipilimumab was the first checkpoint inhibitor in stage III melanoma demonstrating improved RFS and improved overall survival (OS) compared with placebo (8), but has not been implemented widely due to high toxicity rates [54% grade 3–4 immune-related adverse events (irAE); ref. 12]. Adjuvant nivolumab and pembrolizumab (PD-1 blockade) were better tolerated and yielded a higher RFS compared with ipilimumab (10, 13) and interferon (IFN)α-2b (14) in resectable stage IIIB/C–IV melanoma, and compared with placebo (9) in resectable stage III melanoma, respectively. For patients with stage III BRAFV600E/K-mutated melanoma specifically, targeted therapy with dabrafenib plus trametinib has been approved as adjuvant therapy for stage III–IV melanoma due to significant RFS improvement (7), although an indirect treatment comparison with adjuvant anti–PD-1 indicated a more durable benefit with anti–PD-1 (15, 16).

Notwithstanding the different available adjuvant therapy options, more than 40% of patients develop a melanoma recurrence within 4 to 5 years after surgery and no OS benefit for adjuvant anti–PD-1 antibodies or BRAF/MEK inhibitors has been shown to date, illustrating the unmet need for more effective treatment modalities for stage III melanoma.

For immunotherapies, more efficacy could be achieved by its neoadjuvant application prior to surgery.

Immune activation upon ICB requires a broad tumor neoantigen repertoire, increasing T-cell receptor signaling and resulting in activation of different intratumoral and migrating tumor-recognizing T-cell clones. Multiple (pre)clinical models have shown that expansion of these T-cell clones in the peripheral blood is increased when the tumor is still present at the time of ICB initiation (neoadjuvant) as compared with after resection of the tumor and its microenvironment (adjuvant; refs. 17–19). In melanoma, an early phase I trial (20) showed that tumor clones present in the tumor and in peripheral blood expanded more strongly in the peripheral blood after neoadjuvant versus adjuvant ICB. Further, with neoadjuvant ICB, more tumor-present T-cell clones that were not detectable at baseline in peripheral blood became detectable at week 6, which was associated with an improved RFS (20). A recent phase II trial in melanoma also indicated the superiority of neoadjuvant ICB, as patients treated with neoadjuvant pembrolizumab had improved event-free survival (EFS) compared with the adjuvant-treated patients (21).

Besides survival benefit, neoadjuvant administration of ICB offers several other advantages, such as reduction of the tumor burden resulting in enhanced resectability of the tumor lesions. Moreover, the possibility to read out the pathologic response provides early prognostic information, because a pathologic response has been shown to be a surrogate marker associated with EFS/RFS and OS after neoadjuvant ICB in stage III melanoma (22, 23). Therefore, this early response evaluation can guide further treatment decisions regarding, for example, extent of surgery, need for adjuvant therapy, and intensity of the follow-up scheme. Finally, this fast efficacy readout, a homogeneous patient population, and the availability of pre- and posttreatment tumor and blood samples make the neoadjuvant therapy setting an ideal platform for biomarker identification and the investigation of new therapies (Fig. 1). A major concern that has been raised against neoadjuvant therapies is the risk of disease progression to unresectable disease. In current neoadjuvant trials in stage III melanoma, this risk varies between 2% and 17% (18, 21, 24, 25), but an adjuvant trial observed progression in 18% of the patients within 7 weeks after surgery (26), suggesting that early progression is reflecting a more aggressive melanoma subtype or camouflaged stage IV disease. In these patients, surgery probably would not prevent progression or improve OS, but expose patients to unneeded surgical morbidity. Neoadjuvant irAEs and their management might also hamper or complicate surgery, potentially resulting in more surgery-related adverse events or surgery delay, which has not been observed to date (25, 27).

Due to the abovementioned advantages of neoadjuvant ICB, multiple new treatments and regimens are currently being tested in the neoadjuvant setting. The concept not only is being tested in stage III melanoma but has also expanded to the earlier stage II melanoma. However, it is currently unknown which patients would benefit from a more intense treatment schedule (e.g., combinations of anti–PD-1 + anti-CTLA4/anti-LAG3/anti-TIGIT) or a mild one (e.g., anti–PD-1 mono­therapy), which highlights the importance and need for biomarker-based personalization, selecting only patients who will benefit from that specific treatment. Stage III melanoma is currently one of the front-runners in biomarker research in neoadjuvant ICB, serving as a poster child for ICB treatment personalization for other stages and cancer types.

This review discusses the current landscape of conducted and ongoing neoadjuvant immunotherapy trials in stage III melanoma, known and promising biomarkers, and our vision of the future of biomarker-driven, personalized neoadjuvant immunotherapy in early-stage melanoma.

During the last decade, neoadjuvant ICB has been investigated in several clinical trials in melanoma (Table 1). The first trial (2010) tested 2 cycles of neoadjuvant ipilimumab 10 mg/kg every 3 weeks followed by surgery in 35 patients with resectable stage III/IV melanoma and observed 1-year progression-free survival (PFS) of 47% (28). No patient achieved a complete pathologic response (pCR; 0% viable residual tumor), but in five (15%) patients only microscopic residual disease was found, which nowadays would be called a major pathologic response (MPR; ≤10% viable residual tumor) according to the International Neoadjuvant Melanoma Consortium (INMC) classification (29). A subsequent trial testing the combination of neoadjuvant ipilimumab 3 mg/kg (n = 15) or 10 mg/kg (n = 15) in combination with high-dose IFNα in stage III melanoma showed no difference in efficacy between the two arms (pCR rate 36% vs. 29%, respectively) but a higher toxicity rate in the 10 mg/kg arm. However, the combination with IFNα seemed to be superior to ipilimumab monotherapy in the previous trial, with a 12-month PFS of 79% in the combination arm (30).

The phase Ib OpACIN trial was the first trial to prospectively compare neoadjuvant versus adjuvant ICB (2 + 2 cycles ipilimumab plus nivolumab neoadjuvant bracketing surgery, n = 10, vs. 4 cycles adjuvant after surgery, n = 10) in stage III melanoma. Expansion of T-cell clones in peripheral blood was higher at week 6 in the neoadjuvant arm compared with the 6 weeks of adjuvant therapy, confirming the hypothesis of a stronger and broader immune activation upon neoadjuvant immunotherapy. Pathologic responses (<50% viable residual tumor) were observed in seven out of nine (78%) patients of the neoadjuvant cohort, with an MPR rate of 62% (20). After 4 years of follow-up, RFS rates remained stable, with 80% for the neoadjuvant and 60% for the adjuvant group (24). The treatment regimen of ipilimumab 3 mg/kg and nivolumab 1 mg/kg was, however, toxic, with 90% grade 3–4 irAEs in both groups. This was an unexpected high toxicity rate given that trials testing this combination in stage IV melanoma demonstrated toxicity rates of 55% to 59% (2, 31). Similar results were found in a trial comparing 9 weeks of neoadjuvant ipilimumab plus nivolumab (n = 11) to nivolumab monotherapy (n = 12), in which combination therapy resulted in a higher efficacy (pCR rate 45% vs. 25%, respectively), but again more grade 3–4 irAEs (73% vs. 8%) were observed (18). The trial was stopped prematurely because two patients in the nivolumab group developed stage IV disease in the neoadjuvant treatment period, precluding surgery.

The monotherapy response rates were confirmed 1 year later by another trial testing 1 cycle (3 weeks) of neoadjuvant pembrolizumab followed by 1-year adjuvant pembrolizumab in stage III melanoma achieving an MPR rate of 30% (32). However, the 2-year RFS rate was only 63%, suggesting that 1 cycle of anti–PD-1 might be too short for durable tumor control or combination therapies (e.g., with ipilimumab) are needed for treating tumors with unfavorable characteristics (10).

We therefore adhered to the neoadjuvant combination of ipilimumab and nivolumab and designed the multicenter OpACIN-neo trial in order to find a dosing scheme with less toxicity and preserved efficacy. Neoadjuvant ipilimumab 1 mg/kg plus nivolumab 3 mg/kg was found to be the most optimal scheme with a pathologic response rate of 77% and 20% grade 3–4 irAEs at week 12 (33).

After 3 years of follow-up, a strong association between pathologic response and RFS was observed, with an RFS of 95% in patients with a pathologic response compared with 37% in patients with a pathologic nonresponse (pNR; P < 0.001; ref. 23).

In the PRADO extension cohort of OpACIN-neo, 99 patients were included to receive 2 cycles of neoadjuvant ipilimumab 1 mg/kg plus nivolumab 3 mg/kg. The pathologic response rate of 72% and 22% grade 3–4 irAEs during the first 12 weeks of therapy confirmed the efficacy of the previously found optimal dosing scheme. Additionally, response-directed personalized surgical and adjuvant treatment regimens were tested in this trial, based on the pathologic response in the index lymph node (ILN), which has previously been shown to be representative of the whole lymph node bed (34). In the case of MPR, therapeutic lymph node dissection (TLND) and adjuvant therapy were omitted. In patients with pathologic partial response (pPR; ≤50% viable tumor cells) and pNR (>50% viable tumor cells), treatment was escalated with a standard TLND, and patients with pNR received additional adjuvant BRAF/MEK inhibition or nivolumab. In the MPR group, more local relapses were observed than expected (2-year RFS of 93% compared with 97% in OpACIN-neo), but with similar rates of distant metastasis–free survival (2-year distant metastasis–free survival of 98% in PRADO vs. 97% in OpACIN-neo), suggesting that salvage TLND for patients who developed local relapses after ILN resection was sufficient to prevent the distant spread of the disease. For patients who achieved a pNR, 2-year RFS was improved by the addition of adjuvant therapy (76% compared with 36% in OpACIN-neo; refs. 24, 25).

Additional neoadjuvant combinations with anti–PD-1 have already been tested. Pembrolizumab plus high-dose IFNα-2b in patients with resectable stage III–IV melanoma showed a pCR rate of 43% and a 2-year RFS rate of 60% (35). An indirect comparison with neoadjuvant ipilimumab plus nivolumab treatment demonstrated that pembrolizumab plus IFN yielded lower RFS rates and higher toxicity and discontinuation rates (24, 33). Whether this scheme could be attractive in a certain subgroup of patients needs to be evaluated. A promising neoadjuvant combination in stage III melanoma is nivolumab plus relatlimab (anti-LAG3), which demonstrated a pathologic response rate of 70% including 57% pCR. The grade 3/4 toxicity rate of 26%, with 23% irreversible adrenal insufficiencies, was likely driven by the long adjuvant part of this treatment schedule, as no grade 3–4 irAEs were observed during the neoadjuvant part. A 2-year RFS rate of 92% was observed in patients with a pathologic response versus 55% in those without a pathologic response, raising the question whether these excellent RFS rates could be preserved when omitting the adjuvant therapy in patients achieving an MPR, possibly making this scheme more tolerable (36).

In patients with a BRAF-mutated melanoma, the phase II NeoTrio trial compared neoadjuvant pembrolizumab monotherapy to sequential dabrafenib plus trametinib followed by pembrolizumab and to the combination of dabrafenib plus trametinib and pembrolizumab. The triple-combination therapy showed the highest pathologic response rate of 80% and a pCR rate of 50% but caused significant toxicity, with 55% grade 3–4 irAEs compared with 25% in the sequential regime and 5% in the pembrolizumab monotherapy scheme. Although pathologic response rates were higher in the combination groups, EFS rates were similar (37).

In the DONIMI trial, different combinations of neoadjuvant nivolumab ± ipilimumab with the histone deacetylase inhibitor domatinostat were tested based on a baseline IFNγ signature algorithm. Addition of domatinostat did not improve the response upon ipilimumab plus nivolumab in patients with a low IFNγ signature in their baseline tumor biopsy. In patients with a high IFNγ signature, nivolumab monotherapy seemed sufficient, as 90% achieved a pathologic response in the nivolumab monotherapy group (38).

In the recently presented NIVEC trial that tested the addition of intratumoral oncolytic virus injection to neoadjuvant nivolumab, a high MPR of 65% was observed. The 1-year EFS rate was 75% (39), which is comparable to anti–PD-1 monotherapy (21). Of note, only 8% grade 3–4 irAEs were reported. In a second trial testing another neoadjuvant oncolytic virus plus PD-1 blockade, the EFS was also not superior to neoadjuvant pembrolizumab alone (40). In this trial, pembrolizumab in combination with anti-TIGIT indicated improved outcomes compared with the two other treatment arms (40). Yet, larger cohorts need to confirm this observation, as the cohorts were not perfectly balanced, for example, with regard to BRAF mutation status.

The first larger randomized, multicenter phase II trial (SWOG-S1801, NCT03698019) comparing neoadjuvant plus adjuvant pembrolizumab (n = 154) versus adjuvant pembrolizumab (n = 159) in resectable stage III/IV melanoma was presented last year. After a median follow-up of 14 months, the estimated 2-year EFS rate was superior for the neoadjuvant arm (72% vs. 49%; ref. 21), endorsing the theory that neoadjuvant checkpoint inhibition is able to induce a broader and stronger antitumor immune response (20, 41).

Numerous clinical trials investigating neoadjuvant ICB in stage III melanoma are currently ongoing (Table 2). Worth highlighting are two company-driven platform trials (NCT05116202 and NCT04303169) testing novel checkpoint inhibitor combinations, often on a backbone of PD-1 blockade. Anti-TIGIT, anti-TIM3, or cytokines like IL2 or IL12 have been shown to be additive with anti–PD-1 ± anti-CTLA4 in several experimental models or early neoadjuvant trials (42–45) and are currently being tested or planned to be tested.

The only phase III trial testing neoadjuvant ICB in melanoma is currently the NADINA trial (NCT04949113), which compares neoadjuvant ipilimumab plus nivolumab versus standard adjuvant nivolumab monotherapy. In line with the PRADO trial, patients with an MPR in the neoadjuvant arm do not receive adjuvant therapy, whereas patients without MPR do receive additional adjuvant therapy (nivolumab or dabrafenib plus trametinib). The first readout is expected for the end of 2023.

In order to maximize the risk–benefit ratio in a curative-intent situation, all efforts should be made to personalize treatment regimens. Personalized treatment regimens can not only improve RFS and OS but also decrease toxicity and thereby improve the quality of life of patients. In addition, choosing the right therapy and limiting therapy switching could also reduce health care costs and thus might enable such therapies for countries with less funded health care services.

The neoadjuvant therapy setting enables personalization in different phases of treatment. First, because previous trials have shown a pathologic response upon neoadjuvant ICB as a strong surrogate marker for long-term RFS, baseline biomarkers that are predictive for pathologic response could guide the choice of neoadjuvant treatment regimens, including mono- versus combination therapies. Second, early on-treatment (changes of) biomarkers might be used to adjust (i.e., intensify or abate) the neoadjuvant treatment regimen prior to surgery. Finally, the pathologic response upon the neoadjuvant treatment could direct the extent of surgery and omit adjuvant treatment, as was already tested in the PRADO trial (25).

The efficacy of immunotherapy is the result of a complex interplay between the immune system, the tumor cells, and their microenvironment (TME), including tumor antigen uptake, antigen presentation, activation of immune cells in the draining lymph node, homing to the tumor, and execution of immune-mediated tumor-cell killing (46), previously summarized in the cancer immunogram (47). Meanwhile, additional new biomarkers have been discovered for all disease stages of melanoma, but for this review, we will restrict ourselves to markers that we consider potentially relevant for the neoadjuvant treatment setting (Fig. 2).

The tumor mutational burden (TMB) is the number of somatic mutations harbored by tumor cells, which varies greatly across cancer types. TMB is often used as a proxy for neoantigen burden in biomarker research, because recognition of tumor neoantigens is crucial for eliciting an antitumor immune response. Extensive research has demonstrated TMB to be a predictive biomarker for response and prolonged survival after ICB in different tumor types (48–51). This has ultimately led to the tissue-agnostic FDA approval of TMB as a diagnostic biomarker for treatment with pembrolizumab of patients with unresectable or metastatic solid tumors harboring high TMB [≥10 mutations per megabase (mut/Mb)] who progressed on prior therapies (52).

However, the implementation of TMB as a predictive biomarker is still facing significant hurdles (53–55). The gold-standard method for TMB determination, whole-exome sequencing (WES), is expensive and time-consuming. Therefore, efforts have been made to implement assays that can reliably extrapolate TMB from targeted panel–based sequencing data (56–58). Currently, several molecular diagnostic companies have each developed their own methodologies to reliably replace TMB quantification by WES (59). Moreover, the use of a fixed pan-cancer threshold of 10 mut/Mb for the approval of pembrolizumab across solid cancer types limits the utility, as two retrospective analyses using this cutoff could not reproduce the predictive ability of TMB for response to ICB in all cancer types (55, 60). A TMB cutoff of the highest 20% for each cancer type showed a better association between TMB and improved OS after ICB in almost all cancer types (49). Of note, TMB also varies widely between subtypes of melanoma. For example, desmoplastic melanoma has been shown to have a higher TMB than acral or mucosal melanoma (61, 62), resulting in a higher response rate upon ICB in desmoplastic melanoma and lower response rates in acral and mucosal melanoma (63, 64).

Neoantigen burden on its own might have an even stronger predictive value than TMB, because not all mutations give rise to neoantigens, and not all neoantigens are presented and/or recognized by immune cells (53, 65). Yet, so far, “traditional” neoantigen predictions that mainly focus on peptide–major histocompatibility complex (MHC) binding have shown no better prediction of response to or survival on ICB as compared with TMB (51, 66–68), indicating additional features defining the immuno­genicity of neoantigens. For example, neoantigens derived from clonal mutations might elicit a more effective antitumor immune response than subclonal neoantigens (69). Clonal TMB was revealed to be a stronger predictor of ICB response than total TMB in a large meta-analysis (70). Furthermore, traditional TMB analyses are mainly based on the calculation of nonsynonymous single-nucleotide variants, whereas indel mutations generate more “foreign” neoantigens and are considered to be more immunogenic (70). Somatic copy-number alterations (SCNA; changes to the chromosomal structure that result in gain or loss in copies of sections of DNA) can have a negative association with response to or survival on ICB when occurring in antigen presentation genes or in other relevant immune pathways, as it has been speculated that SCNAs may interfere with neoantigen loading on MHCs or may result in loss of genes [e.g., human leukocyte antigen (HLA) genes] that are needed for an immune response (71). Finally, the differential agretopicity index (DAI), a metric that calculates the predicted MHC binding affinity of the “wild-type” peptide relative to the mutated peptide, has outperformed TMB and neoantigen burden for survival prediction after ICB in different advanced melanoma and non–small cell lung cancer (NSCLC) cohorts (72).

In patients with stage III melanoma treated with neoadjuvant ICB, a high TMB was demonstrated to be associated with pathologic response and EFS (refs. 18, 24, 73; Table 3). Interestingly in the PRADO trial, TMB was associated only with response but not EFS (73), possibly due to the addition of adjuvant therapy in nonresponding patients. The different neoantigen/TMB subtypes have not been tested in the neoadjuvant melanoma setting but could become important biomarkers extending or substituting sole TMB analyses in the future.

Although there is no doubt about the potential value of TMB as a predictive marker for neoadjuvant immunotherapy, further standardization, harmonization, and cancer type–specific TMB cutoffs are warranted before broad implementation in the daily clinic.

Mutational processes caused by specific etiologies or exposures induce specific “mutation signatures,” which can represent biomarkers indicative of therapy response. In advanced melanoma, higher UV mutational signature scores were predictive for response and survival after immunotherapy, in particular in patients with low-to-intermediate TMB in the tumor (74, 75), suggesting that this can be a potential discriminator within TMB-low cohorts. The predictive value is likely owed to an increased hydrophobicity of the neoantigens, hence increased immunogenicity due to better presentation on MHC molecules and better recognition by T cells (76, 77). The UV signature has not yet been evaluated as a predictive biomarker in the neoadjuvant melanoma setting.

Other signatures not only are less specific for melanoma but also might have implications in other cancer types. For example, the baseline Apolipoprotein B mRNA Editing Enzyme, Catalytic Polypeptide-like (APOBEC) signature was predictive for MPR in a trial testing neoadjuvant ipilimumab plus nivolumab in patients with head and neck squamous cell carcinoma (HNSCC; ref. 78), and predictive for ICB responses in advanced NSCLC and urothelial carcinoma (79, 80), but not in our melanoma cohorts. Also more prevalent in other tumor types are alterations in the DNA mismatch repair pathway, leading to microsatellite instability (MSI). Five clinical trials in different tumor types showed durable responses to pembrolizumab in patients with MSI-high tumors, resulting in pembrolizumab being approved by the FDA for the treatment of any advanced, MSI-high solid tumor (52). In a neoadjuvant ICB trial in MSI-high colorectal cancer, the pathologic response rate to neoadjuvant ipilimumab plus nivolumab was 99% (81). If these pathologic responses translate into long-term EFS benefit, one might envision that this biomarker could be the basis for the omission of tumor resection and a less strict follow-up in MSI-high colorectal cancer.

Genomic alterations of specific genes (“driver mutations”) contribute to tumor growth by pathogenetic changes in cellular function and may influence the ability of the tumor to bypass immune surveillance. In stage IV melanoma, patients with a BRAF-mutated tumor had a significantly improved survival with combined ipilimumab plus nivolumab compared with nivolumab monotherapy, which was not observed in patients with a BRAF wild-type tumor (2). In the neoadjuvant melanoma setting, two trials testing neoadjuvant ipilimumab plus nivolumab showed no significant difference in pathologic response (24, 25). Subgroup analyses in the phase II SWOG-S1801 trial testing neoadjuvant pembrolizumab (21) indicate that both patients with a BRAF-mutated and BRAF wild-type melanoma have a more favorable outcome upon neoadjuvant pembrolizumab. Whether this holds true for ipilimumab plus nivolumab needs to be evaluated in the ongoing phase III NADINA trial (82). Other mutations that have been associated with ICB response/resistance and their potential immunogenicity in advanced melanoma (but have not been investigated in the neoadjuvant setting) are NRAS, SERPINB3/SERPINB4, PTEN, BCLAF1, and TP53 (83–86).

High rates of tumor-infiltrating CD8⁺ T cells, CD4⁺ T cells, and FoxP3⁺ cells have been associated with response to neoadjuvant ICB in several melanoma and NSCLC trials (17–19, 24, 30, 36). Aside from the presence or density of these tumor-infiltrating lymphocytes, their phenotype should also be considered, as expression of transcription factor TCF7, tumor reactivity marker CD39, or checkpoint PD-1 on the CD8 T-cell surface have all shown an association with improved response to ICB (87–91).

Furthermore, as the activation and expansion of specific antigen-reactive T-cell clones are required for an effective T-cell response, the diversity and clonality of the intratumoral or peripheral T-cell repertoire are also thought to be associated with ICB response (19, 92, 93). A trial in patients with stage III melanoma demonstrated higher T-cell clonality and diversity in pre- and on-treatment tumor samples of patients with response to nivolumab monotherapy, whereas patients treated with ipilimumab and nivolumab showed a more diverse pattern of T-cell clonality and diversity, lacking an association with response (18). In the OpACIN trial, a lower productive T-cell clonality in baseline tumor samples and a lower number of newly detected T-cell clones at week 6 in the peripheral blood were found in patients who relapsed after adjuvant or neoadjuvant ipilimumab plus nivolumab. Of note, neoadjuvant therapy induced greater expansion of these T cells (ref. 20; Table 3). In line with this observation, another group found that newly detected T-cell clones in the TME itself, and not expansion of preexisting T-cell clones, was associated with response to PD-1 blockade in patients with basal or squamous cell carcinoma (94).

It is now generally considered that a diverse T-cell repertoire at baseline and a more clonal T-cell repertoire during therapy could predict improved response to ICB, but validation in larger cohorts in the neoadjuvant setting is needed.

Dendritic cells (DC), in particular the basic leucine zipper transcription factor ATF-like 3 (BATF3) DCs, play an important role in cross-presenting antigens to CD8⁺ T cells and attracting them into the tumor (95). The role of this DC subtype for outcome upon neoadjuvant immunotherapy is reflected by the BATF3⁺-DC gene signature (96). Patients with stage III melanoma were more likely to relapse after neoadjuvant or adjuvant ICB when they had a low expression of the Batf3⁺–DC signature in their pretreatment tumor biopsy (97). In addition, CXC-chemokine ligand 9 (CXCL9) and CXCL10, which are mainly produced by BATF3⁺ DCs, recruit T cells and B cells into the TME (96, 98), have also been associated with improved response to ICB in metastatic melanoma (32, 98, 99).

Indoleamine 2,3-dioxigenase 1 (IDO1) can suppress these DCs, but also natural killer and T effector cells, by catalyzing tryptophan into kynurenine and upregulating regulatory T cells and neovascularization (100). High IDO1 is associated with resistance to anti–PD-1 in NSCLC (101). Recently, a new anti-IDO/PD-L1 vaccine in combination with nivolumab showed promising results in patients with stage IV melanoma (102), potentially leading to the renaissance of IDO targeting. Whether IDO1 can function as a predictive biomarker for response in the neoadjuvant stage III melanoma setting needs to be evaluated.

Lymphoid formations [tertiary lymphoid structures (TLS)] can be formed in nonlymphoid tissue upon chronic inflammation but also in tumors. They induce an influx of immune cells into the tumor and have been associated with improved prognosis in multiple cancers (103). The ectopic lymphoid tissue consists of aggregates of immune cells (103) and B cells in the TLS, which have been shown to be predictive for response to ICB in melanoma, renal cell carcinoma, and sarcoma (18, 104, 105). The chemokine CXCL13 is thought to be a major mediator in TLS formation and B-cell attraction into the TLS (90, 106). CXCL13 has been identified as a biomarker for response upon ICB in bladder cancer, potentially superior to PD-L1 expression or the IFNγ signature (106). In melanoma, CXCL13 was associated with improved RFS after neoadjuvant anti–PD-1 (32). Further analyses in larger cohorts are warranted to elucidate the relevance of this marker.

Tumor-associated macrophages (TAM) are important in multiple ways during the antitumor immune response and can be proinflammatory (M1-like macrophages) or anti-inflammatory (M2-like macrophages; refs. 107, 108). A decrease in M2-like macrophages has been shown to be associated with pathologic response after neoadjuvant ICB in patients with stage III melanoma (36), suggesting that blocking the M2-like macrophage skewing could increase pathologic response after neoadjuvant treatment. Multiple approaches influencing macrophage activity are currently being investigated (109)—for example, by repolarization of TAMs into M1-like phenotype (e.g., by CSF1R inhibitors or CD40 agonists), inhibition of the tumor-promoting function (e.g., by TIM3 blockade), decreasing their survival (e.g., by CSF1 inhibition), suppressing macrophage recruitment (e.g., by CCL2/CCR2 inhibition), designing novel macrophages (e.g., by chimeric antigen receptor–expressing macrophages), or removing blockage of phago­cytosis (109).

Another example of improving the antitumor function of macrophages is targeting CD47 on tumors or its receptor signal receptor protein-alpha (SIRP-α) on the macrophages, which have been shown to mediate phagocytosis inhibition (110). High expression of CD47 or SIRP-α has been associated with impaired outcomes in multiple malignancies (111, 112). In preclinical models, targeting CD47 enhanced tumor cell phagocytosis by M1 and M2 macrophages (113) and dual targeting of PD-1 and CD47 showed an increase in antitumor immune response (114), increasing the possible relevance of CD47 or SIRP-α being predictive biomarkers in the neoadjuvant ICB setting. Yet, so far, no data are available.

Although various checkpoints have been extensively tested in the neoadjuvant melanoma setting, often on the backbone of anti–PD-1 (Table 2), the use of checkpoint (ligand) expression as a biomarker is restricted. The tumor expression of PD-L1 (one of the two ligands of PD-1) has been approved as a companion diagnostic for anti–PD-1 therapy in several cancer types including NSCLC, HNSCC, urothelial carcinoma, and triple-negative breast cancer (52). However, the results in melanoma are conflicting (115). In stage III melanoma, some trials showed a significant association between PD-L1 and (pathologic) response or RFS after neoadjuvant treatment (18, 20, 25), whereas others did not find this association (33, 36), making PD-L1 expression an unreliable marker for neoadjuvant therapy personalization (Table 3). Expression of PD-1, CTLA4, or LAG3 has been shown to correlate with response upon targeting in late-stage disease, but data in the neoadjuvant space are pending or not convincing (36, 116–120).

Finally, a currently underexamined mechanism of the cancer immune evasion is hypersialylation and the binding of these glycans to immune-inhibitory sialic acid–binding immunoglobulin-type lectins (siglec; refs. 121–123). In meta­static melanoma, expression of siglec-3 and -7 binding sialoglycan ligands has been associated with anti–PD-1 resistance (124). We are currently investigating these siglecs as predictive biomarkers in our cohorts. Multiple new therapies interacting with this siglec–sialic acid axis are currently being tested (121), making hypersialylation a promising new therapeutic target and increasing the relevance of the understanding of the role as a biomarker.

In contrast to the presence of single immune cell subsets or checkpoint molecules, immune gene expression signatures could offer a wider representation of an ongoing antitumor immune response within the TME. The 18-gene tumor inflammation signature (TIS) represents an activated but suppressed adaptive antitumor immune response and was first described by Ayers and colleagues (125). Higher expression of the TIS is strongly correlated with ICB response in multiple cancer types and independent of TMB (126, 127). The TIS has been developed into a validated clinical assay and could be used as a pan-tumor predictive biomarker.

In the neoadjuvant melanoma setting, a more confined signature described by Ayers and colleagues, called the “preliminary IFNγ signature,” has been tested extensively. An IFNγ signature algorithm proved indeed to be a predictive baseline biomarker for pathologic response and relapse in several neoadjuvant trials (refs. 24, 25, 38, 73; Table 3). A combination of TMB and IFNγ has been shown to be highly predictive for pathologic response upon neoadjuvant ipilimumab plus nivolumab in stage III melanoma, with pathologic responses between 90% and 100% in high IFNγ and high TMB in the baseline tumor biopsy compared with 39% to 42% in patients with low IFNγ and low TMB (24, 73). Based on these results, the DONIMI trial was the first trial to prospectively use an IFNγ signature algorithm for patient stratification to different neoadjuvant treatment regimens. Even though domatinostat did not show an additive effect, the trial did confirm the predictive value of the IFNγ signature with 14/20 (70%) MPR in the IFNγ-high group and only 5/20 (25%) MPR in the IFNγ-low group (38). Moreover, the evaluation of early on-treatment changes in the IFNγ signature indicated the relevance of the IFNγ signature algorithm for therapy adjustments during neoadjuvant immunotherapy. Patients with a low IFNγ signature in the baseline biopsy and a high IFNγ signature after one dose of ipilimumab and nivolumab (IFNγ signature low >high) had a pathologic response rate of 80%, while in patients who continued to have a low IFNγ signature (IFNγ signature low >low) in their on-treatment biopsy 0% achieved a pathologic response (38). These results suggest that early on-treatment biopsies during neoadjuvant therapy can help to identify patients who will not benefit from the current neoadjuvant treatment options and could benefit from an on-treatment escalation with novel combinations, such as anti–PD-1 + anti-CTLA4 + IL2 (45), anti–PD-1 ± anti-CTLA4 with intermittent BRAF/MEK inhibition (128), or anti–PD-1 ± anti-CTLA4 + anti-TIGIT or anti-LAG3 (129, 130).

The previously described association of the Batf3⁺-DC gene signature with outcome upon neoadjuvant checkpoint inhibition could be explained by insufficient CD4 help in tumors that do not respond to the treatment. Indeed, a low CD4/IL2 signature in tumor material from neoadjuvant-treated patients was associated with pNR, but when IL2 was added to tumor fragments of patients with a pNR, their profile changed to that of responding tumors (45). Considering the idea of CD4 inducing an IL12-driven DC maturation, one might postulate that the addition of IL12 might be effective in patients with a low CD4/IL2 or TIS/IFNγ signature in their tumor. In a study in 10 melanoma patients receiving a combination of neoadjuvant intratumoral IL12 and anti–PD-1, the researchers observed high pathologic response rates (MPR in 87%), suggesting addition of IL12 could be beneficial in patients with a low inflammatory gene signature in their baseline biopsy (44).

The TME consists, aside from tumor cells and immune cells, also of tumor stroma: connective tissue and vasculature exercising supportive functions and playing an important, underrated role in tumor growth, metastasis, and therapeutic resistance (131).

A well-investigated example is cancer-associated fibroblasts (CAF), which secrete extracellular matrix factors, promoting tumor growth, survival, and migration and are able to form a network preventing intratumoral CD8⁺ T-cell migration (132). CAFs are associated with an impaired immune response and drug resistance after ICB (133). UV radiation, the primary etiologic factor for skin cancers such as melanoma, causes a change of the dermal fibroblasts into a CAF phenotype. CAF activity is characterized by a six-gene signature, and this CAF signature is predictive for response to anti–PD-1 in metastatic melanoma (134). Targeting the CAFs could increase the response to the ICB, for example, by inhibition of CAF activation via targeting TGFβ or CXCR4 (135, 136) or by reprogramming the CAFs with vitamin D/A receptor antagonists (136–138).

In addition, LRRC15 expression on the CAFs surrounding the tumor cells (demonstrated in the specific LRRC15⁺ CAF signature) is highly expressed in multiple tumor types and associated with poor response to anti–PD-L1 therapy in patients with bladder cancer and NSCLC (139). The (LRRC15⁺) CAF signature could serve as a biomarker for new targeting initiatives (140), even in early-stage melanoma, because the CAFs also play an important role in primary melanomas.

Endothelial cells and pericytes in the tumor stroma play an important role in angiogenesis (131). CAFs are thought to secrete vascular endothelial growth factor receptor (VEGFR) and to induce expression of leucine-rich alpha-2-glycoprotein 1 (LRG1; refs. 141, 142), both known to mediate tumor neoangiogenesis and are associated with impaired responses to ICB (143–146). In patients with stage III melanoma treated with neoadjuvant ipilimumab and nivolumab, circulating VEGFR-2 levels are associated with pNR (24) and circulating LRG1 with relapse in patients with a pNR (24, 147). Whether patients with high VEGFR-2 or LRG1 expression would benefit from, for example, lenvatinib (blocking VEGFR1–3; ref. 148) or LRG1-targeting initiatives (bioRxiv 2020.07.25.218149) should be further investigated.

Due to selective and static measurements such as tumor biopsies, sampling bias could result in default prediction of treatment response. Capturing the spatial and temporal complexity of the tumor is essential for response prediction to ICB (149), which could be overcome by using repetitive liquid biopsies during treatment and follow-up. Determination of the presence of microscopic residual disease after neoadjuvant systemic therapy and surgery might be the hallmark for decisions on subsequent adjuvant therapy indications, with circulating tumor DNA (ctDNA) being the most powerful tool for detecting residual disease. The presence of pre- or postsurgery ctDNA has been associated with poor response to ICB in multiple cancer types such as melanoma and urothelial carcinoma (150–154). In patients with stage III melanoma treated with adjuvant ICB, the presence or increase of ctDNA after surgery has been associated with decreased RFS and DMFS (154, 155), indicating that there is still a tumor present after surgery. Although the role of ctDNA has not yet been confirmed in neoadjuvant ICB trials, we hypothesize that ctDNA could also assist in the selection of adjuvant or neoadjuvant treatment, because it provides additional information to the IFNγ signature and TMB (156). In stage II colorectal cancer, researchers compared ctDNA-based adjuvant therapy (treating only the patients with detectable ctDNA levels after surgery) with standard adjuvant therapy and found that fewer patients required adjuvant chemotherapy, whereas relapse rates remained similar (157). These findings imply that ctDNA could also serve as a marker for patient selection. This is currently being investigated in stage II melanoma in the DETECTION study (NCT04901988), which is treating stage IIB/C melanoma patients with a postsurgery elevated ctDNA with either adjuvant ICB or only at the time of confirmed melanoma metastasis (158).

Next to being a biomarker on its own, ctDNA can also be used to determine the relative TMB (blood TMB), which has been shown to be reliable for predicting response to ICB in metastatic NSCLC (159–161). In stage III melanoma, this could provide an insight into the TMB at baseline, without requiring a baseline biopsy.

Posttreatment circulating PD1⁺ CD8⁺ T cells have been shown to be predictive for ICB response in advanced metastatic NSCLC and for RFS in patients with stage III melanoma treated with adjuvant anti–PD-1 (162, 163). Circulating PD-1 has indeed also been detected, using the OLINK assay, increasing strongly after neoadjuvant ipilimumab and nivolumab, but its baseline expression had no predictive value for response (24).

Upon neoadjuvant nivolumab and relatlimab in resectable stage III–IV melanoma, high rates of circulating eomesodermin (EOMES)-expressing CD8⁺ T cells after treatment were associated with favorable outcomes (ref. 36; Table 3). These EOMES⁺ CD8⁺ T cells are thought to play an important role in the tumor infiltration of CD8⁺ T cells and thus the antitumor immune response (164, 165). However, further research is needed to determine whether the CD8⁺ T-cell EOMES expression is useful as a baseline biomarker.

Extracellular vesicles, for example, exosomes, are thought to be important in intercellular communication and to play an important role in tumor progression and metastasis (166). PD-L1–expressing exosomes have been postulated to mediate tumor-mediated systemic immune suppression (167). The preclinical work by Poggi and colleagues showed that not PD-L1 expression on the tumor itself, but expression on exosomes, mediated PD-L1/PD-1 tumor immune escape (167). Thus, tumors that express PD-L1 but produce exosomes to a lesser extent might be less susceptible to PD-L1/PD-1 blockade than tumors that do produce exosomes, possibly explaining the incongruences on tumor PD-L1 expression as a biomarker for response to neoadjuvant ICB. Indeed, PD-L1–expressing exosomes have been shown to be associated with response to anti–PD-1 (168) and to anti-CTLA4 in metastatic melanoma (166), but need to be confirmed in larger cohorts and the neoadjuvant setting.

Circulating cytokines have been proposed as another way of measuring the activity of the immune system. For example, high circulating IFNγ is associated with response in melanoma and NSCLC (169, 170), whereas IL6 and IL8 are associated with impaired response to ICB in multiple tumors such as melanoma (171–175). IL6 is also thought to play an important role in the irAE development (176, 177). In patients with stage III melanoma treated with neoadjuvant ipilimumab, high levels of IL10 at baseline were associated with disease progression, and high levels of IL17 were associated with toxicity (ref. 178; Table 3). In patients treated with neoadjuvant ipilimumab and nivolumab, circulating CX3CL1 was associated with nonresponse (24), but an independent confirmation cohort was missing. The predictive value of circulating cytokines and chemokines for response or toxicity should be further investigated.

Finally, an autoantibody signature is developed from baseline serum autoantibodies in patients with resectable stage III–IV melanoma treated with nivolumab, ipilimumab, or a combination of nivolumab and ipilimumab to predict the likelihood of recurrence and the risk of developing significant toxicity (179). The signatures for recurrence and toxicity showed little overlap, indicating a different pathophysiology (179). To determine if autoantibody biomarkers also apply to patients who received neoadjuvant treatment and if they can predict pathologic response, more research is required.

HLA genes encode cell-surface proteins that are responsible for antigen presentation to T cells and are known to be the most polymorphic in humans (180). This variation is located mainly in the antigen-binding groove, altering the peptide-binding specificity of HLA molecules. A study in >1,500 patients with advanced cancer showed that a more diverse array of HLA-I molecules (i.e., maximal HLA-I heterozygosity at loci “A”, “B,” and “C” vs. homozygosity for at least one locus) was associated with increased survival after ICB, possibly due to a broader presentation of tumor antigens to CD8⁺ T cells (181). The combination of HLA heterozygosity and TMB enhanced the association with increased survival. Analysis of specific HLA-I super types showed that HLA-B44 was associated with improved survival and the HLA-B62 super type with decreased survival in patients with advanced melanoma (181).

Mechanisms interfering with the antigen-presenting pathway via the HLA system have been associated with resistance to ICB therapies. Examples are loss of heterozygosity (LOH) of HLA-I genes (181, 182), downregulation of HLA-I expression (183), and mutations that disrupt the function of the β2 microglobulin (B2M) molecule that stabilizes the HLA-I complex (184).

To date, HLA heterozygosity, super types, and LOH have not been tested in the neoadjuvant melanoma setting due to too-small cohorts, but if neoadjuvant immunotherapy becomes standard therapy, HLA aberrations should be further investigated.

Over the past decades, it has become evident that the gut microbiome has a complex and diverse role in many processes in the body, including alteration of the immune system and thus the antitumor immune response. A protumor microbiome causes hyperinflammation, altered cytokine levels, and release of genotoxic chemicals such as carcinogens and mutagens, whereas an antitumor microbiome could increase immune surveillance, TLS, and molecular mimicry (185). The molecular mimicry between the tumor-associated antigens and the bacterial antigens of the microbiome increases the potential for antitumor T-cell response (186).

In several studies, predominantly implemented in NSCLC and melanoma, it has been shown that the composition of the gut microbiome (and especially its diversity) affects the sensitivity to ICB and the risk of irAEs (187–189). Antibiotic use decreases this diversity, subsequently reducing response to ICB in melanoma, NSCLC, and RCC (190–193). In patients with stage III melanoma treated with neoadjuvant ipilimumab and nivolumab, the Ruminococcaceae-dominated microbiomes were associated with higher response rates and lower toxicity compared with Bacteroidaceae-dominated microbiomes (194). The prevalence of the Bacteroidaceae-dominated microbiome was more prominent in Australia and the United States compared with the more frequent Ruminococcaceae-dominated microbiome in The Netherlands (194), suggesting that certain patients could benefit from lifestyle interventions depending on geographic location. In line with this notion, a cross-cohort study identified a panel of species, including Bifidobacterium pseudocatenulatum, Roseburia spp., and Akkermansia muciniphila, associated with response to ICB, but no single species could be regarded as a fully consistent biomarker across studies (195). Therefore, there is currently no clear-cut biomarker that could be used to predict response in neoadjuvant immunotherapy.

In an era of rapidly growing knowledge of cancer evolution and resistance mechanisms, and increasing availability of anticancer-specific therapies, the need for biomarker-driven personalized trials becomes more and more obvious. Due to its immunogenic properties, melanoma has been one of the most investigated cancer types regarding immunotherapies and is currently at the forefront of clinical neoadjuvant ICB research, providing optimal conditions for biomarker research.

In our opinion, biomarker research in the neoadjuvant melanoma setting should focus on baseline biomarkers identifying the most optimal neoadjuvant treatment compound or combination for each individual patient with regard to efficacy and toxicity. Naturally, each personalized therapy should aim for the highest probability of MPR, facilitating de-escalation of subsequent surgical procedures and adjuvant treatment regimens because pathologic response has been shown to be strongly associated with survival (22, 25, 196). Extensive surgery or adjuvant therapies should be reserved for only those patients in whom the tumor does not respond despite their personalized neoadjuvant treatment regimen. We speculate that this could even be pushed further, by omitting any form of surgery when a deep response is confirmed by imaging and/or multiple biopsies showing (near) complete pathologic responses.

In addition, we should also focus on toxicity, as the potential of immunotherapies is being explored in earlier stages of disease with longer life expectancies and a stronger emphasis on quality of life. Biomarker research for toxicity prediction is still in its infancy but is expected to gain attention because more and more patients are being cured. Thus, efficacy and toxicity should be used as twin objectives to guide patients’ treatment decisions, balancing the probability of both pillars on each individual's situation, prognosis, and preference (Fig. 3). The ultimate goal for personalized biomarker-driven neoadjuvant therapy is a highly effective and minimally toxic systemic therapy for each individual patient, with a short-term treatment duration and limited impact on quality of life.

Until we reach that goal, novel treatment compounds and treatment combinations should be primarily tested in patients who are thought to have a low chance of response to currently available therapies. The poor prognosis of these patients could justify the investigation of new combinations in these patients with early-stage cancer, facilitating accelerated treatment innovation, instead of waiting for results of trials in heterogeneous late-stage cancer patient populations.

Trials investigating these novel treatment combinations should be adaptive—with the possibility to stop early for futility—and fast and efficient in order to make optimal use of the relatively limited resources (patients, patient samples, and time). An example is an adaptive umbrella trial where in one cancer type, multiple agents with specific molecular targets can be tested based on specific biomarkers. These multiple treatment arms can be implemented and adapted under a master protocol in order to enhance logistic and regulatory efficiency. Another example of an adaptive trial design has previously been proposed by our group: “the Lombard Street approach” (197), which focuses on identifying biomarkers that predict response upon a certain treatment and using this biomarker in a subsequent trial to treat patients with favorable biomarkers with the identified therapy and patients with unfavorable biomarkers with new treatment regimens.

In conclusion, we believe that current biomarker knowledge in the neoadjuvant melanoma field could serve as a poster child and thus as a tool in treatment personalization for all stages of melanoma and other tumor types, allowing all patients to receive the appropriate medication. The ultimate objective of all cancer research will continue to be curing every tumor without compromising the patient's quality of life.

I.L.M. Reijers reports other support from Signature Oncology outside the submitted work. C.U. Blank reports personal fees from advisory roles for Bristol Myers Squibb, MSD, Roche, Novartis, GSK, AstraZeneca, Pfizer, Lilly, GenMab, Pierre Fabre, and Third Rock Ventures and grants/research funding from Bristol Myers Squibb, Novartis, NanoString, and 4SC during the conduct of the study; is a cofounder of and owns shares in Immagene BV and Signature Oncology; and is listed as an inventor on several related patents (including submitted): WO 2021/177822 A1, N2027907, and P091040NL2. No disclosures were reported by the other author.

The figures were created with BioRender.com.