Primary Tumor Characteristics as Biomarkers of Immunotherapy Response in Advanced Melanoma: A Retrospective Cohort Study

Simple Summary Melanoma survival rates have been vastly improved by the use of immune checkpoint inhibitors. However, some patients do not respond to the treatment or experience progression of disease. Because of this variability of response, markers predicting efficacy of these treatments is of interest. This study aimed to clarify the role of primary tumor characteristics in melanoma treatment and survival. The authors demonstrated that cutaneous melanoma and its subtypes were significantly associated with response, progression-free survival, and overall survival compared with acral or unknown primary melanoma. Other primary features did not demonstrate an association on multivariable analyses. Thus, primary features, other than cutaneous primary, should likely not influence metastatic treatment selection. Abstract Identifying patients likely to benefit from immune checkpoint inhibitor (ICI) treatment remains a crucial goal for melanoma. The objective of this study is to assess the association between primary tumor features and immunotherapy response and survival in advanced melanoma patients. In this single-center retrospective cohort study, disease characteristics, response to immunotherapy, PFS, and OS were assessed among melanoma patients (excluding mucosal and uveal primaries) treated with ICI. Among 447 patients, 300 (67.1%) received anti-PD-1 monotherapy and 147 (32.9%) received ipilimumab/nivolumab. A total of 338 (75.6%) had cutaneous melanoma, 29 (6.5%) had acral melanoma, and 80 (17.9%) had melanoma of unknown primary. Ulceration and stage at initial presentation were associated with inferior outcomes on univariate analysis. However, on multivariate analysis, this result was not observed, but cutaneous melanoma and each of its subtypes (superficial spreading, nodular, other, unknown) were positively associated with response, longer PFS, and longer OS. Metastatic stage (M1c, M1d) at presentation (OR = 1.8, p < 0.05) and BRAFV600E mutation status (OR = 1.6, p < 0.001) were associated with shorter PFS. This study is limited by its retrospective and single-center design. Cutaneous melanoma and its subtypes were significantly associated with response, PFS, and OS compared with acral or unknown primary melanoma.

Despite the growing use of ICI across tumor types, there is a lack of well-defined prognostic or predictive markers of ICI efficacy and survival.While efforts have been made to identify factors associated with response (and lack thereof), such as tumor factors, immune microenvironment, and host factors [6][7][8][9][10][11][12], the impact of primary tumor characteristics on therapy response is unclear.One study found that primary melanoma thickness, ulceration, and stage were associated with brain metastases [13].Thus, the primary tumor characteristics could reflect the underlying disease biology (e.g., hypoxia, immune response, aggressiveness) that impacts other aspects of metastatic behavior including the immunotherapy response.Another study demonstrated that primary melanomas located on sun-exposed skin treated with first-line anti-PD1 therapy had improved progression-free survival (PFS) compared to those on intermittently sun-exposed areas or sun-protected areas [14].
In this single-center retrospective cohort study, we assessed clinical features, primary tumor characteristics, response, and survival outcomes in 447 patients with unresectable or metastatic melanoma treated with ICI.

Overview
We conducted a single-center retrospective cohort study to assess the characteristics of primary melanomas and ICI outcomes.The study was approved by the Vanderbilt University Medical Center institutional review board.Patients receiving one or more doses of anti-PD-1 monotherapy or combination of ipilimumab and nivolumab were included.Patients with melanoma of mucosal or uveal origin were excluded due to their distinct tumor biology and published lower responses to anti-PD-1 [15][16][17].

Data Collection
Retrospective clinical data were collected, including date of initial diagnosis, anatomic site of primary, subtype (cutaneous, acral, or unknown primary), histologic subtype (superficial spreading [SSM], nodular, other, unknown), Breslow thickness, ulceration status (0/1), number of mitoses, presence of lymphovascular invasion (LVI), and tumor-infiltrating lymphocytes (TILs).The "other" cutaneous histologic subtype category aggregated multiple subtypes including desmoplastic, spindle cell, lentigo maligna, amelanotic, spitzoid, and balloon cell.Pre-immunotherapy treatment details, BRAF and other mutation status, number of positive sentinel nodes (0, 1-2, 3+), M status, and stage (AJCC 8th edition [18]) after the staging workup were collected.LDH level and stage (AJCC) at the time of immunotherapy initiation were collected.Immunotherapy outcomes were evaluated according to the therapeutic response (RECIST criteria), progression details, and date of last follow-up, which was defined as the date of last survival follow-up or death, whichever occurred later.

Statistical Analyses
Descriptive statistics were used to summarize patient and treatment characteristics.Univariate analyses assessed the association between each primary tumor characteristic and the response to ICI, progression-free survival (PFS), and overall survival (OS).Categorical variables and continuous variables were analyzed using a chi-squared test and a Wilcoxon rank-sum test, respectively.PFS and OS were tested using log-rank tests.Continuous variables were summarized by mean, median, and interquartile range (IQR).To estimate the independent effect of each clinical variable on the outcome, we conducted multivariable analyses using logistic regression for the response and Cox regression for the PFS and OS.In the Cox models, variables that did not meet the proportional hazard assumption were treated as time-dependent covariates, allowing their effect on the outcome to change over time.The multivariate analyses results were reported as odds ratios (ORs) or hazard ratios (HRs) with 95% confidence intervals (CIs).A backwards selection identified the independent variables that were most strongly associated with the outcome to avoid overfitting and improve model interpretation.To assess the impact of treatment, a reduced model without treatment status was compared to the full model using a likelihood ratio test.All of the statistical analyses were conducted using R v4.2.2.

Response to ICI
The multivariable analyses with backward selection revealed that gender, TILs, treatment, Breslow thickness, stage at presentation, M status, LVI, age, prior therapy, ulceration, mitoses, and elevated LDH were not significantly associated with the response to ICI (p > 0.05).In the reduced logistical regression model, the superficial spreading subtype (OR = 6.5, p < 0.01), nodular subtype (OR = 10.5, p < 0.01), other cutaneous subtype (OR = 9.0, p < 0.01), unknown cutaneous subtype (OR = 7.4, p < 0.01), and unknown subtype (OR = 7.1, p < 0.01) were associated with the response (reference group: acral).The BRAF V600E mutation (OR = 0.4, p < 0.01) and 1-2 positive sentinel nodes (reference group: 0 positive sentinel nodes; OR = 0.50, p < 0.01) were negatively associated with the response.When treatment was added to the model, it was not a significant predictor of the therapeutic response (OR = 1.1, p = 0.7) and did not significantly affect the model likelihood ratio test or discrimination indexes.We also wondered whether the loss of association of ulceration with response was due to a more aggressive metastatic presentation in patients with ulceration.Ulceration was associated with higher metastatic stage at presentation with a marginally significant p-value (p = 0.06) but not with higher LDH (p = 0.45) (Table 4).

Discussion
This is the first study to comprehensively assess whether primary tumor characteristics and initial presentation impacted the subsequent immunotherapy response in the metastatic setting.Though a more advanced stage of initial presentation, ulceration, and histologic subtype were associated with inferior outcomes in the univariate analyses, they were less or not significant in the multivariable analysis.Instead, cutaneous melanoma and each of its subtypes (superficial spreading, nodular, other, unknown) were significantly associated with improved clinical outcomes.Other features of the primary, including Breslow depth, TILs, mitoses, and LVI, did not impact the immunotherapy response after controlling for multiple variables.
There remains a need to identify the markers of response and survival to better select patients likely to benefit from ICI.The melanoma biology influences the immune response by regulating antigen expression and presentation, thereby playing a critical role in determining the likelihood of an immunotherapy response.The increased tumor cell expression of MHC-II and PD-L1, influenced by tumor infiltrating lymphocytes, has shown promise [19][20][21][22][23], although they remain imperfect [7,24,25].A high tumor mutational burden (TMB) is a potential marker of response, although it has a limited predictive capacity [26,27].Similarly, an inflamed tumor microenvironment, as expressed by CD8+ T cells and interferon-γ signatures, has also been correlated with a response to ICI therapies [28,29].Therapeutic resistance has also been associated with transcriptional programs of mesenchymal differentiation, angiogenesis, and hypoxia, as well as JAK-STAT mutations [30,31].While these biomarkers have promise, it is possible that the clinical and pathologic features of primary tumors could also have predictive value.
The ulceration and stage of presentation were associated with poorer PFS and OS in the univariate analyses, with similar but more modest effects seen with high mitotic rates, though these associations did not persist in the multivariable models.Thus, it is likely that factors like ulceration and a high mitotic rate are not only associated with higher rates of metastatic progression and death in the pre-ICI era [32] but also with other adverse metastatic features, such as visceral organ involvement, bulky disease, and high LDH, which correlate with ICI failure [33,34].Indeed, ulceration was associated with higher metastatic stage at presentation with a marginally significant p-value, although not with higher LDH.
The lack of an independent association between most of the primary tumor features and outcomes following ICI therapy for metastatic disease suggests that these features may not have a major predictive role in this setting.However, they add valuable prognostic information following initial diagnosis.Since most of these adverse prognostic factors were identified prior to immunotherapy development, it is noteworthy that they presage not only inferior outcomes in an era lacking effective systemic therapies but also poor responses to active modern treatment options.
All of the cutaneous subtypes were more likely to respond compared with the acral melanoma, though "other" cutaneous subtypes (lentigo maligna, desmoplastic, spindle cell, etc.) had the longest PFS.Specifically, the ORR was greatest for the desmoplastic (N = 8, RR: 75%) and amelanotic subtypes (N = 3, RR: 67%), followed by lentigo maligna (N = 11, RR: 45%) and spindle cell (N = 8, RR: 38%).It is well known that the desmoplastic melanoma has a higher response in part due to its high mutation burden [35].However, larger studies on less common atypical subtypes (such as amelanotic and lentigo maligna) Nare warranted to assess for other unexpected beneficial associations.
Several known prognostic factors were recapitulated.The metastatic stage (M1c, M1d) was associated with shorter PFS, while age and prior therapy correlated with shorter OS [36,37].The BRAF V600E mutation negatively impacted ICI response, possibly due to prior BRAF/MEK inhibitors in some patients.The optimal first-line regimen for BRAFmutant melanomas remains contested, although it should consist of immunotherapy for most patients [38].Ongoing studies aim to determine combination and sequence strategies (NCT02908672, NCT02967692, NCT02631447).
One prior study assessed primary tumor characteristics as potential biomarkers of immunotherapy response.LVI was significantly associated with an immunotherapy response and prolonged PFS and OS in advanced melanoma patients [39].Our findings are somewhat in line with this, as LVI was associated with prolonged OS.Notably, our study has a significantly larger cohort and assessed multiple other clinical variables.
The limitations of this study included its retrospective and single-center nature, and relatively small sample size.Prospective, multi-center studies with a larger and more diverse sample size should be conducted to enhance generalizability.
These findings suggest that the clinical and pathological variables of primary tumors are associated with immunotherapy response, PFS, and OS in melanoma, though these do not appear to be independent of the adverse prognostic features of the metastatic presentation.Thus, primary features, other than cutaneous primary, should likely not be integrated in predictive models [40] for metastatic treatment, but may provide additional prognostic data surrounding initial presentation.Future research should aim to replicate and validate these results to help stratify patients by likelihood of response to ICI therapy.

Conclusions
Features of the primary tumor were associated with clinical outcomes when assessing as univariate analyses (primarily ulceration and initial stage of presentation).However, these associations were largely not significant after adjusting for known prognostic variables, with the exception of cutaneous histology vs. acral and mucosal primaries.This suggests that adverse prognostic features of initial presentation may correlate with later inferior outcomes with modern therapies, although these variables are captured with known prognostic features at the time of metastatic disease.Informed Consent Statement: Informed consent was obtained for all subjects involved in the study.

Table 2 .
Confounding factors and outcomes of interest.

Table 3 .
Cont.For continuous variables, the values are categorized into quantiles for the log-rank test.(p-values and FDRadjusted p-values for log-rank tests of clinical variables by PFS). *