Introduction
In follicular lymphoma (FL) there is a clinical need for pre-treatment identification of high-risk patients, recognisable as the subset of patients (15–30%) with early progression and poor survival outcomes despite therapeutic advances [
1‐
3]. Current FL prognostic biomarkers, such as the Follicular Lymphoma International Prognostic Index (FLIPI) [
4,
5], are well validated but lack the necessary precision for clinical decision-making. To improve risk stratification, subsets of tumour-infiltrating lymphocytes (TILs) have been studied, but there is no consensus on the observed effect [
1,
6‐
8]. Immune subsets of prognostic interest include, among others: CD68
+ lymphoma-associated macrophages [
9‐
12], CD3
+ T cells [
10,
12‐
14], CD4
+ T helper cells [
12‐
14], CD8
+ cytotoxic T cells [
12‐
15], CD4
+FOXP3
+ T regulatory cells (Tregs) [
12,
14,
16‐
18], CD21
+ dendritic cells [
12], mast cells [
19], and PD-1 expressing T-cells [
12,
14,
20]. These cells engage in crosstalk through multiple pathways [
21], and therefore a holistic observation of immune infiltrate diversity could be more informative than examining isolated components.
In ecological sciences the Shannon diversity (or entropy) index quantifies biodiversity in terms of “evenness” [
22]. For example, if three species are found in an area, and one accounts for 99% of the population, this community would be considered less diverse than one where the three species are found in approximately equal abundances. Entropy is calculated from the proportion of each species in the community and increases when diversity is higher. It has found applications in histopathology to quantify heterogeneity of HER2 expression [
23] and chromosome 8q24 copy number variation [
24] in breast cancer. If we consider each cell phenotype as a species, this metric can be applied to quantify immune infiltrate diversity.
Similarly, it is possible to quantify the diversity of not only phenotypes but also their spatial interactions, which is recognised for its potential as a biomarker [
25] for many tumour types including FL [
16,
26]. The hypothesised interactions distribution (HID) method [
27] can identify spatial interactions defined as co-occurrences of different cell types within 30 μm. The diversity of these spatial interactions can also be investigated using entropy.
We aimed to develop a methodology to quantitatively assess immune infiltrate diversity in the tumour microenvironment of FL and test its potential utility as a prognostic biomarker. To this end, an automated multiplex immunofluorescence and image analysis pipeline were developed to simultaneously identify cells positive for CD4, CD68, CD8, CD21, FOXP3, and PD-1. We show that increased diversity of immune infiltrate populations and interactions is associated with improved overall survival (OS) in a cohort of FL patients.
Discussion
This study introduced a 6-plex immunofluorescence protocol for concurrent observation of immune subsets and an image analysis pipeline to accurately detect cell types and objectively measure tumour microenvironment diversity. This new approach provides a versatile and adaptable platform that could be extended to other tumour types. The proposed pipeline benefits from precise marker localisation as well as conservation of valuable tissue material through multiplexing. The improved accuracy and reliability of quantitative immunofluorescence compared to conventional immunohistochemistry, and its cost-effectiveness compared to in situ hybridisation, provide scope and rationale for wider clinical adoption.
Developing baseline prognostic biomarkers for risk stratification is a major area of research in FL, driven by an urgent need to develop effective therapies capable of improving the outcomes of high-risk disease. Using this pipeline, we report that increased diversity of immune infiltrate populations and interactions in FL is potential biomarkers of favourable OS. Diversity was quantified through a novel approach using Shannon’s entropy, a metric describing species biodiversity in ecological sciences. The diversity of spatial interactions remained significant after Bonferroni correction for multiple comparisons in multivariable analysis of OS. Therefore, this biomarker could improve risk stratification, offering additional prognostic value when combined with FLIPI assessment. The diversity biomarkers also outperformed simple cell density measurements. Indeed, none of the immune infiltrate cell densities remained significantly associated with survival endpoints in multivariable analysis (Table
2), similar to results reported by others [
14] for rituximab-treated patients. This evidence supports applicability of the diversity biomarker for risk stratification in FL.
Survival analysis (Tables
1,
2,
3) in this study treated all variables as continuous, to avoid the loss of information from arbitrary dichotomisation [
43]. However, since dichotomisation is sometimes required for clinical decision-making, we also carried out Kaplan–Meier analysis by selecting a cut-off to split the patients in two groups. Clinical studies often select the median or quartiles as the cut-off, even though is no underlying statistical reasoning for this selection [
44]. We adopted the Contal and O’Quigley approach [
40], as it provides a non-arbitrary cut-off selection and supplies a corrected
p value, taking into account the inflated type-I error that may result from testing multiple cut-offs. Further validation of this cut-off in additional cohorts would be beneficial. However, since we performed Cox regression analyses without dichotomising the variables, the reported prognostic value does not rely upon a specific cut-off selection.
Previous studies investigating tumour immune microenvironment diversity in other types of cancer have demonstrated the importance of diversity in T-cell populations, as measured by T-cell receptor (TCR) next-generation (NGS) sequencing, in a way that is agnostic to the types of T-cells that are quantified [
45]. Increased TCR diversity has been associated with improved clinical benefit in metastatic melanoma [
46] and favourable overall survival in metastatic breast cancer [
47]. Furthermore, clonal TCR diversity has been shown to increase after immunotherapy treatments (e.g. cryo-immunotherapy for breast cancer [
46] and Sipuleucel-T immunotherapy for prostate cancer [
48]) and is investigated as a potential endpoint for response to therapy [
47]. A diverse T-cell repertoire is thought to increase the likelihood that a useful anti-tumour T-cell population is present [
46], leading to favourable outcomes. In this study we expand the concept of diversity to include T-cells and macrophages and propose that a diverse repertoire of immune cells in the microenvironment of FL could similarly increase the likelihood of relevant anti-tumour pathways being active.
In this study, CD68
+ macrophages were significantly correlated with favourable OS in univariable analysis. A favourable trend of increased CD68
+ density was observed for PFS and POD24. This effect could be attributed to one of the mechanisms of action of the anti-CD20 rituximab treatment, whose immune-mobilising effects include the induction of antibody-dependent cell phagocytosis [
49]. Consequently, cells coated with rituximab are recognised by macrophages as targets and killed [
50]. The favourable effect of macrophages has been previously demonstrated in a rituximab-treated cohort [
10]. However, this effect depends strongly on the type of treatment, as in cohorts treated without rituximab [
9,
11] increased numbers of tumour-associated macrophages correlated with unfavourable outcome.
To ensure reproducibility of results, we quantitatively validate the staining assay and cell detection algorithms and share publicly the image dataset [
28]. The FL cohort included treatment pathways and prognostic outcomes reflective of current modern practice. The TMA technology employed is equivalent to whole section assessments in lymphomas [
51], enabling rapid processing of large number of samples. Furthermore, the diversity metrics demonstrated low intra-patient heterogeneity (CoV = 7.7–8.3%), indicating robustness when assessed using triplicate TMA core samples.
A limitation of this study is the use of a single cut-off to score positive and negative cells for each stain. Robust cut-offs were selected by two different users of the computer-assisted scoring system. However, this approach may sometimes underperform because of the inherent variation of staining intensities in positive cells. Notably, in FL two functionally different PD-1
+ cell phenotypes have been observed [
20], characterised by different levels of PD-1 expression: PD-1
+high T follicular helper cells found inside the follicles actively support FL B-cell growth, while the PD-1
+low cells found outside the follicles represent exhausted T-cells. The PD-1
+ T-helper cells found within the follicles are also known to express CD4 less strongly (30.7% lower CD4 intensity) compared to other CD4
+ cells in the interfollicular areas [
52]. The present study attempted to select single cut-offs able to pick up both the weakly and strongly positive cells. Use of multiple cut-offs was avoided as nuanced intensity variations can be challenging to capture using manual gating in a reproducible manner. Future work could adopt automated clustering [
53] of cells based on their intensities or rely on additional functional markers in the multiplex panel (e.g. TIM3 for exhausted phenotypes or CXCR5 for T follicular helper cells [
20]) to differentiate between PD-1 subsets.
Cox regression for PFS and logistic regression for POD24 in the rituximab-treated subset did not demonstrate significant prognostic value for any of the immune infiltrate biomarkers. However, the limited size and variable treatment increase the risk of false-negative results. Therefore, the effect of tumour microenvironment diversity on early relapse merits further investigation before it could be ruled out. A significantly lower total immune infiltrate ratio was observed in patients that had POD24, as seen in Fig.
3. Even though the causality underpinning this observation is not well understood, the total immune infiltrate ratio may be further investigated as an indicator of POD24 that can be calculated at baseline, before treatment has started. Additional markers of B-cells and particularly of FL B-cells would be beneficial for a meaningful study of the “DAPI only” cells. The lack of a FL tumour marker is a limitation of this study, which prioritised the analysis of the non-neoplastic, immune microenvironment. By assessing 7 markers (DAPI, CD4, CD8, CD68, FOXP3, CD21, PD-1), we have reached the limit in number of markers permitted by the multiplex staining and imaging platform used in this study.
In summary, automated assessment of immune infiltrate and its diversity, based on multiplex immunofluorescence, warrants further exploration for prognostic biomarker development in FL. Future work may involve validation of diversity measurements using orthogonal assays, such as gene expression profiling. This pipeline is ready to be tested in larger series, with the potential to significantly improve risk stratification and treatment adaptation for high-risk FL in the future.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.