An artificial intelligence (AI) algorithm analysing non-small cell lung cancer (NSCLC) samples has revealed that the degree of immune infiltration in different regions of the tumour could predict the likelihood of relapse, say UK investigators.
The research, published by Nature Medicine on May 27th, looked at the spatial distribution of immune cells, as well as DNA and RNA expression and other measures, to allow an AI tool to build up an immune map of tumours from 100 patients.
It showed that some regions were packed with immune cells, known as 'hot' areas, while others apparently had no immune infiltration, dubbed 'cold' areas, potentially reflecting the ability of the tumour to hide itself from the immune system.
Crucially, the number of these immune cold regions in a tumour was directly related to the risk of relapse, independent of other factors.
Dr Yinyin Yuan, team leader in computational pathology at The Institute of Cancer Research, London, led the study.
She said in a news release that their tool "could in the future help pick out those patients with lung cancer who are at highest risk of their cancer coming back".
She added that the study has allowed new insights into "how lung cancers can cloak themselves to escape the attention of the immune system, and in doing so can continue to evolve and develop".
"Cancer's ability to evolve and to come back after treatment is one of the biggest challenges facing cancer researchers and doctors today."
Dr Yuan told Medscape News UK that, for now, the algorithm has been developed using tumour samples resected during surgery, with no prior treatment, and so is only applicable to patients with resectable disease.
Consequently, it "remains to be tested for tissues obtained in other settings", although the "principle remains the same".
"A potential limiting factor is that the tissues obtained could be smaller than surgical samples, and therefore less representative of the tumour," Dr Yuan added. "Nevertheless, this is something we are actively pursuing in our ongoing studies."
The study forms part of the TRAcking Non-small Cell Lung Cancer Evolution Through Therapy (TRACERx) study, a 9-year, £14 million project that has already offered several insights into new ways to diagnose and treat lung cancer.
Professor Charles Swanton, from University College London and The Francis Crick Institute, leads the TRACERx initiative.
He said that, "focusing on the intricacies of the tumour alone isn't enough, we need to explore the environment that each tumour is growing in and to understand its influences".
"TRACERx has given us the platform to explore this in exceptional detail and has shown us that multiple approaches at different stages of the disease are needed to outwit cancer."
Co-author Dr Mariam Jamal-Hanjani, senior clinical lecturer and consultant medical oncologist at UCL Cancer Institute, London, added: "Unfortunately, lung cancer survival remains among the lowest of all cancer types despite new treatment options for patients.
"This can be due to lung cancer cells evolving, becoming resistant to treatment causing the disease to return or worsen."
She said that the TRACERx study "is enabling us to map the evolution of lung cancer from diagnosis to cure after surgery or death, and is already revealing how we can offer patients better care and treatment".
Michelle Mitchell, chief executive of Cancer Research UK, which funds the TRACERx project, said that it is "the biggest single investment we've ever made in a lung cancer research programme, and is a massive strategic focus for Cancer Research UK".
"The breakthroughs we're seeing through TRACERx are only just the beginning, and I look forward to seeing a brighter future for people with lung cancer thanks to our investment."
For the study, Dr Yuan and colleagues studied tumour samples from the first 100 NSCLC prospective TRACERx patients, of whom 62 were men, and the median age was 68 years.
The patients had a variety of histology subtypes, with lung adenocarcinoma (LUAD) the most common, in 61 patients.
The team determined the presence of cancer cells, lymphocytes, fibroblasts and endothelial cells, as well as macrophages and other cells in stained images, while immunohistochemistry was used to identify T-cell subsets.
This was combined with multiregional exome and RNA-sequencing data, and the researchers used an AI framework to develop a deep-learning pipeline that could develop spatial profiles of immune infiltration into the tumour.
They tested the performance of the pipeline against estimates of tumour-infiltrating lymphocytes and single cell annotations by pathologists, as well as on an external validation cohort of 4324 samples of LAUD tumours resected from 970 patients at a single centre between 1998 and 2014.
The results showed that there was a high degree of immune variability between tumour regions within the same patients, independent of pathological stage.
Based on the median lymphocyte percentage within a region, the team were able to classify regions as immune 'hot', defined as a quarter standard deviation above the lymphocyte median, and immune 'cold', or a quarter standard deviation below the median. The remaining 20% were intermediate.
There were significant differences in pathology tumour-infiltrating lymphocyte estimates between immune hot and cold regions, and immune hot regions had significantly higher levels of RNA sequence-estimated immune infiltrate.
Interestingly, cancer subclones from immune cold regions were found to be more closely related in terms of their mutations, having diversified more recently than subclones from immune hot regions.
Analysis revealed that tumours with more than one cold region were at significantly greater risk of relapse than tumours with one or fewer cold regions, independent of the total number of regions sampled, and the tumour size and stage.
The external validation cohort confirmed this finding, with the number of immune cold samples per patient correlating with relapse both as a dichotomised and continuous variable.
This relationship was stronger than that for the average and variability of lymphocyte percentage per tumour, the number of immune hot regions, the proportion of immune cold regions to the total number of regions sampled, and the T-cell subpopulation percentages and ratios.
Citing potential limitations to their study, the team says that it will be "imperative to validate our findings on a larger multiregional cohort of untreated NSCLC tumours".
They nevertheless add: "These data illuminate the clinical significance of immune cold regions that may reflect immune-evading subclones, warranting further investigation into mechanisms that could contribute to the spatial variability of immune cells."
TRACERx is funded by Cancer Research UK.