NEW YORK – During the second part of the virtual annual meeting of the American Association for Cancer Research last week, researchers highlighted several projects under the auspices of the National Cancer Institute Cancer Moonshot initiative that showed the power of large tumor atlases for generating novel information on several different cancers.
In a session on Wednesday, Sage Bionetworks' Justin Guinney talked about the Human Tumor Atlas Network, which was originally established in 2018 as one of 12 projects funded by the Moonshot initiative to generate 3D molecular and spatial atlases of diverse human tumors. Importantly, the projects within HTAN are meant to characterize crucial transitions in cancer, from tumor initiation to metastasis, as well as response and resistance to therapy.
The HTAN involves 10 research centers — half are developing pre-cancer atlases and half are creating human tumor atlases — and a data coordinating center. The pre-cancer project and the human tumor atlas are in pilot phase.
At the AACR meeting, Guinney provided the first look at the data from the HTAN, which is currently readying its first full data release. To characterize how cancer changes over time, all the HTAN centers are using molecular profiling technology, with a focus on single-cell and multiplex imaging, as these technologies offer a fine-grade window into the tumor and its microenvironments, Guinney said.
Compared to The Cancer Genome Atlas, which launched in 2005 and has molecularly characterized 33 kinds of cancer by analyzing several hundred samples per tumor type, the HTAN is focused on analyzing fewer samples in 10 tumor types. The TCGA also uses a fixed set of bulk-tissue assays and processes the data centrally, while the HTAN has an open set of assays and processes the data site by site.
During his presentation, Guinney highlighted two studies using data from the HTAN. In the first study, researchers from Memorial Sloan Kettering Cancer Center are generating a single-cell atlas of small-cell lung cancer, or SCLC, from single-cell RNA-sequencing data done on 155,098 cells taken from 19 patients.
Previous studies of SCLC had classified the disease into three subtypes, based on differential expression of specific transcription factors: SCLA-A, as defined by the expression of ASCL1; SCLA-N, as defined by the expression of NEUROD1; and SCLA-P, as defined by the expression of POU2F3. However, since these insights had been gained using bulk-tissue profiling techniques, the researchers wanted to see what else could be gleaned from more granular single-cell profiling or immune microenvironment investigations.
The subsequent study they conducted uncovered a new subtype of SCLC, with a tumor cell population characterized by high expression of PLCG2. The researchers also found stem-like and pro-metastatic features that recur across SCLC subtypes and are associated with worse survival.
In a session on Friday, Charles Rudin, chief of the thoracic oncology service at MSK, elaborated on this HTAN project and discussed how single-cell profiling of human tumors confirmed some theories and yielded some surprises, including the new subtype, termed Cluster 22. Specifically, the expression of PLCG2 in Cluster 22 led the researchers to ask whether this feature could be a phenotypic driver of the disease. Indeed, through cell-line experiments in the lab, they found that raising the expression of PLCG2 was associated with cell migration, invasion, and metastasis. Researchers analyzed patient data and also found that high expression of PLCG2 is associated with poor prognosis.
The researchers then looked at the cancer cells' associated microenvironment. A characterization of the immune infiltrates in SCLC showed that there was an enrichment of NK cells within the lymphocytes and a remarkably high number of CD8 effector T cells compared to normal lung cells. Further, Rudin said, a myeloid fraction analysis of SCLC showed that the pro-fibrotic macrophage population was significantly enhanced when compared to normal lung cells. Importantly, this was enriched in the Cluster 22 subtype.
The second HTAN study Guinney highlighted was the Serial Measurements of Molecular Architectural Responses to Therapy clinical trial, within which Oregon Health and Science University researchers used a wider array of assays on a single breast cancer patient. It had two goals: to understand mechanisms of resistance and then inform the next choice of therapy for the patient.
The researchers employed longitudinal multimodal data capture on samples from the patient, including DNA, exome, and RNA sequencing. That data was then passed to a clinical tumor board for analysis. The longitudinal testing, according to Guinney, allowed the tumor board to track the progression of the patient's cancer over 42 months and use the sequencing data to inform necessary changes to the course of therapy.
From DCIS to invasive breast cancer
In a session on Friday, several researchers presented data from individual HTAN studies that highlighted the value of human tumor atlases for identifying drivers of cancer progressions and therapeutic resistance.
Stanford University researcher Michael Angelo discussed his group's efforts to study the progressive changes in the structure and composition of breast tumor stroma and track the transition to invasive breast cancer. The long-term goals of this ductal carcinoma in situ, or DCIS, HTAN Initiative were to integrate multimodal data using a predictive computational model of neoplastic progression, identify molecular changes in the development and transition from DCIS to invasive cancer, test predictors of progression to invasive disease, and spatially map and co-localize molecular and imaging data, Angelo said.
The researchers started with a breast pre-cancer atlas, consisting of a series of tissue microarrays containing samples from patients diagnosed with DCIS, and matched invasive breast cancer samples when they existed, combined with information on patients' clinical courses. Pathologists then reviewed these tissue microarrays and carried out multiomic data workflows to further annotate breast cancer progression, including analyses of histology, genomic gains and losses, transcriptomics, genomic copy-number alternations, and cellular composition and structure.
The researchers then digitized the tumor microenvironment, Angelo said. Compartment annotation and single-cell mapping led to 38-plex tissue images, which were put through a cell segmentation and mapping bioinformatics pipeline that marked each different cell type. Tissue compartment annotation maps were also overlaid on those images to show where each of these cells were in the breast.
Using these maps, the researchers were then able to ask themselves how ductal and stromal compartments evolved in breast cancer, what cells tended to co-localize or avoid one another, and whether they were typically found within or near ducts, vessels, or stroma. With further analysis, researchers identified two phenotypic features to distinguish normal breast tissue from recurrent breast cancer. These two clusters seemed to be mutually exclusive in their representation, Angelo said.
There was also a third cluster enriched just in DCIS, particularly in ductal macrophages and stromal mast cells.
Using this data, the researchers trained a classifier program to predict invasive cancer relapse, and came up with a ranked list of the 20 most important features that could predict risk of progression. The list involved 14 distinct cell populations, heavily predominated by stromal and myoepithelial cell types. They also found that spatial features were a crucial determinant of risk, Angelo said.
Mapping progression in colon cancer
Also at the AACR meeting, Vanderbilt-Ingram Cancer Center's Martha Shrubsole presented the Moonshot project COLON MAP, a human colorectal pre-cancer atlas that has identified distinct molecular features underlying two major subclasses of pre-malignant tumors. Shrubsole said she and her colleagues were interested in finding evidence of an alternative origin of tumorigenesis beyond stem cells in human colon tumors, and to "explore whether these tumors exhibited a different immune tone and microenvironment."
They set out to develop an integrative single-cell atlas of the host and microenvironment in colorectal neoplastic transformation by performing 3D modeling of progression in sporadic colon adenoma to colon cancer; single-cell RNA-seq to identify cell state markers; spatial profiling analysis; multiregional exome sequencing; and biofilm analysis.
Their analyses showed that adenoma polyp cells and serrated polyp cells did indeed exhibit different pathway profiles. For example, the serrated cells lacked WNT pathway activation and had low stem potential, whereas the adenoma cells had WNT pathway activation and high stem potential. The serrated cells also activated gene networks related to damage response and metaplasia.
Further analyses provided evidence of non-stem origins for serrated cells, such as a shared phylogeny with differentiated cells, Shrubsole said. A velocity analysis also showed that they likely came from differentiated cells and not from stem cells.
The researchers then looked to see if the features they identified in the polyps were also in the patients' cancers. They found that colorectal cancers with high microsatellite instability, or MSI-H, seemed to carry metaplastic signatures of serrated polyps, whereas microsatellite stable, or MSS, cancers correlated with adenoma-specific WNT subtype cells.
However, they also found that the stem cell program was reactivated in MSI-H colorectal cancer cells, compared with serrated polyps. This was different from MSS colorectal cancer, which showed a lack of stem and metaplastic features. MSI-H colorectal cancers also showed heterogeneous regions of stem-like and metaplastic cells that correlated with the local immune environment.
Ultimately, Shrubsole said, there were tumor-intrinsic differences between conventional and serrated pathways of tumorigenesis, with conventional adenomas arising from activation of WNT and associated regenerative programs and serrated polyps arising from mucinous metaplasia in response to damage. The origins of conventional adenomas are likely to be normal stem cells, while serrated polyps may develop from committed absorptive cells.
The group's next steps are to study the role of microbial exposures through the use of biofilm samples, and to build the number of samples in the atlas to confirm earlier findings.
Finally, Nir Hacohen, codirector of the cell circuits program at the Broad Institute, presented his group's work on mismatch repair-deficient, or MMRd, and MMR-proficient, or MMRp, colorectal cancer, which represent two immunological states. MMRd tumors typically have high tumor mutational burden, or TMB, and are more likely than MMRp tumors to respond well to immunotherapy. Meanwhile, around 85 percent of colorectal cancer patients have MMRp tumors, which are known to have low TMB and are typically unresponsive to immunotherapy.
Hacohen and his colleagues aimed to determine the difference in immune cell composition and organization between MMRd and MMRp tumors and study how malignant, immune, and stromal cells organized into functional units to form the tumor microenvironment.
They began by performing scRNAseq on about 400,000 cells from a 62-patient cohort with primary, untreated MMRd and MMRp colorectal cancer and found seven cell lineages organized in 88 clusters. The major immune composition differences between MMRd and MMRp cells, Hacohen said, was that the former comprised mainly CXCL 13-positive T cells and activated gamma-delta-like T cells, whereas the latter comprised mainly IL17-positive T cells. The data also suggested that the MMRd CXCL13 cells were tumor-specific.
The researchers also found multicellular interaction networks based on covariation of program activities across patient specimens. There were seven such hubs in MMRd and nine hubs in MMRp colorectal cancer cells. For example, one hub was a cellular interaction network that drove inflammation in human MMRd and MMRp cancer cells but was not present in normal colon cells. Another hub provided putative anti-tumor immunity in MMRd cancer cells.
In conclusion, Harcohen said, the immune hubs are a functional unit of the immune response. The researchers found different types of hubs, such as anti-tumor immunity, inflammation, and tissue repair and growth. They also found different states of hubs, such as those that have context-dependent features and those that exist in different temporal phases.
Harcohen noted that the hubs could be helpful in organizing researchers' knowledge of colorectal cancer and in helping to recognize similar patterns across patient-specific tumor landscapes.