Challenges and Opportunities in Building a Global Representative Single-Cell and Spatial Atlas in Cancer

Cancers are heterogeneous, complex environments in which cellular interactions play a role in disease initiation, progression, and therapeutic response. Histopathology, bulk sequencing, and bulk transcriptomics have provided much insight into the cellular heterogeneity of many cancers. New insights into unique and disparate cell populations, cell biology, and the overall landscape of tumors have excelled with the explosion of single-cell (single nuclei) RNA sequencing (scRNA-seq). These datasets have been transformative in identifying cells of origin that initiate cancer and metastatic clones that escape treatment. Patient samples used to generate and identify these unique cell-specific effects are highly biased to European ancestry (1, 2). Large consortium efforts, such as the Human Cell Atlas (3), the Human Breast Cell Atlas (4), and the Human Tumor Atlas Network (5), have generated a single-cell map of all organs and tissues for application to diseases such as cancer, but their initial efforts did not capture ancestrally diverse samples. Other organizations have tried to address this by funding and sourcing genomic projects aimed at including genetically diverse populations, but these projects are limited and require a larger investment and infrastructure. With the paucity of single-cell data from diverse ancestral populations, the disparity in applying these rich genomic findings to other ancestries only continues to grow.

Beyond simple inclusion to generating datasets from diverse populations, there is an enormous logistical problem in accessing relevant and usable samples. In the United States, the health disparity for disparate ancestral populations is complex with many minoritized individuals, of whom many are from the Global South, and diverse genetic ancestral populations having low socioeconomic status and limited access to health care (6). Single-cell technologies have been optimized for use in privileged R1 (very high research doctoral universities) research universities and centers and not in low- and middle-income (LMI) countries. The majority of these technologies support properly flash-frozen samples that require access to proximal liquid nitrogen and high-end storage solutions. An outstanding problem is not simply in funding single-cell research to include more samples from diverse individuals, but also to propel these technologies to have more flexible use cases, such as formalin-fixed, paraffin-embedded (FFPE) samples. Storing samples in FFPE requires less infrastructure and would increase and improve collection methods to extend to more rural, remote LMI populations. Such methodologic and technological advances are emerging, but more use cases are required to understand the quality of the data and inference to leverage existing single-cell datasets. Given that the vast majority of single-cell data have been generated using fresh or flash-frozen (7) sample preparation compared with the new single-nuclei FFPE datasets, these new methods will have to be vetted for consistency and equivocal cell representation across a variety of tissue types across different facilities. Because FFPE storage methods may differ across populations based on their resources, while these methods become more established, large-scale efforts are required to ensure that any difference observed in datasets comes from true genetic biology versus sample collection (technical) methods. As we begin to generate single-cell and spatial datasets across diverse ancestral populations, it is imperative we have robust analytical methods for FFPE-derived data related to their different genetic makeups.

Based on the limitations of single-cell technology and accessibility due to cost for different, diverse populations, the implementation of new emerging spatial biology seems more enticing and adaptable to existing collection and storage methods globally. For example, globally, the standard method for diagnosis of a larger variety of diseases is conducted on FFPE samples viewed and annotated by trained pathologists. These methods are readily adapted in remote areas because FFPE is an easy and cheap method to store tissues long term, with no requirement for refrigeration or special handling. This process is robust and reproducible across sites and is used for most cancer diagnostic and treatment assessment. FFPE archival blocks are often used for bulk transcriptomic or genetic analysis because of their accessibility (8). Hence, harnessing FFPE tissues into single-cell and spatial methods will greatly improve the cancer research community's ability to reduce hurdles in obtaining ethically, racially, and genetically diverse samples.

Molecular cancer diagnostic tests are a clear example of the limited adoption in LMI countries (9). These tests focus on basic pathology and molecular approaches, such as PCR gene paneling and IHC (10). For decades, these methods have been routinely available in well-equipped and funded academic/medical centers in the United States and are now just beginning to be adopted in LMI countries (11), but a lack of uptake in LMI countries is due to limited infrastructure and technical capacity. Global health practitioners are in agreement that this process has come too slowly at the cost of premature deaths, inadequate treatments, and community impact worsening (global) health disparities (9). With the modernization of many facilities to operate COVID-19 PCR testing, the adoption rate of PCR-based cancer diagnostic tests will hopefully also accelerate. This pandemic highlights the drastic need for on-site diagnostics and demonstrates that these molecular technologies can be leveraged to advance cancer care in LMI regions (12). The question that remains is how and to what extent infrastructure can grow to accommodate more elaborate facilities for newer, complex single-cell and spatial omic technologies whose requirements far exceed those for COVID or routine molecular cancer diagnostics and genomic sequencing. Beyond instrumentation and facility needs, getting staff properly trained in these LMI regions is challenging and remains the largest hurdle because this requires broad investment in foundational training programs that include science, technology, engineering, and math (STEM) training in early education.

Although costs to perform these sequencing assays have dramatically decreased, single-cell and spatial omic technologies are still cost-prohibitive for LMI countries. A single-cell experiment for one small experiment can be upward of the $10,000 to $20,000 range, which is further heightened by the increased costs of shipments, customs, and profit margin for local companies in LMI countries. This level of expense far exceeds the available funds for scientists in most LMI institutions. Although these technologies are pricey in high- and middle-income (HMI) research institutes in which there are funding agencies able to help facilitate these experiments, in LMI countries, this cost challenge is daunting and requires a dramatic shift in not simply laboratory infrastructure but a commitment to a steady stream of funding for reagent and personnel costs beyond instrumentation and facility needs. Getting staff properly trained in these LMI remote regions is challenging and remains the largest hurdle because this requires broad investment in educational programs very early in STEM training. There needs to be a government investment in the education and training of basic genomic and bioinformatic skills beginning in secondary schools or earlier.

Personnel requirements extend beyond just laboratory techniques, as there are fundamental variations through cultural differences, consenting process, sample acquisition, and other peculiarities across these types of research tool implementations. These LMI communities are often isolated, and sample collection is mitigated by lack of access; hence, transport and logistics of sample processing intensify the ease of executing these complex molecular technologies. Collection and recruitment are also affected by a cultural distrust that can exist within diverse and underrepresented populations. Recruitment efforts led by researchers of similar racial or ethnic backgrounds paired with community involvement have been shown to be more effective (13). In some global communities, consent may be high for tumor collection but quite low for blood or saliva collection (14). Existing analysis pipelines for tumor burden calling and germline DNA are essential; in these cases, tumor collection without paired normal samples will yield inconclusive results, and this is further hindered by the lack of an ancestry-relevant reference dataset (15) for comparison. Additionally, these communities may be isolated, and collection is mitigated by a lack of physical access to conventional libraries. This example highlights the challenges from the bedside to the bench and ultimately a call for analytical tools to diversify in their application to different sample types (states; Fig. 1).

A strategic plan to train within LMI countries and diverse communities within HMI countries will need to involve key scientific leaders who themselves are culturally sensitive, diverse, and from these populations, paired with active engagement with community leaders and members. Funding has been disproportionately biased toward cancers that primarily afflict non-Hispanic individuals (16). Hence, these diverse scientific leaders should be supported by federal and foundation funding award opportunities, not only to serve in reviewing proposals but also to be funded as awardees themselves. Community engagement plans should be prioritized for funding opportunities on a scale that starts with scientific researchers who have established partnerships with LMI countries and minoritized communities. Through these efforts, their existing infrastructure—for example, health fairs, community centers, and church events—can be harnessed to participate with scientists and learn and contribute to the science being conducted.

This problem of single-cell and spatial omics analysis is daunting, as the adoption hurdle of these extensive computational methods is difficult even in HMI countries (17). These technologies produce large datasets with complex downstream analyses, such as the reconstruction of differentiation trajectory and analysis of cell–cell interaction, and require constantly evolving computational tools and modeling. New computational approaches are required to integrate ancestral genotyping with the single-cell differences detected in different global populations (15). Different omics platforms have different types of attributes and distributions, making it both challenging to integrate them all and difficult to keep abreast of emerging analyses even in well-structured, highly accessible environments like HMI medical centers. Overcoming these hurdles will require fundamental training in computation early on in LMI countries to establish workforces capable of learning and adapting these analytical tools. But an investment in computational training opens the doors for LMI regions to have accessibility and ownership of their data. Operationally, computers and servers are likely easier to implement than expensive, complex molecular machinery.

FFPE-friendly single-cell and spatial technologies can allow for more adoption; however, more pathologists are required to ascertain which samples are pathologically and clinically relevant. One problem yet to be overcome is the lack of pathologists globally, especially in LMI regions. For example, there is one pathologist for every 500,000 people in Africa (18). The growth of digital pathology may alleviate this by allowing FFPE slides to be analyzed at a quicker pace with more accuracy. Artificial intelligence and machine learning (ML) technology have made remarkable progress in biomedical research and are already used in clinical practice in which labs use digital pathology for some tumor diagnoses (19). ML quickens the pace of single-cell analysis and decreases the burden of computational memory. Its use can be implemented from most personal computers and presents a unique ability for digital pathology. Committed infrastructural support from federal and global research funding will be required for these large computational datasets to be processed and analyzed. In addition, the development of computational tools utilizing and leveraging already existing pipelines can help to overcome the challenges of building ancestry-specific algorithms within. Sponsorship of trainees and scientists from LMI countries, both within the Global North and South, needs to be expanded and scaled up for true equitable access to data generation and interpretation. In addition, this requires (i) harmonized datasets to be integrated and (ii) an accessible data portal that has been developed for easy queries and data visualization for generalists. Computational training, adoption, and application in LMI countries could provide a sustainable structure to ancestral single-cell cancer research and diagnostics. With increased training in LMI regions, there still will be the challenge of overcoming structural problems such as limitations in access to the internet, tools for single-cell and spatial omic data analysis, current scientific literature, and high computing processors.

Cancer clinical trials are not representative of the patients and groups who experience health disparity. Knowing that these health disparities have an underlying ancestry-based onset, clinical trials and our current treatment options for standard of care may not be targeting the correct biological manifestations of cancer in these populations. The use of single-cell and spatial omics to understand the ancestral contribution to the pathogenesis of cancer will be a tremendously powerful tool when paired with therapeutics and clinical trials. Cancer disparities and equity research in the single-cell spatial realm will aid our understanding of why, despite incidence rates decreasing for most cancers, mortality rates continue to rise in these populations. Single-cell and spatial research will allow for more targeted therapeutic development and preventative screening through the study of disease etiology. This highlights the importance of a roots-up approach to single-cell and spatial omics with regard to ancestry. A strategic plan that incorporates community engagement with leaders in places where populations have a disproportionate burden of disease and worse outcomes paired with formidable infrastructure targeted to increasing diversity in the discovery sets used for cancer research is crucial to carrying out the goal of more inclusive, representative precision medicine.