We analyzed the spatial expression patterns of pharmacogenes in six human breast tumor samples and found pharmacogene expression to be heterogeneous within tumor regions. We also showed statistically significant differences in expression of pharmacogenes in tumor regions compared to surrounding non-tumor regions. We observed that the most heterogeneously expressed genes were involved in ROS handling and detoxification mechanisms. The heterogeneous expression of these pharmacogenes may have important implications for cancer therapy due to their ability to impact drug distribution and efficacy throughout the tumor.
The Visium Spatial Transcriptomics platform measures the expression of the whole transcriptome within intact fresh-frozen tissue sections while simultaneously preserving the spatial context of gene expression. The technology relies on spatially barcoded poly-T capture probes that hybridize with mRNA. Due to this non-specific mRNA capture mechanism, highly expressed genes may be captured more frequently than genes expressed at relatively lower levels, limiting sensitivity. Among all six tissue specimens, 214 pharmacogenes were detected, albeit many at lower levels compared to the most highly expressed genes. While the relative expression of pharmacogenes across different histologic regions were comparable across all six tissue samples, larger numbers of uniquely barcoded reads were reported in samples with higher sequencing depth. Future development of capture probes specifically designed for pharmacogenes may improve the sensitivity of their detection.
Heterogeneity was defined as the interquartile range of unique-UMI reads measured across each UMI-barcoded pharmacogene for every spatial spot. Gene expression in each spot may not be uniform; for example, spots within stromal tissue sections will have markedly lower gene expression when compared to spots within tumor regions. To account for such variability in gene expression across the entire tissue, we normalized the read counts to the total number of UMIs detected in each spot. We have presented the data both with and without such normalization because the process of normalization may undermine some of the observed heterogeneity in pharmacogenes. For example, overexpression of other cancer-related genes relative to a given pharmacogene in a tumor region may result in a high total UMI count leading to a masking effect caused by normalization.
Our study evaluated spatial heterogeneity in tumor pharmacogene expression but did not evaluate the downstream impact of this heterogeneity. However, the literature supports the hypothesis that spatial heterogeneity in tumor pharmacogene expression contributes to chemotherapeutic resistance mechanisms. For example, certain drug and metabolite transporters, such as those included in the multidrug resistance protein (MRP) group of the large ABC family, are known to play a role in chemoresistance when overexpressed as they contribute to the efflux of chemotherapeutic compounds from the cell (34). One of these transporters, ABCC5, was among our most heterogeneously expressed pharmacogenes. ABCC5 transports cyclic nucleotides, including the metabolites of 5-fluorouracil (5-FU), a common anticancer agent used in both breast cancer and colon cancer treatment. ABCC5-transfected cells have been shown to have a nearly 9-fold increase in 5-FU resistance (34). Another transporter in this same family also found to be heterogeneously expressed is ABCB8; it transports compounds from the mitochondria to the cytosol. One study showed inhibition of ABCB8 with shRNA resulted in increased doxorubicin-induced mitochondrial DNA damage (35). This evidence for ABCC5 and ABCB8, combined with our detection of spatial heterogeneity in expression of these genes within breast tumors, provides further mechanistic clarity for how 5-FU or doxorubicin treatment could result in seeding resistant tumor cells following chemotherapy.
Several of the pharmacogenes with the highest heterogeneity (e.g. GPX4, GSTP1, MGST3, and SOD1) are also known to impact drug response. Glutathione peroxidases (GPX) are enzymes that protect cells from oxidative stress by catalyzing the reduction of peroxides. GPX4 has been shown to be critical for survival of lapatinib-resistant cancer cells, but not lapatinib-naïve cancer cells (32). In colorectal cancer, increased GPX3 expression resulted in increased resistance to oxaliplatin and cisplatin (36). This evidence, combined with our detection of spatial heterogeneity in GPX3 and GPX4 expression within tumors, could explain why certain cells survive platinum-agent or lapatinib treatment, leading to resistant tumors.
GSTP1 was also highly heterogeneously expressed in our data. Enzymes in the glutathione s-transferase (GST) family catalyze the conjugation of polar glutathione groups that enhance systemic elimination of chemotherapeutic agents and toxic metabolites. GST activity has been shown to be inducible by treatment with vincristine, doxorubicin, or topotecan (37). When GSTP1 enzymatic activity is impaired, as is the case with the rs1695 missense variant (38), platinum-based chemotherapy-induced granulocytopenia was shown to be more common in a meta-analysis of 12 case control trials (39). Additionally, GSTP1 expression was found to be higher in adriamycin-resistant cells, and higher GSTP1 expression was also found in breast cancer tissues from subjects with progressive/stable disease vs. those with partial/complete response. Interestingly, this finding was true for tissue collected before and after anthracycline/taxane treatment, indicating intrinsic mechanisms of resistance. These data indicate that spatial differences in GSTP1 expression could be involved in chemoresistance and seeding of resistant tumor cells following chemotherapy treatment.
MGST3 was another highly heterogeneously expressed gene in our data. This enzyme is involved in immune function by catalyzing the conjugation of reduced glutathione and leukotriene A4, producing leukotriene C4. Overexpression of MGST3 was found in cisplatin resistant lung adenocarcinoma cells compared to non-resistant progenitor cells, and when MGST3 expression was increased (via antagonism with its microRNA regulator, mir-432-5p) the progenitor cells demonstrated increased survival to cisplatin treatment (40). Although it is unclear if this relates to immune modulation, these data and other similar studies (41, 42) indicate that spatial differences in MGST3 expression could be involved in chemoresistance and seeding of resistant tumor cells following chemotherapy treatment.
Another highly heterogeneously expressed gene in our data was superoxide dismutase 1 (SOD1). SOD1 eliminates damaging superoxide radicals by converting them to less toxic molecular oxygen and hydrogen peroxide. SOD activity counteracts superoxide-induced autophagy (43) and inhibition of SOD1 with siRNA or small molecule inhibitors results in increased cisplatin sensitivity in ovarian cancer cells (44, 45). These studies combined with our results suggest spatial differences in SOD1 expression could be involved in chemoresistance and seeding of resistant tumor cells following chemotherapy treatment.
We acknowledge some limitations to our study. First, we have a small sample size and more information will likely be gained from bigger studies. Second, the lower expressed genes may contain additional variability due to the low read numbers. Third, the resolution is not small enough to capture single cell resolution, so there may be some overlap in cell types.