The recent publication of in situ Hi-C data of the monocytic THP-1 cells and their differentiated macrophage induced by the addition of phorbol myristate acetate (PMA), together with that of the in situ Hi-C study of human primary monocytes and the granulocyte-macrophage colony stimulating factor (GM-CSF)-induced macrophages, provides a unique opportunity to validate the observations obtained from this commonly used immortalized cell line with its native counterpart.
Following a well-established method , we first compared larger-length scale, compartment-level interactions, after first verifying that the A-compartments are generally enriched for active genes and the B-compartments are enriched for inactive genes for each of the four cell types (Supplementary Fig. S1) . We found that there is a significant difference (20%) in compartment-level interactions between the primary monocytes and monocytic THP-1 cells (Supplementary Fig. S2a), which is consistent with previous analysis . A comparable difference (19%) in compartment-level interactions was also found between the primary, GM-CSF-induced macrophages and the macrophagic THP-1 cells (Supplementary Fig. S2b). Such a difference in compartment-level features is generally considered as evidence of substantially reorganized chromatin [12, 13], similar to what is observed following, for example, differentiation of embryonic stem cells . Thus, these results indicate that these immortalized cells have, in fact, evolved to a fundamentally different chromatin structure from their native counterparts.
To determine whether these longer-range structural differences are consequential for gene expression, we compared the level of gene expression within those regions that are in different compartments in the primary and THP-1 cells. For the monocytic-type cells, those genes that are found in the A-compartments in the primary cells but in the B-compartments in the THP-1 cells exhibit a higher level of expression in the primary monocytes than in the THP-1 cells (Fig. 1a). These genes are enriched for many different gene ontological terms, suggesting that there are many biological pathways affected by these structural differences (Fig. 1b, Supplementary Table 1). However, it is worth noting that many terms are associated with basic immune functions, such as the innate immune response, positive regulation of chemokine production, and defense response to viruses, all important processes in monocytes (Fig. 1b). For example, several genes that encode for human alpha defensins (DEFA1, DEFA1B, and DEFA3), which are involved in host defense [14, 15], all exhibit essentially no expression in the monocytic THP-1 cells where they are in the B-compartment, but have significant expression in primary monocytes where they are in the A-compartment (Fig. 1c). This suggests that the immortalized cells have lost at least some of their original monocytic functions as a consequence of, or in association with, these alterations in chromatin structure. In addition, the genes in the B-compartments in the primary monocytes that have relocated to the A-compartments in the THP-1 cells are found to be upregulated in the THP-1 cells (Fig. 1d). These THP-1 up-regulated genes are also enriched for many biological pathways, including response to starvation, cell adhesion, and response to nutrients (Fig. 1e, Supplementary Table 1). It is possible that these genes exhibit greater expression in the THP-1 cells owing to their highly proliferative nature in culture, while for the primary cells, since proliferation is largely absent, these genes are not needed and are thus inactive.
Similarly, when the primary, GM-CSF-induced macrophage cells and the macrophagic THP-1 cells were compared, we also found significant relocation of genes between different compartments (Supplementary Fig. S3a, b). Likewise, the genes in these different compartments are enriched for many biological pathways, suggesting that these structural changes are also associated with differences in the functional states between these cells (Supplementary Fig. S3c, d; Supplementary Table 1). Thus, like the monocytic-type cells, these results suggest that the macrophagic THP-1 cells are also fundamentally different from their native counterpart.
We next examined for differences in the more local-level interactions of loops and TADs, the latter of which are very highly conserved (> 92%) between many cell types [16, 17]. Since statistically valid comparisons of such high-resolution features requires data at a comparable sequencing depth, we first obtained multiple (5) down-sampled datasets of the original THP-1 data, each at a similar level (90 million reads) as the primary cell datasets (see Methods). We used TopDom to identify the TADs within each dataset at the highest map resolution possible with this data (40 kb, see Methods) and calculated the extent to which the TAD locations were identical between different datasets. We found that there was good agreement among the down-sampled monocytic THP-1 datasets (74.1 ± 0.4%), with the less than perfect agreement between these datasets consistent with the lowered sampling depth. In contrast, only 26.1 ± 0.3% of the TADs in the primary monocytes were the same as those in the down-sampled THP-1 datasets, which is significantly different (p < 0.001, normality test) from that within the down-sampled datasets (Supplementary Fig. S4a). Similarly, comparing just the down-sampled macrophagic THP-1 datasets, there was good agreement in the identified TADs (71.2 ± 0.6%). However, only 51.5 ± 0.3% of the TADs in the primary, GM-CSF-induced macrophage cells were the same as those in the down-sampled macrophagic THP-1 cells (Supplementary Fig. S4b), which is also significantly different (p < 0.001, normality test) from the measurements of just the down-sampled datasets. Similar differences, for both monocytic and macrophagic-type cells, were also observed at lower map resolutions as well (Supplementary Fig. S4c, d). This degree of difference is truly exceptional insomuch as, for example, the differentiation of embryonic stem cells was associated with changes in chromatin structure at the compartment-level but with essentially no changes at the TAD-level . This underscores the magnitude of the structural changes in the chromatin that the THP-1 cells have undergone from their presumed original state.
We last compared the differences at the loop-level, using the program Peakachu, which has been shown to robustly identify loops at a 10 kb resolution even for data that is at a lower sequencing depth (down to 30 million reads) than our primary cell data . This method computes the probability that there is a loop in each pixel in the Hi-C map using a machine-learning framework, with bona-fide loops identified as those pixels with probability values above 0.97, a threshold value recommended by the original authors (see Methods) . With the same consideration as that for the TAD analysis, we first compared the number of loops identified among the THP-1 down-sampled datasets, and found excellent agreement among the monocytic THP-1 (885 ± 8) and the macrophagic THP-1 (826 ± 17) down-sampled datasets (Supplementary Fig. S5). By contrast, we found that there are 451 and 997 loops in the primary monocytes and primary, GM-CSF-induced macrophages, respectively. Thus, there is a statistically significant difference in the number of loops between the primary and cell lines with both types of cells (p < 0.001, normality test), with a quite pronounced increase of nearly 2-fold in the number of loops in the monocytic THP-1 cells.
We finally examined the similarity in the loop locations between the different datasets (Fig. 2, Supplementary Fig. S6). For this, we compared the extent to which the loops identified in each dataset (down-sampled THP-1 or primary cells) are at the same location as in the full THP-1 datasets. Using the threshold cutoff value of 0.97, the loops identified in the down-sampled THP-1 datasets were indeed highly similar to those in the full datasets, with 88.3 ± 0.6% and 89.9 ± 1.4% of loops identical in the monocytic and macrophagic cells, respectively. However, at the same threshold, there was significantly less (p < 0.001, normality test) identity of loop locations between the primary monocytes (80.1%) and the primary, GM-CSF-induced macrophage cells (80.1%) with the corresponding full THP-1 datasets (Fig. 2c). A similar level of difference between the down-sampled datasets and the primary cell datasets was observed over a broad range of threshold values (Fig. 2c), showing the robustness in this difference. Thus, in both the number and location of the loops, there are significant differences between the primary cells and the THP-1 cell line.