Indexes characterizing meta-clusters and STARTRAC analysis

To characterize the tissue distribution of meta-clusters, odds ratios (OR) were calculated and used to indicate preferences. Specifically, for each combination of meta-cluster i and tissue j, a 2 by 2 contingency table was constructed, which contained the number of cells of meta-cluster i in tissue j, the number of cells of meta-cluster i in other tissues, the number of cells of non-i meta-clusters in tissue j, the number of cells of non-i meta-clusters in other tissues. Then Fisher’s exact test was applied on this contingency table, thus OR and corresponding p-value could be obtained. P-values were adjusted using the BH method implemented in the R function p.adjust. We found that all ORs > 1.5 or ORs < 0.5 had adjusted p-values < 1e-10. Hence, a higher OR with a value > 1.5 indicated that meta-cluster i was more preferred to distribute in tissue j, a lower OR with a value < 0.5 indicated that meta-cluster i was preferred not to distribute in tissue j.

Integrating the TCR data and cluster assignment of the T cells, the STARTRAC indexes (9) quantify the magnitude of clonal expansion, migration potential, and state transition potential. The clonal expansion index uses Shannon entropy to quantify the distribution evenness of the TCR repertoire (all clonotypes) of a meta-cluster. Then the expansion index is calculated as 1-evenness, similar to standard TCR clonality measurement (56), then normalized to the range from 0 to 1, with the high value indicating high clonality of the meta-cluster. To assess the significance of the observed expansion index, a permutation procedure was applied. Specifically, the clonotype IDs of cells in each patient were randomly permuted, and the permuted expansion index was then calculated. Such an operation was repeated 1000 times. The observed expansion index was compared with the distribution of permuted expansion indexes, then a p-value was reported. It should be noted that the number of cells sequenced in each sample was still limited, and the clonality may be underestimated, especially those meta-clusters with lower cell numbers. However, the expansion index was not used to compare the meta-clusters or identify which meta-clusters have “higher” clonality than others, but used to identify meta-clusters with “high” clonality while correcting for randomness. Thus, meta-clusters with high expansion indexes were highlighted.

A proliferation index was also defined, which complemented the TCR-based clonal expansion index and indicated the ongoing proliferation activity of a meta-cluster. For each cell, a proliferating score was calculated as the average z-score scaled expression values of the common proliferation markers (ZWINT, E2F1, FEN1, FOXM1, H2AFZ, HMGB2, MCM2, MCM3, MCM4, MCM5, MCM6, MKI67, MYBL2, PCNA, PLK1, CCND1, AURKA, BUB1, TOP2A, TYMS, DEK, CCNB1, and CCNE1) (57). Then the proliferating cells were defined as those with proliferating scores larger than a threshold that was chosen using the function getOutliers from R package extremevalues. The proliferation index was calculated as the frequency of proliferating cells in a meta-cluster.

To define the migration index, the Shannon entropy of each clonotype was calculated, which characterizes the broadness of the clonotype distributed across tissues (blood, normal, and tumor). A high entropy value indicates that the given T cell clone has more mobility. Then for a meta-cluster, the migration index was defined as the average of the entropies of all clonotypes, weighted by the frequencies of the clonotypes in the meta-cluster. When limiting the tissues to a tissue pair of interest, the pairwise version of the migration index (pMigr) could be defined. Similarly, the state transition index and the pairwise state transition index (pTrans) were used to quantify the broadness of the clonotypes distributed across multiple (or paired for pTrans) meta-clusters. A higher value of the transition index indicates the higher state transition potential of the meta-cluster. For each meta-cluster, the ranking of its pTrans informs us which other meta-clusters are more connected with it in differentiation. For the two CD4⁺  meta-cluster of interest, i.e.  IFNG ⁺  Tfh/Th1 and  TNFRSF9 ⁺  Treg, in the rankings of their  pTrans,  IL21 ⁺  Tfh and other Treg meta-clusters showed the highest pTrans respectively (fig. S25B, fig. S26C), supporting the  state transition between the two Tfh-related meta-clusters and between TNFRSF9 ⁺  Treg and other Treg meta-clusters. In other cases, we used a cut-off of 0.1 to extract the highly connected meta-cluster pairs and visualized them in a heatmap (Fig. 2D). Shannon entropy > 0.1 was equivalent to the situation where the frequency distribution was more balanced than that when the cell number ratio between two meta-clusters was 10% vs. 90%. This indicated that a clonotype shared by two meta-clusters with a pTrans > 0.1 could be considered as evidence supporting the state transition between the two meta-clusters. Further, the cut-off 0.1 is significantly larger than the pTrans of naïve T cells. Thus, the 0.1 cut-off was empirically set to highlight the highly connected meta-clusters.