selected 1,216 profiled cells after filtration, together with the indexed protein levels of the well-characterized lymphoid marker CD10, as well as two putative markers from our Drop-seq data, CSF3R and CD52. These putative markers derive from the ‘primed’ lymphoid and ‘primed’ Neu/Mo programs respectively, and have not to our knowledge been previously used to subdivide early CD38- hematopoietic progenitors. A principal component analysis (PCA) with genes from the nine transcriptional programs (Figure 3A) returned a separation between lymphoid versus Neu/Mo transitions, suggesting additional cellular heterogeneity within the CD34+ CD38- CD45RA+ gate (figure?).
As expected, we observed that CD34+ CD38- CD45RA+ cells strongly enriched for LMPP, Neu/Mo, lymphoid, and HSC/MPP groups, with negligible mapping to EMP or downstream precursors. We observed an even stronger enrichment of lymphoid progenitors when examining the subset of cells that stained positive for CD10 (Figures 5C), validating its suitability as a marker for lymphoid commitment, as well as our coupling and projection strategy. However, we observed that CD10- cells remained transcriptionally heterogeneous, and projected to uncommitted, myeloid (Neu/Mo), and lymphoid fates at roughly equal proportions. To our knowledge, this represents the first description of transcriptomic heterogeneity within this group of cells, which have been previously thought to represent a homogeneous LMPP population. To explore whether we could further subdivide this gate based on novel markers, we examined the distribution of CD52 and CSF3R protein expression for cells projecting to distinct transcriptomics clusters (Figure 5D; additional markers in Figure S5B). In agreement with our predictions from the Drop-seq data, CD52 protein expression was significantly (p<1e-5; Kolmogorv-Smirnov test) higher in lymphoid-primed cells compared to myeloid-primed cells, while CSF3R protein levels exhibited the opposite pattern (p<1e-4; two-sided Kolmogorov-Smirnov test). While significant, no protein marker was in perfect agreement with our clustering, and it is unlikely that CSF3R alone or CD52 could be used to sort transcriptomically pure populations. Thus, we have shown that early fate determination programs are heterogeneously activated in LMPP cells, and identified novel markers that are consistent with this segregation. However, echoing recent studies in mouse (Nestorowa et al., 2016; Paul et al., 2015), it is challenging to define FACS panels that can perfectly represent the complexity of a single cell transcriptomics readout.