Abstract

   Type-1 conventional dendritic cells (cDC1s) are rare immune cells
   critical for the induction of antigen-specific cytotoxic CD8^+ T-cells.
   The genetic program driving human cDC1 specification remains largely
   unexplored. We have previously identified PU.1, IRF8 and BATF3
   transcription factors as sufficient to induce cDC1 fate in mouse
   fibroblasts but reprogramming of human somatic cells was limited by low
   efficiencies. Here, we investigated single-cell transcriptional
   dynamics during human cDC1 reprogramming. Human induced cDC1s (hiDCs)
   generated from embryonic fibroblasts gradually acquire a global cDC1
   transcriptional profile and activate antigen presentation signatures.
   Importantly, other DC subsets are not induced at the single-cell level.
   We extracted gene modules associated with successful reprogramming and
   identified inflammatory signaling and the cDC1-inducing transcription
   factor network as key drivers of the process. Combining IFN-γ, IFN-β
   and IFN-α with constitutive expression of cDC1-inducing transcription
   factors lead to improvement of reprogramming efficiency by 190-fold.
   hiDCs uptake dead cells, secrete inflammatory cytokines and perform
   antigen cross-presentation, a key cDC1 function. This approach allowed
   efficient hiDC1 generation from adult fibroblasts and mesenchymal
   stromal cells. Mechanistically, PU.1 shows dominant and independent
   chromatin targeting at early phases of reprogramming, recruiting IRF8
   and BATF3 to shared binding sites. The cooperative binding at open
   enhancers and promoters leads to silencing of fibroblast genes and
   activation of a cDC1 program. These findings provide mechanistic
   insights into human cDC1 specification and reprogramming and represent
   a platform for generating patient-tailored cDC1s, a long-sought DC
   subset for vaccination strategies in cancer immunotherapy.

Introduction

   Dendritic cells (DCs) orchestrate innate and adaptive immune responses.
   Classically, human DCs are divided in 4 main subsets: conventional type
   1 (cDC1s) and type 2 (cDC2s), plasmacytoid (pDCs), and monocyte-derived
   dendritic cells (moDCs), each defined by specific transcriptional
   regulators, surface markers and functional properties ([46]1). The cDC1
   subset, evolutionary conserved across mammalian species, is controlled
   by a network of transcription factors, including PU.1, IRF8 and BATF3
   ([47]2, [48]3). In mice, PU.1 promotes chromatin remodelling at the
   Irf8 locus required to initiate expression and drive cDC1
   differentiation from progenitors ([49]4). High levels of IRF8 are then
   required throughout cDC1 specification and maintained by an
   auto-activation loop that becomes BATF3-dependent during the pre-cDC1
   to cDC1 transition ([50]5, [51]6). The inter-dependency of these
   transcriptional regulators and the mechanisms underlying chromatin
   engagement at the onset of human cDC1 specification remains largely
   unexplored.

   Recent studies have revealed the heterogeneous nature of human DCs and
   identified new DC sub-populations with single-cell RNA sequencing
   (scRNA-seq) ([52]7). Human cDC1s are characterized by surface
   expression of CD141, CADM1, CLEC9A and XCR1 ([53]1), secrete
   immune-modulatory cytokines, including IL-12, interferons (IFNs) and
   CXCL10, and perform antigen cross-presentation to CD8^+ T-cells,
   critical processes for inducing anti-tumor immune responses ([54]8,
   [55]9). Accordingly, cDC1 abundance in human tumors was associated with
   patient survival and responsiveness to checkpoint inhibitors ([56]9,
   [57]10). However, human cDC1s are very rare in vivo, so their study and
   use for cancer immunotherapy require methods to generate functional
   cDC1s in vitro. Human CD34^+ bone marrow (BM) progenitors were used to
   derive CD141^+ cDC1s in vitro in the presence of FLT3L, SCF, GM-CSF and
   IL-4 ([58]8). More recently, FLT3L was combined with Notch signalling
   to favour cDC1 differentiation ([59]11, [60]12). The generation of
   cDC1-, cDC2- and pDC-like cells from induced pluripotent stem cell
   (iPSC) cultures was also demonstrated ([61]13). However, these
   protocols are complex, require feeder layers and generate multiple DC
   subsets with conflicting functions. Thus, new strategies to generate
   homogeneous populations of human cDC1s in vitro are warranted ([62]14).

   Direct reprogramming offers a novel strategy to generate cells for
   regenerative medicine and immunotherapy ([63]15). Reprogramming of
   fibroblasts into various cell types including macrophages, neurons and
   cardiomyocytes was achieved without transiting through pluripotency
   ([64]16). We have recently shown that the transcription factors PU.1,
   IRF8 and BATF3 (PIB) are sufficient to convert mouse embryonic
   fibroblasts (MEFs) into cDC1-like cells ([65]17). MEF-derived induced
   DCs (iDCs) acquire cDC1 phenotype, transcriptional program and
   functionality, including the ability to cross-present antigens to CD8^+
   T-cells. Moreover, we showed that overexpression of PIB in human
   fibroblasts yielded a small population of cDC1-like cells, highlighting
   the need to understand the principles governing DC reprogramming and
   optimize its efficiency in human cells. The heterogeneity and
   asynchronicity of direct reprogramming make it difficult to uncover
   fundamental mechanisms using population-based analysis. Several reports
   have described the utility of scRNA-seq to identify intermediate
   populations that emerge during reprogramming ([66]18, [67]19), map
   clonal dynamics ([68]20), and dissect successful and unsuccessful
   trajectories ([69]19), leading to strategies for increased
   reprogramming efficiency ([70]18, [71]20, [72]21). Given the emerging
   potential of scRNA-seq to uncover DC heterogeneity and the biology of
   direct reprogramming, we asked whether this approach could be used to
   inform and improve human DC reprogramming.

Results

Transcriptional profiling of human DC reprogramming at the single cell level

   To induce DC fate in Human Embryonic Fibroblasts (HEFs), we used a
   polycistronic construct encoding PU.1, IRF8 and BATF3 separated by 2A
   sequences (PIB), cloned in a Doxycycline (Dox)-inducible lentiviral
   vector (TetO-PIB) ([73]17) ([74]Fig. 1A). After transducing HEFs with
   PIB, we observed the emergence of a CD45^+ cell population at day 3 and
   a small population of CD45^+HLA-DR^+ cells from day 6 to day 9 with
   DC-like morphology ([75]Fig. 1B-D) that we named human induced DCs
   (hiDCs). To elucidate transcriptional changes, we performed scRNA-seq
   using the 10X Chromium system. 45,870 cells were profiled from 3
   donors, including peripheral blood cDC1s, cDC2s, pDCs, non-transduced
   HEFs (d0), hiDCs at day 3 (CD45^+, d3), day 6 (CD45^+, d6) and day 9
   (CD45^+HLA-DR^-, d9 DR^-; CD45^+HLA-DR^+, d9 DR^+) ([76]Fig. 1E,
   [77]fig. S1, fig. S2A, [78]table S1). t-Distributed stochastic neighbor
   embedding (t-SNE) visualization highlighted four clusters: HEFs
   (ACTA2^+), cDC1s (CLEC9A^+), cDC2s (CD1C^+) and pDCs (IL3RA^+)
   ([79]Fig. 1E, [80]fig. S2B, [81]table S2). Genes encoding CD45 (PTPRC)
   and HLA-DR (HLA-DRA) used to purify hiDCs were not detected in HEFs
   ([82]fig. S2B). While hiDC d3 and d6 did not map specifically to any
   clusters, hiDCs d9 mapped with cDC1s, with DR^+ being closer to cDC1s
   than DR^- ([83]Fig. 1E, [84]fig S2A). These data suggest human cDC1
   reprogramming requires a timeframe of 9 days and the CD45^+HLA-DR^-
   represent a partial reprogrammed cell state. Accordingly, genes
   upregulated at later stages of reprogramming showed enrichment in
   immune pathways and ontologies, suggesting the progressive acquisition
   of a cDC1 transcriptional program ([85]fig. S2C). Next, we integrated
   our single cell data with a publicly available DC dataset ([86]7) using
   scPred ([87]22). As expected, we observed that 53.8% of purified cDC1s
   were assigned to the DC1 subset, 66.9% and 29.8% of cDC2s to the DC2
   and DC3 subsets, respectively, and 66.5% of pDCs to the DC6 subset
   ([88]fig. S3). While HEFs were unaffiliated, 1.3% of d3, 5.3% of d6,
   14.4% of d9 DR^- and 36.7% of d9 DR^+ hiDCs were assigned specifically
   to the DC1 subset, suggesting that PU.1, IRF8 and BATF3 progressively
   impose a cDC1-like transcriptional program in HEFs with a degree of
   heterogeneity that does not cross subset boundaries ([89]Fig. 1F-G).
   DC1-affiliated cells express higher levels of the cDC1-specific genes
   CADM1 and WDFY4 when compared to their unaffiliated counterparts
   ([90]7)([91]Fig. 1H). Accordingly, hiDCs expressed the cDC1 surface
   markers CD141 (30.1±14.9%), CD38 (34.2±12.9%) and CLEC9A (10.6±1.7%)
   ([92]fig. S4A-C). cDC2 (CD1C and CD11b) or pDC (CD123) markers were not
   detected ([93]fig. S4B) ([94]7). CD135/FLT3 receptor and the myeloid
   marker CD33 were detected at low levels. As expected, expression of
   T-cell (CD3), macrophage (CD14), granulocyte (CD15), B cell (CD19) and
   NK cell (CD56) markers was not detected ([95]fig. S4D). Next, we
   extracted the most variable genes across the dataset and grouped them
   in 5 clusters ([96]Fig. 1I). Cluster 1 contains genes highly expressed
   in HEFs and silenced during reprogramming. Cluster 2 highlights early
   transcriptional changes during reprogramming and cluster 3 includes the
   cDC1-specific genes C1orf54, ANPEP, TACSTD2 and SLAMF8 ([97]7, [98]23,
   [99]24) highly expressed in d9 hiDCs and cDC1s ([100]Fig. 1J). Cluster
   4 and cluster 5 contain genes highly expressed in cDC2s and pDCs,
   respectively. Interestingly, d9 hiDCs express high levels of antigen
   processing and cross-presentation genes including PSMB9, TAP1 and HLA-C
   ([101]Fig. 1K-M). Indeed, surface expression of the co-stimulatory
   molecules CD40, CD80 and CD86 was detected in hiDCs, 33.6±14.3%,
   30.7±1.1%, and 16.5±3.5%, respectively ([102]fig S4E-F), suggesting
   that reprogrammed cells have acquired T-cell stimulatory capacity.
   Together, these data show that PIB factors impose a cDC1 signature in
   human fibroblasts.

Fig. 1. PU.1, IRF8 and BATF3 induce global DC1 gene expression program in
human fibroblasts.

   [103]Fig. 1
   [104]Open in a new tab

   (A) HEF were co-transduced with Dox-inducible lentiviral particles
   encoding M2rtTA (UbC-M2rtTA) and PU.1, IRF8 and BATF3 (PIB, TetO-PIB).
   Purified PIB-transduced HEFs (hiDC) were profiled by single-cell
   RNA-seq at day 3 (CD45^+), day 6 (CD45^+) and day 9 (CD45^+HLA-DR^-,
   hiDC d9 DR^-; CD45^+HLA-DR^+, hiDC d9 DR^+). HEF and peripheral blood
   cDC1 (HLA-DR^+CD11C^+CD141^+), DC2 (HLA-DR^+CD11C^+CD141^-CD1C^+) and
   pDC cells (HLA-DR^+CD11C^-CD123^+) were included as controls. (B) Flow
   cytometry analysis of hiDCs at day 3 and 9 after addition of Dox. (C)
   Kinetics of hiDC d9 DR^- (top) and hiDC d9 DR^+ (bottom) cell emergence
   (n=5-8, mean ± SD). (D) Scanning electron microscopy of hiDCs and
   M2rtTA control at day 9. Scale bars, 10 μm. (E) t-SNE plot of
   single-cell transcriptomes showing 45,870 single cells. (F) Integration
   of single-cell data with published DC subset data (DC1-DC6 ([105]7))
   using scPred ([106]22). Heat map shows the percentage of single cells
   affiliated to DC1-DC6 subsets or unaffiliated. (G) t-SNE of single
   cells affiliated by scPred. (H) Violin plots showing gene expression
   distribution of cDC1-specific genes in hiDCs d9 and cDC1s. Log values
   of gene counts are shown. (I) Heat map showing differentially expressed
   genes across profiled populations and grouped in 5 clusters. (J) Violin
   plots showing expression distribution of genes selected from cluster 3.
   (K) Top five Reactome pathway enrichment analysis for each gene cluster
   using Enrichr. (L) Heat map and (M) violin plots showing expression of
   genes associated with antigen cross-presentation pathway across
   profiled populations.

Pathways associated with successful and unsuccessful cDC1 reprograming

   We used Monocle 3 to reconstruct the cDC1 reprogramming trajectory
   ([107]25). HEFs and cDC1s were placed in the beginning and end of
   pseudotime, respectively ([108]Fig. 2A-B). While d9 hiDCs were placed
   at the end together with cDC1s, d3 and d6 hiDCs were located in the
   middle highlighting the stepwise transition of single cell
   transcriptomes during cDC1 reprogramming. Importantly, affiliated d9
   hiDCs were positioned later in pseudotime when compared to their
   unaffiliated counterparts, suggesting that trajectory reconstruction is
   capturing a successful cDC1 reprogramming path. We also observed a
   “dead-end” trajectory with unaffiliated hiDCs mapping closer to HEFs,
   suggesting that these cells fail to enter the successful reprogramming
   path. To infer expression dynamics and directionality of individual
   cells during reprogramming, we applied scVelo analysis ([109]26) to the
   reprogramming trajectory. hiDC velocities are mainly aligned with the
   cDC1 reprogramming trajectory projected by Monocle 3 ([110]Fig. 2C).
   Nevertheless, we observed that hiDCs mapping closer to the dead-end
   trajectory display velocities pointing towards HEFs, in agreement with
   previous reports describing unsuccessful reprogramming paths marked by
   the expression of genes associated with the original cell state
   ([111]19–[112]21). Reconstruction of gene expression dynamics using
   latent time showed downregulation of cell cycle genes during cDC1
   reprogramming ([113]Fig. 2D-E), suggesting that hiDCs exit cycle. We
   also observed enrichment in cytokine and IFN type-I and II (IFN-γ)
   signaling pathways later in latent time. To address differences between
   hiDC d9 DR^+ and cDC1s, we performed gene set enrichment analysis
   (GSEA) and observed that reprogrammed cells were enriched in IFN-γ,
   IFN-α, Oncostatin M, IL-6 and IFN-α signaling pathways ([114]fig. S5A).
   The downregulation of cell cycle and upregulation of IFN gene
   signatures go in line with our previous findings in the mouse system
   ([115]17). Indeed, we identified 742 genes commonly upregulated during
   mouse and human DC reprogramming that include genes implicated in IFN
   response ([116]fig. S5B, C), suggesting a species-conserved role for
   IFN signaling in cDC1 reprogramming. To map stage-specific gene changes
   along reprogramming, we clustered genes differentially expressed along
   pseudotime in 21 modules ([117]Fig. 2F). Unaffiliated hiDCs fail to
   downregulate several gene modules enriched in HEFs, including modules
   1, 2, 4 and 7 (unsuccessful reprogramming). Affiliated hiDCs are
   enriched in modules 3, 9, 13, 15 and 17 (successful reprogramming)
   highly expressed in cDC1s. Extraction of genes encoding surface
   molecules and transcriptional regulators from these modules highlighted
   fibroblast genes enriched in unsuccessful reprogramming (CD248 and
   PRRX1) and DC genes upregulated in successful reprogrammed cells,
   including the cDC1 marker CD226 ([118]23) and IRF7 ([119]27) ([120]Fig.
   2G, [121]figure S6A-B). Accordingly, hiDC d9 DR^+ expressed higher
   levels of CD226 than hiDC d9 DR^- ([122]Fig. 2H). To validate the
   utility of CD226 to identify cDC1-like cells, we purified CD226^+ hiDCs
   d9 DR^+ and compared their DC1 affiliation with CD226^- hiDC d9 DR^+
   cells. Interestingly, profiling of CD226^+ cells resulted in increased
   DC1 affiliation (from 19.5% to 40.9%) ([123]Fig. 2I), suggesting that
   CD226 allows the isolation of hiDCs with refined cDC1 identity.
   Transcription factor enrichment analysis for unsuccessful reprogramming
   genes identified previously described barriers to direct reprogramming,
   including TWIST1, TWIST2, PRRX1, PRRX2 and OSR1 ([124]28) ([125]fig.
   S6C, [126]table S2). Interestingly, the same analysis on successful
   reprogramming genes reinforced the importance of PU.1, IRF8 and BATF in
   the establishment of successful cDC1 reprogramming gene signatures
   ([127]fig. S6D). To assess whether additional regulators would enhance
   reprogramming, we supplemented PU.1, IRF8 and BATF3 with individual
   transcription factors associated with successful cDC1 reprogramming
   ([128]fig. S6B, S6D, [129]table S2) and observed that IRF7 and BATF
   increased cDC1 reprogramming efficiency ([130]Fig. 2J). IRF7 is a
   transcriptional regulator downstream inflammatory signalling ([131]27).
   BATF is highly homologous to BATF3 and was shown to compensate BATF3
   during cDC1 development ([132]29). These results suggest that cDC1
   reprogramming efficiency may be increased with inflammatory signalling
   and the intrinsic modulation of transcription factors enriched in
   successful cDC1 reprogramming.

Fig. 2. Pseudo-temporal ordering of single cells highlights gene modules and
pathways associated with successful and unsuccessful cDC1 reprogramming.

   [133]Fig. 2
   [134]Open in a new tab

   (A) Monocle 3 reconstruction of single-cell trajectories for HEF, hiDC
   at day 3, day 6, day 9 (d9 DR^- and d9 DR^+) unaffiliated or affiliated
   with DC1 and scPred-filtered cDC1s. (B) cDC1 reprogramming trajectory
   colored by relative trajectory position (pseudotime) (left). Whisker
   box plots showing pseudotime distribution by cell type (right). (C)
   scVelo ([135]26) single-cell velocities visualized in a tSNE plot.
   Arrows indicate direction and thickness indicates speed along cDC1
   reprogramming trajectory. (D) Heat map highlighting 6 gene clusters
   (A-F) showing gene expression dynamics along scVelo latent time. (E)
   Top 5 Reactome pathway enrichment analysis of clusters A-F. (F) Heat
   map showing mean expression values of gene modules by cell-type
   differentially expressed along cDC1 reprogramming trajectory. (G)
   Violin plots showing expression distribution of genes associated with
   unsuccessful and successful DC reprogramming. Log values of gene counts
   are shown. (H) Flow cytometry analysis of CD226 expression in d9 DR^-
   and d9 DR^+ hiDCs. (I) Classification with published DC subset data
   (DC1-DC6 ([136]7)) using scPred ([137]22). (J) Percentage of
   CD45^+HLA-DR^+ cells at day 9 generated by co-transduction of
   Dox-inducible PIB with indicated transcription factors (n=4, mean ±
   SD). M2rtTA- and PIB-transduced cells were included as controls. **
   p<0.005; **** p<0.00005.

Increasing cDC1 reprogramming efficiency

   We hypothesize that inflammatory cytokines could synergize with PIB to
   impose cDC1 identity. Thus, we added 17 hematopoietic cytokines
   ([138]table S1) to the cultures 2 days after PIB induction in HEFs and
   measured reprogramming efficiency at day 9. IFN-γ had the most
   significant impact promoting 20-fold increase in CD45^+HLA-DR^+ cell
   generation (7.9±2.2% versus 0.4±0.2%) ([139]Fig. 3A). Interestingly,
   other inflammatory cytokines also increased reprogramming efficiency,
   including IL-1β (3-fold), IL-6 (2.5-fold), Oncostatin M (4-fold), TNF-α
   (3-fold) and IFN-β (4-fold), providing experimental validation to the
   single-cell analysis. Similarly to mouse DC reprogramming ([140]17),
   FLT3L, IL-4 and GM-CSF used for in vitro differentiation of DCs from
   progenitors ([141]12) or monocytes ([142]30) did not impact
   reprogramming efficiency. As expected, we did not observe
   CD45^+HLA-DR^+ cells in non-transduced HEF cultures supplemented with
   cytokines. Considering the role of IFNs as regulators of HLA gene
   expression, we quantified the emergence of total CD45^+ cells as a
   measure of hematopoietic commitment ([143]fig. S7A). IFN-γ and IFN-β
   induced 2- and 3-fold increase in CD45^+ cells respectively, confirming
   that IFN signaling facilitates DC reprogramming independently of HLA-DR
   activation. HiDC generation remained constant with doses of IFN-γ
   ranging from 1.5 to 100 ng/ml, showing that low concentrations are
   effective ([144]fig. S7B). We then tested combining IFN-γ with other
   inflammatory cytokines ([145]Fig. 3B, [146]fig. S7C-D). IFN-γ with
   IFN-β or IFN-α induced an additional 2.5-fold and 2-fold increase and
   combining the three cytokines resulted in 28.1±3.4% efficiency
   ([147]Fig. 3B, [148]fig. S7C-D). Accordingly, hiDCs express genes
   encoding surface receptors for Type I and II IFN and TNF ([149]fig.
   S7E). These data demonstrate that DC reprogramming efficiency is
   increased with the provision of inflammatory cytokines.

Fig. 3. Inflammatory cytokine signaling and enforced expression of
transcription factors enable human cDC1 reprogramming at high efficiency.

   [150]Fig. 3
   [151]Open in a new tab

   (A) Quantification of hiDCs (CD45^+HLA-DR^+) at day 9 obtained in the
   presence of single cytokines and (B) combinations of two or three
   cytokines. Non-transduced HEF were included as control (n=2-5, mean ±
   SD). (C) Quantification of reprogrammed cells (tdT^+MHC-II^+ in live
   GFP^+ cells) obtained by transducing Clec9a-tdTomato (tdT) reporter
   mouse embryonic fibroblasts with lentiviral particles encoding GFP and
   PIB driven by Dox-inducible (TetO) or constitutive promoters (UbC,
   SFFV, PGK, EF1S, EF1 and EF1i promoters). (D) Quantification of hiDCs
   at day 9 generated with TetO-PIB or SFFV-PIB, in the presence or
   absence of IFN-γ, IFN-β and IFN-α (n=3-9, mean ± SD). (E)
   Quantification of hiDC yield after 9 days per input fibroblast. (F)
   hiDCs at day 9 generated in the four conditions were FACS sorted and
   profiled by scRNA-seq. Heat map shows the percentage of single cells
   affiliated to the DC1-DC6 subsets ([152]7) by scPred ([153]22). (G)
   Heat map showing relative expression of genes commonly upregulated in
   hiDCs in the four conditions and in cDC1s. (H) Violin plots showing
   expression distribution of antigen-presentation genes. Log values of
   gene counts are shown. (I) Single-cell distribution along the cDC1
   reprogramming pseudotime by condition. (J) GSEA for successful cDC1
   reprogramming (genes included in modules 13, 3, 15, 17, and 9) between
   hiDC day 9 and HEF. Normalized enrichment score (NES) and False
   discovery rate (FDR) q-values are shown. * p<0.05; ** p<0.005; ***
   p<0.0005; **** p<0.00005.

   Using trajectory reconstruction of cDC1 reprogramming, we identified
   PU.1, IRF8 and BATF3 as drivers of successful reprogramming gene
   signatures. Thus, we hypothesized that enforced expression of this
   combination of transcription factors using a stronger constitutive
   promoter would increase reprogramming efficiency. We cloned the PIB
   polycistronic cassette followed by IRES-GFP into constitutive
   lentiviral backbones with multiple promoters and evaluated cDC1
   reprogramming efficiency in Clec9a-tdTomato MEFs ([154]17) ([155]Fig.
   3C). PIB overexpression driven by spleen focus-forming virus (SFFV)
   promoter induced superior efficiency (46.6±16.7% tdTomato^+ MHC-II^+
   cells) ([156]Fig. 3C, [157]fig. S7F). Indeed, SFFV-driven
   overexpression of reprogramming factors increased efficiency of
   reprogramming fibroblasts to iPSCs ([158]31). In HEFs we observed the
   emergence of 21.3±6.1% hiDCs with SFFV overexpression ([159]Fig. 3D).
   Interestingly, this system induced CD45 expression in the majority
   (>80%) of fibroblasts, suggesting that reprogramming is kick-started in
   a larger cohort of cells ([160]Fig. 3D). We then combined SFFV-driven
   induction with IFN-γ, IFN-β and IFN-α and observed 76.9±11.9%, of
   hiDCs, a remarkable 190-fold increase in cDC1 reprogramming efficiency
   when compared to the original protocol ([161]Fig. 3D). We also observed
   an increase in the absolute number of hiDCs generated with the improved
   protocol ([162]Fig. 3E). Then, we evaluated the impact of
   anti-inflammatory cytokine signaling during cDC1 reprogramming. While
   IL-10 did not impact reprogramming efficiency, TGF-β reduced 2-fold
   ([163]fig. S7G). Neither of the cytokines had an impact in CD40
   expression. Next, we profiled hiDCs obtained with inducible (TetO-PIB)
   and constitutive (SFFV-PIB) with and without cytokines and used scPred
   for integration ([164]Fig. 3F, [165]table S3). Interestingly, 61.4% and
   53.2% of hiDCs generated with SFFV were affiliated to DC1 lineage, with
   and without cytokines respectively, in contrast to only 33.4% and 22.0%
   generated with the inducible system. These data suggest that cytokine
   signaling and enforced expression of PU.1, IRF8 and BATF3 synergize for
   successful reprogramming, increasing both reprogramming efficiency and
   cDC1 identity. Commonly upregulated genes are associated with IFN
   signaling and antigen processing and presentation pathways, including
   B2M, HLA-A, HLA-DRA, HLA-DPA1, CD74 and NAAA ([166]Fig. 3G-H, [167]fig.
   S8A-B). Importantly, hiDC generated with SFFV-PIB and cytokines were
   positioned later in pseudotime ([168]Fig. 3I), were enriched in
   successful reprogramming gene signatures ([169]Fig. 3J) and expressed
   higher levels of CD226, CD38 and CADM1 surface markers correlated with
   successful cDC1 reprogramming ([170]fig. S8C, [171]Table S2).
   Collectively, we demonstrated that scRNA-seq coupled with
   dimensionality reduction and trajectory inference techniques inform the
   dynamics of cDC1 reprogramming and identify critical pathways for
   establishing human cDC1 fate.

Functional human reprogrammed cDC1-like cells

   DCs orchestrate adaptive immunity by uptaking and presenting antigens
   to T-cells. Thus, we evaluated the levels of costimulatory molecules
   induced upon activation with agonists for toll-like receptor 4 (TLR4)
   [Lipopolysaccharide (LPS)], TLR3 [Polyinosinic-polycytidylic acid (Poly
   I:C)] and TLR7/8 [Resiquimod (R848)] in hiDCs, required for efficient
   T-cell activation. While hiDCs responded to higher extent to TLR4 than
   cDC1s, they both upregulated co-stimulatory molecules after TLR3 or
   combined stimuli ([172]Fig. 4A). To evaluate phagocytic capacity, we
   performed short incubation with labeled dead cells. We observed that
   47.5 ± 12.0%, 19.4 ± 7.7% and 10.9 ± 3.5% of hiDCs, hiDCs generated in
   the presence of IFN-γ, IFN-β and IFN-α (hiDC+cyt) and cDC1s,
   respectively, incorporated dead cell material ([173]Fig. 4B-D,
   [174]movie S1), a critical feature of cross-presenting DCs.
   Interestingly, we observed that CD226^+ hiDCs performed better in dead
   cell phagocytosis when compared to CD226^- hiDCs, suggesting that CD226
   marks functional hiDCs ([175]Fig. 4E). DC maturation and phagocytosis
   are often inversely correlated ([176]10). Accordingly, hiDCs generated
   in the presence of IFN-γ, IFN-β and IFN-α expressed higher levels of
   co-stimulatory molecules and showed decreased capacity to incorporate
   dead cells ([177]Fig. 4A-C).To confirm that hiDCs also provide the
   third signal required for T-cell activation, we evaluated cytokine
   secretion ([178]Fig. 4F). We first observed that hiDCs and cDC1s
   responded to TLR3 challenge by secreting the human cDC1-specific
   cytokine IFN-λ1 ([179]9). This comes in contrast to moDCs that were
   unresponsive to TLR3 agonists ([180]32). In addition, hiDCs also
   responded to TLR4 and 3 by secreting IL12p70, CXCL10 and TNF-α.
   Moreover, we observed that IFN-γ, IFN-β and IFN-α signaling increased
   the magnitude of cytokine secretion. We then asked whether hiDCs
   cross-present antigens to CD8^+ T-cells. We co-cultured HEFs, moDCs,
   cDC1s and hiDCs pulsed with CMV protein with CD8^+ T cells isolated
   from CMV^+ donors. As readout for T-cell activation, we quantified
   IFN-γ secretion after co-culture ([181]Fig. 4G). As expected, we
   observed that cDC1s, in contrast to moDCs or HEFs, efficiently
   cross-presented CMV antigens to CD8^+ T cells. Strikingly, we observed
   that hiDCs generated with or without cytokines established the ability
   to cross-present antigens to CD8^+ T-cells. Together, these data
   support that reprogrammed hiDC are functional cross-presenting DCs.

Fig. 4. Optimized DC reprogramming protocol allows generation of functional
human cDC1-like cells.

   [182]Fig. 4
   [183]Open in a new tab

   (A) Median fluorescence intensity (MFI) of CD40 and CD80 expression in
   hiDCs (CD45^+HLA-DR^+) at day 9 generated with SFFV-PIB in the absence
   or presence of IFN-γ, IFN-β and IFN-α (hiDC+cyt) and in peripheral
   blood CD141^+CLEC9A^+ cDC1. Cells were stimulated overnight with TLR
   agonists LPS, Poly I:C, R848 or the three combined (All) (n=3-7, mean ±
   SD). (B and C) Quantification of CellVue Far Red-labeled dead cell
   engulfment by hiDC and hiDC+cyt at day 9 after 2-hour incubation. HEF
   cultured without and with cytokines (HEF+cyt) and CD141^+CLEC9A+ cDC1
   were included as controls (n=3-12, mean ± SD). (D) Time lapse
   microscopy of dead cell engulfment (red) by FACS purified hiDCs
   (arrowhead). Scale bar, 100 μm. (E) Dead cell engulfment by
   CD45^+HLA-DR^+CD226^+ and CD45^+HLA-DR^+CD226^- hiDC and hiDC+cyt
   (n=6-7, mean ± SD). (F) Cytokine secretion of purified hiDC and
   hiDC+cyt at day 9 after overnight incubation with TLR agonists. HEFs
   cultured without and with cytokines, monocyte-derived DCs (moDC) and
   purified CD141^+CLEC9A^+XCR1^+ cDC1 were included as controls (n=3-12,
   mean ± SD). (G) HEF, HEF+cyt, hiDC, hiDC+cyt, moDC and cDC1 were primed
   overnight with LPS, Poly I:C and R848, pulsed with CMV protein for 3
   hours, washed and co-cultured with CD8^+ T cells harvested from CMV^+
   donors. Antigen cross-presentation was quantified by measuring IFN-γ
   levels in the supernatant 24 h after co-culture (n=2-4, mean ± SD). *
   p<0.05; ** p<0.005; *** p<0.0005; **** p<0.00005.

Efficient reprogramming of human adult somatic cells

   The generation of cDC1s from easily accessible human cells would enable
   cDC1-based vaccination. Thus, we investigated the feasibility of
   generating hiDCs from adult human dermal fibroblast (HDFs) and
   mesenchymal stromal cells (MSCs). First, we observed that cDC1
   reprogramming efficiency ranged from 20-35% across HDFs obtained from 3
   healthy donors ([184]Fig. 5A-B). When combined with IFN-γ, IFN-β and
   IFN-α, reprogramming efficiency increased ~2-fold as well as CD40 and
   CD80 expression ([185]Fig. 5C). scPred analysis correctly assigned
   60.6% and 59.3% of HDF-derived hiDCs generated with and without
   cytokines, respectively, to the DC1 subset ([186]Fig. 5D, [187]table
   S4). cDC1 identity was further supported by t-SNE visualization
   ([188]Fig. 5E),expression of cDC1-specific (C1orf54, NAAA and HLA-DPA1)
   ([189]Fig. 5F) and antigen processing and presentation genes (CD74,
   HLA-C, B2M, PSMB9 and TAP1) ([190]Fig. 5G). We then asked whether we
   could generate hiDCs from MSCs in xeno-free conditions. Purified MSCs
   from 3 donors were transduced with SFFV-PIB particles and cultured in
   chemically defined, serum-free X-VIVO 15 media ([191]Fig. 5H). We
   observed that 60-75% of MSCs converted into hiDCs co- expressing CD40
   and CD80 ([192]Fig. 5I-K). Interestingly, IFN-γ, IFN-β and IFN-α did
   not improve the generation of hiDCs from MSCs, suggesting that
   inflammatory cytokine signaling may facilitate cDC1 reprogramming in a
   cell-type specific manner. These data highlight the consistency of
   human cDC1 reprogramming across multiple somatic cell types and donors.

Fig. 5. Efficient cDC1 reprogramming of adult fibroblasts and mesenchymal
stromal cells.

   [193]Fig. 5
   [194]Open in a new tab

   (A) Flow cytometry analysis and (B) quantification of hiDC
   (CD45^+HLA-DR^+) at day 9 generated from human dermal fibroblasts (HDF)
   in the absence (SFFV-PIB) or presence of IFN-Y, IFN-β and IFN-α
   (SFFV-PIB+cyt) from three independent donors. HDF cultured in the
   absence (-) or presence of cytokines (+cyt) were included as controls
   (n=3-12, mean ± SD). (C) Expression of CD40 and CD80 in HDF-derived
   hiDC. (D) HDF-derived hiDC were FACS sorted and profiled by scRNA-seq.
   Heat map shows the percentage of single cells affiliated to DC1-6
   subsets ([195]7) by scPred ([196]22). (E) t-SNE of single cell
   transcriptomes showing HDFs, HDF-derived hiDCs and cDC1s. (F) Heat map
   showing relative expression of genes upregulated during cDC1
   reprogramming of HDF and expressed in cDC1s. cDC1 and antigen
   presentation genes are highlighted in bold and shown in (G) as violin
   plots. Log values of gene counts are shown. (H) Strategy to derive hiDC
   from human Mesenchymal Stromal Cells (MSC) under xeno-free conditions.
   MSC were isolated from three healthy donors, FACS-purified
   (Lin^-CD45^-CD271^+), expanded in pHPL media, transduced and cultured
   in X-VIVO 15. (I and J) Quantification of MSC-derived hiDCs at day 9
   generated with or without cytokines (n=3-7, mean ± SD). (K) Flow
   cytometry analysis of CD40 and CD80 in MSC-derived hiDCs. ns - non
   significant; **** p<0.00005.

PU.1-dominant chromatin targeting capacity in cDC1 reprogramming

   To shine light on molecular mechanisms underlying DC reprogramming
   mediated by PU.1, IRF8 and BATF3, we expressed the three reprogramming
   factors in combination or individually in HDFs, and performed chromatin
   immunoprecipitation sequencing (ChIP-seq) at early stages of
   reprogramming (48h, [197]Fig. 6A). First, we observed that PU.1 showed
   the highest chromatin binding activity (75,593 peaks), followed by IRF8
   (18,962 peaks) and BATF3 (11,505 peaks) when factors were co-expressed
   ([198]Fig. 6B, [199]table S5). Interestingly, we observed that >40% of
   PU.1-bound peaks were similar between individual and combined
   expression, suggesting that PU.1 has independent targeting capacity
   that is augmented when IRF8 and BATF3 are available. Accordingly, we
   observed increased PU.1 binding at promoter regions when expressed in
   combination with IRF8 and BATF3 ([200]Fig. 6C). In sharp contrast, IRF8
   and BATF3 peaks were scarce when these transcription factors were
   expressed individually (<3% of peaks when compared to combined
   expression), suggesting that IRF8 and BATF3 require cooperative binding
   with PU.1 to engage chromatin and induce cDC1 fate. The DC1-specific
   genes C1orf54 and ZNF366, and STAT1, important for DC maturation, were
   targeted individually by PU.1 or the three transcription factors when
   combined ([201]Fig. 6D, [202]fig. S9A). Known and de novo motif
   prediction analysis for PU.1 peaks showed strong enrichment for PU.1
   motif when expressed individually or in combination ([203]Fig. 6E,
   [204]fig. S9B). While IRF8 and BATF3 expressed individually showed
   enrichment in IRF and AP-1 motifs respectively, the PU.1 motif was
   highly enriched for these transcription factors when expressed in
   combination. We next explored the regulatory interactions between PU.1,
   IRF8 and BATF3 and observed that these cDC1-inducing factors bind each
   other's loci when co-expressed ([205]fig. S9C), suggesting
   auto-regulation at the transcriptional level ([206]fig. S9D). These
   data go in line with recent findings describing PU.1 as a non-classical
   pioneer transcription factor capable of redistributing partner
   transcription factors in human myeloid and lymphoid cells ([207]33),
   and highlight the cooperative dynamics of the three reprogramming
   factors at early stages of cDC1 reprogramming.

Fig. 6. PU.1 has dominant and independent chromatin targeting capacity,
recruiting IRF8 and BATF3 to shared binding sites.

   [208]Fig. 6
   [209]Open in a new tab

   (A) Strategy to profile chromatin binding sites of PU.1, IRF8 and BATF3
   (PIB) at early stages of reprogramming. HDFs were transduced with PIB
   (left) or individual factors (right) and analyzed by ChIP-seq after 48
   hours. (B) Heatmaps showing genome-wide distribution of PU.1, IRF8 and
   BATF3 when expressed in combination (left) or individually (right).
   Signal is displayed within an 8 kb window centered on individual peaks.
   The number of peaks in each condition is shown. Average signal
   intensity of peaks is depicted (bottom). (C) Pie charts showing genomic
   distribution of PU.1, IRF8 and BATF3 peaks when expressed together or
   individually. (D) PU.1, IRF8 and BATF3 occupancy profile at the C1orf54
   and ZNF366 loci. Boxes highlight PU.1 peaks when expressed individually
   or combined with IRF8 and BATF3. (E) De novo motif prediction analysis
   for PIB or individual transcription factor target sites.

Cooperative binding of PU.1, IRF8 and BATF3 at promoters and enhancers at
open chromatin

   We next investigated the overlap between PU.1, IRF8 and BATF3 chromatin
   targets. 5,383 genomic positions were shared between the three
   reprogramming factors when expressed together, representing 28% and 47%
   of total IRF8 and BATF3 peaks, respectively ([210]Fig. 7A). As
   expected, only a residual overlap (3 genomic positions) was observed
   when the transcription factors were expressed individually ([211]fig.
   S10A). De novo motif prediction for the PIB overlapping peaks showed
   enrichment for PU.1-IRF and BATF motifs ([212]Fig. 7B), which also
   displayed some overlap and similarity (Jaccard similarity index = 0.02)
   ([213]Fig. 7C). Furthermore, differential motif prediction analysis for
   unique and shared peaks highlighted enrichment in BATF-IRF4 motif
   ([214]fig. S10B). These data suggest that PU.1, IRF8 and BATF3 interact
   physically. To test this hypothesis, we performed
   co-immunoprecipitation (co-IP) and observed interactions between the 3
   factors, highlighting an important interplay between reprogramming
   factors ([215]Fig. 7D, [216]fig. S10C). We next plotted differentially
   expressed genes between HDFs and hiDC d9 that were bound by at least
   one of the reprogramming factors and observed that they contain both
   downregulated fibroblast genes and upregulated cDC1-associated genes,
   including SLAMF8 and TACSTD2 ([217]Fig. 7E, [218]fig. S10D, E). To
   dissect whether PU.1, IRF8 and BATF3 binding occurs at open or closed
   chromatin regions, we took advantage of publicly available ChIP-seq
   datasets for histone marks in HDFs. PU.1, IRF8 and BATF3 co-bound
   sites, as well as sites bound by individual transcription factors, were
   mainly enriched in chromatin marks associated with active promoters and
   enhancers including H3K4me3, H3K4me1 and H3K27ac ([219]Fig. 7F,
   [220]fig. S10F). To better characterize PIB chromatin engagement, we
   used ChromHMM chromatin segmentation ([221]34). We observed that PIB
   co-bound peaks were enriched mainly at promoter and enhancer regions
   ([222]Fig 7G). We also observed a small fraction (12%) of peaks
   associated with bivalent chromatin marked either by H3K4me1, H3K4me3
   and H3K27me3 or H3K4me1 and H3K27me3. Together, our data support a
   model where PU.1 binds mainly to active promoters and enhancers at open
   chromatin sites and recruits IRF8 and BATF3 to silence fibroblast genes
   and gradually impose the human cDC1 transcriptional program ([223]Fig.
   7H).

Fig. 7. PU.1, IRF8 and BATF3 cooperatively bind at fibroblast and cDC1 genes
marked with open chromatin marks.

   [224]Fig. 7
   [225]Open in a new tab

   (A) Venn diagram shows genome-wide peak overlap between PU.1, IRF8 and
   BATF3 (PIB). (B) De novo motif prediction analysis for co-bound sites.
   (C) Motif comparison between enriched PU.1-IRF and BATF motifs. Jaccard
   similarity coefficient = 0.02. (D) Immunoblots showing
   immunoprecipitation (IP) for PU.1 (top), IRF8 (middle), and BATF3
   (bottom) in HEK293T cells 24 hours after transfection with PIB (left).
   Co-immunoprecipitation (Co-IP) performed with 1, 2 and 5 million (M)
   cells after transfection with PIB (right). Non-immunoprecipitated input
   (10%), and IgG isotype antibody were used as controls. (E) Heat map
   showing differentially expressed genes between HDFs and hiDC at day 9
   that are bound by either PU.1, IRF8 and BATF3 or the three factors
   (intersect). (F) Heatmaps of normalized read coverages of chromatin
   marks in HDFs for co-bound sites. The signal is displayed within an 8
   kb window and centered on transcription factor binding sites. Average
   signal intensity of histone marks at PIB intersection peaks is shown in
   the upper panel. (G) Heatmaps for chromatin state enrichment showing
   the percentage of genome occupancy for PU.1 (P), IRF8 (I) and BATF3 (B)
   when expressed together. Genome occupancy of co-bound sites (PIB
   intersect) is shown on the right. (H) Model for the mechanism of action
   of PIB to set in motion cDC1 reprogramming.

Discussion

   In this study, we use scRNA-seq to dissect reprogramming trajectories
   and identify pathways associated with successful and unsuccessful cDC1
   reprogramming. Overexpression of PIB imposes a cDC1-like cell fate and
   does not induce transcriptional programs of other DC subsets at the
   single-cell level. This goes in line with our previous findings, where
   PIB overexpression imposed a cDC1-like fate in mouse fibroblasts
   ([226]17). A fraction of hiDCs did not downregulate fibroblast-specific
   genes and therefore did not achieve complete reprogramming. Similarly,
   a refractory reprogramming path sustaining fibroblast gene expression
   was identified in endoderm progenitor, neuronal and cardiac
   reprogramming ([227]19–[228]21). It will be interesting to test whether
   pharmacological inhibition or gene knock-down of transcriptional
   regulators associated with unsuccessful reprogramming lower barriers of
   cDC1 reprogramming ([229]18) and address whether those barriers are
   conserved in other reprogramming systems. We used single-cell
   transcriptional profiling to resolve pathways associated with
   successful cDC1 reprogramming and improve efficiency. We identified
   IRF7 and BATF as additional regulators of the successful reprogramming
   path. Interestingly, IRF7 is a master regulator of type-I IFN-dependent
   immune responses ([230]27). IFN-β signaling increased cDC1
   reprogramming efficiency, pointing out to a central role of type-I IFN
   signaling in hiDC generation. Additionally, a BATF3-independent pathway
   for the generation of cDC1s in mice has been identified during
   infection by intracellular pathogens requiring compensation by BATF
   ([231]29). Our data suggest that BATF3 and BATF may synergize
   molecularly during cDC1 reprogramming. Moreover, trajectory
   reconstruction enabled the identification of CD226, a cDC1-specific
   surface marker ([232]23, [233]35), that allowed the isolation of hiDCs
   with improved functional properties. Successful cDC1 reprogramming gene
   signatures were also associated with the reprogramming factors. DC
   specification relies on different doses of PU.1 and IRF8, with the cDC1
   lineage requiring the highest of both ([234]2). Indeed, effects
   downstream IRF8 were shown to be dose-dependent ([235]36). In this
   study, we showed that enforced expression of PU.1, IRF8 and BATF3 using
   the strong SFFV promoter allowed major improvements in cDC1
   reprogramming efficiency. In reprogramming, high levels of the
   transcription factors Ascl1 or Mef2c, Gata4 and Tbx5 are required for
   successful neuronal ([236]19) or cardiomyocyte reprogramming,
   respectively ([237]18). Interestingly, previous studies showed that
   SFFV promoter allows high transgene expression and efficient iPSCs
   generation ([238]31), pointing out to a general cell fate conversion
   mechanism. We also identified the association of inflammatory signaling
   and successful cDC1 reprogramming. We confirmed synergy between
   extrinsic IFN-γ, IFN-β and IFN-α signaling with intrinsic expression of
   reprogramming factors. The cytokine cocktail may increase efficiency by
   inherently promoting a cDC1-specific transcriptional program or by
   inducing an inflammatory program that benefits cell fate conversion.
   Indeed, TLR3 stimulation was shown to enhance chromatin remodeling and
   reprogramming towards iPSCs ([239]37), while TLR3 knock down impaired
   the generation of cardiomyocytes by direct reprogramming ([240]21). As
   an exception, cytokines did not increase efficiency from MSCs,
   suggesting that reprogramming is already happening at optimal levels or
   that MSCsare less responsive to inflammatory stimuli. Collectively,
   single-cell analysis revealed the gradual and asynchronous nature of
   exclusive cDC1 reprogramming, highlighted a successful reprogramming
   path with associated pathways and factors that allowed a 190-fold
   increase in reprogramming efficiency. We can envision that a similar
   single-cell approach can be applied to other reprogramming systems
   where translation to the human system is limited by low efficiency.

   The rarity of natural cDC1s has limited genomic location studies in
   human cells. Nevertheless, studies in mice have shown that IRF8
   interacts with ETS and AP-1 factors during DC ontogeny. For instance,
   IRF8 is recruited to target sites via the interaction with PU.1 or
   other ETS transcription factors, due to the low affinity of IRF8 to
   interferon-response elements ([241]38). Moreover, AP-1 TFs, including
   BATF3, were shown to interact with IRF4 and IRF8 at AP-1–IRF composite
   sites ([242]29). Here, our chromatin immunoprecipitation and protein
   interaction studies show that the three transcription factors interact
   and form a "cDC1 reprogramming complex" at the early stages of cDC1
   reprogramming. Importantly, these data imply that induction of human
   cDC1 identity involves direct interplay between the three transcription
   factors. Accordingly, it was recently reported that IRF8 binding at
   mouse cDC1-specific genes is enriched in both AP1–IRF and ETS–IRF
   composite elements ([243]39). Chromatin engagement models of
   transcription factor-mediated reprogramming have been suggested to rely
   on pioneer factor capacity of binding closed chromatin and to directly
   induce target cell-type gene expression programs ([244]40) or to depend
   on the cooperative interaction enabling the inactivation of the
   starting cell-type enhancers and the gradual induction of target
   cell-type genes ([245]41). Our cDC1 reprogramming data support a
   cooperative and step-wise, immune induction model ([246]41, [247]42).
   PU.1 functions as a non-classical pioneer transcription factor as it
   binds to promoters and enhancers at open chromatin regions recruiting
   IRF8 and BATF3. In CD4^+ T cells, reorganization of chromatin relies on
   BATF pioneer function coupled with ETS1-dependent recruitment of CTCF
   ([248]43). In cDC1 reprogramming however, BATF3 shows an almost
   complete dependency on PU.1 and IRF8 expression for chromatin binding.
   This suggests that either BATF and BATF3 utilize different chromatin
   engagement mechanisms or in cDC1 specification, BATF3 may cooperate
   with PU.1 for CTCF recruitment and chromatin looping. In the future, it
   will be interesting to further map chromatin binding dependencies of
   each transcription factor by profiling engagement in pairs during cDC1
   reprogramming. We believe cDC1 reprogramming provides a unique system
   to explore molecular mechanisms underlying human cDC1 specification.

   The lack of methods allowing the generation of a pure population of
   functional cDC1-like cells has also been a major bottleneck in the
   clinical development of cDC1-based cancer vaccines ([249]14).
   Classically, moDCs were used to generate DC vaccines due to the ease at
   obtaining sufficient number of cells from peripheral blood ([250]44).
   However, moDCs show functional deficiencies, including their limited
   T-cell priming ability, that may explain suboptimal clinical responses
   ([251]45). Here, we show that hiDCs express co-stimulatory molecules
   and secrete inflammatory cytokines, i.e. IFN-λ, IL12p70 and CXCL10 upon
   TLR stimulation. IFN-λ secretion by tumor-associated cDC1s was shown to
   induce IL12p70 and CXCL10 secretion, involved in the recruitment and
   activation of NK and T cells ([252]9, [253]46). Indeed, expression of
   these cytokines within the tumor is correlated with favorable clinical
   outcome ([254]9). Both natural cDC1s and hiDCs responded to TLR3
   stimulation and we also observed that hiDCs were responsive to TLR4
   stimulation. Accordingly, cDC1s differentiated from CD34^+
   hematopoietic progenitors, pointing out to a general feature of in
   vitro generated cDC1-like cells ([255]12). In the future, it will be
   interesting to investigate whether this functional difference confers a
   functional advantage to hiDCs in driving antigen-specific immune
   responses. We show that hiDCs uptake dead cells, process and
   cross-present antigens to CD8^+ T-cells, which supports cDC1
   affiliation at a functional level. Methods to derive cDC1-like cells
   from BM and iPSC cultures are laborious, time consuming, require feeder
   layers and yield a mixture of different DC subtypes ([256]8,
   [257]11–[258]13). cDC1 reprogramming takes up to 9 days, is robust,
   reproducible from different donors, does not require feeder layers or
   isolation and expansion of CD34^+ progenitors and can be readily
   adapted to good manufacturing practices (GMP). Nonetheless, cDC1
   reprogramming does not transit through an expandable cell state, which
   limits the absolute number of generated cDC1-like cells. This
   limitation may be surpassed in the future by up-scaling the number of
   input cells for reprogramming and by exploring additional reprogramming
   entities, i.e. small molecules and microRNAs, to allow
   near-deterministic cDC1 induction ([259]47). In addition, an important
   future consideration will be the dose and timing for cytokine exposure
   for optimal generation of cDC1s. Indeed, we showed that hiDCs generated
   in the presence of IFN-γ, IFN-β and IFN-α acquire a mature phenotype
   characterized by high expression of co-stimulatory molecules and
   reduced dead cell engulfment capacity, which might limit the
   therapeutic potential of hiDCs. Finally, we showed that PIB allows cDC1
   reprogramming from different starting cell types, demonstrating
   cross-species and cross-cell type conservation of minimal transcription
   factor networks underlying the cDC1 lineage, potentially enabling cDC1
   specification from a broad range of cell-types. This was also observed
   in other direct reprogramming processes including hemogenic, neuronal
   and hepatic ([260]16, [261]48). We envision to use our network of three
   transcription factors for the “forward programming” of iPSCs into cDC1
   or to directly convert other human somatic cells such as PBMCs,
   monocytes or even tumour cells to promote presentation of their own
   (neo)antigens. Thus, our study provides the foundation for the
   systematic comparison of hiDCs generated from multiple human cell-types
   to available in vitro sources of human cDC1-like cells.

   Overall, we demonstrated that PU.1, IRF8 and BATF3 impose cDC1 fate in
   human somatic cell types, providing a tractable system to study human
   cDC1 specification and function. We provided evidence that human
   antigen cross-presenting cells can be induced by direct reprogramming,
   supporting the generation of potent patient-specific cDC1s for
   immunotherapy.

Materials and Methods

Study Design

   This study aimed to investigate the single-cell transcriptional
   dynamics during cDC1 reprogramming of human fibroblasts, dissecting
   factors and pathways triggered during successful reprogramming to
   optimize the generation of human functional cDC1s. We also aimed to
   provide mechanistic insights into the chromatin engagement of the
   reprogramming factors to impose cDC1 fate. For each experiment, the
   number of biological replicates and statistical details are described
   in figure legends.

Viral transduction and reprogramming

   HEFs, HDFs and Clec9a-tdTomato MEFs were seeded at a density of 40,000
   cells per well and MSCs at a density of 50,000 cells per well on 0.1%
   gelatin coated 6-well plates. On the following day, cells were
   incubated overnight with either a ratio of 1:1 TetO-PIB and M2rtTA,
   SFFV-PIB-GFP or SFFV-GFP lentiviral particles in media supplemented
   with 8 μg/ml polybrene. Cells were transduced overnight twice in
   consecutive days and media replaced in between. After the second
   transduction, media was replaced by normal growth media (day 0). When
   using TetO-PIB, media was supplemented with Dox (1 μg/ml). Media was
   changed every 2-3 days for the duration of cultures. When stated, media
   was supplemented with LPS (100 ng/ml, Enzo), Poly I:C (25 μg/ml,
   InvivoGen) and R848 (3 μg/ml, InvivoGen) overnight. When stated,
   cytokines were added at day 2 and kept for the duration of cultures;
   cytokine concentrations used in this study as well as antibodies, cell
   lines and plasmids are listed in [262]Table S1. For reprogramming in
   xeno-free conditions, MSCs were cultured in X-Vivo 15 (Lonza) after
   transduction and media changed every 2-3 days for the duration of cell
   cultures.

Single-cell RNA sequencing

   HEFs, hiDCs at day 3, 6 and 9 (CD45^+HLA-DR^- and CD45^+HLA-DR^+),
   cDC1s, cDC2s and pDCs from peripheral blood (from 3 individual donors)
   were FACS sorted for scRNA-seq. Purified cells were loaded on a 10×
   Chromium (10× Genomics) according to manufacturer's protocol. scRNA-seq
   indexed libraries were prepared using Chromium Single Cell 3' v2 and v3
   Reagent Kit (10× Genomics) according to manufacturer's protocol. hiDCs
   at day 9 reprogrammed in the presence and absence of cytokines from
   HEFs and HDFs and CD45^+HLA-DR^+CD226^+ hiDCs were also profiled.
   Library quantification and quality assessment was determined using
   Agilent Bioanalyzer using the High Sensitivity DNA analysis kit
   (Agilent). Indexed libraries were pooled at equimolarity and sequenced
   on an Illumina NextSeq 500. Coverage of approximately 130,000 reads per
   single cell was obtained. Details regarding scRNA-seq data analysis can
   be found in [263]supplementary materials.

ChIP sequencing

   HDFs were transduced with TetO-PIB or vectors encoding individual
   factors (TetO-PU.1, TetO-IRF8, or TetO-BATF3) and M2rtTA. Transduced
   HDFs were expanded and ChIP was performed 48 hours after the addition
   of Dox using antibodies for human PU.1, human IRF8 and human BATF3
   ([264]table S1). Details regarding ChIP-seq and analysis can be found
   in [265]supplementary materials.

Human antigen cross-presentation assay

   HEFs, moDCs, magnetic-activated cell sorting (MACS)-enriched cDC1s (see
   [266]supplementary methods) and hiDCs at reprogramming day 8 were
   stimulated with LPS (3 ng/ml), Poly I:C (25 μg/ml) and R848 (3 ng/ml).
   After overnight stimulation, cells were washed in PBS containing 2% FBS
   and pulsed with 2 μl/ml of CMV protein (Miltenyi Biotec). After 3
   hours, cells were washed and co-cultured with MACS-enriched CD8^+ T
   cells isolated from CMV-seropositive donors. CMV positivity was
   verified by flow cytometry using a CMV Dextramer (Immudex). 1x10^5
   CD8^+ T cells and 5x10^4 DCs were co-cultured in 96-well plates in
   200μl of X-VIVO 15. After 24 h, T-cell activation was measured by
   quantifying IFN-γ levels in the supernatants using ELISA (BD).
   Absorbance was read at 490 nm in a GloMax Discover Microplate Reader
   (Promega).

Statistical analysis

   Comparisons among groups were performed using one-way ANOVA followed by
   Dunnett multiple comparisons test with Prism 9 software (GraphPad).
   P-values are shown when relevant (*p<0.05; **p<0.01, ***p<0.001,
   ****p<0.0001, ns - non-significant).

Supplementary Material

   Supplementary Data File S1
   [267]EMS145501-supplement-Supplementary_Data_File_S1.xlsx^ (2.7MB,
   xlsx)
   Supplementary Data File S2
   [268]EMS145501-supplement-Supplementary_Data_File_S2.xlsx^ (1MB, xlsx)
   Supplementary Data File S3
   [269]EMS145501-supplement-Supplementary_Data_File_S3.xlsx^ (554.6KB,
   xlsx)
   Supplementary Data File S4
   [270]EMS145501-supplement-Supplementary_Data_File_S4.xlsx^ (15.2MB,
   xlsx)
   Supplementary Data File S5
   [271]EMS145501-supplement-Supplementary_Data_File_S5.xlsx^ (79.4KB,
   xlsx)
   Supplementary Materials
   [272]EMS145501-supplement-Supplementary_Materials.pdf^ (4.2MB, pdf)
   Supplementary movie S1
   [273]Download video file^ (10.6MB, mp4)
   Table S1
   [274]EMS145501-supplement-Table_S1.xlsx^ (18.7KB, xlsx)

One Sentence Summary.

   Programming functional human cDC1s

Acknowledgments