PMCID
stringclasses 8
values | Title
stringclasses 8
values | FullText
stringclasses 8
values |
|---|---|---|
PMC2872605
|
Kinase-Dead BRAF and Oncogenic RAS Cooperate to Drive Tumor Progression through CRAF
|
We describe a mechanism of tumorigenesis mediated by kinase-dead BRAF in the presence of oncogenic RAS. We show that drugs that selectively inhibit BRAF drive RAS-dependent BRAF binding to CRAF, CRAF activation, and MEK–ERK signaling. This does not occur when oncogenic BRAF is inhibited, demonstrating that BRAF inhibition per se does not drive pathway activation; it only occurs when BRAF is inhibited in the presence of oncogenic RAS. Kinase-dead BRAF mimics the effects of the BRAF-selective drugs and kinase-dead Braf and oncogenic Ras cooperate to induce melanoma in mice. Our data reveal another paradigm of BRAF-mediated signaling that promotes tumor progression. They highlight the importance of understanding pathway signaling in clinical practice and of genotyping tumors prior to administering BRAF-selective drugs, to identify patients who are likely to respond and also to identify patients who may experience adverse effects.The RAS–ERK (extracellular-signal regulated protein kinase) MAPK (mitogen-activated protein kinase) signaling pathway regulates cell responses to environmental cues (Marshall, 1995) and plays an important role in human cancer (Gray-Schopfer et al., 2007). The pathway comprises the RAS small guanine-nucleotide binding protein and the protein kinases RAF, MEK (mitogen and extracellular-regulated protein kinase kinase), and ERK. RAS is attached to the inner face of the plasma membrane and is activated downstream of growth factor, cytokine, and hormone receptors. Active RAS recruits RAF to the membrane for activation through a complex process involving changes in phosphorylation and binding to other enzymes and scaffold proteins (Kolch, 2000). RAF phosphorylates and activates MEK, which phosphorylates and activates ERK. The complexity of this pathway is increased by the multiplicity of its components. There are three RAS (HRAS, NRAS, and KRAS), three RAF (ARAF, BRAF, and CRAF), two MEK (MEK1 and MEK2), and two ERK (ERK1 and ERK2) genes that encode proteins with nonredundant functions. Furthermore, the pathway is not linear. BRAF binds to and activates CRAF in a RAS-dependent manner that appears to require CRAF transphosphorylation by BRAF (Garnett et al., 2005; Rushworth et al., 2006; Weber et al., 2001), providing subtle pathway regulation that is not fully understood. ERK phosphorylates many substrates and the duration and intensity of its activity affects how cells respond to extracellular signals (Marshall, 1995). Thus, the pathway must be carefully controlled to ensure appropriate responses to environmental cues. In normal cells, outcomes include survival, proliferation, senescence, and differentiation, but in cancer the constitutive pathway activation favors proliferation and survival. RAS–ERK signaling is particularly important in melanoma. Somatic mutations occur in BRAF, NRAS, and KRAS in 43%, 20%, and 2% of melanomas respectively (www.sanger.ac.uk/genetics/CGP/cosmic/). The mutations in RAS trap it in a GTP-bound, active conformation and mostly involve glycine 12 (G12), glycine 13 (G13), and glutamine 61 (Q61). A glutamic acid substitution for the valine at position 600 (BRAF) accounts for over 90% of the mutations in BRAF in cancer. However, over 100 other rare mutations have been described, most of which cluster to the glycine-rich loop and activation segment in the kinase domain. These regions normally trap BRAF in an inactive conformation by forming an atypical intramolecular interaction, and it is thought that the mutations disrupt this interaction, thereby allowing the active conformation to prevail (Wan et al., 2004). Functional studies have shown that most of the mutations in BRAF are activating and enhance its ability to directly phosphorylate MEK (Wan et al., 2004; Garnett and Marais, 2004). Curiously however, some mutants have impaired activity and although they cannot directly phosphorylate MEK, they appear to retain sufficient activity to bind to and transphosphorylate and activate CRAF in a RAS-independent manner (Garnett et al., 2005), allowing these mutants to activate the pathway indirectly through CRAF. More puzzling are mutations that occur at aspartic acid 594 (D594). The carboxy oxygen of this highly conserved residue (the “D” of the DFG motif) plays a critical role in chelating Mg and stabilizing ATP binding in the catalytic site (Johnson et al., 1998). As in other kinases, mutation of this residue causes inactivation and thus cancer mutants such as BRAF cannot phosphorylate MEK, activate CRAF, or stimulate cell signaling (Ikenoue et al., 2003; Wan et al., 2004). These mutants therefore appear catalytically and biologically inactive and yet 34 have been found in human cancer (www.sanger.ac.uk/genetics/CGP/cosmic/). Furthermore, while BRAF mutations (over 10,000 described) occur in a mutually exclusive manner with RAS mutations, four of the 34 kinase-dead mutants are coincident with RAS mutations, a highly significant enrichment (p < 10; Fisher's Exact Test) that suggests functional interaction. It has been shown that BRAF is 500-fold activated, can stimulates constitutive MEK–ERK signaling in cells (Gray-Schopfer et al., 2007) and induce melanoma in mice (Dankort et al., 2009; Dhomen et al., 2009), showing that it can be a founder mutation in melanoma. Importantly, BRAF inhibition blocks melanoma cell proliferation and induces apoptosis in vitro and blocks melanoma xenograft growth in vivo (see Gray-Schopfer et al., 2007). These data validate BRAF as a driver of melanomagenesis and as a therapeutic target in melanoma, so drugs to target this pathway have been developed. The first to be tested clinically were the multi-kinase inhibitor sorafenib and the MEK inhibitor PD184352 (CI1040). Disappointingly, both failed to produce objective responses in patients, either because they were not sufficiently potent, or because they caused unacceptable toxicity (Halilovic and Solit, 2008). Recently, more potent and selective BRAF inhibitors have been described. For example, the triarylimidazole SB590885 and the difluorophenylsulfonamine PLX4720 display excellent selectivity for BRAF in vitro and preferentially inhibit BRAF mutant cancer cell proliferation (King et al., 2006; Tsai et al., 2008). More importantly, BRAF-selective drugs have recently entered the clinic and are producing excellent responses in patients with BRAF mutant melanoma (Flaherty et al., 2009; Schwartz et al., 2009). The aim of this study was to better understand the responses that melanoma cells make to BRAF-selective inhibitors and thereby to provide a molecular basis for the design of clinical trials using BRAF drugs. We also wished to examine if kinase-dead BRAF and oncogenic RAS functionally interact in vivo. We selected four drugs for our studies (Figures S1A–S1D). Sorafenib is a class II (inactive conformation binder) drug (Wan et al., 2004) that inhibits BRAF at 40 nM, CRAF at 13 nM, and several other kinases in the low nM range (Wilhelm et al., 2004). It is the least-selective drug that we used. PLX4720 is a class I (active conformation binder) inhibitor that is highly selective and inhibits BRAF at 13 nM (Tsai et al., 2008). 885-A (Figure S1C) is a close analog of the class I inhibitor SB590885 (King et al., 2006) that is also highly selective for BRAF. It inhibits BRAF at 2 nM (Figure S1E), is ineffective against a panel of 64 other protein kinases (Table S1), and preferentially blocks BRAF mutant cancer cell proliferation (Figure S1F). Finally, we also used the potent and selective MEK inhibitor PD184352 (Sebolt-Leopold et al., 1999). As expected, all four drugs blocked ERK activity in BRAF mutant A375 melanoma cells (Figure 1A; see Table S2). Similarly, all four drugs inhibited ERK in SkMel24, SkMel28, D25, and WM266.4 cells, another four lines that express mutant BRAF (Figure S1G). We also tested the drugs in D04, MM415, MM485, and WM852 NRAS mutant cells (Table S2). As expected, PD184352 and sorafenib inhibited ERK in all of these lines (Figure 1A). Surprisingly, however, PLX4720 and 885-A caused an unexpected increase in ERK activity in the NRAS mutant cells (Figure 1A). NRAS or CRAF depletion by RNA interference (RNAi) blocked MEK/ERK activation by PLX4720 and 885-A in NRAS mutant cells (Figure 1B and 1C) and we show that 885-A activated CRAF in these cells (Figure 1D). We previously reported that oncogenic RAS requires CRAF but not BRAF to activate MEK (Dumaz et al., 2006) and consistent with this, BRAF is inactive in NRAS mutant cells (Figure 1E). These data therefore present an intriguing paradox. BRAF is not active and is not required for MEK/ERK activation in RAS mutant cells. Nevertheless, BRAF inhibitors hyperactivate CRAF and MEK in these cells, so we studied the underlying mechanism(s). Wild-type BRAF binds to CRAF in a RAS-dependent manner and although this binding is weak, it leads to CRAF activation (Garnett et al., 2005). Since RAS and CRAF are required for ERK activation by PLX4720 and 885-A, we investigated if these drugs induce BRAF binding to CRAF. Endogenous BRAF was immunoprecipitated from melanoma cells and western blotted for endogenous CRAF. We show that CRAF did not bind to BRAF in untreated or PD184352 treated WM852, D04, MM415, or MM485 cells (Figure 2A), demonstrating that MEK inhibition does not induce binding. In contrast, sorafenib and 885-A induced strong binding of BRAF to CRAF in all four lines (Figure 2A). We also performed the experiment in the inverse manner, immunoprecipitating CRAF and showing that BRAF binding was strongly induced by sorafenib and 885-A (Figure 2A). Curiously, PLX4720 did not appear to induce BRAF binding to CRAF, but previous studies have shown that ERK phosphorylates BRAF in a negative-feedback loop that destabilizes its binding to CRAF (Rushworth et al., 2006). We show that PD184352 stabilizes BRAF binding to CRAF in the presence of PLX4720 (Figure 2B), demonstrating that PLX4720 does induce binding, albeit less strongly than the other drugs. In addition to inducing BRAF binding to CRAF in NRAS mutant cells, 885-A and sorafenib also induce this binding in WM1791c melanoma cells and in SW620 and HCT116 colorectal carcinoma cells (Figure 2C), all of which express mutant KRAS (Table S2). Importantly, no strong binding of BRAF to CRAF was seen in A375 cells even in the presence of PD184352 and the drugs did not induce strong BRAF binding to CRAF in two other BRAF mutant melanoma cell lines (Figure 2D and Figure S2). Thus, sorafenib, 885-A and PLX4720 all induced BRAF binding to CRAF in NRAS or KRAS mutant cells, but not in BRAF mutant cells, showing that BRAF inhibition per se did not induce this binding; it only occurred when BRAF was inhibited in the presence of oncogenic RAS. To confirm the essential role of RAS, we show that a CRAF mutant (CRAF) that cannot bind to RAS (Fabian et al., 1994) did not bind to BRAF (Figure 3A and Figure S3A) and the corresponding mutant of BRAF (BRAF) did not bind to CRAF (Figure 3B and see Figure S3B). We also prepared membrane/cytosol fractionations of RAS mutant cells and show that under normal conditions over 40% of CRAF is in the membrane, whereas BRAF is largely cytosolic (Figure 3C). Notably, 885-A treatment leads to strong recruitment of BRAF to the membrane fraction, whereas CRAF is only weakly affected (Figure 3C). We also show that under normal conditions, EGF did not induce BRAF binding to CRAF in PMWK cells, a line that is wild-type for BRAF and RAS (Table S2). However, in the presence of 885-A, EGF induced robust binding of BRAF to CRAF in PMWK cells and this resulted in sustained pathway activation (Figure 3D). This shows that BRAF binding to CRAF is induced in the presence of both oncogenic RAS and activated wild-type RAS. We note that sorafenib and 885-A induce a mobility shift in BRAF in SDS-gels (Figure 2A). BRAF also undergoes a mobility shift in PLX4720 treated cells in the presence of PD184352 (Figure 2B). This mobility shift is reduced when immunoprecipitated BRAF is treated with calf intestinal alkaline phosphatase (CIP; Figure 3E) and PD184352 pretreatment reduced, but did not ablate the magnitude of the shift induced by 885-A (Figure 3F). Importantly, in vitro CIP treatment and cell pretreatment with PD184352 did not prevent BRAF binding to CRAF (Figures 3E and 3F). Together, these data suggest that the BRAF bound to CRAF is hyperphosphorylated through MEK–ERK-dependent and MEK–ERK-independent mechanisms, but that this phosphorylation is not required for BRAF binding to CRAF. To test directly if BRAF binding to CRAF is driven by 885-A binding to BRAF, we mutated the so-called “gatekeeper threonine” (T529) of BRAF to asparagine (T529N). Since BRAF is not active in RAS mutant melanoma cells (Figure 1E), we measured BRAF activity using transient expression in COS cells (Wan et al., 2004). The results show that BRAF is still activated by HRAS, NRAS and KRAS (Figure 4A and Figure S4A). Importantly, BRAF is ∼170-fold less sensitive to 885-A than wild-type BRAF (17 nM versus 2869 nM; Figure 4B) and 885-A did not stimulate its binding to CRAF (Figure 4C), proving that drug binding to BRAF drives BRAF binding to CRAF. Next, we expressed a kinase-dead version of BRAF (BRAF) in D04 cells and show that it forms a constitutive complex with CRAF (Figure 4D) and that it activates MEK constitutively (Figure 4E, compare lanes 1, 4, and 7). Notably, 885-A does not further enhance MEK activation driven by BRAF (Figure 4E, compare lanes 4, 6 to 7, 9), presumably because it cannot further inhibit this already inactive kinase. Two other kinase-dead BRAF mutants, the classical catalytic lysine mutant (BRAF), and BRAF, a mutant found in human cancer (Wan et al., 2004), also activate MEK in D04 cells (Figure 4F). Thus, it is BRAF inhibition and not drug binding that drives BRAF binding to CRAF. This experiment also shows that MEK activation driven by kinase-dead BRAF is inhibited by sorafenib (Figures 4E and 4F). Indeed, cell responses to sorafenib appear to be paradoxical. We show that although sorafenib inhibits ERK (Figure 1A), it induces BRAF binding to CRAF (Figure 2A), CRAF activation (Figure 4G) and CRAF phosphorylation on S338 (Figure 4G, inset), a critical event in CRAF activation (Mason et al., 1999). To test directly the role of CRAF in cells when BRAF is inhibited, we mutated its gatekeeper threonine to asparagine (CRAF). Notably, CRAF still binds to BRAF in sorafenib and 885-A treated cells (Figure 4H), demonstrating that drug binding to CRAF is not required for BRAF binding to CRAF. More importantly, in the presence of CRAF, sorafenib activates rather than inhibits the pathway (Figure 4H, compare lanes 3 and 7). We therefore posit that sorafenib induces paradoxical activation of CRAF because it inhibits BRAF and drives CRAF activation, but simultaneously binds to and inhibits CRAF. In agreement with this model, we show that two other pan-RAF inhibitors, ZM336372 and RAF265 also induce BRAF binding to CRAF, but without activating ERK (see Figure S4B). Our data establish that inhibition of BRAF in the presence of oncogenic RAS hyperactivates CRAF, MEK, and ERK. To investigate the consequences of this in vivo, we used conditionally targeted alleles of oncogenic Kras (Kras) and kinase-dead Braf (Braf) in transgenic mice. These alleles use Cre-recombinase/LoxP-Stop-LoxP (LSL) technology to regulate inducible expression of mutant proteins from the endogenous mouse genes to ensure normal levels of protein expression. The Kras allele has been described (Jackson et al., 2001), and we recently developed the Braf allele. Briefly, exon 15 of endogenous Braf was targeted to mutate D594 to alanine (D594A; see Figure 5A). To prevent expression of Braf in all cells, an LSL cassette was inserted between exon 14 and the mutated exon 15. This contains a minigene for exons 15–18 of Braf, a transcription terminator and a Neo selection marker to ensure that only Braf is expressed. Removal of the LSL cassette by Cre-recombinase reveals the mutated exon 15 and Braf is expressed. These mice were crossed to Tyr::CreERT2 mice (Yajima et al., 2006), in which the tyrosinase promoter is used to express tamoxifen-activated Cre-recombinase (CreERT2) in the melanocytes. Since CreERT2 is activated by tamoxifen, this approach provides exquisite spatial and temporal control over Kras and Braf expression. Kras, Braf, and Tyr::CreERT2 mice were crossed to generate Kras;Tyr::CreERT2, Braf;Tyr::CreERT2, or Kras;Braf;Tyr::CreERT2 mice. In all cases, the conditionally targeted alleles were balanced over a corresponding wild-type allele. Mice were treated with tamoxifen at 2–3 months of age to induce mutant protein expression. We have recently shown that in this model, Braf induces skin hyperpigmentation, nevus formation, and melanoma (Dhomen et al., 2009). In contrast, Braf did not induce skin hyperpigmentation, nevi (data not shown) or tumors (Figure 5C). Kras induced weak tail darkening after 5–6 months (Figure 5B) but did not induce either nevi (data not shown) or tumors (Figure 5C). However, when Braf and Kras were combined, they induced a conspicuous skin phenotype. Within 2–3 months the ears (data not shown), tails (Figure 5B), and paws (Figure 5D) darkened visibly. The mice did not develop nevi, but within 6 months, they all developed large, rapidly growing oligo-pigmented tumors (Figures 5C and 5E). The tumors displayed evidence of ulceration (Figure 5F) and were composed largely of spindle cells that exhibit features of malignancy, including cellular atypia, nuclear pleomorphism, and conspicuous nucleoli (Figure 5G). They were highly proliferative as evidenced by large numbers of mitotic figures in the superficial and deep aspects of the lesions (∼6 mitosis/10HPF; Figure 5H) and positive staining for Ki67 throughout (Figure 5I). The tumors were strongly and diffusely positive for S100 (Figure 6A) and expressed the melanocyte markers tyrosinase, Dct, Pax3, and silver (Figure 6B), consistent with a diagnosis of melanoma. Genomic DNA analysis of the tumors and cell lines derived from them confirmed that Braf had been recombined to Braf (Figure 6C). However, for technical reasons we could not detect Kras recombination (data not shown), so used RT-PCR to amplify and sequence Kras mRNA. We show that only wild-type Kras is expressed in the kidneys, whereas the tumors expressed both wild-type Kras and Kras (Figure 6D). Importantly, we show constitutive binding of Braf to Craf in cells from the Kras/Braf tumors (Figure 6E). As a control, we used cells from melanoma induced by Kras overexpression. Briefly, when Kras was overexpressed in melanocytes in mice using the β-actin promoter (β-actin:LSL:Kras; Meuwissen et al., 2001), it induced rapid onset melanoma (median time to onset 2 months, 100% penetrance within 3 months) in the absence of Braf (manuscript submitted). Importantly, in cells from these tumors, Braf does not bind to Craf (Figure 6E). Thus, it is only kinase-dead Braf and not wild-type Braf that binds to Craf in the presence of oncogenic Kras. In this study, we show that inhibition of BRAF by chemical or genetic means in the presence of oncogenic or growth-factor activated RAS induces BRAF binding to CRAF, leading to CRAF hyperactivation and consequently elevated MEK and ERK signaling. The mechanism we describe is another paradigm of RAF activation downstream of RAS and based on our findings, we propose the following mechanism by which this occurs. We posit that in RAS mutant cells, BRAF maintains itself in an inactive conformation through its own kinase activity, either through auto-phosphorylation, or by phosphorylating a partner protein that then keeps it inactive (Figure 7A). We are currently using mass-spectrometry and mutagenic approaches to elucidate the underlying mechanism. We propose that when BRAF is inhibited, it escapes this auto-inhibited state and is recruited to the plasma membrane by RAS, where it forms a stable complex with CRAF. Critically, we posit that because it is inhibited, BRAF does not directly phosphorylate MEK, but rather it acts as a scaffold whose function is to enhance CRAF activation, thereby allowing CRAF to hyperactivate the pathway (Figure 7B). We do not know the stoichiometry of the components in these complexes, but since BRAF and CRAF must both bind to RAS for complex formation, it seems likely that at least two RAS proteins are needed to stimulate formation of the complex (Figure 7B). It is unclear why PLX4720 only induces weak binding of BRAF to CRAF, but this may stem from its unique property of displacing the α-C helix of BRAF when it binds (Tsai et al., 2008) and suggests that this helix is important for BRAF binding to CRAF, something that will only be resolved when the BRAF:CRAF crystal structure is solved. We have attempted to identify other proteins that may be required to stabilize the BRAF–CRAF complexes. Our unpublished mutagenesis data suggests that 14-3-3 is required to stabilize these drug-induced complexes (data not shown) and this is consistent with previous observations demonstrating that 14-3-3 mediates BRAF binding to CRAF (Garnett et al., 2005; Rushworth et al., 2006). Although this appears to contradict our observation that dephosphorylation does not disrupt the complex, because 14-3-3 binds to BRAF and CRAF in a phosphorylation-dependent manner, we presume that 14-3-3 protects these sites from dephosphorylation. We have also used RNAi to examine the potential role of other proteins implicated in BRAF-CRAF complex formation or pathway activation, including the scaffold proteins KSR, Sprouty2 and RKTG and the small G protein RHEB, but our preliminary results have not revealed obvious roles for these proteins. Our studies have parallels to the recently described heterodimers between DRAF and KSR in Drosophila (Rajakulendran et al., 2009). Notably, flies have only one RAF isoform and it appears to be an ortholog of BRAF rather than ARAF or CRAF. Our inability to demonstrate an obvious role for KSR in mediating BRAF binding to CRAF or CRAF activation by BRAF suggests that the mechanism underlying dimerization here may be different from those described in flies, but clearly additional studies are required to investigate further the role of scaffold proteins in mediating the phenomena we report. In contrast to the BRAF-selective inhibitors, the pan-RAF inhibitors appear to induce paradoxical activation of CRAF. They induce BRAF binding to CRAF and CRAF activation, but do not activate MEK–ERK signaling. We posit that this is because these agents target both BRAF and CRAF. Thus, although their inhibition of BRAF will stimulate CRAF activation, they will simultaneously inhibit CRAF (Figure 7C). This model is supported by our observation that CRAF converts sorafenib from a pathway inhibitor to a pathway activator and we argue that the paradoxical activation of CRAF by these inhibitors is mediated by BRAF, rather than disrupted feedback inhibition as previously suggested (Hall-Jackson et al., 1999). Recently, paradoxical activation of PKB/AKT and PKCɛ was also described (Cameron et al., 2009; Okuzumi et al., 2009). While ATP-competitive inhibition can block kinase function, they do not block the upstream events that activate the target kinase. For instance, PKB/AKT inhibitors block the function of this kinase, but occupation of the ATP-pocket by these inhibitors was sufficient to induce the priming phosphorylation usually required for its full activation (Okuzumi et al., 2009). Inhibitor binding to PKCɛ has been shown to have a similar effect (Cameron et al., 2009). Importantly, the paradoxical activation of PKB/AKT and PKCɛ did not result in pathway activation because of the continued presence of the inhibitors (Frye and Johnson, 2009). In contrast, although BRAF inhibitors also block BRAF kinase activity, this relieves auto-inhibition and results in BRAF hyperphosphorylation, BRAF binding to CRAF, pathway activation and oncogenesis, all presumably because BRAF can heterodimerize with CRAF. Our study also highlights the critical difference between BRAF-selective and pan-RAF drugs. Whereas BRAF-selective drugs cause pathway activation in a RAS-dependent manner, this does not occur with pan-RAF drugs. Our results provide important insight into the genetics of human cancer. Excluding V600 mutants, D594 mutants are the third most common in BRAF in cancer (34 out of 443 cases or ∼7.7%; www.sanger.ac.uk/genetics/CGP/cosmic/). Furthermore, as mentioned in the Introduction, while BRAF and RAS mutations are generally mutually exclusive, 4 of the 34 (11.8%) tumors with D594 mutations also have mutations in RAS. This is a highly significant enrichment for the coincidence of these mutations (p < 10; Fisher's Exact Test) and suggests a functional interaction. We now provide strong circumstantial evidence of such an interaction using transgenic mice. By themselves, Braf and Kras do not induce melanoma, but they cooperate to induce rapid onset melanoma. This highly significant result (p < 0.0002) provides a rational explanation for the coincidence of these mutations in human cancer. Furthermore, we show that the BRAF inhibitors also hyperactivate this pathway in growth factor stimulated cells, providing an explanation of why kinase dead BRAF mutations are not always coincident with RAS mutations; presumably in some tumors the cooperating mutation is upstream of RAS. Our results also suggest several potential mechanisms by which resistance to RAF targeting drugs could develop in patients. BRAF mutant tumors could become resistant to BRAF-selective drugs, if they acquire a mutation in RAS or an upstream component that activates RAS, or if the drugs select a population of cells harboring pre-existing mutations in RAS. Theoretically this would cause BRAF-mediated CRAF activation, which may not only induce resistance, but could potentially promote tumor growth. In line with this, increased expression of CRAF can mediate acquired resistance to pan-RAF drugs in BRAF mutant cancer cells in vitro (Montagut et al., 2008), establishing that CRAF can mediate resistance under some circumstances. Our in vitro studies also suggest that a potential mechanism of resistance in patients with RAS mutant tumors being treated with pan-RAF drugs is acquisition (or selection for cells with pre-existing mutations) of a CRAF mutation such as a gatekeeper mutant that prevents drug binding. Again this would potentially result in BRAF-mediated activation of CRAF (Figure 7D) and possibly accelerated tumor growth. Although our studies are restricted to cell lines and transgenic mice, they do have important immediate clinical implications. They strongly argue that BRAF-selective inhibitors should not be administered to patients with RAS mutant tumors, because long-term use could accelerate tumor growth. Intriguingly, 10%–15% of patients treated with BRAF-selective drugs develop squamous cell carcinoma (SCC)(Flaherty et al., 2009; Schwartz et al., 2009). Although MEK–ERK signaling has not yet been implicated in this response, 22% of SCCs harbour oncogenic mutations in RAS (9% HRAS, 8% NRAS, 5% KRAS: www.sanger.ac.uk/genetics/CGP/cosmic/), raising the intriguing possibility that the BRAF-selective drugs act as tumor promoters in premalignant skin cells harboring existing mutations in RAS and/or activation of upstream components that activate RAS. While sorafenib is equipotent for wild-type and BRAF (Wilhelm et al., 2004), the BRAF inhibitors we used are approximately 10-fold more active against BRAF (King et al., 2006; Tsai et al., 2008). Nevertheless, our data establish that they target wild-type BRAF in RAS mutant cells. The problem of mutant v.s. wild-type protein specificity is likely to be difficult to resolve, because whereas full inhibition of BRAF may be necessary for clinical response in BRAF mutant tumors, activation of only a small proportion of wild-type BRAF could be sufficient to activate the pathway in RAS mutant cells. Thus, to achieve efficacy against BRAF but avoid activation of wild-type BRAF in RAS mutant cells, the drugs will need to be exquisitely selective for the mutant protein. Alternatively, pan-RAF drugs may be effective because they will target both BRAF and CRAF activated by BRAF in RAS mutant tumors. Furthermore, our data suggest that CRAF or MEK selective drugs should be used in RAS mutant tumors, because they do not induce BRAF-CRAF complexes and will not activate the pathway if the tumors acquire mutations such as CRAF that block drug binding. Perhaps RAF and MEK inhibitors should be combined to provide the best responses and prevent emergence of resistance, but these issues need to be balanced against the urgency of the clinical problem being addressed. In summary, we show that inhibition of BRAF in RAS mutant cancer cells leads to MEK hyperactivation through CRAF. We have elucidated another mechanism by which BRAF activates MEK–ERK signaling, not only to drive tumorigenesis and tumor progression, but also potentially to allow development of de novo or acquired resistance to RAF-targeted therapies. Clearly, BRAF is a remarkably versatile oncogene that can promote MEK–ERK activation and tumor progression through several mechanisms and these will require different therapeutic strategies for effective disease management. Notably, many of the mutations that occur in other kinases in cancer are also predicted to cause inactivation (www.sanger.ac.uk/genetics/CGP/cosmic/). Our data raise the possibility that these could also act as idiosyncratic gain-of-function mutations that drive tumorigenesis. This study also raises important clinical questions and highlights the importance of fully understanding how signaling networks function to fully comprehend how patients may respond to targeted drugs. They also highlight the importance of genetic screening for patients, not only to identify those who are likely to respond, but to exclude those who could experience adverse effects and thereby ensure successful implementation of personalized medicine. Expression vectors for epitope-tagged BRAF and CRAF have been described (Wan et al., 2004). For western blotting the following antibodies were used: rabbit anti-ppMEK1/2 and mouse anti-myc 9B11 (Cell Signaling Technology); mouse anti-NRAS (C-20), rabbit anti-ERK2 (C-14), rabbit anti-ARAF (C-20), mouse anti-BRAF (F-7) (Santa Cruz Biotechnology); mouse anti-Tubulin, and mouse anti-ppERK1/2 (Sigma); mouse anti-CRAF (for western blotting) (BD Transduction Laboratories). For immunoprecipitation, the following antibodies were used: rabbit anti-myc (Abcam); rabbit anti-CRAF (C-20;Santa Cruz Biotechnology); mouse anti-BRAF (F-7) (Ab from Santa Cruz Biotechnology). Calf intestinal phosphatase (CIP) was from New England Biolabs (NEB). PD184352, sorafenib and PLX4720 were synthesized in-house; 885-A was synthesized by Evotec AG (Abingdon, UK). All drugs were prepared in DMSO. Synthetic routes are available on request. Human cell lines were cultured in DMEM (A375, WM852, HCT116, SW620, and PMWK) or RPMI (D04, MM485, MM415, and WM1791c) supplemented with 10% fetal bovine serum. For protein depletion, 3 × 10 D04 cells were transfected with 5nM CRAF (5′-AAGCACGCTTAGATTGGAATA-3′) or NRAS (5′-CATGGCACTGTACTCTTCTCG-3′) specific, or scrambled control (5′-AAACCGTCGATTTCACCCGGG-3′) siRNA using INTERFERin as recommended by the manufacturer (Polyplus Transfection SA). For transient expression studies, D04 cells were transfected using the Amaxa Nucleofector System as recommended by the manufacturer (Lonza). COS-7 cells were propagated, transfected, and extracted as described (Wan et al., 2004). For generation of stable lines, D04 cells were transfected with pMCEF-FLAG-CRAF or pMCEF-FLAG-CRAF using Effectene as recommended by the manufacturer (Invitrogen) and selected in G418 (1 mg/ml). Cell lysates were prepared with NP40 buffer as described (Wan et al., 2004). For immunoprecipitation, lysates were incubated with 2 μg BRAF F-7, 5 μg CRAF C-20 or 2 μg rabbit anti-myc antibodies, captured on Protein G sepharose 4B beads (Sigma) and analyzed by western blotting using standard protocols. Specific bands were detected using fluorescent-labeled secondary antibodies (Invitrogen; Li-COR Biosciences) and analyzed using an Odyssey Infrared Scanner (Li-COR Biosciences). For CIP treatment, immunoprecipitates were washed twice with NP40 lysis buffer, once in CIP buffer (50 mM Tris-Cl [pH 7.5], 150 mM NaCl, 10 mM MgCl2, and 1 mM EDTA), and incubated with CIP with or without 0.2 mM Na3VO4 and 7 mM EDTA. The immunoprecipitates were washed in CIP buffer and western blotted. Coupled RAF kinase assays were performed with immunoprecipitated CRAF or BRAF as described (Wan et al., 2004). Membrane fractionation was as described (Garnett et al., 2005). Experiments were performed under Home Office license authority in accordance with United Kingdom Coordinating Committee on Cancer Research Guidelines (Workman et al., 1988) and with local Ethics Committee approval. To activate CreERT2, mice were treated with four doses (10mg each) of topically applied tamoxifen as described (Dhomen et al., 2009). Genotyping was performed by PCR. Braf and Braf was analyzed as described for Braf and Braf respectively and Tyr::CreERT2 was analyzed as described (Dhomen et al., 2009). Kras was analyzed using primers 5′-CGCAGACTGTAGAGCAGCG-3′ and 5′-CCATGGCTTGAGTAAGTCTGC-3′. For expression analysis, RNA was prepared (QIAGEN RNEasy, QIAGEN) and first-strand cDNA synthesis was performed with 500ng total RNA and random hexanucleotides (Random Primers, Invitrogen). Specific genes were amplified under linear conditions for analysis as described (Dhomen et al., 2009). For Kras cDNA sequencing, a 238 bp fragment of Kras cDNA was PCR amplified using primers 5′-GGCGGCAGCGCTGTGGCGGCG-3′ and 5′-CGTAGGGTCATACTCATCCAC-3′ and sequenced using automated dideoxy sequencing. For immunohistochemistry (IHC), tissues were fixed and analyzed as described (Dhomen et al., 2009). Positive (a well characterized sample of mouse melanoma) and negative (omission of the primary antibody and substitution with preimmune serum) controls were included in each slide run. Immunohistochemical staining was analyzed by two of the authors on a multi-headed microscope. Tumor cell lines were established by mechanically dissociating tumors in DMEM/20%FCS/Primocin (0.1mg/ml - InvivoGen) and clonal lines were selected by limiting dilution. Extended Experimental ProceduresReagentsFor western blotting the following antibodies were used: rabbit anti-ppMEK1/2 and mouse anti-myc 9B11 (Cell Signaling Technology); mouse anti-NRAS (C-20), rabbit anti-ERK2 (C-14), rabbit anti-ARAF (C-20), mouse anti-BRAF (F-7) (Santa Cruz Biotechnology); mouse anti-Tubulin, and mouse anti-ppERK1/2 (Sigma); mouse anti-CRAF (for western blotting) (BD Transduction Laboratories). For immunoprecipitation, the following antibodies were used: rabbit anti-myc (Abcam); rabbit anti-CRAF (C-20; Santa Cruz Biotechnology); mouse anti-BRAF (F-7) (Santa Cruz Biotechnology). Calf intestinal phosphatase (CIP) was from New England Biolabs (NEB). PD184352, sorafenib and PLX4720 were synthesized in-house; 885-A was synthesized by Evotec AG (Abingdon, UK). All drugs were prepared in DMSO. Synthetic routes are available on request.Expression ConstructsThe expression vectors for wild-type human CRAF and wild-type human BRAF, pEFm/CRAF and pEFm/BRAF respectively have been described (Marais et al., 1995). Briefly, the vector backbone is pUC19 and the elongation factor 1α (EF1α) promoter is used to drive exogenous protein expression. The vector includes the first intron from human EF1α to assist mRNA processing during expression. The β-globin 5′ and 3′ untranslated regions (UTRs) are used to provide a strong translation start site (5′ UTR), and also to provide mRNA stability and a poly adenylation signal (3′ UTR). The vector introduces an amino-terminal myc-epitope tag (EQKLISEEDL) onto the exogenously expressed protein. The BRAF coding region includes the alternatively spliced exons 1 and 2 but not exons 8b or 10a and various modifications were introduced to provide additional restriction sites (without changing the amino acid sequence) and alterations to the 3′-UTR to allow easier manipulation of this construct. Standard PCR-directed mutagenesis approaches were used to generate the various mutations used in the study and all mutations were verified by automated dideoxy sequencing. The expression vector pMCEF/FLAG/CRAF uses the same expression cassette, but the backbone also possesses a neo resistance cassette to facilitate selection in the presence of G418. In addition, a version of this vector was used that incorporates a FLAG (DYKDDDKGS), rather than a myc-epitope tag.Cell CultureHuman cell lines were cultured in DMEM (A375, WM852, HCT116, SW620, PMWK, SKMel24, SKMel28, D25) or RPMI (D04, MM485, MM415, WM1791c) supplemented with 10% fetal bovine serum. African green monkey kidney COS-7 cells were cultured in DMEM supplemented with 10% fetal bovine serum. All cell lines were incubated at 37°C and 10% CO2. For inhibitor treatment, the drugs were dissolved in DMSO and added to the medium for 2-5 hr. When two compounds were used, the first was added 30 min prior to the second. For cell growth assays, cells were seeded in 96-well plates and treated with drug in quadruplicate in a 10-point titration assay for 5 days. The amount of growth (% DMSO controls) was determined using sulphorhodamine B reagent (Monks et al., 1991) as follows. 1,000-10,000 cells (depending on cell type) were plated into 96-well plates in 100 μL medium. After 24 hr, compounds prepared in DMSO (10mM stocks) were serially diluted in culture medium at 2× the final required concentrations and 100μL were added to the cells to nine final concentrations of 0.005 μM-100 μM. After a further 5 days, the cells are fixed in trichloroacteic acid (10%), and stained with sulforhodamine-B (0.1%). After rinsing, the bound stain was disolved using 100 μL 10 mM Tris (pH 8.0) and the absorbance at 540 nm determined. The data were analyzed by nonlinear regression to a four-parameter logistic equation (Graphpad Prism, Graphpad Software Inc., San Diego, CA, USA) and the GI50 value determined.siRNA Transfections3 × 10 D04 cells per 35 mm diameter well were seeded in 2ml growth medium the day before transfection. The cells were either mock-transfected or transfected with 6nM CRAF-specific (5′-AAGCACGCTTAGATTGGAATA-3′) or NRAS-specific (5′-CATGGCACTGTACTCTTCTCG-3′) siRNA using INTERFERin as recommended by the manufacturer (Polyplus Transfection SA). Briefly, 0.6 μl of 20 μM siRNA and 6 μl of INTERFERin were combined in a total of 200 μl serum free medium in RNase-free tubes. The mix was vortexed for 10 s and incubated for 5-10 min before adding the complexes dropwise to the cells. The cells were serum-starved the day after transfecting and extracts were prepared 48 hr after transfection.DNA TransfectionsFor transient protein expression in D04 cells, Lonza Nucleofector Technology (Lonza, Cologne AG) was used. 2 μg of DNA was mixed with 1x10 cells resuspended in 100 μl of Nucleofection Solution V in an Amaxa-certified cuvette and transfected using program T030. The cells were re-plated into 35mm diameter tissue culture wells and incubated for 48 hr before preparation of cell extracts.For generation of stable lines, D04 cells were transfected using Effectene (Invitrogen) and selected in G418. 3-4x10 cells were plated in 35 mm diameter wells and incubated overnight. 0.4 μg of DNA diluted into 100 μl of DNA condensation buffer (EC) and 3.2 μl enhancer were mixed vigorously and incubated for 2-5 min. 10 μl Effectene reagent was added and the mixture was incubated for another 5-10 min. The cells were washed with 2ml PBS and 1.6ml fresh serum containing medium was added. The DNA complexes were diluted with 600 μl of culture medium and the mixture added to the cells drop-wise. After six hours, the medium was replaced with 2ml of fresh growth medium. After 48 hr, the cells were replated into several 10cm dishes in a 10-fold dilution series and incubated in G418 (1mg/ml) for selection. The medium was refreshed weekly and after 2-3 weeks, single colonies were selected and expanded.For transient expression in COS-7 cells, Lipofectamine (Invitrogen) was used. 2x10 cells were plated into 35mm diameter wells and incubated overnight. 75 to 200 ng of expression plasmid (depending on construct) was mixed with empty vector to a total of 700 ng DNA in 16μl PBS. Typically, 3 μl of Lipofectamine in 13 μl of PBS was added to the DNA on the surface of a bacterial plate and incubated (Lipofectamine is inactivated by binding to polypropylene) for 15 min at room temperature. The cells were washed twice with 1ml serum-free DMEM, and then overlaid with 800 μl of serum free DMEM. 200 μl of serum free DMEM was added to the DNA:Lipofectamine mix, and the total volume was added to the cells. After six hours, the complexes were removed and replaced with 2ml of normal culture medium. Cell extracts were prepared two days following transfection.Preparation of Cell LysatesCulture medium was aspirated from cells and cells were placed on ice and washed three times in ice-cold PBS. Depending on the assays, the cells were scraped into 50-200 μl Nonidet P40 (NP40) extraction buffer (Table S3) and incubated on ice for five minutes. The cells were sheared by passing through a pipette tip several times and the samples were centrifuged at 20,000 × g for 5 min at 4°C and the soluble fraction was harvested.RAF CoimmunoprecipitationsImmunoprecipitations were performed in 300 μl cell lysates from one 35mm diameter well for endogenous protein or from 2-3 wells for transfected protein. Endogenous BRAF or CRAF were immunoprecipitated with 2μg BRAF F-7 or 5μg CRAF C-20 respectively and myc-tagged BRAF and CRAF with 2 μg rabbit anti-myc antibody. The antibody-protein complex was captured using 20 μl of a 1:1 Protein G sepharose 4B beads (Sigma-Aldrich) mixture in NP40 lysis buffer (Table S3) and immunoprecipitates (IPs) were mixed for 2 hr at 4°C on a rotation wheel. Thereafter, the IPs were washed three times with 300 μl of NP40 lysis buffer (Table S3) before analysis on standard sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). Specific bands were detected using fluorescent-labeled secondary antibodies (Invitrogen; Li-COR Biosciences) and analyzed using an Odyssey Infrared Scanner (Li-COR Biosciences). For CIP treatment, immunoprecipitates were washed twice with NP40 lysis buffer (Table S3) and once in CIP buffer (Table S3). Thereafter immunoprecipitates were incubated with 30 μl CIP buffer containing 5 units of CIP in the presence or absence of 0.2 mM Na3VO4 phosphatase inhibitor and 7mM EDTA. Controls were incubated in CIP buffer without CIP. The reactions were performed at 30°C for 30 min before analysis on SDS-PAGE.Cell Fractionation ExperimentsD04 cells were plated on two 10cm dishes per treatment and grown to confluency. After treatment, cells were washed three times with cold PBS, washed once with 20mM HEPES pH 7.4 and then lysed by scraping in 20mM HEPES pH 7.4 supplemented with protease inhibitors (1ml per 2 plates). Cells were disrupted by passing them through a 9G syringe ten times, followed by another ten times through a 19G syringe (Terumo Medical). Lysates were centrifuged at 900 × g for five minutes to pellet the nuclear proteins. The supernatant was transferred to fresh 1.5ml ultracentrifuge tubes (Beckman Coulter) and 200 μl removed as a total lysate control. The remainder was centrifuged at 100,000 × g for 30 min at 4°C to separate the cytosolic fraction from the membrane fraction. The supernatant containing the cytosolic fraction was transferred to a fresh 1.5ml tube and the pelleted membrane fraction washed once in 20 mM HEPES pH 7.4 before resuspension in 200 μl 20mM HEPES pH 7.4/1% Triton X-100. For analysis on SDS-PAGE, the concentration of protein was determined by Bradford protein assay (Bio-Rad Laboratories) using purified bovine serum albumin (BSA) as a standard as described by the manufacturer. Equal amounts of protein were loaded for the cytosolic fraction and total cell lysate. Three times as much protein was loaded for the membrane fraction.RAF Kinase AssaysThe in vitro kinase activity of endogenous RAF proteins or myc-tagged RAF proteins transiently expressed in COS-7 cells was measured using a coupled kinase cascade assay with GST-MEK, GST-ERK and myelin basic protein (MBP) (Sigma-Aldrich) as sequential substrates. ERK activation was quantified by measuring the incorporation of [P]-orthophosphate (PerkinElmer) into MBP. For measurement of endogenous BRAF kinase activity, D04 or A375 cells were seeded in 10cm dishes and harvested in 300 μl of NP40 buffer (Table S3) as described above. Protein concentrations were determined and equal amounts of protein were immunoprecipitated as described above.For measurement of mutant BRAF kinase activities, COS-7 cells were transiently transfected with myc-tagged BRAF and cells in one 35 mm diameter well were harvested in 200μl of NP40 buffer (Table S3). The relative concentrations of exogenously expressed RAF in these cell lysates were determined by quantitative western blotting using the myc antibody (Cell Signaling Technology) specified above. Bands were quantified using the Odyssey infrared imaging system (LI-COR Biosciences). Equivalent amounts of RAF were immunoprecipitated using rabbit myc antibody (Abcam) as specified above.Endogenous and transiently transfected RAF proteins were immunoprecipitated in a total of 300 μl NP40 buffer (Table S3) for 2 hr at 4°C and immunoprecipitates were washed sequentially three times with chilled wash buffer (Table S3) containing decreasing concentrations of KCl (1M KCl, 0.1M KCl and no KCl). The first-step reaction was initiated by addition of 20 μl MKK buffer (containing GST-MEK and GST-ERK, Table S3) to the beads and incubated at 30°C for 10 min in the case of myc-tagged BRAF or 30 min for endogenous BRAF and CRAF. Reactions were terminated by the addition of 20 μl KILL buffer (Table S3), which contains EDTA to chelate Mg ions and inhibit kinase activity. The reaction supernatants were collected from the beads and transferred into fresh tubes for the second step reaction. 5 μl of supernatant was incubated with 25 μl MBP buffer containing [γ-P]ATP (PerkinElmer) for ten minutes at 30°C in triplicate to measure ERK activity. The second reaction was terminated by spotting 20 μl of reaction mix onto a 1cm piece of P81 paper (VWR International), which was then dropped into 400ml 25mM orthophosphate solution. The papers were washed three times in 400 ml 25 mM orthophosphate solution to remove the unincorporated ATP and the [P]-orthophosphate incorporated into MBP was determined using Cerenkov counting. For transfected samples, the background counts were determined using lysates of cells transfected with the empty vector. For endogenous protein, samples in which no RAF was immunoprecipitated were used. Background values were removed and to ensure linearity, assays were used at below 50% saturation. To determine BRAF and BRAF sensitivity to 885-A, immunoprecipitated BRAF was preincubated with drug in KCl-free buffer for 10 min at room temperature prior to the first-step reaction.To measure the activity of purified BRAF, a 96-well DELFIA-based assay system was used. Full-length rabbit MEK1 protein was expressed with a GST tag at the N-terminus and a C-terminal histidine tag in Escherichia coli JM109 bacteria and purified by nickel-agarose affinity chromatography. Full length BRAF protein was generated by infection of SF9 insect cells with a recombinant baculovirus expressing full-length human BRAF with an N-terminal histidine tag and purified as above. For the kinase assays, all incubations were at room temperature with shaking. 4 μg GST-MEK1, 100-200ng purified BRAF and 1 μl inhibitor at the required concentrations (0.001 to 100 μM final concentration) were added to the wells of glutathione-coated plates and preincubated for 10 min. ATP in DELFIA assay buffer (20 μL; Table S3), to give a final concentration of 100 μM, was added to each well, and the plates were incubated for 45 min. The plates were washed 3X with 200 μl 0.1% tween20/water. Primary antibody (rabbit anti-phospho MEK1/2 diluted 1/2000, Cell Signaling Technologies) and Eu-labeled anti-rabbit secondary antibody (diluted 1/1000, Perkin-Elmer) were preincubated for 30 min and 100 μl was added to the plates and incubated for a further hour. The plates were washed as before, and 100 μl DELFIA enhancement solution (Perkin-Elmer, Turku, Finland) was added. The plates were sealed and incubated for 30 min and europium counts measured on Spectramax M5 plate reader (Molecular Devices, Wokingham, UK). IC50 values were determined using GraphPad Prism (Graphpad Software Inc., San Diego, CA, USA).Transgenic MiceExperiments were performed under Home Office license authority with local Ethics Committee approval. To activate CreERT2, four doses of tamoxifen (Sigma; 10mg each in 100% ethanol) were applied topically to the shaven skin on the backs of the mice every other day for 7 days. Genotyping was performed by PCR using DNA prepared with DNeasy kits (QIAGEN). Braf and Braf were analyzed using primers: A) 5′-GCCCAGGCTCTTTATGAGAA-3; B) 5′-AGTCAATCATCCACAGAGACCT-3′; and C) 5′-GCTTGGCTGGACGTAAACTC-3′. A+B detects the wild-type BRAF allele (466 bp product) and Braf, the Cre-recombinase recombined allele (518pb product). A+C detects the targeted allele Braf (140 bp). Tyr::CreERT2 was analyzed using primers 5′-GAAGCAACTCATCGATTG-3′ and 5′-TGAAGGGTCTGGTAGGATCA-3′. Kras was analyzed using primers 5′-CGCAGACTGTAGAGCAGCG-3′ and 5′-CCATGGCTTGAGTAAGTCTGC-3′.mRNA expression analysis was performed by RT-PCR. RNA was prepared using the RNEasy kit (QIAGEN). First-strand cDNA synthesis was performed with 500ng total RNA and random hexanucleotides (Random Primers, Invitrogen). Tyrosinase (Tyr), was detected using primers 5′-TGGTTCCTTTCATACCGCTC-3′ and 5′-CAGATACGACTGGCTTGTTCC-3′; Dct with 5′-GCAAGATTGCCTGTCTCTCC-3′ and 5′-AGTCCAGTGTTCCGTCTGCT-3′; Pax3 with 5′-CCAGGATGATGCGGCCCGGCCCGGG-3′ and 5′-AGGATGCGGCTGATAGAACTCACTG-3′; and silver/gp100 (Si) with 5′-GGAGAGGTGGCCAGGTATC-3′ and 5′-CAGTAATGGTGAAGGTTGAAC-3′. The control Gapdh was detected with 5′- GATGGCCCCTCGGAAAGCT-3′ 5′-CCAGTGAGCTTCCCGTTCAGC-3′. To sequence Kras cDNA, a 238 bp product from Kras cDNA was PCR amplified using primers 5′-GGCGGCAGCGCTGTGGCGGCG-3′ and 5′-CGTAGGGTCATACTCATCCAC-3′ and directly sequenced using these primers and automated dideoxy sequencing.For immunohistochemistry (IHC), tumors were fixed in 10% buffered formalin and embedded in paraffin. Sections (3-10μm) were stained with hematoxylin and eosin using standard protocols. For S100 and Ki67 staining, antigen retrieval was performed in citrate buffer (pH 6.0, 30 min) and revealed using a rabbit polyclonal antibody (Dako, 1/1000), the Rabbit Envision Peroxidase kit and the AEC substrate chromogen (Dako) for S100, and a rat monoclonal antibody (Dako,1/25), the rat Vectastain ABC kit (Vector Labs, USA) and DAB as chromagen for Ki67. Positive (a well characterized sample of mouse melanoma) and negative (omission of the primary antibody and substitution with preimmune serum) controls were included in each slide run. Immunohistochemical staining was analyzed by two of the authors on a multi-headed microscope. Tumor cell lines were established by mechanically dissociating tumors in DMEM/20%FCS/Primocin (0.1mg/ml - InvivoGen) and clonal lines were selected by limiting dilution. For western blotting the following antibodies were used: rabbit anti-ppMEK1/2 and mouse anti-myc 9B11 (Cell Signaling Technology); mouse anti-NRAS (C-20), rabbit anti-ERK2 (C-14), rabbit anti-ARAF (C-20), mouse anti-BRAF (F-7) (Santa Cruz Biotechnology); mouse anti-Tubulin, and mouse anti-ppERK1/2 (Sigma); mouse anti-CRAF (for western blotting) (BD Transduction Laboratories). For immunoprecipitation, the following antibodies were used: rabbit anti-myc (Abcam); rabbit anti-CRAF (C-20; Santa Cruz Biotechnology); mouse anti-BRAF (F-7) (Santa Cruz Biotechnology). Calf intestinal phosphatase (CIP) was from New England Biolabs (NEB). PD184352, sorafenib and PLX4720 were synthesized in-house; 885-A was synthesized by Evotec AG (Abingdon, UK). All drugs were prepared in DMSO. Synthetic routes are available on request. The expression vectors for wild-type human CRAF and wild-type human BRAF, pEFm/CRAF and pEFm/BRAF respectively have been described (Marais et al., 1995). Briefly, the vector backbone is pUC19 and the elongation factor 1α (EF1α) promoter is used to drive exogenous protein expression. The vector includes the first intron from human EF1α to assist mRNA processing during expression. The β-globin 5′ and 3′ untranslated regions (UTRs) are used to provide a strong translation start site (5′ UTR), and also to provide mRNA stability and a poly adenylation signal (3′ UTR). The vector introduces an amino-terminal myc-epitope tag (EQKLISEEDL) onto the exogenously expressed protein. The BRAF coding region includes the alternatively spliced exons 1 and 2 but not exons 8b or 10a and various modifications were introduced to provide additional restriction sites (without changing the amino acid sequence) and alterations to the 3′-UTR to allow easier manipulation of this construct. Standard PCR-directed mutagenesis approaches were used to generate the various mutations used in the study and all mutations were verified by automated dideoxy sequencing. The expression vector pMCEF/FLAG/CRAF uses the same expression cassette, but the backbone also possesses a neo resistance cassette to facilitate selection in the presence of G418. In addition, a version of this vector was used that incorporates a FLAG (DYKDDDKGS), rather than a myc-epitope tag. Human cell lines were cultured in DMEM (A375, WM852, HCT116, SW620, PMWK, SKMel24, SKMel28, D25) or RPMI (D04, MM485, MM415, WM1791c) supplemented with 10% fetal bovine serum. African green monkey kidney COS-7 cells were cultured in DMEM supplemented with 10% fetal bovine serum. All cell lines were incubated at 37°C and 10% CO2. For inhibitor treatment, the drugs were dissolved in DMSO and added to the medium for 2-5 hr. When two compounds were used, the first was added 30 min prior to the second. For cell growth assays, cells were seeded in 96-well plates and treated with drug in quadruplicate in a 10-point titration assay for 5 days. The amount of growth (% DMSO controls) was determined using sulphorhodamine B reagent (Monks et al., 1991) as follows. 1,000-10,000 cells (depending on cell type) were plated into 96-well plates in 100 μL medium. After 24 hr, compounds prepared in DMSO (10mM stocks) were serially diluted in culture medium at 2× the final required concentrations and 100μL were added to the cells to nine final concentrations of 0.005 μM-100 μM. After a further 5 days, the cells are fixed in trichloroacteic acid (10%), and stained with sulforhodamine-B (0.1%). After rinsing, the bound stain was disolved using 100 μL 10 mM Tris (pH 8.0) and the absorbance at 540 nm determined. The data were analyzed by nonlinear regression to a four-parameter logistic equation (Graphpad Prism, Graphpad Software Inc., San Diego, CA, USA) and the GI50 value determined. 3 × 10 D04 cells per 35 mm diameter well were seeded in 2ml growth medium the day before transfection. The cells were either mock-transfected or transfected with 6nM CRAF-specific (5′-AAGCACGCTTAGATTGGAATA-3′) or NRAS-specific (5′-CATGGCACTGTACTCTTCTCG-3′) siRNA using INTERFERin as recommended by the manufacturer (Polyplus Transfection SA). Briefly, 0.6 μl of 20 μM siRNA and 6 μl of INTERFERin were combined in a total of 200 μl serum free medium in RNase-free tubes. The mix was vortexed for 10 s and incubated for 5-10 min before adding the complexes dropwise to the cells. The cells were serum-starved the day after transfecting and extracts were prepared 48 hr after transfection. For transient protein expression in D04 cells, Lonza Nucleofector Technology (Lonza, Cologne AG) was used. 2 μg of DNA was mixed with 1x10 cells resuspended in 100 μl of Nucleofection Solution V in an Amaxa-certified cuvette and transfected using program T030. The cells were re-plated into 35mm diameter tissue culture wells and incubated for 48 hr before preparation of cell extracts. For generation of stable lines, D04 cells were transfected using Effectene (Invitrogen) and selected in G418. 3-4x10 cells were plated in 35 mm diameter wells and incubated overnight. 0.4 μg of DNA diluted into 100 μl of DNA condensation buffer (EC) and 3.2 μl enhancer were mixed vigorously and incubated for 2-5 min. 10 μl Effectene reagent was added and the mixture was incubated for another 5-10 min. The cells were washed with 2ml PBS and 1.6ml fresh serum containing medium was added. The DNA complexes were diluted with 600 μl of culture medium and the mixture added to the cells drop-wise. After six hours, the medium was replaced with 2ml of fresh growth medium. After 48 hr, the cells were replated into several 10cm dishes in a 10-fold dilution series and incubated in G418 (1mg/ml) for selection. The medium was refreshed weekly and after 2-3 weeks, single colonies were selected and expanded. For transient expression in COS-7 cells, Lipofectamine (Invitrogen) was used. 2x10 cells were plated into 35mm diameter wells and incubated overnight. 75 to 200 ng of expression plasmid (depending on construct) was mixed with empty vector to a total of 700 ng DNA in 16μl PBS. Typically, 3 μl of Lipofectamine in 13 μl of PBS was added to the DNA on the surface of a bacterial plate and incubated (Lipofectamine is inactivated by binding to polypropylene) for 15 min at room temperature. The cells were washed twice with 1ml serum-free DMEM, and then overlaid with 800 μl of serum free DMEM. 200 μl of serum free DMEM was added to the DNA:Lipofectamine mix, and the total volume was added to the cells. After six hours, the complexes were removed and replaced with 2ml of normal culture medium. Cell extracts were prepared two days following transfection. Culture medium was aspirated from cells and cells were placed on ice and washed three times in ice-cold PBS. Depending on the assays, the cells were scraped into 50-200 μl Nonidet P40 (NP40) extraction buffer (Table S3) and incubated on ice for five minutes. The cells were sheared by passing through a pipette tip several times and the samples were centrifuged at 20,000 × g for 5 min at 4°C and the soluble fraction was harvested. Immunoprecipitations were performed in 300 μl cell lysates from one 35mm diameter well for endogenous protein or from 2-3 wells for transfected protein. Endogenous BRAF or CRAF were immunoprecipitated with 2μg BRAF F-7 or 5μg CRAF C-20 respectively and myc-tagged BRAF and CRAF with 2 μg rabbit anti-myc antibody. The antibody-protein complex was captured using 20 μl of a 1:1 Protein G sepharose 4B beads (Sigma-Aldrich) mixture in NP40 lysis buffer (Table S3) and immunoprecipitates (IPs) were mixed for 2 hr at 4°C on a rotation wheel. Thereafter, the IPs were washed three times with 300 μl of NP40 lysis buffer (Table S3) before analysis on standard sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). Specific bands were detected using fluorescent-labeled secondary antibodies (Invitrogen; Li-COR Biosciences) and analyzed using an Odyssey Infrared Scanner (Li-COR Biosciences). For CIP treatment, immunoprecipitates were washed twice with NP40 lysis buffer (Table S3) and once in CIP buffer (Table S3). Thereafter immunoprecipitates were incubated with 30 μl CIP buffer containing 5 units of CIP in the presence or absence of 0.2 mM Na3VO4 phosphatase inhibitor and 7mM EDTA. Controls were incubated in CIP buffer without CIP. The reactions were performed at 30°C for 30 min before analysis on SDS-PAGE. D04 cells were plated on two 10cm dishes per treatment and grown to confluency. After treatment, cells were washed three times with cold PBS, washed once with 20mM HEPES pH 7.4 and then lysed by scraping in 20mM HEPES pH 7.4 supplemented with protease inhibitors (1ml per 2 plates). Cells were disrupted by passing them through a 9G syringe ten times, followed by another ten times through a 19G syringe (Terumo Medical). Lysates were centrifuged at 900 × g for five minutes to pellet the nuclear proteins. The supernatant was transferred to fresh 1.5ml ultracentrifuge tubes (Beckman Coulter) and 200 μl removed as a total lysate control. The remainder was centrifuged at 100,000 × g for 30 min at 4°C to separate the cytosolic fraction from the membrane fraction. The supernatant containing the cytosolic fraction was transferred to a fresh 1.5ml tube and the pelleted membrane fraction washed once in 20 mM HEPES pH 7.4 before resuspension in 200 μl 20mM HEPES pH 7.4/1% Triton X-100. For analysis on SDS-PAGE, the concentration of protein was determined by Bradford protein assay (Bio-Rad Laboratories) using purified bovine serum albumin (BSA) as a standard as described by the manufacturer. Equal amounts of protein were loaded for the cytosolic fraction and total cell lysate. Three times as much protein was loaded for the membrane fraction. The in vitro kinase activity of endogenous RAF proteins or myc-tagged RAF proteins transiently expressed in COS-7 cells was measured using a coupled kinase cascade assay with GST-MEK, GST-ERK and myelin basic protein (MBP) (Sigma-Aldrich) as sequential substrates. ERK activation was quantified by measuring the incorporation of [P]-orthophosphate (PerkinElmer) into MBP. For measurement of endogenous BRAF kinase activity, D04 or A375 cells were seeded in 10cm dishes and harvested in 300 μl of NP40 buffer (Table S3) as described above. Protein concentrations were determined and equal amounts of protein were immunoprecipitated as described above. For measurement of mutant BRAF kinase activities, COS-7 cells were transiently transfected with myc-tagged BRAF and cells in one 35 mm diameter well were harvested in 200μl of NP40 buffer (Table S3). The relative concentrations of exogenously expressed RAF in these cell lysates were determined by quantitative western blotting using the myc antibody (Cell Signaling Technology) specified above. Bands were quantified using the Odyssey infrared imaging system (LI-COR Biosciences). Equivalent amounts of RAF were immunoprecipitated using rabbit myc antibody (Abcam) as specified above. Endogenous and transiently transfected RAF proteins were immunoprecipitated in a total of 300 μl NP40 buffer (Table S3) for 2 hr at 4°C and immunoprecipitates were washed sequentially three times with chilled wash buffer (Table S3) containing decreasing concentrations of KCl (1M KCl, 0.1M KCl and no KCl). The first-step reaction was initiated by addition of 20 μl MKK buffer (containing GST-MEK and GST-ERK, Table S3) to the beads and incubated at 30°C for 10 min in the case of myc-tagged BRAF or 30 min for endogenous BRAF and CRAF. Reactions were terminated by the addition of 20 μl KILL buffer (Table S3), which contains EDTA to chelate Mg ions and inhibit kinase activity. The reaction supernatants were collected from the beads and transferred into fresh tubes for the second step reaction. 5 μl of supernatant was incubated with 25 μl MBP buffer containing [γ-P]ATP (PerkinElmer) for ten minutes at 30°C in triplicate to measure ERK activity. The second reaction was terminated by spotting 20 μl of reaction mix onto a 1cm piece of P81 paper (VWR International), which was then dropped into 400ml 25mM orthophosphate solution. The papers were washed three times in 400 ml 25 mM orthophosphate solution to remove the unincorporated ATP and the [P]-orthophosphate incorporated into MBP was determined using Cerenkov counting. For transfected samples, the background counts were determined using lysates of cells transfected with the empty vector. For endogenous protein, samples in which no RAF was immunoprecipitated were used. Background values were removed and to ensure linearity, assays were used at below 50% saturation. To determine BRAF and BRAF sensitivity to 885-A, immunoprecipitated BRAF was preincubated with drug in KCl-free buffer for 10 min at room temperature prior to the first-step reaction. To measure the activity of purified BRAF, a 96-well DELFIA-based assay system was used. Full-length rabbit MEK1 protein was expressed with a GST tag at the N-terminus and a C-terminal histidine tag in Escherichia coli JM109 bacteria and purified by nickel-agarose affinity chromatography. Full length BRAF protein was generated by infection of SF9 insect cells with a recombinant baculovirus expressing full-length human BRAF with an N-terminal histidine tag and purified as above. For the kinase assays, all incubations were at room temperature with shaking. 4 μg GST-MEK1, 100-200ng purified BRAF and 1 μl inhibitor at the required concentrations (0.001 to 100 μM final concentration) were added to the wells of glutathione-coated plates and preincubated for 10 min. ATP in DELFIA assay buffer (20 μL; Table S3), to give a final concentration of 100 μM, was added to each well, and the plates were incubated for 45 min. The plates were washed 3X with 200 μl 0.1% tween20/water. Primary antibody (rabbit anti-phospho MEK1/2 diluted 1/2000, Cell Signaling Technologies) and Eu-labeled anti-rabbit secondary antibody (diluted 1/1000, Perkin-Elmer) were preincubated for 30 min and 100 μl was added to the plates and incubated for a further hour. The plates were washed as before, and 100 μl DELFIA enhancement solution (Perkin-Elmer, Turku, Finland) was added. The plates were sealed and incubated for 30 min and europium counts measured on Spectramax M5 plate reader (Molecular Devices, Wokingham, UK). IC50 values were determined using GraphPad Prism (Graphpad Software Inc., San Diego, CA, USA). Experiments were performed under Home Office license authority with local Ethics Committee approval. To activate CreERT2, four doses of tamoxifen (Sigma; 10mg each in 100% ethanol) were applied topically to the shaven skin on the backs of the mice every other day for 7 days. Genotyping was performed by PCR using DNA prepared with DNeasy kits (QIAGEN). Braf and Braf were analyzed using primers: A) 5′-GCCCAGGCTCTTTATGAGAA-3; B) 5′-AGTCAATCATCCACAGAGACCT-3′; and C) 5′-GCTTGGCTGGACGTAAACTC-3′. A+B detects the wild-type BRAF allele (466 bp product) and Braf, the Cre-recombinase recombined allele (518pb product). A+C detects the targeted allele Braf (140 bp). Tyr::CreERT2 was analyzed using primers 5′-GAAGCAACTCATCGATTG-3′ and 5′-TGAAGGGTCTGGTAGGATCA-3′. Kras was analyzed using primers 5′-CGCAGACTGTAGAGCAGCG-3′ and 5′-CCATGGCTTGAGTAAGTCTGC-3′. mRNA expression analysis was performed by RT-PCR. RNA was prepared using the RNEasy kit (QIAGEN). First-strand cDNA synthesis was performed with 500ng total RNA and random hexanucleotides (Random Primers, Invitrogen). Tyrosinase (Tyr), was detected using primers 5′-TGGTTCCTTTCATACCGCTC-3′ and 5′-CAGATACGACTGGCTTGTTCC-3′; Dct with 5′-GCAAGATTGCCTGTCTCTCC-3′ and 5′-AGTCCAGTGTTCCGTCTGCT-3′; Pax3 with 5′-CCAGGATGATGCGGCCCGGCCCGGG-3′ and 5′-AGGATGCGGCTGATAGAACTCACTG-3′; and silver/gp100 (Si) with 5′-GGAGAGGTGGCCAGGTATC-3′ and 5′-CAGTAATGGTGAAGGTTGAAC-3′. The control Gapdh was detected with 5′- GATGGCCCCTCGGAAAGCT-3′ 5′-CCAGTGAGCTTCCCGTTCAGC-3′. To sequence Kras cDNA, a 238 bp product from Kras cDNA was PCR amplified using primers 5′-GGCGGCAGCGCTGTGGCGGCG-3′ and 5′-CGTAGGGTCATACTCATCCAC-3′ and directly sequenced using these primers and automated dideoxy sequencing. For immunohistochemistry (IHC), tumors were fixed in 10% buffered formalin and embedded in paraffin. Sections (3-10μm) were stained with hematoxylin and eosin using standard protocols. For S100 and Ki67 staining, antigen retrieval was performed in citrate buffer (pH 6.0, 30 min) and revealed using a rabbit polyclonal antibody (Dako, 1/1000), the Rabbit Envision Peroxidase kit and the AEC substrate chromogen (Dako) for S100, and a rat monoclonal antibody (Dako,1/25), the rat Vectastain ABC kit (Vector Labs, USA) and DAB as chromagen for Ki67. Positive (a well characterized sample of mouse melanoma) and negative (omission of the primary antibody and substitution with preimmune serum) controls were included in each slide run. Immunohistochemical staining was analyzed by two of the authors on a multi-headed microscope. Tumor cell lines were established by mechanically dissociating tumors in DMEM/20%FCS/Primocin (0.1mg/ml - InvivoGen) and clonal lines were selected by limiting dilution.
|
PMC5785775
|
Generating Focused Molecule Libraries for Drug Discovery with Recurrent Neural Networks
|
In de novo drug design, computational strategies are used to generate novel molecules with good affinity to the desired biological target. In this work, we show that recurrent neural networks can be trained as generative models for molecular structures, similar to statistical language models in natural language processing. We demonstrate that the properties of the generated molecules correlate very well with the properties of the molecules used to train the model. In order to enrich libraries with molecules active toward a given biological target, we propose to fine-tune the model with small sets of molecules, which are known to be active against that target. Against Staphylococcus aureus, the model reproduced 14% of 6051 hold-out test molecules that medicinal chemists designed, whereas against Plasmodium falciparum (Malaria), it reproduced 28% of 1240 test molecules. When coupled with a scoring function, our model can perform the complete de novo drug design cycle to generate large sets of novel molecules for drug discovery.Chemistry is the language of nature. Chemists speak it fluently and have made their discipline one of the true contributors to human well-being, which has “change[d] the way you live and die”. This is particularly true for medicinal chemistry. However, creating novel drugs is an extraordinarily hard and complex problem. One of the many challenges in drug design is the sheer size of the search space for novel molecules. It has been estimated that 10 drug-like molecules could possibly be synthetically accessible. Chemists have to select and examine molecules from this large space to find molecules that are active toward a biological target. Active means for example that a molecule binds to a biomolecule, which causes an effect in the living organism, or inhibits replication of bacteria. Modern high-throughput screening techniques allow testing of molecules on the order of 10 in the lab. However, larger experiments will get prohibitively expensive. Given this practical limitation of in vitro experiments, it is desirable to have computational tools to narrow down the enormous search space. Virtual screening is a commonly used strategy to search for promising molecules among millions of existing or billions of virtual molecules. Searching can be carried out using similarity-based metrics, which provides a quantifiable numerical indicator of closeness between molecules. In contrast, in de novo drug design, one aims to directly create novel molecules that are active toward the desired biological target. Here, like in any molecular design task, the computer has to(i)create molecules,(ii)score and filter them, and(iii)search for better molecules, building on the knowledge gained in the previous steps. create molecules, score and filter them, and search for better molecules, building on the knowledge gained in the previous steps. Task i, the generation of novel molecules, is usually solved with one of two different protocols. One strategy is to build molecules from predefined groups of atoms or fragments. Unfortunately, these approaches often lead to molecules that are very hard to synthesize. Therefore, another established approach is to conduct virtual chemical reactions based on expert coded rules, with the hope that these reactions could then also be applied in practice to make the molecules in the laboratory. These systems give reasonable drug-like molecules and are considered as “the solution” to the structure generation problem. We generally share this view. However, we have recently shown that the predicted reactions from these rule-based expert systems can sometimes fail. Also, focusing on a small set of robust reactions can unnecessarily restrict the possibly accessible chemical space. Task ii, scoring molecules and filtering out undesired structures, can be solved with substructure filters for undesirable reactive groups in conjunction with established approaches such as docking or machine learning (ML) approaches. The ML approaches are split into two branches: Target prediction classifies molecules into active and inactive, and quantitative structure–activity relationships (QSAR) seek to quantitatively predict a real-valued measure for the effectiveness of a substance (as a regression problem). As molecular descriptors, signature fingerprints, extended-connectivity (ECFP), and atom pair (APFP) fingerprints and their fuzzy variants are the de facto standard today. Convolutional networks on graphs are a more recent addition to the field of molecular descriptors. Jastrzebski et al. proposed to use convolutional neural networks to learn descriptors directly from SMILES strings. Random forests, support vector machines, and neural networks are currently the most widely used machine learning models for target prediction. This leads to task iii, the search for molecules with the right binding affinity combined with optimal molecular properties. In earlier work, this was performed (among others) with classical global optimization techniques, for example genetic algorithms or ant-colony optimization. Furthermore, de novo design is related to inverse QSAR. While in de novo design a regular QSAR mapping X → y from molecular descriptor space X to properties y is used as the scoring function for the global optimizer, in inverse QSAR one aims to find an explicit inverse mapping y → X, and then maps back from optimal points in descriptor space X to valid molecules. However, this is not well-defined, because molecules are inherently discrete (the space is not continuously populated), and the mapping from a target property value y to possible structures X is one-to-many, as usually several different structures with very similar properties can be found. Several protocols have been developed to address this, for example enumerating all structures within the constraints of hyper-rectangles in the descriptor space. Gómez-Bombarelli et al. proposed to learn continuous representations of molecules with variational autoencoders, based on the model by Bowman et al., and to perform Bayesian optimization in this vector space to optimize molecular properties. While promising, this approach was not applied to create active drug molecules and often produced syntactically invalid molecules and highly strained or reactive structures, for example cyclobutadienes. In this work, we suggest a complementary, completely data-driven de novo drug design approach. It relies only on a generative model for molecular structures, based on a recurrent neural network, that is trained on large sets of molecules. Generative models learn a probability distribution over the training examples; sampling from this distribution generates new examples similar to the training data. Intuitively, a generative model for molecules trained on drug molecules would “know” how valid and reasonable drug-like molecules look and could be used to generate more drug-like molecules. However, for molecules, these models have been studied rarely, and rigorously only with traditional models such as Gaussian mixture models (GMM). Recently, recurrent neural networks (RNNs) have emerged as powerful generative models in very different domains, such as natural language processing, speech, images, video, formal languages, computer code generation, and music scores. In this work, we highlight the analogy of language and chemistry, and show that RNNs can also generate reasonable molecules. Furthermore, we demonstrate that RNNs can also transfer their learned knowledge from large molecule sets to directly produce novel molecules that are biologically active by retraining the models on small sets of already known actives. We test our models by reproducing hold-out test sets of known biologically active molecules. To connect chemistry with language, it is important to understand how molecules are represented. Usually, they are modeled by molecular graphs, also called Lewis structures in chemistry. In molecular graphs, atoms are labeled nodes. The edges are the bonds between atoms, which are labeled with the bond order (e.g., single, double, or triple). One could therefore envision having a model that reads and outputs graphs. Several common chemistry formats store molecules in such a manner. However, in models for natural language processing, the input and output of the model are usually sequences of single letters, strings or words. We therefore employ the SMILES format, which encodes molecular graphs compactly as human-readable strings. SMILES is a formal grammar which describes molecules with an alphabet of characters, for example c and C for aromatic and aliphatic carbon atoms, O for oxygen, and −, =, and # for single, double, and triple bonds (see Figure 1). To indicate rings, a number is introduced at the two atoms where the ring is closed. For example, benzene in aromatic SMILES notation would be c1ccccc1. Side chains are denoted by round brackets. To generate valid SMILES, the generative model would have to learn the SMILES grammar, which includes keeping track of rings and brackets to eventually close them. In morphine, a complex natural product, the number of steps between the first 1 and the second 1, indicating a ring, is 32. Having established a link between molecules and (formal) language, we can now discuss language models. Examples of molecules and their SMILES representation. To correctly create smiles, the model has to learn long-term dependencies, for example, to close rings (indicated by numbers) and brackets. Given a sequence of words (w1, ..., wi), language models predict the distribution of the (i+1)th word wi+1. For example, if a language model received the sequence “Chemistry is”, it would assign different probabilities to possible next words: “fascinating”, “important”, or “challenging” would receive high probabilities, while “runs” or “potato” would receive very low probabilities. Language models can both capture the grammatical correctness (“runs” in this sentence is wrong) and the meaning (“potato” does not make sense). Language models are implemented, for example, in message autocorrection in many modern smartphones. Interestingly, language models do not have to use words. They can also be based on characters or letters. In that case, when receiving the sequence of characters chemistr, it would assign a high probability to y, but a low probability to q. To model molecules instead of language, we simply swap words or letters with atoms, or, more practically, characters in the SMILES alphabet, which form a (formal) language. For example, if the model receives the sequence c1ccccc, there is a high probability that the next symbol would be a “1”, which closes the ring, and yields benzene. More formally, to a sequence S of symbols si at steps ti ∈ T, the language model assigns a probability of1where the parameters θ are learned from the training set. In this work, we use a recurrent neural network (RNN) to estimate the probabilities of eq 1. In contrast to regular feedforward neural networks, RNNs maintain state, which is needed to keep track of the symbols seen earlier in the sequence. In abstract terms, an RNN takes a sequence of input vectors x1:n = (x1, ..., xn) and an initial state vector h0, and returns a sequence of state vectors h1:n = (h1, ..., hn) and a sequence of output vectors y1:n = (y1, ..., yn). The RNN consists of a recursively defined function R, which takes a state vector hi and input vector xi+1 and returns a new state vector hi+1. Another function O maps a state vector hi to an output vector yi.234 The state vector hi stores a representation of the information about all symbols seen in the sequence so far. As an alternative to the recursive definition, the recurrent network can also be unrolled for finite sequences (see Figure 2). An unrolled RNN can be seen as a very deep neural network, in which the parameters θ are shared among the layers, and the hidden state ht is passed as an additional input to the next layer. Training the unrolled RNN to fit the parameters θ can then simply be done by using backpropagation to compute the gradients with respect to the loss function, which is categorical cross-entropy in this work. (a) Recursively defined RNN. (b) The same RNN, unrolled. The parameters θ (the weight matrices of the neural network) are shared over all time steps. As the specific RNN function, in this work, we use the long short-term memory (LSTM), which was introduced by Hochreiter and Schmidhuber. It has been used successfully in many natural language processing tasks, for example in Google’s neural machine translation system. For excellent in-depth discussions of the LSTM, we refer to the articles by Goldberg, Graves, Olah, and Greff et al. To encode the SMILES symbols as input vectors xt, we employ the “one-hot” representation. This means that if there are K symbols, and k is the symbol to be input at step t, then we can construct an input vector xt with length K, whose entries are all zero except the kth entry, which is one. If we assume a very restricted set of symbols , input c would correspond to xt = (1, 0, 0), 1 to xt = (0, 1, 0), and to xt = (0, 0, 1). The probability distribution Pθ(st+1|st, ..., s1) of the next symbol given the already seen sequence is thus a multinomial distribution, which is estimated using the output vector yt of the recurrent neural network at time step t by5where yt corresponds to the kth element of vector yt. Sampling from this distribution would then allow generating novel molecules: After sampling a SMILES symbol st+1 for the next time step t + 1, we can construct a new input vector xt+1, which is fed into the model, and via yt+1 and eq 5 yields Pθ(st+2|st+1, ..., s1). Sampling from the latter generates st+2, which serves again also as the model’s input for the next step (see Figure 3). This symbol-by-symbol sampling procedure is repeated until the desired number of characters have been generated. Symbol generation and sampling process. We start with a random seed symbol s1, here c, which gets converted into a one-hot vector x1 and input into the model. The model then updates its internal state h0 to h1 and outputs y1, which is the probability distribution over the next symbols. Here, sampling yields s2 = 1. Converting s2 to x2 and feeding it to the model leads to updated hidden state h2 and output y2, from which we can sample again. This iterative symbol-by-symbol procedure can be continued as long as desired. In this example, we stop it after observing an EOL () symbol, and obtain the SMILES for benzene. The hidden state hi allows the model to keep track of opened brackets and rings, to ensure that they will be closed again later. To indicate that a molecule is “completed”, each molecule in our training data finishes with an “end of line” (EOL) symbol, in our case the single character (which means that the training data is just a simple SMILES file). Thus, when the system outputs an EOL, a generated molecule is finished. However, we simply continue sampling, thus generating a regular SMILES file that contains one molecule per line. In this work, we used a network with three stacked LSTM layers, using the Keras library. The model was trained with back-propagation through time, using the ADAM optimizer at standard settings. To mitigate the problem of exploding gradients during training, a gradient norm clipping of 5 is applied. For many machine learning tasks, only small data sets are available, which might lead to overfitting with powerful models such as neural networks. In this situation, transfer learning can help. Here, a model is first trained on a large data set for a different task. Then, the model is retrained on the smaller data set, which is also called fine-tuning. The aim of transfer learning is to learn general features on the bigger data set, which also might be useful for the second task in the smaller data regime. To generate focused molecule libraries, we first train on a large, general set of molecules, then perform fine-tuning on a smaller set of specific molecules, and after that start the sampling procedure. To verify whether the generated molecules are active on the desired targets, standard target prediction was employed. Machine learning based target prediction aims to learn a classifier c: M → to decide whether a molecule m ∈ molecular descriptor space M is active or not against a target. The molecules are split into actives and inactives using a threshold on a measure for the substance effectiveness. pIC50 = −log10(IC50) is one of the most widely used metrics for this purpose. IC50 is the half maximal inhibitory concentration, that is, the concentration of drug that is required to inhibit 50% of a biological target’s function in vitro. To predict whether the generated molecules are active toward the biological target of interest, target prediction models (TPMs) were trained for all the tested targets (5-HT2A, Plasmodium falciparum and Staphylococcus aureus). We evaluated random forest, logistic regression, (deep) neural networks, and gradient boosting trees (GBT) as models with ECFP4 (extended connectivity fingerprint with a diameter of 4) as the molecular descriptor. We found that GBTs slightly outperformed all other models and used these as our virtual assay in all studies (AUC[5-HT2A] = 0.877, AUC[Staph. aur.] = 0.916). ECFP4 fingerprints were generated with CDK version 1.5.13. scikit-learn, XGBoost, and Keras were used as the machine learning libraries. For 5-HT2A and Plasmodium, molecules are considered as active for the TPM if their IC50 reported in ChEMBL is <100 nM, which translates to a pIC50 > 7, whereas for Staphylococcus, we used pMIC > 3. The chemical language model was trained on a SMILES file containing 1.4 million molecules from the ChEMBL database, which contains molecules and measured biological activity data. The SMILES strings of the molecules were canonicalized (which means finding a unique representation that is the same for isomorphic molecular graphs) before training with the CDK chemoinformatics library, yielding a SMILES file that contained one molecule per line. It has to be noted that ChEMBL contains many peptides, natural products with complex scaffolds, Michael acceptors, benzoquinones, hydroxylamines, hydrazines, etc., which is reflected in the generated structures (see below). This corresponds to 72 million individual characters, with a vocabulary size of 51 unique characters. 51 characters is only a subset of all SMILES symbols, since the molecules in ChEMBL do not contain many of the heavy elements. As we have to set the number of symbols as a hyperparameter during model construction, and the model can only learn the distribution over the symbols present in the training data, this implies that only molecules with these 51 SMILES symbols seen during training can be generated during sampling. The 5-HT2A, the Plasmodium falciparum, and the Staphylococcus aureus data sets were also obtained from ChEMBL. As these molecules were intended to be used in the rediscovery studies, they were removed from the training data before fitting the chemical language model. To evaluate the models for a test set T, and a set of molecules GN generated from the model by sampling, we report the ratio of reproduced molecules , and enrichment over random (EOR), which is defined as6where n = |GN ∩ T| is the number of reproduced molecules from T by sampling a set GN of |GN| = N molecules from the fine-tuned generative model, and m = |RM ∩ T| is the number of reproduced molecules from T by sampling a set RM of |RM| = M molecules from the generic, unbiased generative model trained only on the large data set. Intuitively, EOR indicates how much better the fine-tuned models work when compared to the general model. In this work, we address two points: First, we want to generate large sets of diverse molecules for virtual screening campaigns. Second, we want to generate smaller, focused libraries enriched with possibly active molecules for a specific target. For the first task, we can train a model on a large, general set of molecules to learn the SMILES grammar. Sampling from this model would generate sets of diverse, but unfocused molecules. To address the second task, and to obtain novel active drug molecules for a target of interest, we perform transfer learning: We select a small set of known actives for that target and we refit our pretrained chemical language model with this small data set. After each epoch, we sample from the model to generate novel actives. Furthermore, we investigate if the model actually benefits from transfer learning, by comparing it to a model trained from scratch on the small sets without pretraining. We employed a recurrent neural network with three stacked LSTM layers, each with 1024 dimensions, and each one followed by a dropout layer, with a dropout ratio of 0.2, to regularize the neural network. The model was trained until convergence, using a batch size of 128. The RNN was unrolled for 64 steps. It had 21.3 × 10 parameters. During training, we sampled a few molecules from the model every 1000 minibatches to inspect progress. Within a few 1000 steps, the model starts to output valid molecules (see Table 1). To generate novel molecules, 50,000,000 SMILES symbols were sampled from the model symbol-by-symbol. This corresponded to 976,327 lines, from which 97.7% were valid molecules after parsing with the CDK toolkit. Removing all molecules already seen during training yielded 864,880 structures. After filtering out duplicates, we obtained 847,955 novel molecules. A few generated molecules were randomly selected and depicted in Figure 4. The Supporting Information contains more structures. The created structures are not just formally valid but also mostly chemically reasonable. A few randomly selected, generated molecules. Ad = Adamantyl. In order to check if the de novo compounds could be considered as valid starting points for a drug discovery program, we applied the internal AstraZeneca filters. At AstraZeneca, this flagging system is used to determine if a compound is suitable to be part of the high-throughput screening collection (if flagged as “core” or “backup”) or should be restricted for particular use (flagged as “undesirable” since it contains one or several unwanted substructures, e.g., undesired reactive functional groups). The filters were applied to the generated set of 848 k molecules, and they flagged most of them, 640 k (75%), as either core or backup. Since the same ratio (75%) of core and backup compounds has been observed for the ChEMBL collection, we therefore conclude that the algorithm generates preponderantly valid screening molecules and faithfully reproduces the distribution of the training data. To determine whether the properties of the generated molecules match the properties of the training data from ChEMBL, we followed the procedure of Kolb: We computed several molecular properties, namely, molecular weight, BertzCT, the number of H-donors, H-acceptors, and rotatable bonds, logP, and total polar surface area for randomly selected subsets from both sets with the RDKit library version 2016.03.1. Then, we performed dimensionality reduction to 2D with t-SNE (t-distributed stochastic neighbor embedding, a technique analogous to PCA), which is shown in Figure 5. Both sets overlap almost completely, which indicates that the generated molecules very well recreate the properties of the training molecules. t-SNE projection of 7 physicochemical descriptors of random molecules from ChEMBL (blue) and molecules generated with the neural network trained on ChEMBL (green), to two unitless dimensions. The distributions of both sets overlap significantly. Furthermore, we analyzed the Bemis–Murcko scaffolds of the training molecules and the sampled molecules. Bemis–Murcko scaffolds contain the ring systems of a molecule and the moieties that link these ring systems, while removing any side chains. They represent the scaffold, or “core” of a molecule, which series of drug molecules often have in common. The number of common scaffolds in both sets divided by the union of all scaffolds in both sets (Jaccard index) is 0.12, which indicates that the language model does not just modify side chain substituents but also introduces modifications at the molecular core. To generate novel ligands for the 5-HT2A receptor, we first selected all molecules with pIC50 > 7 which were tested on 5-HT2A from ChEMBL (732 molecules), and then fine-tuned our pretrained chemical language model on this set. After each epoch, we sampled 100,000 chars, canonicalized the molecules, and removed any sampled molecules that were already contained in the training set. Following this, we evaluated the generated molecules of each round of retraining with our 5-HT2A target prediction model (TPM). In Figure 6, the ratio of molecules predicted to be active by the TPM after each round of fine-tuning is shown. Before fine-tuning (corresponding to epoch 0), the model generates almost exclusively inactive molecules. Already after 4 epochs of fine-tuning the model produced a set in which 50% of the molecules are predicted to be active. Epochs of fine-tuning vs ratio of actives. In order to assess the novelty of the de novo molecules generated with the fine-tuned model, a nearest neighbor similarity/diversity analysis has been conducted using a commonly used 2D fingerprint (ECFP4) based similarity method (Tanimoto index).Figure 7 shows the distribution of the nearest neighbor Tanimoto index generated by comparing all the novel molecules and the training molecules before and after n epochs of fine-tuning. For each bin, the white bars indicate the molecules generated from the unbiased, general model, while the darker bars indicate the molecules after several epochs of fine-tuning. Within the bins corresponding to lower similarity, the number of molecules decreases, while the bins of higher similarity get populated with increasing numbers of molecules. The plot thus shows that the model starts to output more and more similar molecules to the target-specific training set. Notably, after a few rounds of training not only are highly similar molecules produced but also molecules covering the whole range of similarity, indicating that our method could deliver not only close analogues but also new chemotypes or scaffold ideas to a drug discovery project. To have the best of both worlds, that is, diverse and focused molecules, we therefore suggest to sample after each epoch of retraining and not just after the final epoch. Nearest-neighbor Tanimoto similarity distribution of the generated molecules for 5-HT2A after n epochs of fine-tuning against the known actives. The generated molecules are distributed over the whole similarity range. Generated molecules with a medium similarity can be interesting for scaffold-hopping. Plasmodium falciparum is a parasite that causes the most dangerous form of malaria. To probe our model on this important target, we used a more challenging validation strategy. We wanted to investigate whether the model could also propose the same molecules that medicinal chemists chose to evaluate in published studies. To test this, first, the known actives against Plasmodium falciparum with a pIC50 > 8 were selected from ChEMBL (Table 2). Then, this set was split randomly into a training (1239 molecules) and a test set (1240 molecules). The chemical language model was then fine-tuned on the training set. 7500 molecules were sampled after each of the 20 epochs of refitting. EOR: Enrichment over random. This yielded 128,256 unique molecules. Interestingly, we found that our model was able to “redesign” 28% of the unseen molecules of the test set. In comparison to molecules sampled from the unspecific, untuned model, an enrichment over random (EOR) of 66.9 is obtained. With a smaller training set of 100 molecules, the model can still reproduce 7% of the test set, with an EOR of 19.0. To test the reliance on pIC50 we chose to use another cutoff of pIC50 > 9, and took 100 molecules in the training set and 1022 in the test set. 11% of the test set could be recreated, with an EOR of 35.7. To visually explore how the model populates chemical space, Figure 8 shows a t-SNE plot of the ECFP4 fingerprints of the test molecules and 2000 generated molecules that were predicted to be active by the target prediction model for Plasmodium falciparum. It indicates that the model has generated many similar molecules around the test examples. t-SNE plot of the pIC50 > 9 test set (blue) and the de novo molecules predicted to be active (green). The language model populates chemical space around the test molecules. To evaluate a different target, we furthermore conducted a series of experiments to reproduce known active molecules against Staphylococcus aureus. Here, we used actives with a pMIC > 3. MIC is the mean inhibitory concentration, the lowest concentration of a compound that prevents visible growth of a microorganism. As above, the actives were split into a training and a test set. However, here, the availability of the data allows larger test sets to be used. After fine-tuning on the training set of 1000 molecules (Table 3, entry 1), our model could retrieve 14% of the 6051 test molecules. When scaling down to a smaller training set of 50 molecules (the model gets trained on less than 1% of the data!), it can still reproduce 2.5% of the test set, and performs 21.6 times better than the unbiased model (Table 3, entry 2). Using a lower learning rate (0.0001, entry 3) for fine-tuning, which is often done in transfer learning, does not work as well as the standard learning rate (0.001, entry 2). We additionally examined whether the model benefits from transfer learning. When trained from scratch, the model performs much worse than the pretrained and subsequently fine-tuned model (see Figure 9 and Table 3, entry 4). Pretraining on the large data set is thus crucial to achieve good performance against Staphylococcus aureus. EOR: Enrichment over random. Fine-tuning learning rate = 10. No pretraining. 8 generate-test cycles. Different training strategies on the Staphylococcus aureus data set with 1000 training and 6051 test examples. Fine-tuning the pretrained model performs better than training from scratch (lower test loss [cross entropy] is better). The experiments we conducted so far are applicable if one already knows several actives. However, in drug discovery, one often does not have such a set to start with. Therefore, high throughput screenings are conducted to identify a few hits, which serve as a starting point for the typical cyclical drug discovery process: Molecules get designed, synthesized, and then tested in assays. Then, the best molecules are selected, and based on the gained knowledge new molecules are designed, which closes the cycle. Therefore, as a final challenge for our model, we simulated this cycle by iterating molecule generation (“synthesis”), selection of the best molecules with the machine learning based target prediction (“virtual assay”), and retraining the language model with the best molecules (“design”) with Staphylococcus aureus as the target. We thus do not use a set of known actives to start the structure generation procedure (see Figure 10). Scheme of our de novo design cycle. Molecules are generated by the chemical language model and then scored with the target prediction model (TPM). The inactives are filtered out, and the RNN is retrained. Here, the TPM is a machine learning model, but it could also be a robot conducting synthesis and biological assays, or a docking program. We started with 100,000 sampled molecules from the unbiased chemical language model. Then, using our target prediction model, we extracted the molecules classified as actives. After that, the RNN was fine-tuned for 5 epochs on the actives, sampling ≈10,000 molecules after each epoch. The resulting molecules were filtered with the target prediction model, and the new actives appended to the actives from the previous round, closing the loop. Already after 8 iterations, the model reproduced 416 of the 7001 test molecules from the previous task, which is 6% (Table 3, entry 5), and exhibits an EOR of 59.6. This EOR is higher than if the model is retrained directly on a set of 50 actives (entry 2). Additionally, we obtained 60,988 unique molecules that the target prediction model classified as active. This suggests that, in combination with a target prediction or scoring model, our model can at least simulate the complete de novo design cycle. Our results indicate that the general model trained on a large molecule set has learned the SMILES rules and can output valid, drug-like molecules, which resemble the training data. However, sampling from this model does not help much if we want to generate actives for a specific target: We would have to generate very large sets to find actives for that target among the diverse range of molecules the model creates, which is indicated by the high EOR scores in our experiments. When fine-tuned to a set of actives, the probability distribution over the molecules captured by our model is shifted toward molecules active toward our target. To study this, we compare the Levenshtein (string edit) distance of the generated SMILES to their nearest neighbors in the training set in Figure 11. The Levenshtein distance of, e.g., benzene c1ccccc1 and pyridine c1ccncc1 would be 1. Figure 11 shows that while the model often seems to have made small replacements in the underlying SMILES, in many cases it also made more complex modifications or even generated completely different SMILES. This is supported also by the distribution of the nearest neighbor fingerprint similarities of training and rediscovered molecules (ECFP4, Tanimoto, Figure 12). Many rediscovered molecules are in the medium similarity regime. Histogram of Levenshtein (string edit) distances of the SMILES of the reproduced molecules to their nearest neighbor in the training set (Staphylococcus aureus, model retrained on 50 actives). While in many cases the model makes changes of a few symbols in the SMILES, resembling the typical modifications applied when exploring series of compounds, the distribution of the distances indicates that the RNN also performs more complex changes by introducing larger moieties or generating molecules that are structurally different, but isofunctional to the training set. Violin plot of the nearest-neighbor ECFP4-Tanimoto similarity distribution of the 50 training molecules against the rediscovered molecules in Table 3, entry 2. The distribution suggests that the model has learned to make typical small functional group replacements, but can also reproduce molecules which are not too similar to the training data. Because we perform transfer learning, during fine-tuning, the model does not “forget” what it has learned. A plausible explanation why the model works is therefore that it can transfer the modifications that are regularly applied when series of molecules are studied, to the molecules it has seen during fine-tuning. In this work, we have shown that recurrent neural networks based on the long short-term memory (LSTM) can be applied to learn a statistical chemical language model. The model can generate large sets of novel molecules with similar physicochemical properties to the training molecules. This can be used to generate libraries for virtual screening. Furthermore, we demonstrated that the model performs transfer learning when fine-tuned to smaller sets of molecules active toward a specific biological target, which enables the creation of novel molecules with the desired activity. By iterating cycles of structure generation with the language model, scoring with a target prediction model (TPM) and retraining of the model with increasingly larger sets of highly scored molecules, we showed that we do not even need a set of known active molecules to start our procedure with, as the TPM could also be a docking program, or a robot conducting synthesis and biological testing. We see three main advantages of our method. First, it is conceptually orthogonal to established molecule generation approaches, as it learns a generative model for molecular structures. Second, our method is very simple to set up, to train, and to use; it can be adapted to different data sets without any modifications to the model architecture; and it does not depend on hand-encoded expert knowledge. Furthermore, it merges structure generation and optimization in one model. A weakness of our model is interpretability. In contrast, existing de novo design methods settled on virtual reactions to generate molecules, which has advantages as it minimizes the chance of obtaining “overfit”, weird molecules, and increases the chances to find synthesizable compounds. To extend our work, it is just a small step to cast molecule generation as a reinforcement learning problem, where the pretrained LSTM generator could be seen as a policy, which can be encouraged to create better molecules with a reward signal obtained from a target prediction model. In addition, different approaches for target prediction, for example, docking, could be evaluated. Deep learning is not a panacea, and we join Gawehn et al. in expressing “some healthy skepticism” regarding its application in drug discovery. Generating molecules that are almost right is not enough, because in chemistry, a miss is as good as a mile, and drug discovery is a “needle in the haystack” problem—in which also the needle looks like hay. Nevertheless, given that we have shown in this work that our model can rediscover those needles, and other recent developments, we believe that deep neural networks can be complementary to established approaches in drug discovery. The complexity of the problem certainly warrants the investigation of novel approaches. Eventually, success in the wet lab will determine if the new wave of neural networks will prevail.
|
PMC3629873
|
Preliminary evaluation of the CellFinder literature curation pipeline for gene expression in kidney cells and anatomical parts
|
Biomedical literature curation is the process of automatically and/or manually deriving knowledge from scientific publications and recording it into specialized databases for structured delivery to users. It is a slow, error-prone, complex, costly and, yet, highly important task. Previous experiences have proven that text mining can assist in its many phases, especially, in triage of relevant documents and extraction of named entities and biological events. Here, we present the curation pipeline of the CellFinder database, a repository of cell research, which includes data derived from literature curation and microarrays to identify cell types, cell lines, organs and so forth, and especially patterns in gene expression. The curation pipeline is based on freely available tools in all text mining steps, as well as the manual validation of extracted data. Preliminary results are presented for a data set of 2376 full texts from which >4500 gene expression events in cell or anatomical part have been extracted. Validation of half of this data resulted in a precision of ∼50% of the extracted data, which indicates that we are on the right track with our pipeline for the proposed task. However, evaluation of the methods shows that there is still room for improvement in the named-entity recognition and that a larger and more robust corpus is needed to achieve a better performance for event extraction. Database URL: http://www.cellfinder.org/Biomedical literature curation is the process of automatically and/or manually compiling biological data from scientific publications and making it available in a structured and comprehensive way. Databases that integrate information derived in some way from scientific publications include, for instance, model organism databases (1), protein–protein interactions (2) and gene–chemical–disease relationships (3). Typical literature curation workflows include the following steps (4): triage (selection of relevant publications), biological entities identification (e.g. genes/proteins, diseases, etc.), extraction of relationships (e.g. protein–protein interactions, gene expression, etc.), association of biological processes with experimental evidence, data validation and recoding into the database. Therefore, literature curation requires a careful reading of publications by domain experts, which is known to be a time-consuming task. Additionally, the increasing growth of available publications prevents a comprehensive manual curation of intended facts and previous studies show that it is not feasible (5). Recent advances in text mining methods have facilitated its application in most of the literature curation stages. Challenges have contributed to the improvement and availability of a variety of methods for named-entity prediction (6), and more specifically for gene/protein prediction and normalization (7, 8). Also binary relationships (9) and event extraction (10) have been improved, and its current performance allows its use on large scale projects (11). Finally, integrated ready-to-use workbenches have also been available, such as @Note (12), Argo (13), MyMiner (14) and Textpresso (15), although the performance and scalability to larger projects is still dubious for some of them. A comparison between some of them is found in this survey on annotation tools for the biomedical domain (16). Previous reports (17, 18) and experiments (19) have confirmed the feasibility of text mining to assist literature curation and recent surveys (4, 20) show that, indeed, it is already part of many biological databases workflows. For instance, text mining support is being explored for the triage stage in FlyBase (21), for curation of regulatory annotation in (22) and also in the AgBase (23), Biomolecular Interaction Network Database (BIND) (24), Immune Epitope Database (IEDB) (25) and The Comparative Toxicogenomics Database (CTD) (26) databases. Additionally, many solutions have been proposed for the CTD database during a recent collaborative task (27). Further, Textpresso has been widely used to prioritize document and for Gene Ontology (GO) terms (28) annotation in WormBase and The Arabidopsis Information Resource (TAIR) (29). Named-entity recognition has also been included in the curation workflow of Mouse Genome Informatics (MGI) (30) for gene/protein extraction, and in Xenbase (31) for gene and anatomy terms, for instance. Finally, few databases have tried automatic relationships extraction methods: protein phosphorylation information has been extracted using rule-based pattern templates (32), recreation of events has been carried out for the Human Protein Interaction Database (HHPID) database (33) and revalidation of relationships for the PharmGKB database (34). We present the first description of the curation pipeline for the CellFinder database (http://www.cellfinder.org/), a repository of cell research, which aims to integrate data derived from many sources, such as literature curation and microarray data. It is based on a novel ontology [Cell: Expression, Localization, Development, Anatomy (CELDA) (http://cellfinder.org/about/ontology)], which allows standardization and integration to other available ontologies on the cell and anatomy domains. Hence, the CellFinder platform provides a framework for comprehensive descriptions of human tissues, cells and commonly used model organisms on molecular and functional levels, in vivo and in vitro. The CellFinder pipeline for literature curation integrates state-of-art freely available tools for the document triage, recognition of a variety of entity types and extraction of biological processes. Curation is carried out for full text documents available at the PubMed Central Open Access (PMC OA) subset (http://www.ncbi.nlm.nih.gov/pmc/tools/openftlist/), and manual intervention from curators is currently only necessary for querying new documents for curation and validation of the derived biological processes. In both cases, web-based tools are being used, which allow their integration into the CellFinder web site. We are not aware of prior usage of available systems for the automatic extraction of biological events. For instance, Xenbase manually annotates gene expression events (31), whereas others databases use proprietary systems (34) or tools, which do not allow re-use for other domains (33). Our literature curation pipeline has been evaluated using a dataset on the kidney cell research. The kidney consists of >26 cell types, which arise and organize into several anatomical structures during a conserved developmental process (35). Kidney disease culminates from a common sclerotic pathway involving epithelial-mesenchymal transition, extracellular matrix remodeling and vascular changes (36). Multiple renal and non-renal (e.g. inflammatory) cell types are involved in these processes, with dynamic gene expression patterns and functions (37). Therefore, to identify relevant research describing cells and their interactions in normal and diseased kidney, we decided to include species-independent experimental and clinical data of renal disease and of kidney development in CellFinder. For the kidney cell use case, information is compiled about characterization of gene expression profiles in cells and other anatomical locations, such as tissues and organs. Hence, named-entity extraction is performed for genes, proteins, cell lines, cell types, tissues and organs. Gene expression events are then extracted between a gene/protein and a certain cell or anatomical part. The sentence below illustrates one such example (PMID 18989465): On the other hand, the podoplanin expression occurs in the differentiating odontoblasts and the expression is sustained in differentiated odontoblasts, indicating that odontoblasts have the strong ability to express podoplanin. On the other hand, the podoplanin expression occurs in the differentiating odontoblasts and the expression is sustained in differentiated odontoblasts, indicating that odontoblasts have the strong ability to express podoplanin. We are aware of only two previous publications, which report extraction of gene expression in anatomical locations from biomedical texts. OpenDMAP (38) uses Protégé and UIMA-based components, and it has been evaluated for three applications: protein transport, protein interactions and cell type-specific gene expression. OpenDMAP extract genes/proteins and cells using A Biomedical Named Entity Recognizer (ABNER) (39) and a short list of trigger words. Relationships between the triple gene-cell-trigger are identified based on manual pattern templates. It reports precision of 64% and recall of 16% from an evaluation of 324 NCBI’s GeneRIFs, which consists of short descriptions of gene functions. A more comprehensive study on the expression of genes in anatomical location was carried out in (40) with the Gene Expression Text Miner system. The work included extending 150 abstracts from the BioNLP corpus (41) with annotations for anatomical parts and cell lines, as well as relationships to the existing gene expression events. Genes/proteins were extracted using GNAT (42), anatomical part and cell line recognition was performed by Linnaeus (43) using 13 anatomical ontologies and one for cell lines. A list of expression triggers was manually built, and association between the entities is also rule-based. Evaluation on the extended 150 abstracts resulted in a precision of almost 60% and a recall of 24%. The next section will describe the CellFinder curation pipeline and the methods that are used in each stage. Results for the experiments performed for most of the steps are shown in the section ‘Results’ followed by discussion on the more important aspects of the pipeline in the section ‘Discussion and future work’. The curation pipeline for the CellFinder database includes the following steps (cf. Figure 1): triage of potential relevant documents, retrieval of full text, linguistic pre-processing, named-entity recognition, post-processing, relationship extraction, manual validation of the results and integration of gene expression events into the database. This section describes details on the methods used in each phase. Figure 1.Overview of the literature curation pipeline for the CellFinder database. It includes the following steps: triage of potential relevant documents, retrieval of full text, preprocessing (sentence splitting, tokenization and parsing), named-entity recognition (genes, proteins, cell lines, cell types, organs, tissues, expression triggers), gene expression events extraction, manual validation of the results and integration into the database. Automatic procedures are shown in red, whereas the manual ones are shown in blue. Overview of the literature curation pipeline for the CellFinder database. It includes the following steps: triage of potential relevant documents, retrieval of full text, preprocessing (sentence splitting, tokenization and parsing), named-entity recognition (genes, proteins, cell lines, cell types, organs, tissues, expression triggers), gene expression events extraction, manual validation of the results and integration into the database. Automatic procedures are shown in red, whereas the manual ones are shown in blue. Document triage is usually the first step in any literature curation workflow and consists of retrieving potential relevant publications for manual curation or for further processing by a text mining pipeline. In the CellFinder project, we aim to curate only full texts documents, which are available for text mining purposes, i.e. the ones included in the PMC OA subset. Although it is a much smaller collection than the whole Medline, this subset currently contains >200 000 documents. In our pipeline, document triage was performed by querying MedlineRanker (44), a machine learning based text categorization system. We have performed eight queries to MedlineRanker as follows: ‘kidney tubular epithelial EMT’, ‘kidney vascular endothelial interstitium’, ‘kidney glomerular basement membrane’, ‘kidney mesangial space podocyte’, ‘kidney development differentiation pronephros’, ‘kidney extra cellular matrix, ‘kidney regeneration mesenchymal precursor’ and ‘corticomedullary junction’. The search terms were aimed to identify cells, genes and structures that relate to cells contained in nephrons and tubules, such as epithelial cells, endothelial cells and podocytes, as well as cell changes associated with mesenchymal–epithelial transition (EMT) and fibrosis, changes in extracellular matrix and relevant proteins and in cells during kidney development, such as mesenchymal precursor cells. Each query retrieved a list of 10 000 (MedlineRanker’s cut-off) potential PMC relevant documents, including many repeated documents found across lists. After a post-processing step, which included verification on whether documents were part of the PMC OA subset and exclusion of repeated entries, a list of 2376 documents was derived. Documents were retrieved from PMC and were processed through our text mining pipeline. Full texts documents were first split by sentences using the OpenNLP toolkit (http://opennlp.apache.org/) and then parsed by the Brown Laboratory for Linguistic Information Processing (BLLIP) parser (https://github.com/dmcc/bllip-parserV) (45) (also known as McClosky-Charniak parser). Part-of-speech tags, tokenization and full parsing were derived from the BLLIP parser output. Dependency trees were built using the Stanford parser (http://nlp.stanford.edu/software/lex-parser.shtml). Part-of-speech, tokenization and parsing information are only necessary for the gene expression extraction (cf. ‘Event Extraction’ below). Named-entity recognition has been performed for five entity types: genes/proteins, cell lines, cell types, anatomical parts and gene expression triggers. Extraction is based on available state-of-art systems and dictionary or ontology-based approaches, without any adaption nor retraining. Methods are similar to the ones investigated in previous experiments performed with the CellFinder corpus (46). To enable data integration into the CellFinder database, all extracted mentions must be normalized to any of the ontologies or terminologies currently supported by our database: Cell Ontology (CL) (47), Cell Line Ontology (CLO) (48), EHDAA2 (49), Experimental Factor Ontology (EFO) (50), Foundational Model of Anatomy (FMA) (51), GO (52), Adult Mouse Anatomy (MA) (53) and Uberon (54). We identify genes using GNAT (42), a system for extraction and normalization of gene and protein mentions. GNAT assigns confidence scores (up to 1.0) to the gene/protein candidates. Based on previous experiments (46), we have decided for a threshold score of 0.25 for filtering out potentially wrong gene/protein predictions. GNAT provides identifiers for all gene mentions with respect to the EntrezGene database (55). Cell lines are recognized based on the version 4.0 of Cellosaurus (ftp://ftp.nextprot.org/pub/current_release/controlled_vocabularies/ cellosaurus.txt), a manually curated vocabulary of cell lines provided by the Swiss Institute of Bioinformatics. Synonyms from Cellosaurus were automatically expanded according to space and hyphens, such as ‘BSF-1’, ‘BSF 1’ and ‘BSF1’, resulting in a list of >41 000 synonyms for 15 245 registered cell lines. Matching of the derived list of synonyms and the full texts is performed by Linnaeus (43). For the recognition of cell types and anatomical parts, we use Metamap (56), a system for Unified Medical Language System (UMLS) concept extraction. We configured Metamap to generate acronym variants and restricted results by the following semantic types: ‘Cell’ for cell types and ‘Anatomical Structure’, ‘Body Location or Region’, ‘Body Part, Organ or Organ Component’, ‘Body Space or Junction’, ‘Body Substance’, ‘Body System’, ‘Embryonic Structure’, ‘Fully Formed Anatomical Structure’ and ‘Tissue’ for anatomical parts. Metamap uses natural language processing techniques for breaking the text into phrases and further match them to UMLS concepts. From the potential matches returned by Metamap, we record not only the ones with highest score but also those that have the longest matching with the respective phrase. Cell types have also been extracted using an ontology-based approach in which synonyms from the CL are matched against the full texts. It consists on a list of 2786 cell types from 1491 terms and matching is again performed by Linnaeus (43). Finally, triggers are extracted based on a list of 509 expression triggers, which was built manually. Terms from the list are matched against the full text using Lingpipe (http://alias-i.com/lingpipe/). Metamap includes a step for acronym resolution, which returns a list of the pairs of abbreviations and long forms found as equivalent. However, Metamap sometimes recognizes the plural of some abbreviations but not the singular form or it does not return some abbreviations as a mention, but only the long forms. For instance, for cell types, Metamap recognizes ‘hESCs’ as an acronym for ‘human embryonic stem cells’, but not its singular form ‘hESC’. Further, although it lists the pair ‘hESCs’ and ‘human embryonic stem cells’ as being equivalent, only the long form is returned as a mention. Based on the list of pairs of abbreviations and long forms returned by Metamap, we try to match missed abbreviations and singular forms using Lingpipe. Metamap returns annotations with regard to Concept Unique Identifier (CUI) terms, the original UMLS identifiers. Whenever available, we map them to FMA and GO terms using mappings available at the UMLS database. CUI terms are also mapped to other ontologies and terminologies supported by UMLS, but not by CellFinder, such as the CRISP Thesaurus (http://www.nlm.nih.gov/research/umls/sourcereleasedocs/current/CSP/).To increase the recall of anatomical terms, we mapped UMLS CUI terms to CRISP terms [using mappings available at BioPortal (57)], and then further to other ontologies supported by CellFinder (e.g. CL, CLO, EHDAA2, MA, Uberon). Annotations returned by Metamap, which could not be automatically mapped to any supported ontology, are not removed, as identifiers could still be provided manually before integration of the data into the CellFinder database (not yet supported in the current curation workflow). Blacklists of manually curated mentions and identifiers are used for filtering out potential false predictions for all four entity types. This list was manually built based on the analysis of wrongly extracted annotations from the two corpora used for evaluation (cf. section ‘Results’). The list of mentions contains only one entry for cell line (‘FL’), 39 for anatomical parts (e.g. ‘organism’, ‘tissue’ and ‘analysis’), 31 entries for cell types (e.g. ‘cell’ and ‘stem cell’) and 79 entries for genes/proteins (e.g. ‘anti’, ‘repair’, ‘or in’). The list of identifiers include those which refer to broad concepts such as ‘cell’ (FMA:68646) or ‘tissue’ (FMA:9637). We filter out extracted mentions associated to any of the identifiers in this list. Results from sentence splitting, tokenization, part-of-speech tagging, parsing, dependency tags and named entities are integrated into the so-called ‘Interaction XML’ file format (https://github.com/jbjorne/TEES/wiki/TEES-Overview) (58) used by the Turku Event Extraction System (TEES) (59). TEES is an event extraction system, which uses multiclass Support Vector Machine on a rich graph-based feature set for trigger, edge and negation detection. Despite recent improvement of relation extraction methods (10), TEES seems to be the only available system suitable to be re-trained with novel corpora from any domain without the need of performing changes in its source code. We trained TEES in a gold-standard set of 20 full text annotated documents, 10 on human embryonic stem cell research (hereafter called CF-hESC), whose entities annotations have been previously published (46) and a new set of 10 full texts documents on kidney stem cell research (hereafter called CF-Kidney). Both corpora have been manually annotated with the five entity types (gene/proteins, cell lines, cell types, anatomical parts, expression triggers) and gene expression events (cf. example in Figure 2). These events are composed of a trigger, which is always linked to two arguments, a gene/protein (hereafter called ‘Gene’ argument) and a cell line, cell type or anatomical part (hereafter called ‘Cell’ argument). We split both corpora into three parts (training, development and test) and perform experiments using one corpus or a combination of both for training. Details on the corpora are shown in Table 1. Figure 2.Examples of gene expression events for the kidney stem cell corpus (PMID 17389645, PMCID PMC1885650). Each expression trigger (dark yellow) is always related with only one gene/protein (in blue) and only one cell (in yellow) or anatomical part (in red). However, the corpus was also annotated with entities, which do not take part in any event. Visualization of the corpus was provided by Brat annotation tool (60). Table 1.Statistics on the corporaFeaturesCF-hESCCF-KidneyTrainingDevelopmentTestTrainingDevelopmentTestDocuments622622Sentences13792595391578618383Sentences with entities9441633021344527314Sentences with events1472640240210122Entities41585831260483434431748 Genes/proteins126416335514401338782 Cell lines198721411181 Cell types155617952491725972 Anatomical parts92113717321161380617 Expression triggers2193267350458276Relationships944160390114414041320 Expression-Gene/protein47284195572702660 Expression-CellLine13636145 Expression-CellType4355612241139886 Expression-anatomy241837147299574Information is shown for the training, development and test data sets of the CF-hESC and CF-Kidney data sets. It includes number of documents, sentences, sentences with entities and sentences with events. Number of annotations is presented by entity type, and the number of events also shown according to the entities participating in the relationships. Examples of gene expression events for the kidney stem cell corpus (PMID 17389645, PMCID PMC1885650). Each expression trigger (dark yellow) is always related with only one gene/protein (in blue) and only one cell (in yellow) or anatomical part (in red). However, the corpus was also annotated with entities, which do not take part in any event. Visualization of the corpus was provided by Brat annotation tool (60). Statistics on the corpora Information is shown for the training, development and test data sets of the CF-hESC and CF-Kidney data sets. It includes number of documents, sentences, sentences with entities and sentences with events. Number of annotations is presented by entity type, and the number of events also shown according to the entities participating in the relationships. TEES receives the Interaction XML file as input and returns a new XML file, which includes predictions for the ‘Cell’ and ‘Gene’ relationships. The later are subsequently combined to compose complete gene expression events by checking the presence of both a ‘Gene’ and a ‘Cell’ relationship linked to the same trigger. TEES relationships are restricted to entities present in the same sentence; therefore, the same restriction is valid for all derived events. We applied TEES-trained models on the kidney cell data set of 2376 full texts. Results were manually validated using Bionotate (61), a collaborative open-source text annotation tool. Bionotate presents a snippet of text along with annotated entities, a question, and a list of possible answers. Curators were instructed to give one answer per snippet, and although Bionotate allows changing the span of the named entities, for this experiment, curators were asked only to answer the question. Bionotate selects snippets randomly among all those included in its repository. A snippet is no longer presented to the user when a certain number of agreements (equal answers) have been reached. For this experiment, one answer from any of our expert curators suffices. We have converted the output from TEES event extractor system to the XML format of the Bionotate. Snippets are composed of the sentence in which the event occurs and the two previous and subsequent sentences, for a better understanding of the context (cf. Figure 3). Additionally, a link to the respective PubMed entry is provided, in case those curators needed to check the abstract or full text of the publication before answering the question. The questions assessed whether there was a gene expression event taking place in the snippet, including its negation, whether the named entities were correctly recognized or if the publication was relevant for the kidney cell research. This resulted in the following possible answers: Yes, an event is taking place and all entities are correct. Yes, but the text says the gene expression is NOT taking place. No, no event is taking place although all entities are correct. No, this is not a gene expression trigger. No, this is not a gene. No, this is not a cell or anatomical part. No, both gene and cell or anatomical part are incorrect. No, the snippet (publication) does not seem to be relevant for CellFinder. Figure 3.Screen-shot of Bionotate configured for the validation of the gene expression events. Three named-entities are always pre-annotated: a trigger (in green), a gene (in blue) and a cell line, cell type or anatomical part (in red). The answers assess whether the biological event is taking place, its negation, the accuracy of the named-entity recognition and the relevancy of the publication from where the snippet was derived. Screen-shot of Bionotate configured for the validation of the gene expression events. Three named-entities are always pre-annotated: a trigger (in green), a gene (in blue) and a cell line, cell type or anatomical part (in red). The answers assess whether the biological event is taking place, its negation, the accuracy of the named-entity recognition and the relevancy of the publication from where the snippet was derived. In this section, we describe the evaluation performed for the methods used in the various stages of the text mining pipeline. We also present an overview of the data, which have been extracted by our curators with the help of the pipeline. The triage phase has not been directly evaluated, except for the answer number 8 during the manual validation of results (cf. ‘Manual validation’ in this section). Evaluation of the named-entity recognition and event extraction will be shown in terms of precision (P), recall (R) and f-score (F). Precision represents the ratio of the correct predictions of a particular system among all the returned ones. On the other hand, recall corresponds to the ratio of gold-standard annotations, which were actually returned by the system. Finally, the f-score is a harmonic average of both measures and shows the overall performance of a system. During the pre-processing step, sentence splitting in all 2376 full text documents resulted in a total of 581 350 sentences. Parsing and dependency tags conversion was successfully for 578 572 of them. The parsing information is only used by the TEES system (cf. ‘Event extraction’ in section ‘Methods and materials’), which means that although named-entity recognition was carried out in all sentences, only those correctly parsed ones were analyzed by TEES. Named-entity extraction was evaluated on the development and test gold-standard documents belonging to the human embryonic and kidney stem cell research (cf. Table 1), but only the development data sets were used for further improvements of methods, such as trigger list or blacklist construction and error analysis (cf. section ‘Discussion and future work’). Table 2 shows the evaluation of each entity type for both corpora. The ‘Exact’ evaluation assesses annotations, which matched regarding span and entity type, whereas ‘Overlap+Type’ allowed overlapping spans for annotations of the same type and ‘Overlap’ let annotations to have different types. The latter is particularly helpful regarding overlapping annotations between cell lines, cell types and anatomical parts, as any of these entity types corresponds to the same argument ‘Cell’ in the gene expression event (cf. Figure 2). Table 2.Evaluation of the automatic named-entity recognition on the CF- hESC and CF-Kidney corporaCorporaMatchEntity types (recall/F-score)GenesC. linesC. typesAnatomyExpressionCF-hESCDevelopmentEx.0.61/0.540.68/0.610.14/0.150.34/0.340.72/0.15OT0.75/0.650.94/0.850.62/0.660.48/0.450.91/0.19Ov.0.82/0.690.94/0.810.70/0.730.72/0.620.97/0.20TestEx.0.68/0.650.40/0.490.25/0.280.30/0.250.45/0.08OT0.76/0.720.58/0.650.58/0.650.43/0.350.54/0.09Ov.0.77/0.710.61/0.690.77/0.820.81/0.710.55/0.10CF-KidneyDevelopmentEx.0.34/0.451.00/0.330.17/0.260.69/0.750.68/0.43OT0.35/0.461.00/0.330.18/0.270.88/0.870.69/0.43Ov.0.46/0.561.00/0.340.77/0.800.90/0.890.76/0.47TestEx.0.69/0.761.00/0.330.89/0.860.67/0.740.80/0.42OT0.70/0.771.00/0.330.93/0.890.69/0.760.80/0.42Ov.0.70/0.771.00/0.330.94/0.910.72/0.770.81/0.42Results are shown for the development and test data sets in the format recall/F-score. Matching is evaluated regarding same span and entity type (Ex.), overlapping span and same type (OT) and overlapping span of any entity type (Ov.). Evaluation of the automatic named-entity recognition on the CF- hESC and CF-Kidney corpora Results are shown for the development and test data sets in the format recall/F-score. Matching is evaluated regarding same span and entity type (Ex.), overlapping span and same type (OT) and overlapping span of any entity type (Ov.). Recall is particularly low for genes/proteins in the development data set of the CF-Kidney corpus owing to a high number of annotations from a few genes/proteins, which have been missed by the system: ‘Gata3’ (155), ‘Ret’ (97) and ‘EpCAM’ (83). Some of these were found by GNAT but with a recall lower than the threshold we have considered. Cell lines are very rare in the CF-Kidney corpus, and the eight identical cell lines of the development data set and the only one of the test data set were correctly extracted (thus recall 1.0). Finally, recall is also particularly low for cell types in the development data set, even when allowing overlaps. Indeed, there is a great variety of cell types (>100), which could not be recognized, especially cell types, which in fact represent gene expressions events, such as ‘NCAM + NTRK2 + cells’ or ‘Gata3−/Ret− cells’. The ontology mapping post-processing step could automatically map a total of 171 (CF-hESC corpus) and 121 (CF-Kidney corpus) additional annotations to an identifier from any of the ontologies supported in CellFinder. They had been previously extracted by Metamap, but they were associated only to the UMLS CUI identifier. However, 1342 (33%) and 961 (16%) of the extracted annotations, respectively, remain assigned only to the UMLS CUI identifier, with respect to the total number of cell types and anatomical parts. The acronym resolution procedure has resulted in a slight increase in recall for cell types and anatomy, without loss of f-score (result not shown). For instance, recall for cell types in the CF-hESC corpus increased from 64 to 70% (result not shown) owing to the extraction of acronyms such as ‘MEF’ (mouse embryonic fibroblast) or ‘EB’ (embryoid body), which have not been previously returned by Metamap. Finally, blacklist filtering of terms also allowed a modest improvement of precision for both corpora (result not shown). For instance, precision for genes/proteins in the CF-hESC corpus increased from 43 to 50% (result not shown) owing to filtering out annotations such as ‘or in’ or ‘membrane’, which had been recognized by GNAT and genes or proteins. The named-entity extraction methods were run on the 2376 full texts and resulted in a total of >2 200 000 mentions for all five entity types. Details on the extracted annotations are presented in Table 3, such as the number of mentions for each entity type, distinct text spans and distinct identifiers. Table 3.Statistics on the extracted named entitiesAnnotationsGenesC. linesC. typesAnatomyExpressionDistinct mentions702 82981 074183 820565 860681 370Distinct spans34 2221825914214 874892Distinct ids34 35311 87511504300For each entity type, the number of annotations, distinct spans and identifiers is shown. Sometimes more than one identifier is assigned to a mention, therefore their high number. Trigger words (Expression) are not normalized to any ontology. Statistics on the extracted named entities For each entity type, the number of annotations, distinct spans and identifiers is shown. Sometimes more than one identifier is assigned to a mention, therefore their high number. Trigger words (Expression) are not normalized to any ontology. To extract gene expression events, we investigated training TEES on three models: CF-hESC corpus (6 full text documents), CF-Kidney corpus (6 full text documents) and a mix of both (12 full text documents) (hereafter called CF-Both). Input to TEES should include three data sets: training, development and test. During the training step, TEES automatically configures its parameters using the development data set and presents an evaluation of its own for the test set. Details on the performance of the relationship extraction is shown in Table 4 for the three training models, as well as for the complete events further performed by the authors. This is the performance of TEES without the influence of the named-entity recognition predictions of our text mining pipeline, as only gold-standard documents are used during the training step. Recall of the relationships range from 60 to 70% while precision is also good, from 60 to almost 90%. Both the recall and precision drop when considering the complete events, and recall is not always as high as the argument with the lower recall. This is due to the fact that TEES predicts the ‘Cell’ and ‘Gene’ relationships independently, and many of them are not associated to the same trigger. Table 4.Evaluation of TEES during trainingData setsRelationshipDevelopmentTestPRFPRFCF-hESCCell0.860.560.680.770.450.57Gene0.910.680.780.820.900.86Event0.600.350.440.380.530.44CF-KidneyCell0.710.500.590.620.680.65Gene0.600.820.690.730.750.74Event0.170.490.250.120.560.20CF-BothCell0.770.550.650.690.640.67Gene0.670.810.730.690.840.76Event0.550.480.510.500.560.53Evaluation is shown for the ‘Cell’ and ‘Gene’ relationships and for the development and test data sets, as described in Table 1. The complete events derived from a ‘Cell’ and a ‘Gene’ argument associated to the same trigger are also shown. For each training run, evaluation is carried out on the corresponding development and test data sets, i.e. two documents for each single corpus (CF-hESC and CF-Kidney) and four documents when training on the joined corpus (CF-Both). Predictions were performed over the gold-standard named-entity annotations. ‘P’ refers to ‘Precision’, ‘R’ to ‘Recall’ and ‘F’ to ‘F-score’. Evaluation of TEES during training Evaluation is shown for the ‘Cell’ and ‘Gene’ relationships and for the development and test data sets, as described in Table 1. The complete events derived from a ‘Cell’ and a ‘Gene’ argument associated to the same trigger are also shown. For each training run, evaluation is carried out on the corresponding development and test data sets, i.e. two documents for each single corpus (CF-hESC and CF-Kidney) and four documents when training on the joined corpus (CF-Both). Predictions were performed over the gold-standard named-entity annotations. ‘P’ refers to ‘Precision’, ‘R’ to ‘Recall’ and ‘F’ to ‘F-score’. In Table 5, we show the performance of TEES relationship extraction when using the predictions obtained in the named-entity recognition step, as well as gene expression events derived from the binary relationships. This is the final performance of our text mining pipeline for the extraction of gene expression events on cell and anatomical locations. Additionally, we include the performance for the prediction of the triplets gene-cell-trigger, which represent every possible combination of annotations from these three arguments in the same sentence. Therefore, it represents the higher possible recall for the event extraction provided the predicted named entities. Table 5.Evaluation of gene expression extractionData setsRelationship/EventDevelopmentTestPredictionsPRFPRFCF-hESCCell0.430.060.100.760.330.4614 551Gene0.350.220.270.760.790.77112 372Events0.500.080.140.270.050.084280Triplets0.060.510.100.050.350.09CF-KidneyCell0.440.020.050.520.570.55109 934Gene0.620.060.100.770.690.735520Event115Triplets0.020.190.040.020.280.05CF-BothCell1.00.010.020.700.640.6769 079Gene0.330.010.010.690.840.763792Event178Triplets0.020.220.040.030.300.05We have trained the TEES system on three data sets: CF-hESC, CF-Kidney and CF-Both. Results for the ‘Cell’ and ‘Gene’ relationships were provided by TEES during processing of the documents. Performance for complete events is evaluated allowing overlapping matches for entity spans, but with equality of entity types and argument types. The triplets correspond to every possible combination of the triggers, genes/proteins, cells or anatomical parts in the same sentence, i.e. the highest possible recall for any relationship extraction system provided the predictions for the entities. The ‘Pred.’ column presents the number of relationships or complete events, which have been extracted from the 2376 full texts on kidney research when using each of the training models. ‘P’ refers to ‘Precision’, ‘R’ to ‘Recall’ and ‘F’ to ‘F-score’. Evaluation of gene expression extraction We have trained the TEES system on three data sets: CF-hESC, CF-Kidney and CF-Both. Results for the ‘Cell’ and ‘Gene’ relationships were provided by TEES during processing of the documents. Performance for complete events is evaluated allowing overlapping matches for entity spans, but with equality of entity types and argument types. The triplets correspond to every possible combination of the triggers, genes/proteins, cells or anatomical parts in the same sentence, i.e. the highest possible recall for any relationship extraction system provided the predictions for the entities. The ‘Pred.’ column presents the number of relationships or complete events, which have been extracted from the 2376 full texts on kidney research when using each of the training models. ‘P’ refers to ‘Precision’, ‘R’ to ‘Recall’ and ‘F’ to ‘F-score’. Results are shown using the approximate span matching, i.e. for each argument, overlapping matches are allowed, but entities should have the same type as well as equality of the argument type (‘Cell’ or ‘Gene’). For the development data set and when using the CF-Kidney corpus for training TEES, whether alone or together with the CF-hESC corpus, no complete event was extracted. This is due to two reasons: (i) the low recall of genes/proteins and cell types for the CF-Kidney corpus (cf. Table 2, evaluation OT) and (ii) the inability of the CF-Kidney model to extract events from documents from other domains, i.e with different cell type nomenclature. Indeed, no gene expression events have been extracted from the two development documents of the CF-hESC corpus included in the development data set of the CF-Both corpus. This probably due to the high complexity and variability of the cell types in the CF-Kidney corpus, with examples such as ‘NCAM− cell’ or ‘EpCAM−NCAM−NTRK2+ cells’. We have run TEES using the three models (CF-hESC, CF-Kidney and CF- Both) on the 2376 documents and the named-entities previously extracted (cf. Table 3). We have obtained only 115 and 178 gene expression events for the CF-Kidney and CF-Both models, respectively, whereas the CF-hESC model retrieved 4280 events. The latter were derived from almost 127 000 binary relationships, i.e. the complete events correspond to only 14% of the original extracted relationships. The last column of Table 5 summarizes the number of relationships and derived events, which have been obtained using each training model. The gene expression events obtained with the three models were converted to the Bionotate XML format, and snippets were loaded into its repository. Curators have manually validated 2741 snippets, which contained events predicted by the three distinct models. Results are summarized in Table 6. The validated data, one file per snippet in the Bionotate’s XML format, is available for download at the CellFinder web site (http://cellfinder.org/about/annotation/). Table 6.Evaluation of the gene expression snippets in BionotateAnswersCF-hESCCF-KidneyCF-BothTotalNo. snippets%No. snippets%No. snippets%No. snippets%1. Yes120449.13429.563.3124445.42. Yes (negation)471.932.600501.83. No (but entities correct)2189.087.010.62278.34. No (trigger wrong)1948.02824.37843.830011.05. No (gene wrong)34614.1119.663.436313.26. No (cell/anatomy wrong)2078.52622.695.12428.87. No (gene/cell/anatomy wrong)552.243.510.6602.28. No (irrelevant document)1777.210.97743.22559.3Total24481001151001781002741100A total of 2741 snippets (gene expression events) were validated. These events were predicted by the three models used for training TEES event extraction system. Percentages for each answer are also shown. Evaluation of the gene expression snippets in Bionotate A total of 2741 snippets (gene expression events) were validated. These events were predicted by the three models used for training TEES event extraction system. Percentages for each answer are also shown. Validation for the events extracted using the CF-hESC model, the best performing one according to the evaluation and the number of predictions, can be summarized as follows. About 51% (answers 1 and 2) of the gene expression events have been extracted correctly, as well as the participating entities. This includes both positive and negative statements of gene expression in cell in anatomical parts. Exactly 17% (answers 3 and 4) of the snippets described processes not related to gene expression, although the gene, cell or anatomy were correctly recognized. Almost 25% (answers 5, 6 and 7) of the extracted events contained a wrong identified gene/protein, cell/anatomy or both of them, which means that precision was higher than the average for the named-entity recognition (cf. Table 2). Finally, 7.2% of the snippets turned out to belong to documents, which are irrelevant to the kidney cell domain, which gives a hint on the performance of the triage step. We have described our preliminary text mining pipeline for the extraction of five entity types and gene expression events. In this section, we discuss the most important results derived from this first experiment with our text mining curation pipeline. In the named-entity recognition step, we have considered only state-of-art and freely available tools, and we did not train specific systems with the gold-standard corpora discussed here. Results for entity extraction are in-line with previous published ones (46), although data sets are different and, therefore, results are not directly comparable. A high recall is preferable over a high precision, as events cannot be predicted if the participating entities have not been previously extracted. On the other hand, a high number of wrong predictions slow down the validation process, and therefore, a balance between precision and recall (given by the f-score) is also desirable. Provided the still low recall for some entities, and the consequent low recall of the event extraction, future work should still focus on the improvement of the named-entity prediction. Regarding genes/proteins extraction, most of the missing annotations could have been recognized by GNAT if we had used a lower threshold. Other tools could also be combined with GNAT, such as GeneTUKit (62) or BANNER (63). Additionally, use of domain-specific post-processing, such as ‘whitelists’ of genes/proteins, would certainly help, and future work will concentrate on these two approaches. Recall for genes/proteins increases considerably for both development data sets when allowing overlaps and an improvement is also perceived when type equality is relieved, which shows that some genes overlap with some cells names or anatomical parts, such as ‘C34’ (a gene) and ‘C34 cell’ (a cell type). Cell lines are not as common as cell types in our corpora, specially in the CF-Kidney corpus where this entity type is almost non-existent (cf Table 1). However, it plays an important role in the cell research, and scientific literature reports many gene expression events, which take place in cell cultures. Restricting our evaluation to the CF-hESC corpus, recall varies from 60 to >90% when allowing overlapping spans (cf. Table 2), but it is still not satisfactory, and dictionary-based methods might not be sufficient. Missing annotations for cell lines are mostly due to the absence of the synonym in any of the available thesaurus or ontologies, such as ‘SD56’, which is not included in Cellosaurus. Thus, future work will include training a machine learning system for cell line recognition, including annotation of additional gold-standard documents. Improvement of the event extraction starts with the improvement of the recall for the named entities. Performance of cell types and anatomical parts are rather variable. A good recall is usually obtained when releasing equality of types, and further experiments should consider unifying the cell types and anatomical parts in our corpora. If fact, previous studies on the CF-hESC corpus show that inter-annotator agreement for these entity types was low (46). Overlaps between cell types and anatomical parts should not be a problem for the gene expression event extraction, as both entity types takes part in the ‘Cell’ argument. Cell types were sometimes poorly recognized for the CF-Kidney data set, owing to the high variability of the nomenclature and the presence of gene expression in its contents, such as ‘NCAM+NTRK2+ cells’ or ‘Gata3−/Ret− cells’. Thus, improvements on cell type extraction should also focus on training machine learning algorithms. Mapping cell types with such a pattern to an identifier is also a challenge, as these terms are not included in any available ontology. The prior identification of the original cell type in which the gene is being expressed can help in the normalization of these cells, an information that is usually present in the text, although not always in the same sentence. Expression triggers are extracted based on a manually curated list, which assures a high recall. Low recall, such as the ones for the development data set of the CF-Kidney corpus are due to unusual trigger words, such as ‘-’ (negative expression), ‘dim’ and ‘bright’. We obtained the gene expression events using the TEES edge detection module, which extracted relationships between expression triggers and a gene/protein, cell or anatomy. TEES allows training the system with novel corpora, and during the training step, examples are generated for all combinations of entities provided in the training corpus. Therefore, a few relationships returned by TEES are related to the wrong entity type. For instance, it extracts some ‘Gene’ arguments associated to cells or anatomical parts and some ‘Cell’ arguments related to genes, although no such examples can be found in any of our gold-standard corpora. TEES extracts the relationships independently. Therefore, the recall of the binary relationships does not correspond to the recall of the complete gene expression event. Future work on event extraction will also include trying additional event extraction systems, such as (64, 65). Use of more annotated documents might also improve the event extraction. Further experiments can also be performed using available corpora, such as the set of annotated abstracts of the Gene Expression Text Miner corpus (40). Additionally, a careful analysis of the wrongly extracted events returned by TEES when using gold-standard annotations (cf. low precision for CF-Kidney corpus in Table 4) could reveal inconsistencies in the manual annotations in our corpora. To avoid huge differences between development and test results, a cross-validation could have been investigated. In summary, a cross-validation in a larger and more robust corpus could provide more stable results. Nevertheless, these preliminaries results on extraction of gene expression in cells and anatomical parts are certainly interesting for the many groups working on event extraction, as this is one of the first curation experiment to use a event extraction system, which had not been developed by the authors. Additionally, it is probably the first external evaluation of TEES on a new corpus, one of the very few event extraction systems available to the public. Finally, the use of corpora from two distinct cell research domains shows how large differences in results are dependent on the corpus and the corresponding learned model. Processing of the data set of 2376 full text documents for kidney cell research resulted in a high number of entities but apparently a low number of extracted events. However, recall is unknown, as well as the number of publications, which described expression of genes in cells and anatomical parts for the kidney cell research. The number of correct gene expression events is certainly low compared with the number of processed documents, but number of irrelevant publications in our collection is also unknown and could be higher than 6%, as reported by answer number 8 of the validation (cf. Table 6). Next event extraction tasks will involve recognition of additional relationships, such as identifying the cell type or tissue from which a certain cell line was derived. Future work will also include additional biological processes, such as cell differentiation. These relationships have already been annotated in the two gold-standard corpora discussed here and involve the same entities whose recognition is already included in our pipeline. Manual validation of 2741 snippets reported that half of them contained correctly recognized entities and gene expression events, which is in line with the precision of TEES shown in Table 5. Curators reported that most mistakes concentrated on incomplete extraction of genes/proteins and cell types, such as the recognition of ‘TGF’ instead of ‘TGF-beta’. Feedback from the validation will help to improve both recall and precision for the named-entity recognition by adding more terms to the blacklists (potential wrong predictions) and by creating ‘whitelists’ (potential missing annotations). Curators reported a positive first experience with Bionotate, although changes in visual interface, short-cuts and functional features have been suggested as future work. Next experiments will also focus on the validation of the identifiers, which were automatically assigned during the named-entity recognition, as well as allowing curators to change the span of the pre-annotated entities, a feature already supported by Bionotate. Validation of the normalized identifiers is an important step before final integration of the results into the CellFinder database. Version 2.0 of Bionotate (66) supports this functionality and will certainly be considered for integration in our pipeline. We presented here our preliminary results for the text mining pipeline for curation of gene expression events in cells in anatomical parts for the CellFinder database. Our pipeline relies only on open-source or freely available tools, and evaluation for each stage has been carried out based on gold-standard corpora. We are not aware of previous database curation pipelines where text mining methods have been used in all of the following stages: triage, named-entity recognition and event extraction. We performed named-entity extraction extraction for genes/proteins, cell lines, cell types, tissues, organs and gene expression triggers. Gene expression events were extracted using machine learning algorithms trained on manually annotated corpora from two domains, human embryonic stem cells and kidney cell research. Results for both the name-entity recognition and event extraction steps are promising, although improvements are still necessary to achieve a higher recall and precision. The text mining pipeline has been used to process 2376 full texts documents on kidney cell research and resulted in a total of >60 000 distinct entities and >4500 gene expression events. Half of the events have been manually validated by experts, and about half of them were classified as describing a gene expression taking place in a cell or anatomical part.
|
PMC11334037
|
Insulin-like growth factor 2 targets IGF1R signaling transduction to facilitate metastasis and imatinib resistance in gastrointestinal stromal tumors
|
Gastrointestinal stromal tumors (GISTs) are typical gastrointestinal tract neoplasms. Imatinib is the first-line therapy for GIST patients. Drug resistance limits the long-term effectiveness of imatinib. The regulatory effect of insulin-like growth factor 2 (IGF2) has been confirmed in various cancers and is related to resistance to chemotherapy and a worse prognosis. To further investigate the mechanism of IGF2 specific to GISTs. IGF2 was screened and analyzed using Gene Expression Omnibus (GEO: GSE225819) data. After IGF2 knockdown or overexpression by transfection, the phenotypes (proliferation, migration, invasion, apoptosis) of GIST cells were characterized by cell counting kit 8, Transwell, and flow cytometry assays. We used western blotting to evaluate pathway-associated and epithelial-mesenchymal transition (EMT)-associated proteins. We injected transfected cells into nude mice to establish a tumor xenograft model and observed the occurrence and metastasis of GIST. Data from the GEO indicated that IGF2 expression is high in GISTs, associated with liver metastasis, and closely related to drug resistance. GIST cells with high expression of IGF2 had increased proliferation and migration, invasiveness and EMT. Knockdown of IGF2 significantly inhibited those activities. In addition, OE-IGF2 promoted GIST metastasis in vivo in nude mice. IGF2 activated IGF1R signaling in GIST cells, and IGF2/IGF1R-mediated glycolysis was required for GIST with liver metastasis. GIST cells with IGF2 knockdown were sensitive to imatinib treatment when IGF2 overexpression significantly raised imatinib resistance. Moreover, 2-deoxy-D-glucose (a glycolysis inhibitor) treatment reversed IGF2 overexpression-mediated imatinib resistance in GISTs. IGF2 targeting of IGF1R signaling inhibited metastasis and decreased imatinib resistance by driving glycolysis in GISTs.Core Tip: Our study found that insulin-like growth factor 2 (IGF2) regulated metastasis and imatinib resistance in gastrointestinal stromal tumors (GISTs). IGF2 interacted with IGF1R to regulate glycolysis. Our results confirm that IGF2 targeting of IGF1R signaling inhibited metastasis and improved imatinib chemosensitivity by driving glycolysis in GISTs and indicated that IGF2 might be used to reverse imatinib resistance in GIST patients. Primary gastrointestinal stromal tumors (GISTs) account for 2% of gastrointestinal tumors. GISTs are encoded by the receptor tyrosine kinase gene KIT or PDGFRA. These mutations cause ligand-dependent activation and constitutive activation of signal transduction mediated by PDGFRA or KIT. The downstream molecular pathways of the KIT mutation include PI3K/AKT, JAK-STAT, Src family kinases, and Ras-ERK). Activation of molecular pathways follows KIT activation and leads to the occurrence of GISTs tumors by activation of cell proliferation and inhibition of apoptosis signals . Imatinib remains the primary treatment of GIST patients with advanced or metastatic tumors. Imatinib significantly improves the prognosis of patients in the advanced stages of the disease, but those undergoing imatinib treatment often encounter challenges associated with both primary and secondary drug resistance, which, unfortunately, restricts long-term efficacy. Insulin-like growth factor 2 (IGF2) is a genomic imprinting gene in growth on the chromosome 11 short arm. IGF2 overexpression is observed in a variety. of cancers and is related to chemotherapy resistance and a worse prognosis[12-14]. Studies of IGF1R have increased recently. Insulin-like growth factor (IGF) is comprised of the two ligands IGF1 and IGF2, their target tyrosine kinase receptors, IGF1 receptor (IGF1R) and the insulin receptor, as well as the IGF2 receptor (IGF2R) and IGF-binding proteins that regulate IGF ligand availability. IGF1R, is a tyrosine kinase receptor with binding affinity for both IGF1 and IGF2 ligands. Upon ligand binding, the activated tyrosine kinase domain initiates signaling cascades that specifically activate the GPTase Ras-Raf-ERK/MAPK and PI3K-AKT/mTOR pathways. These pathways, regulate the proliferation rate and apoptosis of cancer cells. The IGF pathway family gene expression (such as IGF1, IGF2, and IGF1R) has been reported to distinguish subsets of GISTs wild type for KIT and PDGFRA. Although data on IGF1R in GISTs have been reported[20-22], further research on the mechanisms of IGF2 and IGF1R in GISTs is needed. Sequencing data from the Gene Expression Omnibus (GEO) database (GSE225819 and GSE155880) were examined by bioinformatics. We found that IGF2 acted as a cancer-promoting factor and was involved in cell proliferation, apoptosis, liver metastasis, and epithelial-mesenchymal transition (EMT) in GISTs. Moreover, the role of IGF2 in GIST cells and the IGF2-IGF1R regulatory axis contributed to imatinib resistance of GISTs by regulating glycolysis and represents a target for GISTs therapy. Gene expression data based on RNA sequencing were obtained from the GEO. Two eligible datasets (GSE225819, GSE155880) were combined. The aligned reads were calculated by FeatureCounts (subread/2.0, http://subread.sourceforge.net/) and differentially expressed genes (DEGs) were analyzed by the R package DESeq2/3.1.0 (https://bioconductor.org/packages/release/bioc/html/DESeq2.html). A total of 2578 DEGs (1398 downregulated, and 1188 upregulated) were identified by screening GSE225819, including 20 normal samples and 20 GISTs samples with liver metastasis (|log2FC| > 1; P < 0.05) (Supplementary Table 1). Based on Deseq2, 1386 DEGs (939 downregulated, and 447 upregulated) were identified by screened GSE155880 including seven Imatinib-sensitive samples and seven imatinib-resistant GIST patients (|log2FC| > 1; P < 0.05) (Supplementary Table 2). RGM-1 normal human gastric mucosal cells, GIST882, and GIST-T1 cells were cultured in Iscove's modified Dulbecco's medium containing10% fetal bovine serum and 1% antibiotics, The culture temperature was 37 °C with 5% CO2. The imatinib concentration was increased from 1 nM to 100 nM over 10 mon and repeated to obtain imatinib-resistant GIST882 (GIST882-R) and GISTT1 (GISTT1R) cells. GIST882 and GIST-T1 cells were transfected with OE-IGF2, sh-IGF2 plasmids and sh-NC, OE-NC negative controls (RiboBio, Beijing, China) using Lipofectamine 3000 (Invitrogen, Waltham, MA, United States) and cultured for 2 d. Transfection efficiency was determined by western blotting. Imatinib mesylate was purchased from Selleckchem (Houston, TX, United States). GIST-T1 and GIST-882 cells were treated with serial dilutions of 1 μM imatinib in dimethyl sulfoxide for 4 h. We lysed transfected cells with RIPA buffer, the total protein was purified, and the protein concentration was determined with bicinchoninic kits (ThermoFisher Scientific, Waltham, MA, United States). The proteins were resolved by 10% SDS-PAGE and transferred to PVDF membranes for incubation with anti-IGF2 (1:1000, ab177467; Abcam, Cambridge, United Kingdom), anti-vimentin (1:1000, ab92547; Abcam), anti-N-cadherin (1:1000, ab76011; Abcam), anti-E-cadherin (1:1000, ab40772; Abcam), anti-Twist1 (1:1000, ab50887; Abcam), anti-IGF1R (1:1000, ab182408; Abcam), anti-p-IGF1R (1:1000, ab39398; Abcam), anti-PI3K (1:1000, ab302958; Abcam), anti-AKT (1:1000, MA5-14916; Invitrogen), anti-phospho-AKT (1:1000, PA5-95669; Invitrogen), and anti-β-actin (1:1000, ab8227; Abcam) primary antibodies overnight at 4 °C after blocking with skimmed milk (5%). After washing the primary antibodies away, the proteins were incubated with the anti-rabbit secondary antibody (1:5000; SA00001-2; SanYing Biotechnology Inc, Wuhan, China) for 1 h. The protein bands were visualized using an ECL chemiluminescence system, and the protein blots were quantified with Image J. The concentration of IGF2 was measured using ELISA kits (Abcam) according to the manufacturer′s instructions. The samples were prepared from cell culture supernatants and the IGF2 concentration was measured at 450 nm using a microplate reader. We determined GIST cell proliferation by cell counting kit-8 (CCK-8) assay. OE-IGF2- or sh-IGF2-transfected GIST882 and GIST-T1 cells were added to 96-well plates (1 × 10/well). After 1 d, we added CCK-8 reagent (10 μL, Catalog No. AD10; Dojindo Molecular Technologies, Kumamoto, Japan) to each well at room temperature. Absorbance was monitored at 0, 24, 48, 72, and 96 h and the half inhibitory concentration of imatinib was determined at 450 nm. After overnight incubation, the cells were treated with imatinib at 0, 20, 40, 60, and 80 μmol/L for 48 h. CompuSyn software was used to calculate the combination index using the Chou-Talalay method to determine the antagonistic influence. For the migration assay, GIST cells were seeded into 8 µm well Transwell chambers (Corning; Corning, NY, United States). The upper chamber was filled with 200 µL serum-free medium containing 2 × 10 cells and the lower chamber was filled with 500 μL complete medium (10% FBS). After 48 h, the cells were fixed with formaldehyde and stained with 0.2% crystal violet for 10 min. To assay cell invasion, 500 μL culture supernatant was collected from transfected cells and added to the upper Transwell chamber. GIST cells (2 × 10 cells) in about 200 μL serum-free medium were added to the lower chamber. The cells were cultured for 2 d at 37 °C with 5% CO2. After culturing, cells remaining in the lower chamber were removed with cotton swabs and those in the upper chamber were stained with 0.2% crystal violet for 5 min. We used an inverted microscope to count the cells that had migrated through the membrane and invaded the upper chamber. We bought 5-wk-old; male BALB/c nude mice from Vital River Laboratories (Beijing, China) and housed them for 1 wk to adapt to the environment. GIST-T1 cells (5 × 10) transfected with OE-IGF2/OE-NC, sh-IGF2/sh-NC were injected into the inguinal skin and the mice were monitored for growth of the tumor for 7 d before being randomized to four groups and treated with imatinib 50 mg/kg daily. After 4 wk, we killed the mice with an overdose of pentobarbital. All animal experiments were approved by the Animal Ethics Committee of Beijing Viewsolid Biotechnology Co. LTD (Protocol No. VS2126A00170) and all methods followed the ARRIVE guidelines. We fixed the liver tissue of mice in neutral formalin (10%), embedded it in paraffin, cut the tissue into 4 µm sections, and stained it with hematoxylin and eosin (HE). The sections were observed with a microscope. Cells were incubated in commercial seahorse XF assay medium plus pyruvate (1 mmol/L), glucose (10 mmol/L) and glutamine (2 mmol/L) 37 °C for 1 h in a CO2-free incubator. The rate of extracellular acidification was measured before and after addition of oligomycin, glucose, and 2-deoxy-D-glucose (2-DG). FCCP, a mitochondrial uncoupling agent; oligomycin, an ATP synthase inhibitor; 2-DG, a glycolysis inhibitor; rotenone; and antimycin A were added and metabolic energy consumption was assayed with a Seahorse XF96 Analyzer (Agilent, Santa Clara, CA, United States). The concentration of lactate in transfected cells was determined by ELISA with lactate assay kits (MAK064; Sigma-Aldrich, St Louis, MO, United States) according to the manufacturer’s protocol. The optical density of each well was determined at 570 nm (Plate Reader AF2000; Eppendorf, Waltham, MA, United States). GIST cell apoptosis was assayed by flow cytometry (LSRII; BD Biosciences, Franklin Lakes, NJ, United States). using annexin V-FITC apoptosis detection kits. The apoptosis rate was determined by analysis of Q2 and Q3 quadrant cells. We used GraphPad Prism 7.0 for data analysis. Data were reported as mean ± standard deviation of three independent experiments. Single-group comparisons were done with Student’s t-tests. Multiple group differences were compared by analysis of variance. P < 0.05 indicated significance. Based on the limma R package, a total of 2578 (DEGs 1398 downregulated and 1188 upregulated) were screened out from GEO: GSE225819 data, including 20 normal samples and 20 GIST samples with liver metastasis (|log2FC| > 1; P < 0.05), suggesting that these DEGs may be involved in liver metastasis in GIST patients (Figure 1A). The top 10 upregulated genes were PENK, IGF2, GPR20, CTSL, SCRG1, PNMAL1, NKX3-2, ANO1, PLAT, and BCHE. The top 10 downregulated genes were ATP4B, GKN1, MT1G, GKN2, ATP4A, SPINK1, TSPAN8, TFF1, KCNE2, and REG1A (Supplementary Table 1). Based on the Deseq2, 1386 DEGs (939 downregulated and 447 upregulated) were screened out in GSE155880, including seven Imatinib-sensitive samples and seven imatinib-resistant GIST patients (|log2FC| > 1; P < 0.05, Figure 1B). The intersection of the two analyses indicated that only IGF2 was involved in the drug resistance regulation and GIST metastasis in these DEGs (Supplementary Table 2). Moreover, we evaluated IGF2 expression in the GIST cell line. By western blotting, expression levels of IGF2 in GIST882, GIST882-R, GIST-T1, and GIST-T1-R were higher than those in normal RGM-1. Furthermore, IGF2 was significantly over expressed in GIST882-R/GIST-T1-R compared with other cell lines GIST882/GIST-T1 (P < 0.01, P < 0.001; Figure 1C). In addition, the expression levels of IGF2 in culture supernatants were measured using ELISA and compared (Figure 1D). We found that the ELISA and western blot results (P < 0.05, P < 0.001) were similar. IGF2 expression was high in drug-resistant GIST cell lines, suggesting that IGF2 overexpression may be closely related to drug resistance. High expression of insulin-like growth factor 2 in gastrointestinal stromal tumors with liver metastasis and closely related to drug resistance. A: Differentially expressed genes in gastrointestinal stromal tumors (GIST) with liver metastasis tissues and normal gastric tissues (|log2FC| > 1; P < 0.05); B: Differentially expressed genes in imatinib sensitive and in seven Imatinib-resistant GIST patients (|log2FC| > 1; P < 0.05); C: Western blot assay of insulin-like growth factor 2 (IGF2) protein expression in GIST cell lines (GIST882, GIST882-R, GIST-T1, GIST-T1-R); D: ELISA of IGF2 expression in GIST cell lines (GIST882, GIST882-R, GIST-T1, GIST-T1-R). Data are mean ± standard deviation. P < 0.05; P < 0.01; P < 0.001. We transfected GIST882 and GIST-T1 cells with an IGF2 overexpressing plasmid (OE-IGF2) or a shRNA to knock down IGF2 (sh-IGF2). Western blotting detected the efficiency of cell transfection (Figure 2A). IGF2 was highly expressed in OE-IGF2-transfected cells compared with OE-NC cells, while IGF2 expression was low in sh-IGF2-transfected cells (P < 0.001). ELISA also found that IGF2 expression high in OE-IGF2 group compared with OE-NC-GIST882 and GIST-T1 cells and IGF2 was low expressed in sh-IGF2-transfected cells (Figure 2B, P < 0.05, P < 0.01, P < 0.001). The CCK-8 results showed that cell viability was significantly increased after exogenous expression of IGF2, sh-IGF2 transfection inhibited GIST882 and GIST-T1 cell viability (Figure 2C, P < 0.001). Likewise, the Transwell assays found more migrating and invading OE-IGF2-GIST882 and GIST-T1 cells compared with their respective control cells (Figure 2D and E, P < 0.001). We also found that sh-IGF2 transfection inhibited cell viability, migration and invasion. In addition, western blotting detect EMT-related proteins (E-cadherin, vimentin, Twist1, and N-cadherin) expression in cells. Silencing IGF2 increased E-cadherin expression, and inhibited vimentin, Twist1, and N-cadherin expression, but IGF2 overexpression had the opposite experimental findings (Figure 2F, P < 0.001). To further verify the functional role of IGF2 on the growth of GISTs, we performed nude mouse tumorigenesis experiments. OE-IGF2 transfected-GIST-T1 cell lines were injected into the spleen. We found that OE-IGF2 promoted the GIST-T1 cell metastasis in vivo, showing a significant decline in the number of liver metastatic nodules (Figure 2G and H, P < 0.01). Insulin-like growth factor 2 promotes malignant characteristics and metastasis of gastrointestinal stromal tumors. A: Western blot measured the transfection efficiency of OE-insulin-like growth factor 2 (IGF2) or sh-IGF2 in gastrointestinal stromal tumors (GIST) 882 and GIST-T1 cells; B: ELISA of IGF2 expression in OE-IGF2 or sh-IGF2 transfected GIST882 and GIST-T1 cells; C: Cell counting kit-8 assay assessed cell viability in GIST882 and GIST-T1 cells; D: Transwell assay evaluated the migration of OE-IGF2- or sh-IGF2-transfected cells (scar bar = 50 μm); E: Transwell assays of the invasiveness of OE-IGF2 or sh-IGF2 transfected cells. (scar bar = 50 μm); F: Detection of proteins involved in epithelial-mesenchymal transition (vimentin, N-cadherin, E-cadherin, Twist1) in OE-IGF2 or sh-IGF2 transfected cells; G: Liver tissue from tumor xenografts in nude mice injected withOE-IGF2 transfected GIST-T1 cells; H: Liver metastasis determined by hematoxylin-eosin staining. Data are mean ± standard deviation. P < 0.05; P < 0.01; P < 0.001. IGF1R mRNA expression was increased in GIST-T1 and GIST882 cells transfected with OE-IGF2, and IGF1R mRNA expression was decreased after sh-IGF2 transfection (Figure 3A, P < 0.001). PI3K-Akt signaling is the IGF2-IGF1R signal principal downstream target. Expression of IGF2-IGF1R pathway-associated proteins (IGF1R, p-IGF1R, PI3K, AKT, p-AKT) in GIST-T1 cells was measured by western blotting. IGF2 overexpression increased the expression of IGF1R, p-IGF1R, PI3K, AKT, and p-AKT in GIST-T1 cells. The opposite result was noted after IGF2 knockdown (Figure 3B, P < 0.01, P < 0.001). Although sh-IGF2 reduced IGF1R, p-IGF1R, PI3K, AKT, and p-AKT expression in GIST-T1 cells, it was partially restored by overexpression of IGF2R (Figure 3C, P < 0.01, P < 0.001). Insulin-like growth factor 2 activated the IGF1R signaling in gastrointestinal stromal tumors cells. A: Quantitative reverse transcriptase PCR assay of IGF1R mRNA expression in gastrointestinal stromal tumors (GIST) 882 and GIST-T1 cells after OE-insulin-like growth factor 2 (IGF2) or sh-IGF2 transfection; B: Detection of protein levels (IGF1R, p-IGF1R, PI3K, AKT, and p-AKT) involved in the PI3K/AKT in OE-IGF2 or sh-IGF2 transfected-GIST-T1 cells by western blot assay; C: Detection of protein levels (IGF1R, p-IGF1R, PI3K, AKT, and p-AKT) involved in the PI3K/AKT in GIST-T1 cells after sh-IGF2 and OE-IGF2R transfection by western blot assay. P < 0.01; P < 0.001. We analyzed glucose consumption and lactate production in GIST cells. Sh-IGF2 inhibited glucose consumption (Figure 4A), and lactate production in GIST882 and GIST-T1 cells (Figure 4B), but IGF2 overexpression had the opposite experimental findings (P < 0.001). To examine the role of the Warburg effect in liver metastasis of GISTs, we treated OE-NC-GIST882 and OE-IGF2-GIST882 cells with 2-deoxyglucose (2-DG, a glycolysis inhibitor) for 24 hat 0, 4, 8, and 16 mmol/L. 2-DG significantly inhibited glycolysis (Figure 4C, P < 0.05, P < 0.01, P < 0.001) and Transwell assays found that 2-DG treatment inhibited the promoting effect of OE-IGF2 on GIST882 and GIST-T1 cell invasion and migration (Figure 4D and E, P < 0.001). Similarly, OE-IGF2 increased vimentin, Twist1, and N-cadherin expression and inhibited E-cadherin expression in cells, but the expression was partially restored by 2-DG treatment (Figure 4F, P < 0.001). Insulin-like growth factor 2/IGF1R-mediatedglycolysisis required for gastrointestinal stromal tumors with liver metastasis. A: Extracellular acidification rate was measured; B: Lactate production in gastrointestinal stromal tumors (GIST) 882 and GIST-T1 cells transfected with sh-insulin-like growth factor 2 (IGF2) or OE-IGF2 were measured; C: Lactate production in OE-IGF2-GIST882 and GIST-T1 cells cotreated with 2-deoxy-D-glucose (2-DG) (0, 4, 8, and 16 mmol/L); D: Transwell assay of the migration ability of the OE-IGF2-cells cotreated with 2-DG (scar bar = 50 μm); E: Transwell assay of the invasiveness of OE-IGF2-cells cotreated with 2-DG (scar bar = 50 μm); F: Assay of proteins involved in epithelial-mesenchymal transition (vimentin, N-cadherin, E-cadherin, Twist1) in OE-IGF2-cells cotreated with 2-DG. Data are mean ± standard deviation. P < 0.05; P < 0.01; P < 0.001. Figure 1 shows that IGF2 was involved in regulating drug resistance. Next, we will further verify. To test whether IGF2 also regulated drug resistance in GISTs in vivo, we established a xenograft model by inoculating sh-NC or sh-IGF2-GIST-T1 cells into nude mice. In the sh-IGF2-GIST-T1 mouse xenograft model, tumor volume and growth were inhibited by sh-IGF2, and imatinib had the same influence on tumor growth and volume. Combined treatment with imatinib and sh-IGF2 was more effective for reducing tumor progression than single treatment (Figure 5A-C, P < 0.001). The western blot results revealed that expression of IGF1R, p-IGF1R, AKT, PI3K, and p-AKT in tumor tissue was suppressed in both sh-IGF2-transfected cells and after imatinib treatment. Moreover, combined imatinib and sh-IGF2 were more effective than single therapy (Figure 5D, P < 0.001). The above data suggest that IGF2/IGF1R regulate imatinib resistance. Insulin-like growth factor 2/IGF1R regulates imatinib resistance of gastrointestinal stromal tumors by regulating glycolysis. A: Tumor growth in xenografted nude mice; B: Tumor volumes in sh-insulin-like growth factor 2 (IGF2)-gastrointestinal stromal tumors (GIST)-T1 mouse xenograft models treated with imatinib; C: After 35 d, the mice were killed and the tumors were weighed; D: Assay of IGF1R, p-IGF1R, PI3K, AKT, and p-AKT in tumor tissue by western blotting; E: Assay of drug sensitivity in OE-IGF2-GIST882 and GIST-T1 cells treated with 2-deoxy-D-glucose (2-DG); F: Flow cytometry assay of apoptosis of OE-IGF2-GIST882 and GIST-T1 cells treated with 2-DG. Data are mean ± standard deviation. P < 0.05; P < 0.01; P < 0.001. In addition, previous data shows that IGF2 regulates glycolysis in GIST cells. IGF2 regulates cell sensitivity to imatinib through its influence on glycolysis. We used 2-DG to inhibit glycolysis in GIST cells. OE-IGF2 increased drug sensitivity in GIST882 and GIST-T1 cells, but after treatment with 2-DG, transfection with OE-IGF2 no longer changed drug sensitivity in GIST cells (Figure 5E, P < 0.001). Flow cytometric analysis showed that sh-IGF2 suppressed imatinib-induced apoptosis and OE-IGF2 reduced imatinib-induced apoptosis in GIST cells. Treatment with 2-DG and transfection with OE-IGF2 no longer influenced imatinib-induced apoptosis in GIST cells (Figure 5F, P < 0.001). Therefore, the results show that IGF2 regulated imatinib sensitivity in GIST cells by affecting glycolysis. GISTs is the most frequent malignant gastrointestinal sarcoma and causes significant patient harm. Recently, anticancer drug resistance has become a significant challenge to the treatment of GISTs. Treatment with tyrosine kinase inhibitors (TKIs) has led to substantial improvement of survival, both for patients with localized GISTs and those with advanced disease. As the first-line TKI, imatinib offers treatment for advanced and metastatic GISTs, adjuvant therapy in high-risk GISTs and neoadjuvant treatment to downsize large tumors prior to resection. We explored the mechanism of IGF2 in imatinib resistance in GISTs and whether IGF2 enhanced metastasis and imatinib resistance by driving glycolysis by targeting IGF1R signaling transduction. IGF2, identified as an imprinted gene, exhibits a significant impact on cancer progression when its imprinting is lost or relaxed, leading to heightened autocrine IGF2 levels and increased secretion in malignant cells. Numerous studies have revealed the upregulation of IGF2 in various cancers such as hepatocellular carcinoma, correlating with resistance to chemotherapy and a poorer prognosis[12-14]. Our investigation, which focused on DEGs associated with liver metastasis and drug resistance in GISTs, we observed elevated levels of IGF2 in GISTs cases linked to liver metastasis and drug resistance. Our comprehensive analysis included assessment of cell proliferation, viability, migration, and invasiveness. The findings strongly suggest that overexpression of IGF2 induce the proliferation, metastasis, and EMT of GIST cells. IGF1R, is a tyrosine kinase receptor that can be triggered by IGF2 and has a pivotal role in regulating mammalian development, metabolism, and growth. IGF1R is known to be upregulated in various human solid tumors. Its involvement in cell promoting cell proliferation and inhibiting programmed cell death is facilitated by activation of its tyrosine kinase and the subsequent engagement of the Ras/Raf/MEK and PI3K/AKT/mTOR signaling pathways. The IGF2-IGF1R signaling axis assumes critical significance in governing cell proliferation, differentiation, EMT, migration, drug resistance, and maintaining stemness in malignancies. This investigation further demonstrated the activation of IGF1R signaling by IGF2 in GIST cells. It highlights the role of IGF2 as a pivotal chromatin factor that controls the expression level of IGF1R and modulates downstream signaling by the PI3K/AKT pathway. IGF2 also upregulated the expression of glycolytic and mitochondrial respiration markers. IGF2 overexpression has also been shown to cause metabolic reprogramming in breast cancer. As expected, we also that IGF2 mediated the glycolysis in GISTs by targeting IGF1R signaling. Increased expression of IGF2 is a common occurrence in various cancers and has been associated with increased resistance to chemotherapy, leading to a poorer prognosis. Regarding GISTs, the standard first-line therapeutic approach involves the use of imatinib. Imatinib, a potent TKI, is the primary treatment for GISTs, and significantly contributes to the progression-free survival of GIST patients. Our investigation revealed a noteworthy correlation of increased IGF2 expression with the induction of GISTs resistance to imatinib concurrently with a reduction of imatinib-induced apoptosis in GIST cells. These findings underscore IGF2 as a potential regulator of GISTs imatinib resistance, and a promising target for interventions aimed at reversing such resistance. Intriguingly, our study further showed that IGF2 regulates cellular sensitivity to imatinib by modulating glycolysis. The study had some limitations of this study. First, except for GIST cells, the role of IGF2 on GIST patient samples needs verification. Even though we found that IGF-2 overexpression increased the resistance of GIST cells to imatinib in cell culture, the clinical effect needs to be verified. Secondly, our results allows speculation that IGF2 was involved in the resistance to chemotherapy and a worse GISTs prognosis. However, the molecular mechanism of IGF2 specific to GISTs requires further investigation. We will consider these issues in future studies. In addition, studies have found that hypoglycemia in patients with non-islet cell tumor-induced GISTs may be aggravated by imatinib. A recent case study reported that a GISTs that produced big-IGF2 also caused severe hypoglycemia. We also hope to investigate that in future experiments. This study investigated IGF2 regulation of metastasis and imatinib resistance in GISTs. IGF2 interacted with IGF1R to regulate glycolysis. Our results found that IGF2 targeting of IGF1R signaling improved metastasis and imatinib chemosensitivity via driving glycolysis in GISTs and support potential use of IGF2 to reverse imatinib resistance in GISTs patients.
|
PMC1462997
|
Derivation of normal macrophages from human embryonic stem (hES) cells for applications in HIV gene therapy
|
Many novel studies and therapies are possible with the use of human embryonic stem cells (hES cells) and their differentiated cell progeny. The hES cell derived CD34 hematopoietic stem cells can be potentially used for many gene therapy applications. Here we evaluated the capacity of hES cell derived CD34 cells to give rise to normal macrophages as a first step towards using these cells in viral infection studies and in developing novel stem cell based gene therapy strategies for AIDS. Undifferentiated normal and lentiviral vector transduced hES cells were cultured on S17 mouse bone marrow stromal cell layers to derive CD34 hematopoietic progenitor cells. The differentiated CD34 cells isolated from cystic bodies were further cultured in cytokine media to derive macrophages. Phenotypic and functional analyses were carried out to compare these with that of fetal liver CD34 cell derived macrophages. As assessed by FACS analysis, the hES-CD34 cell derived macrophages displayed characteristic cell surface markers CD14, CD4, CCR5, CXCR4, and HLA-DR suggesting a normal phenotype. Tests evaluating phagocytosis, upregulation of the costimulatory molecule B7.1, and cytokine secretion in response to LPS stimulation showed that these macrophages are also functionally normal. When infected with HIV-1, the differentiated macrophages supported productive viral infection. Lentiviral vector transduced hES cells expressing the transgene GFP were evaluated similarly like above. The transgenic hES cells also gave rise to macrophages with normal phenotypic and functional characteristics indicating no vector mediated adverse effects during differentiation. Phenotypically normal and functionally competent macrophages could be derived from hES-CD34 cells. Since these cells are susceptible to HIV-1 infection, they provide a uniform source of macrophages for viral infection studies. Based on these results, it is also now feasible to transduce hES-CD34 cells with anti-HIV genes such as inhibitory siRNAs and test their antiviral efficacy in down stream differentiated cells such as macrophages which are among the primary cells that need to be protected against HIV-1 infection. Thus, the potential utility of hES derived CD34 hematopoietic cells for HIV-1 gene therapy can be evaluated.Human embryonic stem cells (hES cells) show great promise for many novel cellular therapies due to their pluripotent nature . These cells have the capacity to give rise to mature cells and tissues that arise from all three germ layers during embryonic development [2-4]. Several pluripotent hES cell lines have so far been derived from the inner cell mass of human blastocysts and can be cultured indefinitely in an undifferentiated state [5-7]. Thus, these cells provide a renewable source of pluripotent stem cells from which many types of differentiated cells could be produced for experimental and therapeutic purposes. Cell differentiation protocols currently exist for the derivation of neurons, cardiomyocytes, endothelial cells, hematopoietic progenitor cells, keratinocytes, osteoblasts, and hepatocytes to name a few . In addition to providing for potential cellular replacement therapies, opportunities exist in programming hES cells to correct a genetic defect and/or to express a therapeutic transgene of interest. Using such approaches, many possibilities exist for treating a number of genetic and immune system disorders . Many novel applications can be foreseen for hES cells in infectious disease research. AIDS is a potential disease that can benefit from exploiting hES cells for cell replacement therapy as they have the capacity to differentiate into various hematopoietic cells. HIV continues to be a major global public health problem with infections increasing at an alarming rate . Given the present lack of effective vaccines and the ineffectiveness of drug based therapies for a complete cure, new and innovative approaches are essential. Gene therapy through intracellular immunization offers a promising alternative approach and possible supplement to current HAART therapy [12-14]. HIV mainly targets cells of the hematopoietic system, namely, T cells, macrophages, and dendritic cells . As infection progresses, the immune system is rendered defenseless against other invading pathogens and succumbs to opportunistic infections. There is a great deal of progress in the area of stem cell gene therapy for AIDS . A primary goal of many ongoing studies is to introduce an effective anti-HIV gene into hematopoietic stem cells [16-18]. As these cells possess the ability to self renew, they have the potential to continually produce HIV resistant T cells and macrophages in the body thus providing long term immune reconstitution. These approaches use CD34 hematopoietic stem cells for anti-HIV gene transduction via integrating viral vectors such as lentiviral vectors [16-18]. Lentiviral vectors have several advantages over conventional retroviral vectors since higher transduction efficiencies can be obtained and there is less gene silencing. The CD34 cells currently used for many therapies are primarily obtained from bone marrow or mobilized peripheral blood . Thus, CD34 progenitor cells are an essential ingredient for HIV gene therapy. In view of the need for CD34 cells for HIV gene therapy as well as for other hematopoietic disorders, if one can produce these cells in unlimited quantities from a renewable source, it will overcome the limitations of securing large numbers of CD34 cells for therapeutic purposes. In this regard, progress has been made in deriving CD34 cells from hES cells (hES-CD34). Different methods currently exist to derive CD34 cells from hES cells with varying efficiencies [20-27]. Recent reports have indicated the capacity of hES cell derived CD34 cells to give rise to lymphoid and myeloid lineages thus paving the way for utilization of these cells for hematopoietic cell therapy [20,27-29]. For the effective utilization of hES-CD34 cells for HIV gene therapy, a number of parameters need to be examined. First, one has to demonstrate that hES-CD34 cells can give rise to macrophages and helper T cells which are the main cells that need to be protected against HIV infection. Recent evidence has shown that hES-CD34 cells can give rise to myelomonocytic cells . However, thorough phenotypic or functional characterization of these cells is lacking. It is also not clear if these cells are susceptible to HIV infection. Similarly, although the hES-CD34 cells were shown to have lymphoid progenitor capacity, only B cell and natural killer (NK) cell differentiation has been examined so far . The capacity to generate T cells remains to be evaluated. With this background, as a first step, our primary goal in these studies is to examine the capacity of hES-CD34 cells to give rise to phenotypically and functionally normal macrophages and whether such cells are susceptible to productive HIV infection. Since lentiviral vectors have been shown to successfully transduce hES cells [30-33], we further investigated the ability of transduced hES cells to differentiate into transgenic macrophages that can support HIV-1 infection. Demonstration of HIV-1 productive infection in these cells will permit future efficacy evaluations of anti-HIV genes in this system. Here we show that normal and lentiviral vector transduced hES-CD34 cells can give rise to phenotypically and functionally normal macrophages that support HIV infection thus paving the way for many novel approaches to evaluate their potential for HIV gene therapy. Undifferentiated hES cell colonies grown in media supplemented with 4 ng/ml bFGF displayed normal morphology of pluripotent human embryonic stem cells with tight and discreet borders on the MEF feeder layers (Fig 1A). Similarly, lentiviral vector transduced hES cell colonies, also displayed normal morphology and growth characteristics (Fig 1A). As expected, the vector transduced colonies displayed green fluorescence due to the presence of the GFP reporter gene. When cultured on irradiated S17 mouse bone marrow stromal cells, both nontransduced and transduced hES cells developed into embryonic cystic bodies (Fig 1A). FACS analysis of single cell suspensions of the cystic bodies showed levels of CD34 cells which ranged from 7–15%. Figure 1B displays a representative FACS profile of hES-CD34 cells. Purified CD34 cells were later cultured in semi-solid methylcellulose medium to derive myeloid colonies. Both nontransduced (denoted as ES in figures) and vector transduced (denoted as GFP ES in figures) hES cell derived CD34 cells gave rise to normal myelomonocytic colonies similar to human fetal liver derived CD34 cells (denoted as CD34 in figures) (Fig 1A). When pooled colonies were cultured further in liquid cytokine media for 12–15 days for differentiation, the cells developed into morphologically distinct macrophages (Fig 1A). When compared, the morphology of macrophages derived from all stem cell progenitor populations appeared similar. These results were found to be consistent in replicative experiments. The transgene GFP expression was also maintained during the differentiation of hES cells into mature macrophages. GFP expression in cystic body derived CD34 cells was around 80% (data not shown) with similar levels seen in differentiated macrophages (Fig 2). Derivation of macrophages from lentiviral vector transduced and normal hES cells. A) Transduced and non-transduced H1 hES cells were cultured on mouse S17 bone marrow stromal cell layers to derive cystic bodies. Cystic body derived CD34 cells were purified by positive selection with antibody conjugated magnetic beads and placed in methocult media to obtain myelomonocytic colonies. Pooled colonies were cultured in liquid cytokine media supplemented with GM-CSF and M-CSF to promote macrophage growth. For comparison, fetal liver derived CD34 cells were cultured similarly to derive macrophages. Representative ES cell colonies, cystic bodies, methocult colonies, and derivative macrophages are shown with GFP expressing cells fluorescing green under UV illumination. B) Representative FACS profile of hES cell derived CD34 cells stained with PE conjugated antibodies. Percent positive CD34 cells are shown with isotype control shown in the left panel. Phenotypic FACS analysis of hES cell derived macrophages. A) Macrophages derived from transduced and nontransduced hES CD34 and fetal liver CD34 cells were stained with antibodies to CD14, HLA-DR, CD4, CCR5, and CXCR4 and the expression of these surface markers was analyzed by FACS. B) Isotype controls for PE and PE-CY5 antibodies. Percent positive cells are displayed in the plots for each respective cell surface marker staining. Dot plots are representative of triplicate experiments. Macrophages play a critical role in immune system function and are also major target cells for many viral infections including HIV-1. Distinct surface phenotypic markers exist on these cells and, thus far, there has been no thorough evaluation of hES cell derived macrophages. Therefore we analyzed hES cell derived macrophages for the presence of characteristic cell surface markers and compared these to the phenotypic profile displayed on fetal CD34 cell derived macrophages. The surface markers analyzed were CD14, a monocyte/macrophage specific marker, HLA-DR (a class II MHC molecule found on antigen presenting cells), CD4, the major receptor for HIV-1 infection, and CCR5 and CXCR4, chemokine receptors which are critical coreceptors essential for HIV-1 entry. EGFP expression was also analyzed to determine the levels of transduction and any transgene silencing that may occur during differentiation. Fetal liver (CD34), nontransduced (ES), and vector transduced (GFP ES) hES cell derived macrophages were all positive for the monocyte/macrophage marker CD14 (99.3%, 88.7%, and 99.2%, respectively) (Fig 2A). However, the mean fluorescent intensity (MFI) was found to be lower on hES cell derived macrophages. Surface expression of HLA-DR was observed at similar levels between macrophages derived from fetal liver CD34 cells (99.6%), nontransduced hES cells (92.8%), and transduced hES cells (98.2%) (Fig 2A). CD4 levels were comparable for all stem cell derived macrophages (99.2%, 83.3%, and 88.7%, respectively) (Fig 2A). CCR5 and CXCR4 cell surface expression was also observed for fetal liver CD34 cell (99.6% and 99.3%), nontransduced hES cell (91.9% and 92.6%), and transduced hES cell (98.9% and 99.3%) derived macrophages (Fig 2A). As compared to fetal liver CD34 cell derived macrophages, hES cell derived macrophages displayed a higher level of expression of CXCR4. Isotype controls for both PE and PECY5 stains are shown in Fig 2B. The above phenotypic data are representative of triplicate experiments. The antigen presenting cell surface specific marker HLA-DR (MHC II) on normal macrophages is critical for presenting antigen to CD4 T cells. A second co-stimulatory molecule, B7.1 is present at low basal levels on resting macrophages and is necessary to activate T cells. Its expression is elevated upon activation with certain stimuli such as LPS. Our results of LPS stimulation of respective macrophages have shown upregulation of B7.1 with values for fetal liver CD34 cell (CD34) (27.9% to 75.4%) nontransduced (ES) (17.8% to 49.4%) and transduced (GFP ES) (35.6% to 65.7%) hES cell derived macrophages (Fig 3A). These values represent a significant upregulation of B7.1 for all three macrophage populations. Functional analysis of hES cell derived macrophages for B7.1 costimulatory molecule upregulation and phagocytosis of E. coli particles: A) Mature macrophages were stimulated with LPS to determine B7.1 upregulation. Twenty-four hours post-stimulation, macrophages were labeled with a PE-CY5 conjugated anti-B7.1 antibody and analyzed by FACS. B7.1 upregulation data are representative of triplicate experiments. Isotype control is shown in the left panel. B) To assess phagocytic function, E. coli Bioparticleswere added directly to the cultured macrophages. Twenty four hours post-addition, cells were analyzed by FACS. Percent positive cells are displayed in the plots for each experiment. These data are representative of triplicate experiments. Another important function of macrophages is their ability to phagocytose foreign material and present antigenic peptides on their cell surface. To evaluate phagocytic function, fluorescently labeled E. coli Bioparticleswere added to macrophage cultures followed by FACS analysis. Nontransduced (94.6%) as well as lentiviral vector transduced (98.7%) hES cell derived macrophages were found to be capable of phagocytosing the Bioparticlesin comparison to fetal liver CD34 cell derived macrophages (95.8%) (Fig 3B). These values are representative of triplicate experiments. Magi-CXCR4 cells with no phagocytic capacity were used as non-phagocytic cell controls and similarly exposed to E. coli Bioparticles(Fig 3B). No uptake of the bacteria could be seen. Thus, uptake of E. coli Bioparticlesby macrophages is indicative of active ingestion. Macrophages, as effector cells, play a key role in the inflammatory response. Activated macrophages secrete various cytokines, two of the major ones being IL-1 and TNF-α. To determine if hES cell derived macrophages have such a capacity, cells were stimulated with LPS. On days 1, 2, and 3 post-stimulation, culture supernatants were analyzed by ELISA to detect IL-1 and TNF-α. As seen in figure 4A, there were no significant differences in IL-1 secretion between the three sets of macrophages. Similarly, nontransduced and transduced hES cell derived macrophages were also capable of TNF-α secretion upon LPS stimulation. However, levels of the respective cytokines detected were slightly lower than those from fetal liver CD34 cell derived macrophages (Fig 4B). The values of cytokine secretion levels represent triplicate experiments. Cytokine IL-1 and TNFα secretion by stimulated hES cell derived macrophages: Macrophages derived from transduced and nontransduced hES and fetal liver CD34 cells were stimulated with 5 μg/ml LPS. On days 1, 2, and 3 post-stimulation, supernatants were collected and assayed by ELISA for (A) IL-1 and (B) TNFα. Experiments were done in triplicate. The above data have shown that hES cell derived macrophages are very similar to normal human macrophages based on phenotypic and functional analysis. In addition to being important cells of the immune system, macrophages are among the major target cells for certain viral infections, particularly for HIV-1. We wanted to determine if hES cell derived macrophages were susceptible to HIV-1 infection compared to standard macrophages. In these studies, we only used an R5-tropic strain of HIV-1 since macrophages are natural targets for this virus. Our results from challenge studies of these cells clearly indicated the capacity of hES cell derived macrophages in supporting a productive infection. Levels of virus increased up to 15 days similar to non-hES derived macrophages showing that the initial viral input was amplified in productive viral infection. However, the levels of viral yield were found to be slightly lower for the ES cell derived macrophages. In the case of GFP-ES macrophages, there was a decline in viral titer. This could be due to possible lower numbers of cells present in the initial cultures. As a first step towards the use of hES cells for hematopoietic stem cell and HIV gene therapies, we have shown here that phenotypically and functionally normal macrophages could be derived from hES-CD34 cells. Both non transduced and lentiviral vector transduced hES cells were found to be capable of generating CD34 cells that give rise to macrophages which could support productive HIV-1 infection. Current sources of CD34 cells consist of human bone marrow, cytokine mobilized peripheral blood, fetal liver, and cord blood . However, the number of cells that can be obtained for manipulations is not unlimited. Therefore, deriving CD34 cells for therapeutic and investigative purposes from hES cells with unlimited growth potential has the advantage of a consistent and uniform source. The ability to obtain phenotypically normal and functionally competent macrophages from hES cells is important to evaluate their potential therapeutic utilities in the future. Additionally, testing of transgenic hES cells derived via lentiviral vector gene transduction is also helpful to determine the stability of the transgene expression and their capacity for differentiation into end stage mature cells such as macrophages. Based on these considerations, both non- transduced and lentiviral vector transduced hES cells were evaluated for their capacity to give rise to CD34 progenitor cells. In colony forming assays using semisolid methylcellulose medium, the morphology of myelomonocytic colonies derived from hES CD34 cells appeared similar to that of fetal liver CD34 cells. When subsequently cultured in cytokine media that promotes macrophage differentiation, morphologically normal macrophages were obtained with hES-CD34 cells similar to that of fetal liver CD34 cells. At higher magnification, the macrophages displayed flat projecting cellular borders with fried egg appearance with distinct refractory lysosomal granules in the cytoplasm (data not shown). Lentiviral vector transduced hES cells also did not display any abnormal growth or differentiation characteristics as compared to nontransduced hES-CD34 cells indicating no adverse effects due to vector integration and expression. Transduced cells gave rise to cystic bodies with similar CD34 cell content and profiles upon development. The transduced hES-CD34 cells also gave rise to apparently normal macrophages that expressed the transgene as shown by GFP expression. These results are consistent with those of others that showed normal differentiation of hES cells to other cell types following lentiviral transduction . A requirement for successful cellular and HIV-1 gene therapy is that mature end stage cells derived from CD34 progenitor cells be phenotypically and functionally normal to maintain and restore the body's immunological function. Accordingly, hES cell derived macrophages were evaluated to determine if they met these criteria. Macrophages display distinct cell surface markers upon end stage differentiation. To determine whether hES cell derived macrophages display these surface markers, FACS analysis was performed to detect the presence of CD14, HLA-DR (MHCII), CD4, CCR5, and CXCR4. As observed in Fig 2A, both nontransduced and transduced hES cell derived macrophages expressed all of these markers with some differences in their levels of expression. HLA-DR, CD4, and CCR5 expression profiles were comparable between all cell types analyzed. Even though all cell types analyzed stained positive for CD14, relative expression of CD14 was slightly lower on hES cell derived macrophages compared to fetal liver CD34 cell derived macrophages. On the contrary, the levels of CXCR4, a chemokine receptor involved in cellular homing, were found to be higher on hES-CD34 cell derived macrophages. This may be due to inherent differences in the cell types and/or due to their physiological state at the time of harvest . Additional hES cell lines need to be evaluated in the future to establish if these differences are consistent. A major functional role of macrophages in vivo is their ability to serve as professional antigen presenting cells. During this process macrophages present antigen peptide fragments complexed with both classes of MHC molecules and deliver a costimulatory signal through the expression of B7 molecules. Upon stimulation with LPS, hES-CD34 cell derived macrophages had shown upregulation of the costimulatory molecule B7.1 similar to cells derived from fetal liver. Furthermore, the hES-CD34 cell derived macrophages also showed a normal capacity to ingest foreign particles in phagocytosis assays using E.coli Bioparticles. In addition to antigen presentation and phagocytosis, macrophages also play a critical role in inflammation and secrete cytokines in response to external stimuli. When exposed to LPS, the hES-CD34 cell derived macrophages secreted two important cytokines IL-1 and TNF-α similar to that of fetal liver derived cells. The above data has established that phenotypically and functionally normal macrophages could be derived from hES-CD34 cells. Macrophages in addition to playing important physiological roles are also major cell targets for certain viral infections, particularly HIV-1. Here we evaluated the susceptibility of hES-CD34 cell derived macrophages to be productively infected with HIV-1. Similar to that of fetal liver CD34 cell derived cells, the hES-CD34 macrophages also supported HIV-1 infection although the levels of viral yield differed somewhat. However this should not be a major concern for testing anti-HIV genes in these cells. In all the above experiments, the vector transduced transgenic macrophages also behaved similarly to that of nontransduced cells showing that they were also physiologically normal. The lack of vector toxicity on cellular maturation is encouraging for future work with transduced hES-CD34 cells to derive other important differentiated cells like T cells and dendritic cells relevant for HIV studies. Although there are numerous studies on hES cell differentiation into many important end stage mature cells, systematic work on hES cell hematopoietic differentiation and thorough characterization of end stage mature cells that participate in critical immune responses has just begun [21,27-29]. Our current results established that physiologically normal macrophages could be derived from hES cells and that these cells have the potential for use in cellular and gene therapies. To our knowledge this is the first demonstration that hES cell derivatives can be used for infectious disease research. Due to the extensive ability for hES cells to self-renew, large numbers of differentiated cells can be derived so that infection studies and evaluation tests can be carried out in a more standardized way. Our results showing that both normal and transgenic derivative macrophages support HIV-1 infection points out to their utility for testing anti-HIV constructs transduced into hES-CD34 cells and pave the way for their application in stem cell based HIV gene therapy. So far a number of studies including our own have tested many gene therapeutic constructs in CD34 cells from conventional sources. These constructs include anti-HIV ribozymes, RNA decoys, transdominant proteins, bacterial toxins, anti-sense nucleic acids, and most recently siRNAs [36-50]. In addition, a number of cellular molecules that aid in HIV-1 infection such as cellular receptors and coreceptors CD4, CCR5 and CXCR4 have also been successfully tested in CD34 cell derived macrophages and T cells . Some of these approaches have progressed into clinical evaluations as well . Based on our current results, many of these novel anti-HIV constructs can also be tested in hES-CD34 cells for their potential application. Although there are advantages of using hES cell derived CD34 cells for potential cellular therapies, transplantation of these cells constitutes an allogenic source with immune rejection as a major issue. However, a recent study using human leukocyte reconstituted mice suggested that hESCs and their derivative cell types were less prone to invoking an allogeneic response . Other recent studies demonstrated successful engraftment of primary and secondary recipients with hES cell derived hematopoietic cells in both immunodeficient mice and in vivo fetal sheep models adding further support that any obstacles could be overcome . Moreover, multiple novel strategies to avoid immune-mediated rejection of hES cell-derived cells have been proposed . It is not too far in the future that even autologous hES cells may be derived from specific individuals for deriving CD34 cells which can be used for cell replacement therapy. Phenotypically normal and functionally competent macrophages could be derived from hES-CD34 cells. Since these cells are susceptible to HIV-1 infection, they provide a uniform source of macrophages for viral infection studies. Based on these results, it is also now feasible to transduce hES-CD34 cells with anti-HIV genes such as inhibitory siRNAs and test their antiviral efficacy in down stream differentiated cells such as macrophages which are among the primary cells that need to be protected against HIV-1 infection. Thus, the potential utility of hES derived CD34 hematopoietic cells for HIV-1 gene therapy can be evaluated. The NIH approved human ES H1 cell line was obtained from WiCell (Madison, Wisconsin). hES cell colonies were cultured on mouse embryonic fibroblasts (MEF) (Chemicon, Temecula, CA) in the presence of DMEM-F12 (Invitrogen, Carlsbad, CA) supplemented with 20% KNOCKOUT serum replacement with 1 mM L-glutamine, 1% Nonessential Amino Acids, 0.1 mM β-mercaptoethanol, 0.5% penicillin/streptomycin, and 4 ng/ml human basic fibroblast growth factor. Culture medium was replaced daily with fresh complete DMEM-F12. Mature colonies were subcultured weekly by digesting with collagenase IV as previously described . A VSV-G pseudotyped lentiviral vector (SINF-EF1a-GFP) containing a GFP reporter gene (kindly supplied by R. Hawley, George Washington University) was used for hES cell transductions as previously described (30, 58). Generation of the pseudotyped vector in 293T cells and its concentration by ultracentrifugation were described previously . For vector transduction, the undifferentiated hES cells were prepared into small clumps of 50–100 cells with enzyme digestion as done for routine passaging of cells. The cell clumps were incubated with the vector for 2 hrs in the presence of polybrene 6 ug/ml. A secondary cycle of transduction was done by adding fresh vector and incubating for another 2 hrs. The general vector titers were 1 × 10and the multiplicity of infection was 10. The transduction efficiency was about 50%. The transduced colonies were cultured on MEF like above. Undifferentiated hES cells were cultured on S17 mouse bone marrow stromal cell monolayers to derive cystic bodies containing CD34+ hematopoietic progenitor stem cells. hES cell cultures were treated with collagenase IV(1 mg/ml) for 10 minutes at 37°C and subsequently detached from the plate by gentle scraping of the colonies. The hES cell clusters were then transferred to irradiated (35 Gy) S17 cell layers and cultured with RPMI differentiation medium containing 15% FBS (HyClone, Logan, UT), 2 mM L-glutamine, 0.1 mM β-mercaptoethanol, 1% MEM-nonessential amino acids, and 1% penicillin/streptomycin. Media was changed every 2 to 3 days during 14–17 days of culture on S17 cells . After allowing adequate time for differentiation, hES cystic bodies were harvested and processed into a single cell suspension by collagenase IV treatment followed by digestion with trypsin/EDTA supplemented with 2% chick serum (Invitrogen, Carlsbad, CA) for 20 minutes at 37°C. Cells were washed twice with PBS and filtered through a 70 uM cell strainer to obtain a single cell suspension. To assess the levels of CD34 cells in the bulk cell suspension, cells were labeled with PE conjugated anti-CD34 antibody (BD Biosciences, San Jose, CA) and analyzed by FACS. To purify the CD34 cells, Direct CD34 Progenitor Cell Isolation Kit (Miltenyi Biotech, Auburn, CA) was used following the manufacturer's protocol. Isolated CD34 hematopoietic progenitor stem cells were then analyzed by FACS as mentioned above to determine cell purity. For comparative experiments, human CD34 hematopoietic progenitor cells were also purified from fetal liver tissue as described above. CD34 cells were cultured initially in semisolid media to derive myelomonocytic colonies followed by liquid culture in cytokine supplemented media as described below. Purified CD34+ progenitor cells (~2.5 × 10to 4.0 × 10) were placed directly into Methocult semisolid medium (Stem Cell Technologies, Vancouver, BC), mixed, and cultured in 35 mm plates. Myeloid colonies were allowed to develop for 12–15 days. Upon differentiation and proliferation, myelomonocytic colonies were harvested by the addition of 5 ml DMEM containing 10% FBS, 10 ng/ml each GM-CSF and M-CSF. Cells (~10) were placed in a 35 mm well and allowed to adhere for 48 hours. At two and four days post-harvest, medium was replaced with fresh complete DMEM supplemented with 10 ng/ml GM-CSF and M-CSF. By 4–5 days, cells developed into mature macrophages which were used for subsequent phenotypic and functional characterization. To determine if nontransduced and lentiviral vector transduced hES cell derived macrophages display normal macrophage surface markers, FACS analysis was performed using respective fluorochrome conjugated antibodies. Fetal liver derived CD34+ cells as well as nontransduced and transduced hES cell derived macrophages were evaluated in parallel. Cells were scraped from their wells, washed two times with PBS, and stained with the following antibodies: PE-CD14, PE-HLA-DR, PECY5-CD4, PECY5-CCR5, PECY5-CXCR4 (BD Biosciences, San Jose, CA). A blocking step was first performed by incubating the cells with the respective isotype control for 30 minutes at 4C before staining with the respective cell surface marker antibodies. Isotype control staining was used to determine background levels. FACS analysis was performed on a Beckman-Coulter EPICS XL-MCL flow cytometer with data analysis using EXPO32 ADC software (Coulter Corporation, Miami, FL). A minimum of 8,000 cells were analyzed in each FACS evaluation. Physiological roles of macrophages include phagocytic and immune related functions. To determine if hES cell derived macrophages were functionally normal, a stimulation assay to determine upregulation of the costimulatory molecule B7.1 was performed. Activated macrophages upregulate the expression of B7.1 upon activation with various stimuli. Accordingly, fetal liver CD34, nontransduced hES, and GFP-alone transduced hES cell derived macrophages were stimulated by the addition of LPS (5 ug/ml) to the cell culture medium. Twenty-four hours post-stimulation, cells were stained with an anti-B7.1 antibody labeled with PE-Cy5 (BD Biosciences, San Jose, CA) and analyzed by FACS. To assess the hES cell derived macrophages' phagocytic function, 5 ug/ml of fluorescently labeled E. coli Bioparticles(Invitrogen, Carlsbad, CA) were added directly to the cell culture medium. Four hours later, macrophages were washed six times with PBS and fresh medium with 10 ng/ml GM-CSF and M-CSF was added. Twenty-four hours later, cells were analyzed by FACS for the presence of ingested Bioparticleswhich can be detected in the PE (FL2) channel. Lentiviral vector transduced Magi-CXCR4 cells, a HeLa cell derivative with no phagocytic capacity, were used as non-phagocytic cell controls and similarly exposed to E. coli Bioparticles Human ES cell derived macrophages were also analyzed for their ability to secrete two major cytokines, IL-1 and TNF-α, upon external stimulation. Accordingly, macrophages were stimulated with 5 ug/ml of LPS during culture. On days 1, 2, and 3 post-stimulation, cell culture supernatant samples were collected and analyzed by a QuantikineELISA kit (R&D Systems, Minneapolis, MN). Non-stimulated supernatants were also analyzed for basal levels of cytokine secretion. To determine if hES cell derived macrophages can be infected with HIV-1 and support viral replication, cells were challenged with a macrophage R5-tropic BaL-1 strain of HIV-1. An m.o.i. of 0.01 in the presence of 4 ug/ml polybrene was used. At different days post-infection, culture supernatants were collected and assayed for p24 antigen by ELISA. To quantify viral p24 levels, a Coulter-p24 kit (Beckman Coulter, Fullerton, CA) was used. The author(s) declare that they have no competing interests. JA and SB contributed equally to this work. SB was responsible for deriving CD34 cells from the hESC and culturing macrophages. JA performed the phenotypic, functional and infection assays on the differentiated macrophages. DSK provided hES cell protocols and supplied lentiviral vector transduced cells. RA was responsible for the overall experimental design and implementation of the project. hES cell derived macrophages support productive HIV-1 infection: Macrophages derived from transduced and nontransduced hES CD34 and fetal liver CD34 cells were infected with macrophage R5-tropic HIV-1 BaL-1 strain at an m.o.i. of 0.01. Culture supernatants were collected on different days post infection and assayed for viral p24 antigen by ELISA. Data is representative of triplicate experiments.
|
PMC12217045
|
M1 Macrophage‐Derived Extracellular Particles Induce Cell Death in MDA‐MB‐231 Cells
|
Triple‐negative breast cancer (TNBC), a leading cause of female mortality worldwide, presents a treatment challenge due to the lack of targeted receptors. Macrophages, recognized for their role in the immune response, provide a promising avenue for cancer research. Given that macrophages secrete extracellular particles (EPs), which have been implicated in biological processes, including intercellular communication and immune modulation, it is hypothesized that EPs derived from macrophages could have potential anticancer effects. This study examines the effect of M1 macrophage‐secreted EPs on TNBC cells to investigate their potential as a therapeutic. Polarization was induced in RAW 264.7 macrophages and characterized using ELISA, nitrite release, and microscopy. Macrophage‐derived EPs were isolated and characterized using nanoparticle tracking analysis, electron microscopy, and western blotting. The influence of EPs on MDA‐MB‐231 cells, a TNBC model, was assessed using confocal microscopy. Results showed the increasing expression of caspase 3/7 in a time‐dependent manner (0, 24, and 48 h). Cell death was observed in TNBC cells with M1 macrophage‐derived EPs, while cell proliferation was observed when M2 macrophage‐derived EPs interacted with MDA‐MB‐231 cells. Overall, results showed that EPs derived from M1 macrophages could induce cell death in MDA‐MB‐321 cells, opening up potential options for new treatments in TNBC.Breast cancer is the leading cause of death in women worldwide [1, 2]. Among the different types of breast cancer, they are named according to their respective receptor presence, namely human epidermal growth factor receptor (HER) 2 positive, progesterone receptor (PR) positive, estrogen receptor (ER) positive breast cancer, and triple‐negative breast cancer (TNBC). TNBC is the most aggressive type of breast cancer, and it is the leading cause of death in females aged 20–59 due to its metastatic nature . According to Gogate et al., the number of metastatic breast cancer cases is predicted to increase by 54.8% by 2030 in the United States . Currently, pembrolizumab is the only cancer immunotherapy approved by the FDA to treat early‐stage TNBC . To meet the demand of these increasing metastatic breast cancer cases, there is a need to develop more effective and safe therapies that can help overcome TNBC . Extracellular particles (EPs) are nanosized multimolecular bodies secreted by various cell types in the body that include but are not limited to extracellular vesicles (EVs) or vesicle‐like structures, according to the recent guidelines of the International Society of Extracellular Vesicles (ISEV) [6, 7]. The role of EPs is variable, though they are often considered cargo carriers that mediate cell‐to‐cell communication . EPs are composed of vesicular and non‐vesicular structures derived from the cells, and due to their natural ability to encapsulate and carry cellular materials like nucleic acids, proteins, chemokines, and cytokines, they can be leveraged as a potential candidate for cancer therapy. The role of EPs in TNBC therapeutics is a topic of debate , especially since small EVs derived from TNBC cells can create an immunosuppressive environment by attacking critical immune cells like T cells and inducing apoptosis [10, 11]. They have also been shown to affect the differentiation of monocytes into dendritic cells (DCs) and induce myeloid‐derived suppressor cells (MDSCs) [12, 13]. Immature MDSCs have been indicated to help tumors evade immune surveillance and make different cancer treatments ineffective [14, 15]. On the other hand, small EVs derived from various immune cells, such as DCs, natural killer cells (NK cells), T cells, and B cells have shown anti‐cancer or pro‐tumorigenic properties . Tumor progression has been linked with chronic inflammation and dysregulated activity of immune cells and is supported by a complex tumor microenvironment composed of different immune cells, extracellular matrix, non‐immune cells, and vascular structures. Macrophages, a central cell type in tumor biology and immunology, play a crucial role in tumor progression or inhibition based on the signaling molecules they receive from the tumor microenvironment [18, 19]. Tumor‐derived EVs have been shown to change the fate of macrophage polarization, which determines the tumor‐inhibitory or tumor‐promoting effect of macrophage cells . Researchers have found that breast cancer‐derived EVs can influence macrophages by regulating different pathways to support the tumor microenvironment [17, 20, 21]. Generally, macrophages polarized in the M1 phenotype (classically activated) indicate tissue inflammation, and the M2 phenotype (alternatively activated) creates an anti‐inflammatory environment. The M1 phenotype is characterized by increased inducible nitric oxide (iNOS) synthase, an anti‐tumorigenic and inflammatory marker, while the M2 phenotype is more of a wound‐healing phenotype supporting tissue growth, cell migration, and metastasis via upregulation of arginase‐1 and CD‐206 molecules [22, 23]. Other papers have also detailed the role of tumor‐associated M2 macrophages, which have been associated with poor prognosis, chemoresistance, and metastasis [24, 25]. EVs derived from M1 macrophages have been shown to exhibit the ability to transport a diverse array of chemokines, cytokines, and cellular proteins, demonstrating a potentially promising avenue for advancing TNBC treatment strategies [26, 27, 28]. For example, Baek et al. developed effective PEGylated M1 macrophage‐derived exosome mimetic nanovesicles (MNVs) to increase the accumulation of MNVs in the tumor. Their research showed M1 macrophage‐derived exosome mimetic nanovesicles enhanced cancer‐targeting ability in a CT 26 tumor model . Moreover, studies have demonstrated that when 4T1, a breast cancer tumor model, was targeted with docetaxel‐loaded M1‐derived EVs, it prompted the transformation of M0 macrophage phenotypes into the M1 phenotype within the tumor microenvironment . In another study, M1 macrophage‐derived exosomes were targeted against chemoresistance in pancreatic cancer. The exosomes loaded with gemcitabine and deferasirox offer an excellent combination for therapeutic applications to potentially overcome chemoresistance . In addition to these studies, paclitaxel‐loaded M1 macrophages also showed the inhibition of 4T1 breast cancer cells . To leverage the available literature and address current challenges in TNBC therapies, we aimed to explore the effect of M0, M1, and M2 macrophage‐derived EPs on TNBC. Specifically, this study aimed to explore the potential effects of macrophage‐derived EPs on MDA‐MB‐231 cells. Based on literature showing EVs exchange between cell types in the tumor microenvironment , we hypothesized that if cancer cell‐derived EVs can trigger M2 macrophages to support breast cancer tumor growth, we can use macrophage‐derived EPs to check their effect against TNBC. To test our hypothesis, EPs were isolated from RAW 264.7 cells stimulated with lipopolysaccharide (LPS) or interleukin‐4 (IL‐4) and categorized into M0, M1, and M2 macrophages (Figure 1). Experimental design schematic: 1. RAW 264.7 cells were grown in a 6‐well plate to compare M0, M1, and M2 stages. The cells' M1 and M2 states were polarized using lipopolysaccharides (LPS) and interleukin (IL‐4) as stimuli, respectively. The cells and collected cell culture media were subjected to characterization of polarization markers. 2. RAW 264.7 cells were cultured in conditioned media that was fetal bovine serum (FBS)‐depleted to isolate and purify EPs from collected cell culture media in M0, M1, and M2 polarized states. All EPs were characterized after isolation and purification. 3. Isolated EPs were tested against MDA‐MB‐231 cells and caspase 3/7. EP‐isolated conditioned media was added to MDA‐MB‐231 cells to differentiate between cell culture media effects from M0, M1, and M2 RAW 264.7 cells with M0, M1, and M2 isolated EPs. 4. An XTT assay was performed to measure cell viability. The up arrows (↑) indicate increased proliferation and (↓) indicate decreased proliferation. Created with BioRender.com. RAW 264.7 macrophages (ATCC, USA) were cultured using Dulbecco's Modified Eagle Medium DMEM (Gibco, Cat #10569010) with 1% L‐glutamine (Gibco, Cat #25030081) and 10% heat‐inactivated FBS (Gibco, Cat #16140071) at 37°C and 5% CO2. MDA‐MB‐231 cells (ATCC, USA) were cultured using Dulbecco's Modified Eagle Medium DMEM (Gibco, Cat #10569010) with 1% L‐glutamine (Gibco, Cat #25030081) and 10% heat‐inactivated FBS (Gibco, Cat #16140071) at 37°C and 5% CO2. The cells were monitored and passaged when they reached 80% confluency. RAW 264.7 cells were seeded at a density of 4 × 10 cells per well in 6‐well plates to induce polarization into M1 and M2 phenotypes using lipopolysaccharide (LPS) (Thermo Fisher, USA) and mouse interleukin 4 (IL‐4) (eBioscience, Cat#BMS338), respectively. Polarization was induced based on a previously developed protocol . Briefly, RAW 264.7 cells were stimulated with LPS and IL‐4 at the following concentrations: 0, 25, 50, and 100 ng/mL. Macrophages were incubated with inducing agents for 24 h, and polarization was verified under an Olympus IX51 bright‐field microscope. Images were taken using the SeBaView software and an external camera. To verify RAW 264.7 cells polarization, cells were seeded at a density of 4 × 10 per well in a 6‐well plate. The levels of pro‐inflammatory markers IL‐6 and TNF‐α were assessed using an IL‐6 mouse ELISA kit (Invitrogen, Cat#KMC0061) and a TNF‐α mouse ELISA kit (Invitrogen, Cat#BMS607‐3). We followed the manufacturer's protocol to perform the assay. All the reagents and standards were prepared at room temperature according to the protocol. Next, 50 μL of cell culture media from M0 uninduced (Control), M1 LPS‐induced, and IL‐4‐induced M2 cells was collected to take the readings, and color change was measured using a multi‐mode microplate reader at a wavelength of 450 nm (Synergy MX). RAW 264.7 cells at a density of 4 × 10 per well in a 6‐well plate were treated with LPS at different concentrations ranging from 0 to 100 ng/mL. After incubation for 24 h, cell culture media was collected to assess the presence of nitrites using a Griess reagent kit (Invitrogen, Cat#G7921), according to the manufacturer's protocol. Briefly, reagents were combined in a 1:1 ratio and incubated with the samples collected from LPS‐induced RAW 264.7 cells. Nitrite levels were compared among M0 uninduced (Control), LPS‐induced M1 polarized, and IL‐4‐induced M2 polarized cells. Absorbance was determined at 540 nm using a multi‐mode microplate reader (Synergy MX). RAW 264.7 cells (8 × 10 cells) were washed three times with 1× sterile PBS and cultured in serum‐free media (SFM) for 3 days in a T75 flask, as described in [34, 35]. After 72 h, media was collected and subjected to centrifugation for 10 min at 1000 rpm, 4°C, and 30 min at 3000 rpm, 4°C to remove cell debris. Per the manufacturer's protocol, the remaining cell culture media was subjected to EP isolation using an ExoQuick‐TC isolation kit (Systems Biosciences, USA). Briefly, the isolation solution was added to the collected cell culture media in a ratio of 1:5 and incubated for 24 h at 4°C. Following this incubation, the solution was centrifuged at 1500 rpm for 30 min at 4°C to pellet the EPs and purified further according to the manufacturer's instructions. Finally, a series of centrifugation and washing steps were followed to yield EPs. According to this study's purification method and characterization, we refer to our isolated vesicles by the generic term “EPs”. Isolated EPs were further characterized using nanoparticle tracking analysis (NTA) for particle concentration and size. Approximately 0.3 mL of isolated EPs were loaded into the sample chamber of an LM10 unit (Nanosight, Amesbury, UK), and data analysis was performed with the NTA software (Nanosight). In NTA, the paths of EPs act as point scatterers, undergoing Brownian motion in a chamber through which a 632 nm laser beam is passed. Samples were analyzed using the following parameters: the shutter speed was 15 ms, with camera gains between 280 and 560. Software settings for analysis were detection threshold: 12 multi; blur size: auto; frames per second: 23.75; measurement time: 30 ms. When samples contained higher numbers of particles, they were diluted before analysis, and the relative concentration was calculated according to the dilution factor. After purification, EPs were pelleted at 4°C for 100 min at 14000 g and resuspended in a 3.7% glutaraldehyde solution, incubated for 30 min, and re‐centrifuged at 4°C for 100 min. The supernatant was discarded, and EP pellets were washed in 40%, 60%, and 95% ethanol for 15 min each. During the final step, EPs were resuspended in 95% ethanol and deposited on a silicon wafer for SEM analysis. Samples were left overnight to dry under a laminar flow. MDA‐MB‐231 cells at the density of 4 × 10 per well in a 6‐well plate; M1 RAW 264.7 cells derived EPs with 2 × 10 particles/mL were added to MDA‐MB‐231 cells. After 24 h, the silicon wafer was dipped into 3.7% glutaraldehyde for 30 min. Then, the extra solution was removed and transferred to 40%, 60%, and 95% ethanol. Samples were left to dry overnight before observation under SEM. The interaction between the EPs derived from M1 RAW 264.7 cells and MDA‐MB‐231 cells. Samples were then observed under SEM (JEOL JSM‐IT800 Schottky FESEM). Protein levels were detected after isolating EPs from RAW 264.7 cells and characterized by western blot analysis, based on previously published protocols [20, 30, 31]. Briefly, total protein concentration from the samples was measured using a Qubit 4 Fluorometer (Invitrogen, Thermo Fisher). Next, 4× Laemmli buffer was added to samples at a 1:3 ratio and heated to 75°C for 10 min. 19.5 μL of each normalized sample was loaded onto a 12% Mini‐PROTEAN TGX Precast Protein Gel (Bio‐Rad, Hercules, CA). Proteins were separated by electrophoresis, followed by transfer to 0.45 μm nitrocellulose membranes (ThermoFisher). Membranes were blocked with StartingBlock T20 (TBS) blocking buffer for 30 min at 4°C, followed by washing with phosphate buffer saline‐Tween (PBST). RAW 264.7 cell‐derived EP proteins were assessed using anti‐Hsp70 (ab181606, Abcam). Membranes were incubated with primary antibodies (1:1000) overnight at 4°C. Then, membranes were washed 3× for 10 min each in PBST. For primary antibodies, Hsp 70 was incubated with goat anti‐rabbit IgG (A16104, ThermoFisher). Secondary antibodies were diluted in 1:10000, and the membranes were incubated with the secondary antibodies for 30 min at 4°C. The blot was imaged under the iBright Imaging System (Invitrogen, model #FL1500). Each western blot was repeated three times, with the representative images shown in this manuscript. MBA‐MB‐231 cells were stained with BioTracker 405 Blue Mitochondria Dye (Sigma Aldrich, USA, Cat#SCT135). Briefly, MDA‐MB‐231 cells were counted, and the mitochondria tracking dye was added to a final concentration of 40 nM. Cells were incubated at 37°C for 30 min with the dye and seeded in confocal dishes for 24 h. Cells were allowed to attach to the surface of a 4‐chamber confocal dish. The media was replaced the next day with fresh media. Then, EPs isolated from M0 (Control), M1, and M2 polarized RAW 264.7 cells at a concentration of 2 × 10 particles/mL were added to each well in addition to Cell Event Caspase‐3/7 Green ReadyProbes Reagent (Invitrogen, Cat#R37111). Images were taken using an Evident FV3000 confocal microscope using a 20× objective at 0, 24, and 48 h time points. A cell viability assay was performed using a CyQUANT XTT (2,3‐bis(2‐methoxy‐4‐nitro‐5‐sulfophenyl)‐2H‐tetrazolium‐5‐carboxanilide) cell viability assay kit (ThermoFisher, USA). Following the manufacturer's protocol, MDA‐MB‐231 at 5000 cells/well were seeded in a 96‐well plate for 24 h. After 24 h, M0 (control), M1, and M2 RAW 264.7 cell‐derived EPs were added at a concentration of 2 × 10 particles/mL to measure cell viability in each sample allowed to incubate. After 24 h incubation with EPs, 1 mL of electron coupling reagent was mixed with 6 mL of XTT reagent and incubated at 37°C for 4 h in a 5% CO2 incubator. Absorbance was measured at 450 and 660 nm using a Synergy MX multi‐mode microplate reader. The data analysis was performed according to the manufacturer's protocol. Statistical analyses were performed using Microsoft Excel. A Student's t‐test was employed for p‐value determination across all experiments (N = 3). All data are expressed as the mean ± SEM (N = 3). All raw data readings were exported into Excel, where subsequent calculations were performed, and the resulting data were plotted. Macrophages are innate immune cells characterized by their functional characteristics and environmental responses, known as M0 (resting/not activated), M1 (classically activated), or M2 (alternatively active). Lipopolysaccharide (LPS) is an inducer of the M1 macrophage phenotype, while IL‐4 is an inducer of the M2 macrophage phenotype [33, 36]. Accordingly, we used different concentrations of LPS to polarize RAW 264.7 cells and monitored their morphology via optical microscopy. After 24 h, round cells were observed in the non‐induced (M0 macrophages) RAW 264.7 cells, the control group (Figure 2a), while RAW 264.7 cells with LPS (M1 macrophages) looked like dendritic or star‐like shapes (Figure 2b–d), indicating a morphological change. (Figure 2f–h) indicate insignificant changes in shape under the influence of IL‐4 (M2 macrophages) compared to non‐induced RAW 264.7 cells (Figure 2e), the control group. Thus, macrophage polarization was determined morphologically after 24 h under external stimuli of LPS and IL‐4 under different concentrations (0–100 ng/mL). We found that LPS‐induced (M1 macrophages) RAW 264.7 cells showed noticeable differences in morphology compared to non‐induced cells (M0 macrophages) in Figure 2a, while IL‐4‐induced (M2 macrophages) RAW 264.7 cells showed insignificant morphological differences compared to non‐induced (M0 macrophages) cells in Figure 2e. RAW 264.7 cells under the influence of different stimuli: RAW 264.7 cells were observed for M1 polarization in the presence of LPS after 24 h (a) 0, (b) 25, (c) 50, and (d) 100 ng/mL. M2 polarization was observed in the presence of IL‐4 after 24 h (e) 0, (f) 25, (g) 50, and (h) 100 ng/mL. The observation was made using a bright field microscope with a 10× objective and an external camera; images were collected using SeBaView software. LPS‐induced M1 macrophages have been shown to secrete markers, such as tumor necrosis factor‐α (TNF‐α), nitric oxide (NO), and interleukin‐6 (IL‐6), indicating their pro‐inflammatory state . RAW 264.7 cells were challenged with LPS to induce an inflammatory or M1 polarized state. Subsequently, TNF‐α, nitrite production, and IL‐6 levels released by cells were quantified by collecting cell culture media. Results showed that LPS‐induced TNF‐α, nitrite, and IL‐6 concentrations increased in a dose‐dependent manner (Figure 3). TNF‐α and IL‐6 were quantified in cell culture media using ELISA (Figure 3a,b), and the nitrite level was assessed separately in the supernatant using the Griess reaction. We observed the increased nitrite production in a dose‐responsive manner (Figure 3c) compared to M0 or uninduced RAW 264.7 cells. These results confirmed the activated stage of RAW 264.7 cells in a pro‐inflammatory state after LPS induction. RAW 264.7 cell polarization using biochemical analysis: RAW 264.7 cells were cultured in 6‐well plates and incubated for 24 h before collecting cell culture media. Collected media was subjected to enzyme‐linked immunosorbent assay (ELISA), (a) and (b) show the dose‐dependent increase in the levels of IL‐6 and TNF‐α cytokines. (c) Shows the dose‐dependent increase in the levels of nitrites. Results indicate that M0 polarized RAW 264.7 cells, the control, have low nitrites produced compared to LPS‐induced macrophages. N = 3 independent experiments. Error bars = SEM; connecting bars denote the statistical comparison between groups, with p‐value < 0.05 considered statistically significant. Next, EPs derived from three different macrophage states, M0 (uninduced), M1 (induced with LPS), and M2 (induced with IL‐4), were characterized. The characterization was done to identify the type of particle secreted by RAW 264.7 cells under different stimuli. According to the MISEV guidelines [6, 7], EP size was analyzed by NTA and the mode size found among the samples was 27–76 nm, regardless of their polarization state. In addition to their size, it was also observed that the number of particles varies with M0, M1, and M2 polarized states (Figure 4a–c). A scanning electron microscope (SEM) was used to confirm their size and morphology. The EPs had a heterogeneous size range on the order of 100 nm (Figure 4d). Next, protein content in isolated EPs was analyzed using western blot analysis (Figure 4e). After confirming enrichment of heat shock protein 70 (HSP 70), isolated particles from RAW 264.7 cells were classified as EPs. Taken together, these results indicate the successful isolation of EPs from RAW 264.7 cells in M1, M2, and M0 polarization states. Characterization of EPs isolated from polarized RAW 264.7 cells in their M0, M1, and M2 states: (a–c) particle size distribution showing a heterogeneous population of EPs derived from RAW 264.7 macrophages. The plots show the concentration and particle based on a 15 ms camera shutter speed and 23.75 frames per second per sample type. The data were obtained in the NTA 2.0 analytical software and processed using Microsoft Excel. (d) The SEM image of RAW 264.7 cells derived EPs. EP sizes were consistent with NTA analysis. (e) The presence of HSP 70 was demonstrated with western blotting among all the EPs derived from M0, M1, and M2 polarized RAW 264.7 cells. To evaluate the potency of M0 (Control), M1, and M2 macrophage‐derived EPs on MDA‐MB‐231 cells, EPs were incubated with MDA‐MB‐231 cells for 0 to 48 h. Cells were stained with 405 mitochondrial blue dye and caspase 3/7 dye. Figure 5a–i represents the MDA‐MB‐231 cells at the 0 h time point. The initiation of caspase 3/7 in MDA‐MB‐231 cells was observed after 24 h, Figure 6a–i. At the 48 h time point, Figure 7e shows the dominance of green color indicative of caspase 3/7 activation in MDA‐MB‐231 cells incubated with M1‐derived EPs as compared with other EPs (Figure 7b,h). Figure 5e, taken at 0 h, shows no activity of caspase 3 at the beginning of the experiment. In Figure 6e, the white arrows show the slow induction of caspase 3/7 after 24 h with M1‐derived EPs compared to EPs derived from M0 and M2 RAW 264.7 cells. Possible cytoplasmic blebs indicated by white arrows in Figure 7e were also observed in MDA‐MB‐231 cells after 48 h of incubation with M1‐derived EPs. Thus, these results suggest the possibility of M1‐derived EPs yielding a cytotoxic effect against MDA‐MB‐231 cells after 48 h. Evaluating caspase‐3/7 expression induced by M1 RAW 264.7 cell‐derived EPs: Confocal images show MDA‐MB‐231 cells at 0 h incubated with EPs derived from M0 macrophages (a–c), M1 macrophages (d–f), and M2 macrophages (g–i). Cells were stained with BioTracker 405 Blue Mitochondrial Dye to visualize mitochondria (a, d, g), while the GFP channel (502 nm excitation) highlights caspase‐3/7 activation (b, e, h). At the 0 h time point, no caspase‐3/7 activation is observed, confirming that the cells are healthy prior to treatment. The scale bar represents 20 μm. Caspase‐3/7 activation induced by M1 RAW 264.7 cell‐derived EPs at 24 h: Confocal images show MDA‐MB‐231 cells after 24 h of incubation with EPs derived from M0 macrophages (a–c), M1 macrophages (d–f), and M2 macrophages (g–i). Cells were stained with BioTracker 405 Blue Mitochondrial Dye to visualize mitochondria (a, d, g), while the GFP channel (502 nm excitation) highlights caspase‐3/7 activation (b, e, h). At the 24 h time point, caspase‐3/7 activation begins to emerge, indicated by the fluorescence signal in the GFP channel, particularly in cells treated with M1‐derived EPs (e). The scale bar represents 20 μm. Caspase 3/7 was triggered by M1 RAW 264.7 cell‐derived EPs: The induction of cell death at 48 h in MDA‐MB‐231 cells incubated with M0‐derived EPs (a–c), M1‐derived EPs (d–f), and M2‐derived EPs (g–i). Cells were stained with BioTracker 405 Blue Mitochondrial Dye (a, d, g), and the GFP column at 502 nm excitation (b, e, h) shows the caspase 3/7 activation. The arrows in (e) may indicate cytoplasmic blebs/cell death in small dots around cells. The indicated scale bar is 20 μm. After observing the effects of M1 macrophage‐derived EPs against TNBC, an experiment was performed to investigate interactions via SEM. Samples were prepared using a sterile silicon chip and placed in a 6‐well cell culture plate. After 24 h, EPs were found attaching themselves to MDA‐MB‐231 cells (Figure S1b). Figure S1a shows no morphology change in MDA‐MB‐231 cells when EPs were not added; MDA‐MB‐231 cells clumped together in a ball shape when they were attached to a silicon chip without EPs. While Figure S1b shows cell division, we can differentiate morphological differences in Figure S1a,b. In addition, EPs were observed to be spreading on MDA‐MB‐231 cells, with some potential areas of attachment or anchoring. EPs were observed spreading on MDA‐MB‐231 cells, which seemed to be anchoring to cancer cells (Figure S2b,c, arrows). Next, XTT analysis was performed to assess the cytotoxicity of M1‐derived EPs on MDA‐MB‐231 cells. XTT assay results showed only 47% MDA‐MB‐231 cell viability after 24 h of treatment with M1 RAW 264.7 cell‐derived EPs (Figure 8). An MDA‐MB‐231 cell viability of 206% in the sample containing M2 RAW 264.7 cell‐derived EPs was observed, which supports the published literature that suggests that the M2 macrophage‐like phenotype bolsters cancer cell growth . On the other hand, M1‐derived EPs appeared to inhibit the growth of MDA‐MB‐231 cells. Taken together, these results indicate that M1 macrophage‐derived EPs may induce cell death in TNBC cells. MDA‐MB‐231 XTT cell viability assay: MDA‐MB‐231 cells were exposed to EPs derived from M0 (Control), M1, and M2 RAW 264.7 cells. N = 3 independent experiments normalized to the control. Error bars = SEM, connecting bars denote a p‐value < 0.05 and were considered statistically significant. This study indicates the potential for M1 macrophage‐derived EPs to favor tumor interactions and that M1 macrophage‐derived EVs could exhibit anti‐cancer effects. Several studies have explored the potential role of M1 macrophage‐derived EVs' role in TNBC therapy as a potential cytotoxic agent in vitro. For example, macrophage‐derived exosome membranes have been used to load doxorubicin after removing the luminal content and conjugating with PLGA to target c‐Met to show its ability to target TNBC tumors in vivo . However, the effect of EPs derived from macrophages with their native luminal content was never explored. This study identifies and characterize EPs with sizes less than 200 nm in each sample by following MISEV guidelines [6, 7]. A nitrite concentration of 45 μM in collected cell culture media of M1 polarized RAW 264.7 cells was measured [40, 41]. M1 RAW 264.7 cell‐derived EPs with luminal content showed cytotoxicity against MDA‐MB‐231 cells within 48 h by showing time‐dependent increments in caspase‐3/7 expression compared to the M0 uninduced RAW 264.7 cells. This time‐dependent increase in caspase‐3/7 activations in a controlled manner may indicate the EP's ability to release their content in a controlled fashion [42, 43]. Moreover, this study further confirms the ability of M1 RAW 264.7 cell‐derived EPs to decrease cell viability by 53% in 24 h (Figure 8). On the other hand, M2 RAW 264.7 cell‐derived EPs increased the cell viability of MDA‐MB‐231 cells by 206% within 24 h (Figure 8), showing their potential tumor‐supporting effect [22, 23]. In addition, we captured the interaction of EPs with cancer cells that, to our knowledge, have yet to be visualized via SEM. The reason for this anchoring is unclear, but it could be the nature of M1 RAW 264.7 cell‐derived EPs to stimulate the change in morphology of MDA‐MB‐231 cells (Figures S1 and S2). Another reason could be the fusion with the cell membrane of MDA‐MB‐231 cells, which helps EPs internalize or release their content in the MDA‐MB‐231 cell cytosol . This study revealed that M1 RAW 264.7 cell‐derived EPs can interact with MDA‐MB‐231 cells and induce cell death via caspase 3/7 activation. Anchoring of EPs on TNBC cells has been suggested, which indicates that the other mechanisms of M1 macrophage‐derived EPs could be used to interact with tumor cells . Furthermore, this study paves the way for exploring the content of macrophage‐derived EPs in detail, which can impact the fate of MDA‐MB‐231 cells. In the future, M1 macrophage‐derived EPs could be combined with commercially available therapeutics to characterize the potential for synergistic therapeutic effects . The authors declare no conflicts of interest.
|
PMC10287567
|
An integrated cell atlas of the lung in health and disease
|
Single-cell technologies have transformed our understanding of human tissues. Yet, studies typically capture only a limited number of donors and disagree on cell type definitions. Integrating many single-cell datasets can address these limitations of individual studies and capture the variability present in the population. Here we present the integrated Human Lung Cell Atlas (HLCA), combining 49 datasets of the human respiratory system into a single atlas spanning over 2.4 million cells from 486 individuals. The HLCA presents a consensus cell type re-annotation with matching marker genes, including annotations of rare and previously undescribed cell types. Leveraging the number and diversity of individuals in the HLCA, we identify gene modules that are associated with demographic covariates such as age, sex and body mass index, as well as gene modules changing expression along the proximal-to-distal axis of the bronchial tree. Mapping new data to the HLCA enables rapid data annotation and interpretation. Using the HLCA as a reference for the study of disease, we identify shared cell states across multiple lung diseases, including SPP1 profibrotic monocyte-derived macrophages in COVID-19, pulmonary fibrosis and lung carcinoma. Overall, the HLCA serves as an example for the development and use of large-scale, cross-dataset organ atlases within the Human Cell Atlas.Rapid technological improvements over the past decade have allowed single-cell datasets to grow both in size and number. This has led consortia, such as the Human Cell Atlas, to pursue the generation of large-scale reference atlases of human organs. To advance our understanding of health and disease, such atlases must capture variation between individuals that is expected to impact the molecular phenotypes of the cells in a tissue. Whereas the generation of atlases at this scale by single research groups is currently not feasible, integrating datasets generated by the research community at large will enable capture of the diversity of the cellular landscape across individuals. Several foundational studies have started to map the cellular landscape of the healthy human lung. These studies each have a specific bias due to their choice of experimental protocol and technologies, and are therefore not tailored to serve as a universal reference. The studies moreover include only a limited number of samples and individuals, thus lacking the scale and diversity to capture the full cellular heterogeneity present within the lung as well as across individuals. Integrated single-cell atlases provide novel insights not obtained in individual studies. Recent reference atlases have led to the discovery of unknown cell types, the identification of marker genes that are reproducible across studies, the comparison of animal and in vitro models with human healthy and diseased tissue and patient stratification for disease endotypes. However, many currently available integrated atlases are limited in the number of human samples, datasets or cell types per organ, as well as donor metadata (for example, age, body mass index (BMI) and smoking status), or focus mainly on a specific disease. These limitations constrain the potential of atlases to serve as a reference, as they fail to represent and catalog the diversity of cellular phenotypes within the healthy organ and across individuals. Moreover, when integrating data from different sources, it is paramount to correctly separate technical biases from biologically relevant information. Yet, the majority of existing atlases have not assessed the quality of their data integration. Nonetheless, successful integration of the available datasets into a single tissue atlas is a critical step in achieving the goals of the Human Cell Atlas. In this resource, we present an integrated single-cell transcriptomic atlas of the human respiratory system, including the upper and lower airways, from published and newly generated datasets (Fig. 1). The Human Lung Cell Atlas (HLCA) comprises data from 486 donors and 49 datasets, including 2.4 million cells, which we re-annotated to generate a consensus cell type reference. The HLCA expands our understanding of the healthy lung and its changes in disease and can be used as a reference for analyzing future lung data. Together, we provide a roadmap for building and using comprehensive, interpretable and up-to-date organ- and population-scale cell atlases.Fig. 1HLCA study overview.Harmonized cell annotations, raw count data, harmonized patient and sample metadata and sample anatomical locations encoded into a CCF were collected and generated as input for the HLCA core (left). After integration of the core datasets, the atlas was extended by mapping 35 additional datasets, including disease samples, to the HLCA core, bringing the total number of cells in the extended HLCA to 2.4 million (M). The HLCA core provides detailed consensus cell annotations with matched consensus cell type markers (top right), gene modules associated with technical, demographic and anatomical covariates in various cell types (middle right), GWAS-based association of lung conditions with cell types (middle right) and a reference projection model to annotate new data (middle right) and discover previously undescribed cell types, transitional cell states and disease-associated cell states (right, bottom). Harmonized cell annotations, raw count data, harmonized patient and sample metadata and sample anatomical locations encoded into a CCF were collected and generated as input for the HLCA core (left). After integration of the core datasets, the atlas was extended by mapping 35 additional datasets, including disease samples, to the HLCA core, bringing the total number of cells in the extended HLCA to 2.4 million (M). The HLCA core provides detailed consensus cell annotations with matched consensus cell type markers (top right), gene modules associated with technical, demographic and anatomical covariates in various cell types (middle right), GWAS-based association of lung conditions with cell types (middle right) and a reference projection model to annotate new data (middle right) and discover previously undescribed cell types, transitional cell states and disease-associated cell states (right, bottom). To build the HLCA, we collected single-cell RNA sequencing (scRNA-seq) data and detailed, harmonized technical, biological and demographic metadata from 14 datasets (11 published and three unpublished). These datasets include samples from 107 individuals, with diversity in age, sex, ethnicity (harmonized as detailed in Methods), BMI and smoking status (Fig. 2a). Cells were obtained from 166 tissue samples using a variety of tissue donors, sampling methods, experimental protocols and sequencing platforms (Supplementary Tables 1 and 2). Anatomical locations of the samples were projected onto a one-dimensional (1D) common coordinate framework (CCF), representing the proximal (0) to distal (1) axis of the respiratory system, to standardize the anatomical location of origin (Fig. 2a and Supplementary Tables 2 and 3).Fig. 2Composition and construction of the HLCA core.a, Donor and sample composition in the HLCA core for demographic and anatomical variables. Donors/samples without annotation are shown as not available (NA; gray bars) for each variable. For the anatomical region CCF score, 0 represents the most proximal part of the lung and airways (nose) and 1 represents the most distal (distal parenchyma). Donors show diversity in ethnicity (harmonized metadata proportions: 65% European, 14% African, 2% admixed American, 2% mixed, 2% Asian, 0.4% Pacific Islander and 14% unannotated; see Methods), smoking status (52% never, 16% former, 15% active and 17% NA), sex (60% male and 40% female), age (ranging from 10–76 years) and BMI (20–49; 30% NA). b, Overview of the HLCA core cell type composition for the first three levels of cell annotation, based on harmonized original labels. In the cell type hierarchy, the lowest level (1) consists of the coarsest possible annotations (that is, epithelial (48% of cells), immune (38%), endothelial (9%) and stromal (4%)). Higher levels (2–5) recursively break up coarser-level labels into finer ones (Methods). Cells were set to ‘none’ if no cell type label was available at the level. Cell labels making up less than 0.02% of all cells are not shown. Overall, 94, 66 and 7% of cells were annotated at levels 3, 4 and 5, respectively. c, Cell type composition per sample, based on level 2 labels. Samples are ordered by anatomical region CCF score. d, Summary of the dataset integration benchmarking results. Batch correction score and biological conservation score each show the mean across metrics of that type, as shown in Supplementary Fig. 1, with metric scores scaled to range from 0 to 1. Both Scanorama and fastMNN were benchmarked on two distinct outputs: the integrated gene expression matrix and integrated embedding (see output). The methods are ordered by overall score. For each method, the results are shown only for their best-performing data preprocessing. Methods marked with an asterisk use coarse cell type labels as input. Preprocessing is specified under HVG (that is, whether or not genes were subsetted to the 2,000 (HVG) or 6,000 (FULL) most highly variable genes before integration) and scaling (whether genes were left unscaled or scaled to have a mean of 0 and a standard deviation of 1 across all cells). EC, endothelial cell; NK, natural killer; Bioconserv., conservation of biological signal. a, Donor and sample composition in the HLCA core for demographic and anatomical variables. Donors/samples without annotation are shown as not available (NA; gray bars) for each variable. For the anatomical region CCF score, 0 represents the most proximal part of the lung and airways (nose) and 1 represents the most distal (distal parenchyma). Donors show diversity in ethnicity (harmonized metadata proportions: 65% European, 14% African, 2% admixed American, 2% mixed, 2% Asian, 0.4% Pacific Islander and 14% unannotated; see Methods), smoking status (52% never, 16% former, 15% active and 17% NA), sex (60% male and 40% female), age (ranging from 10–76 years) and BMI (20–49; 30% NA). b, Overview of the HLCA core cell type composition for the first three levels of cell annotation, based on harmonized original labels. In the cell type hierarchy, the lowest level (1) consists of the coarsest possible annotations (that is, epithelial (48% of cells), immune (38%), endothelial (9%) and stromal (4%)). Higher levels (2–5) recursively break up coarser-level labels into finer ones (Methods). Cells were set to ‘none’ if no cell type label was available at the level. Cell labels making up less than 0.02% of all cells are not shown. Overall, 94, 66 and 7% of cells were annotated at levels 3, 4 and 5, respectively. c, Cell type composition per sample, based on level 2 labels. Samples are ordered by anatomical region CCF score. d, Summary of the dataset integration benchmarking results. Batch correction score and biological conservation score each show the mean across metrics of that type, as shown in Supplementary Fig. 1, with metric scores scaled to range from 0 to 1. Both Scanorama and fastMNN were benchmarked on two distinct outputs: the integrated gene expression matrix and integrated embedding (see output). The methods are ordered by overall score. For each method, the results are shown only for their best-performing data preprocessing. Methods marked with an asterisk use coarse cell type labels as input. Preprocessing is specified under HVG (that is, whether or not genes were subsetted to the 2,000 (HVG) or 6,000 (FULL) most highly variable genes before integration) and scaling (whether genes were left unscaled or scaled to have a mean of 0 and a standard deviation of 1 across all cells). EC, endothelial cell; NK, natural killer; Bioconserv., conservation of biological signal. Consensus definitions of cell types based on single-cell transcriptomic data across studies—particularly of transitional cell states—are lacking. To enable supervised data integration and downstream integrated analysis, we harmonized cell type nomenclature by building a five-level hierarchical cell identity reference framework (Methods, Supplementary Table 4 and Fig. 2b). We then unified cell type labeling across datasets by mapping the collected cell identity labels for every dataset as provided by the data generator to the hierarchical reference framework, showing varying cell type proportions per sample (Fig. 2c). To optimally remove dataset-specific batch effects, we evaluated 12 different data integration methods on 12 datasets (Fig. 2d and Supplementary Fig. 1) using our previously established benchmarking pipeline. We used the top-performing integration method, scANVI, to create an integrated embedding of all 584,444 cells of 107 individuals from the collected datasets: the HLCA core (Fig. 3a).Fig. 3The HLCA core conserves detailed biology and enables consensus-driven annotation.a, A UMAP of the integrated HLCA, colored by level 1 annotation. b, Cluster label disagreement (label entropy) of Leiden 3 clusters of the HLCA. The HLCA was split into three parts (immune, epithelial and endothelial/stromal) for ease of visualization. Cells from every cluster are colored by label entropy. Clusters with less than 20% of cells annotated at level 3 are colored gray. c, Cell type label composition of the immune cluster with the most label disagreement (left), with original labels (middle left) and matching manual re-annotations (middle right). A zoom-in on the UMAP from b shows the final re-annotations (right). d, UMAPs of the immune, epithelial and endothelial/stromal parts of the HLCA core with cell annotations from the expert manual re-annotation. e, Percentage of cells originally labeled correctly, mislabeled or underlabeled (that is, only labeled at a coarser level) compared with final manual re-annotations. The percentages were calculated per manual annotation, as well as across all cells (right bar). f, UMAP of HLCA clusters annotated as rare epithelial cell types (that is, ionocytes, neuroendocrine cells and tuft cells). Final annotations, original labels and the study of origin are shown (top), as well as the expression of ionocyte marker FOXI1, tuft cell marker LRMP and neuroendocrine marker CALCA (bottom). g, Log-normalized expression of the migratory dendritic cell marker CCR7 in cells identified during re-annotation as migratory dendritic cells, versus other dendritic cells. AT, alveolar type; DC, dendritic cell; FB, fibroblast; Mph, macrophage; MT, metallothionein; SM, smooth muscle; SMG, submucosal gland; TB, terminal bronchiole. a, A UMAP of the integrated HLCA, colored by level 1 annotation. b, Cluster label disagreement (label entropy) of Leiden 3 clusters of the HLCA. The HLCA was split into three parts (immune, epithelial and endothelial/stromal) for ease of visualization. Cells from every cluster are colored by label entropy. Clusters with less than 20% of cells annotated at level 3 are colored gray. c, Cell type label composition of the immune cluster with the most label disagreement (left), with original labels (middle left) and matching manual re-annotations (middle right). A zoom-in on the UMAP from b shows the final re-annotations (right). d, UMAPs of the immune, epithelial and endothelial/stromal parts of the HLCA core with cell annotations from the expert manual re-annotation. e, Percentage of cells originally labeled correctly, mislabeled or underlabeled (that is, only labeled at a coarser level) compared with final manual re-annotations. The percentages were calculated per manual annotation, as well as across all cells (right bar). f, UMAP of HLCA clusters annotated as rare epithelial cell types (that is, ionocytes, neuroendocrine cells and tuft cells). Final annotations, original labels and the study of origin are shown (top), as well as the expression of ionocyte marker FOXI1, tuft cell marker LRMP and neuroendocrine marker CALCA (bottom). g, Log-normalized expression of the migratory dendritic cell marker CCR7 in cells identified during re-annotation as migratory dendritic cells, versus other dendritic cells. AT, alveolar type; DC, dendritic cell; FB, fibroblast; Mph, macrophage; MT, metallothionein; SM, smooth muscle; SMG, submucosal gland; TB, terminal bronchiole. A large-scale integrated atlas provides the unique opportunity to systematically investigate the consensus in cell type labeling across datasets. To identify areas of consensus and disagreement, we iteratively clustered the HLCA core and investigated donor diversity and cell type label agreement in these clusters using entropy scores (see Methods). Most clusters contained cells from many donors (Extended Data Fig. 1a). Clusters with low donor diversity (n = 14) were largely immune cell clusters (n = 13), representing donor- or donor group-specific phenotypes. Similarly, a high diversity of (contradictory) cell type labels (high label entropy) can identify both annotation disagreements between studies and clusters of doublets (Methods). Most clusters (61 out of 94) showed low label entropy, suggesting overall agreement of coarse cell type labels across datasets (Fig. 3b). The remaining 33 clusters exhibited high label entropy, highlighting cellular phenotypes that were differently labeled across datasets (Fig. 3b). For example, the immune cluster with the highest label entropy contained many cells that were originally mislabeled as monocytes and macrophages but were actually type 2 dendritic cells (Fig. 3c and Extended Data Fig. 1b). Thus, populations with high label entropy identify mislabeled cell types, indicating the need for consensus re-annotation of the integrated atlas. As a first step to achieve such a consensus on the diversity of cell types present in the HLCA core, we performed a full re-annotation of the integrated data on the basis of the original annotations and six expert opinions (consensus annotation; Methods and Fig. 3d). Each of the 61 annotated cell types (Supplementary Table 5) was detected in at least four datasets out of 14, often in specific parts of the respiratory system, and different cell types showed varying fractions of proliferating (MKI67) cells (Extended Data Fig. 2a–c). While our consensus cell type annotations partly correspond to original labels (41% of cells), there were also refinements (28%) and substantial re-annotations (31%; Fig. 3e and Supplementary Fig. 2). To robustly characterize the cell types, we established a universal set of marker genes that generalizes across individuals and studies (Methods, Extended Data Fig. 3 and Supplementary Table 6). The fully re-annotated HLCA core thus combines data from a diverse set of studies to provide a carefully curated reference for cell type annotations and marker genes in healthy lung tissue. Rare cell types, such as ionocytes, tuft cells, neuroendocrine cells and specific immune cell subsets, are often difficult to identify in individual datasets. Yet, combining datasets in the HLCA core provides better power for identifying these rare cell types. Ionocytes, tuft and neuroendocrine cells make up only 0.08, 0.01 and 0.02% of the cells in the HLCA core according to the original labels, and were originally identified in only seven, two and four datasets out of 14, respectively. Despite their low abundance, these cells formed three separate clusters of the HLCA core (Fig. 3f). Our re-annotation increases the number of datasets in which these cells are detected up to threefold and identifies both cells falsely annotated as monocytes, tuft cells or neuroendocrine cells, as well as originally undetected rare cells (Fig. 3f and Supplementary Fig. 3a). Importantly, other integration methods tested during our benchmarking, such as Harmony and Seurat’s RPCA, failed to separate these rare cells into distinct clusters (Supplementary Fig. 3b). We were further able to detect six cell identities that were not previously found in the human lung or were only recently described in individual studies. These cell types include migratory dendritic cells (n = 312 cells, expressing CCR7, LAD1 and COL19), hematopoietic stem cells (n = 60, expressing SPINK2, STMN, PRSS57 and CD34), highly proliferative hillock-like epithelial cells not previously reported in adult human lung (n = 4,600, expressing KRT6A, KRT13 and KRT14), the recently described alveolar type 0 cells (n = 1,440, expressing STFPB, SCGB3A2, SFTPC and SCGB3A1) and the closely related preterminal bronchiole secretory cells (n = 4,393, expressing SFTPB, SCGB3A2, SFTPC and SCGB3A1, together with alveolar type 0 cells called transitional club-AT2 cells) and a subset of smooth muscle cells (n = 335) that to our knowledge have not previously been described (Fig. 3d,g and Extended Data Fig. 4a–f). These smooth muscle cells, predominantly found in the airways, express canonical smooth muscle markers (CNN1 and MYH11) and also uniquely and consistently express FAM83D across datasets (Extended Data Fig. 4e,f). The HLCA core thus enables improved detection and identification of rare cell types, as well as the discovery of unknown cell types. Demographic and other metadata covariates affect cellular transcriptional phenotypes. Better insight into the impact of these covariates (for example, sex, BMI and smoking) on cell type gene expression can shed light on the contribution of these factors to progression from healthy to diseased states. In addition, technical covariates such as ribosomal and mitochondrial genes exhibit batch-specific variation in expression (Methods and Supplementary Table 7). The diversity in demographics (for example, smoking status, age, harmonized ethnicity and BMI) and experimental protocols represented in the HLCA core enables us to explore the contribution of each technical or biological covariate to cell type-specific gene expression variation (Methods and Supplementary Fig. 4). For many cell types, anatomical location is the biological variable explaining most of the variance between samples (Fig. 4a). Furthermore, sex is most associated with transcriptomic variation in lymphatic endothelial cells, whereas BMI is most associated with variation in B and T cells, harmonized ethnicity in transitional club-AT2 cells and smoking status in innate lymphoid/natural killer cells. Furthermore, for several cell types (for example, mast, AT1 and smooth muscle cells), the tissue dissociation protocol explains most of the variance of all technical as well as biological covariates recorded. These associations provide a systematic overview of the effects of biological and technical factors on the transcriptional state of lung cell types.Fig. 4Demographic and technical variables driving interindividual variation.a, Fraction of total inter-sample variance in the HLCA core integrated embedding that correlates with specific covariates. Covariates are split into technical (left) and biological covariates (right). Cell types are ordered by the number of samples in which they were detected. Only cell types present in at least 40 samples are shown. Tissue sampling method represents the way a sample was obtained (for example, surgical resection or nasal brush). Donor status represents the state of the donor at the moment of sample collection (for example, organ donor, diseased alive or healthy alive). The heatmap is masked gray where fewer than 40 samples were annotated for a specific covariate or where only one value was observed for all samples for that cell type. b, Selection of gene sets that are significantly associated with anatomical location CCF score, in different airway epithelial cell types. All gene set names are Gene Ontology biological process (GO: BP) terms. Sets upregulated toward distal lungs are shown in green, whereas sets downregulated are shown in blue. The full name of the term marked by an asterisk is ‘Antigen processing and presentation of exogenous peptide antigen via MHC-I’. c, Cell type proportions per sample, along the proximal-to-distal axis of the respiratory system. The lowest and highest CCF scores shown (0.36 and 0.97) represent the most proximal and most distal sampled parts of the respiratory system, respectively (trachea and parenchyma), excluding the upper airways. The dots are colored by the tissue dissociation protocol and tissue sampling method used for each sample. The boxes show the median and interquartile range of the proportions. Samples with proportions more than 1.5 times the interquartile range away from the high and low quartile are considered outliers. Whiskers extend to the furthest nonoutlier point. n = 23, 19, 9 and 90 for CCF scores 0.36, 0.72, 0.81 and 0.97, respectively. d, Selection of gene sets significantly up- (green) or downregulated (blue) with increasing BMI, in four different cell types. For b and d, P values were calculated using correlation-adjusted mean-rank gene set tests (Methods) and false discovery rate corrected using the Benjamini–Hochberg procedure. IL-1, interleukin-1; MHC-I, major histocompatibility complex class I; TNF, tumor necrosis factor. a, Fraction of total inter-sample variance in the HLCA core integrated embedding that correlates with specific covariates. Covariates are split into technical (left) and biological covariates (right). Cell types are ordered by the number of samples in which they were detected. Only cell types present in at least 40 samples are shown. Tissue sampling method represents the way a sample was obtained (for example, surgical resection or nasal brush). Donor status represents the state of the donor at the moment of sample collection (for example, organ donor, diseased alive or healthy alive). The heatmap is masked gray where fewer than 40 samples were annotated for a specific covariate or where only one value was observed for all samples for that cell type. b, Selection of gene sets that are significantly associated with anatomical location CCF score, in different airway epithelial cell types. All gene set names are Gene Ontology biological process (GO: BP) terms. Sets upregulated toward distal lungs are shown in green, whereas sets downregulated are shown in blue. The full name of the term marked by an asterisk is ‘Antigen processing and presentation of exogenous peptide antigen via MHC-I’. c, Cell type proportions per sample, along the proximal-to-distal axis of the respiratory system. The lowest and highest CCF scores shown (0.36 and 0.97) represent the most proximal and most distal sampled parts of the respiratory system, respectively (trachea and parenchyma), excluding the upper airways. The dots are colored by the tissue dissociation protocol and tissue sampling method used for each sample. The boxes show the median and interquartile range of the proportions. Samples with proportions more than 1.5 times the interquartile range away from the high and low quartile are considered outliers. Whiskers extend to the furthest nonoutlier point. n = 23, 19, 9 and 90 for CCF scores 0.36, 0.72, 0.81 and 0.97, respectively. d, Selection of gene sets significantly up- (green) or downregulated (blue) with increasing BMI, in four different cell types. For b and d, P values were calculated using correlation-adjusted mean-rank gene set tests (Methods) and false discovery rate corrected using the Benjamini–Hochberg procedure. IL-1, interleukin-1; MHC-I, major histocompatibility complex class I; TNF, tumor necrosis factor. To better characterize how biological variables affect cellular phenotypes, we modeled their cell type-specific effects on the transcriptome at the gene level (Methods). Sex-related differences in lymphatic endothelial cells are dominated by differential expression of genes located on the X and Y chromosomes, but also include a decrease in IFNAR1 in females (Supplementary Table 8), which may be linked to differential interferon responses between the biological sexes. We furthermore found cell type-specific programs that change with proximal (low CCF score) to distal (high CCF score) location along the respiratory tract (Supplementary Tables 8 and 9). For instance, oxidative phosphorylation (including cytochrome c oxidase genes such as COX7A1), antigen presentation by major histocompatibility complex class I molecules (including proteasome and protease subunit genes such as PSMD14 and PSMB4), signaling by interleukin-1 and tumor necrosis factor α, as well as planar cell polarity, were downregulated toward more distal locations in secretory, multiciliated and basal cells (Fig. 4b). Some gene programs were specific for a subset of airway epithelial cell types (for example, cornification and keratinization, which were programs that were downregulated in distal multiciliated and secretory cells; including genes such as KRT8 and KRT19). The changes in airway epithelial cell states toward the terminal airways are further illustrated by increased expression of developmental pathway genes such as NKX2-1, NFIB, GATA6, BMP4 and SOX9 in multiciliated cells along the proximal-to-distal axis (Fig. 4b), whereas basal cells decrease in number (Fig. 4c). Similarly, several cell types display transcriptomic changes in donors with increasing BMI (Fig. 4d and Supplementary Tables 8 and 9). AT2 cells, secretory cells and alveolar macrophages exhibit downregulation of a range of biological processes (Supplementary Fig. 5), including cellular respiration, differentiation and synthesis of peptides and other molecules. In secretory cells, a downregulation of the insulin response pathway is also associated with higher BMI, in line with the insulin resistance observed in donors with obesity. In alveolar macrophages, inflammatory responses involving JAK/STAT signaling (previously associated with obesity-induced chronic systemic inflammation) are associated with higher BMI. In contrast, in plasma cells, high BMI is associated with downregulation of gene sets associated with immune response and upregulation of gene sets associated with cellular respiration, the cell cycle and DNA repair. This is consistent with obesity being a known risk factor for multiple myeloma—a plasma cell malignancy. Thus, the HLCA enables a detailed understanding of the effects of anatomical and demographic covariates on the cellular landscape of the lung and their relation to disease. Biological and technical factors can also affect cell type proportions. Indeed, all cell types show changes in abundance as a function of anatomical location (Fig. 4c and Extended Data Fig. 5). For example, ionocytes are present at comparable proportions in the airway epithelium, from the larger lower airways (CCF score = 0.36) down to the distal lobular airways (CCF score = 0.81), while being largely absent in the lung parenchyma (CCF score = 0.97). In contrast, neuroendocrine cells are predominantly observed in the larger lower airways but are absent from more distal parts of the bronchial tree (Fig. 4c). In some cases, these proportions are highly dependent on the tissue sampling method and the dissociation protocol used (for example, for smooth muscle FAM83D cells; Extended Data Fig. 5). These observations shed light on the effects of biological and technical factors on the abundance of cell types in different parts of the lung and can help guide important choices in study design. The HLCA core contains an unprecedented diversity of donors, sampling protocols and cell identities, and can serve as a transcriptomic reference for lung research. New datasets can be mapped to this reference to substantially speed up data analysis by transferring consensus cell identity annotations to the new data. We tested this on a recently released multimodal lung dataset (Methods, Fig. 6a and Extended Data Fig. 6). Overall, the transferred labels were correct in the majority of cases, with 68% of the cells correctly labeled, 14% of labels incorrectly labeled and 18% set to unknown due to highly uncertain labeling (Fig. 5b and Methods). Uncertain labels were observed specifically in continuous transitions from one cell type to another and among cellular identities not present in the HLCA core, including rare cell identities (erythrocytes (n = 328), chondrocytes (n = 42), myelinating Schwann cells (n = 7), nonmyelinating Schwann cells (n = 29) and nerve-associated fibroblasts (n = 66); Fig. 5b and Extended Data Fig. 6d). Taken together, these results show that the HLCA core can be used for highly detailed annotation of new datasets, while allowing for the identification of unknown cell types in these datasets based on label transfer uncertainty.Fig. 5The HLCA core serves as a reference for label transfer and data contextualization.a, UMAP of the jointly embedded HLCA core (gray) and the projected healthy lung dataset (colored by label transfer uncertainty). HLCA cell types surrounding regions of high uncertainty are labeled. b, Percentage of cells from the newly mapped healthy lung dataset that are annotated either correctly or incorrectly by label transfer annotation or annotated as unknown, split by original cell type label (number of cells in parentheses). Cell type labels not present in the HLCA are boxed. c, Top, percentage of cells derived from tumor tissue, per endothelial cell cluster from the joint HLCA core and lung cancer data embedding. Only clusters with at least ten tumor cells are shown. Clusters are named based on the dominant HLCA core cell type annotation in the cluster. Middle, box plot showing the expression of EDNRB in endothelial cell clusters, split by tissue source. Bottom, as in the middle plot but for the expression of ACKR1. Numbers of cells per group were as follows: 6,574 (endothelial cell aerocyte capillary), 7,379 (endothelial cell arterial (I)), 10,906 (endothelial cell general capillary (I)), 3,440 (endothelial cell general capillary (II)), 2,859 (endothelial cell general capillary (III)), 6,318 (endothelial cell venous pulmonary) and 7,161 (endothelial cell venous systemic). d, Association of HLCA cell types with four different lung phenotypes based on previously performed GWASs. The horizontal dashed lines indicate a significance threshold of α = 0.05. P values were calculated using linkage disequilibrium score regression (Methods) and multiple testing corrected with the Benjamini–Hochberg procedure. e, Cell type proportions in lung bulk expression samples as estimated from HLCA-based cell type deconvolution, comparing controls (n = 281) versus donors with severe COPD (GOLD stage 3/4; n = 83). f, UMAP of fibroblast-dominated clusters from the jointly embedded HLCA core and mapped healthy lung dataset, colored by spatial cluster, with cells outside of the indicated clusters colored in gray. For all boxplots, the boxes show the median and interquartile range. Data points more than 1.5 times the interquartile range outside the low and high quartile are considered outliers. In c, these are not shown (see Supplementary Fig. 6 for full results), whereas in e, they are shown. Whiskers extend to the furthest nonoutlier point. corr., corrected; FVC, forced vital capacity; MAIT cells, mucosal-associated invariant T cells; NKT cells, natural killer T cells. a, UMAP of the jointly embedded HLCA core (gray) and the projected healthy lung dataset (colored by label transfer uncertainty). HLCA cell types surrounding regions of high uncertainty are labeled. b, Percentage of cells from the newly mapped healthy lung dataset that are annotated either correctly or incorrectly by label transfer annotation or annotated as unknown, split by original cell type label (number of cells in parentheses). Cell type labels not present in the HLCA are boxed. c, Top, percentage of cells derived from tumor tissue, per endothelial cell cluster from the joint HLCA core and lung cancer data embedding. Only clusters with at least ten tumor cells are shown. Clusters are named based on the dominant HLCA core cell type annotation in the cluster. Middle, box plot showing the expression of EDNRB in endothelial cell clusters, split by tissue source. Bottom, as in the middle plot but for the expression of ACKR1. Numbers of cells per group were as follows: 6,574 (endothelial cell aerocyte capillary), 7,379 (endothelial cell arterial (I)), 10,906 (endothelial cell general capillary (I)), 3,440 (endothelial cell general capillary (II)), 2,859 (endothelial cell general capillary (III)), 6,318 (endothelial cell venous pulmonary) and 7,161 (endothelial cell venous systemic). d, Association of HLCA cell types with four different lung phenotypes based on previously performed GWASs. The horizontal dashed lines indicate a significance threshold of α = 0.05. P values were calculated using linkage disequilibrium score regression (Methods) and multiple testing corrected with the Benjamini–Hochberg procedure. e, Cell type proportions in lung bulk expression samples as estimated from HLCA-based cell type deconvolution, comparing controls (n = 281) versus donors with severe COPD (GOLD stage 3/4; n = 83). f, UMAP of fibroblast-dominated clusters from the jointly embedded HLCA core and mapped healthy lung dataset, colored by spatial cluster, with cells outside of the indicated clusters colored in gray. For all boxplots, the boxes show the median and interquartile range. Data points more than 1.5 times the interquartile range outside the low and high quartile are considered outliers. In c, these are not shown (see Supplementary Fig. 6 for full results), whereas in e, they are shown. Whiskers extend to the furthest nonoutlier point. corr., corrected; FVC, forced vital capacity; MAIT cells, mucosal-associated invariant T cells; NKT cells, natural killer T cells. Single-cell studies of disease rely on adequate, matching control samples to allow correct identification of disease-specific changes. To demonstrate the ability of the HLCA core to serve as a comprehensive healthy control and contextualize disease data, we mapped scRNA-seq data from lung cancer samples to the HLCA core (Methods and Extended Data Fig. 7a–c). Using HLCA label transfer, we correctly identified cell states missing from the HLCA core as unknown (cancer cells and erythroblasts). The remaining cells were annotated correctly in 77%, incorrectly in 1% and as unknown in 22% of cases (Extended Data Fig. 7d–g). A finding of the original study was the separation of endothelial cells into tumor-associated and normal cells. Clustering of the projected dataset with the HLCA reference showed that cells expressing the suggested tumor-associated marker ACKR1 were also abundant in healthy tissue from the HLCA core, specifically in venous endothelial cells (both pulmonary and systemic, Fig. 5c and Supplementary Fig. 6a–c). This suggests that ACKR1 is a general marker of venous endothelial cells rather than a tumor-specific endothelial cell marker. Similarly, the reported normal endothelial cell marker EDNRB characterizes aerocyte capillary endothelial cells, both in tumor and in healthy tissue (Fig. 5c and Supplementary Fig. 6d). As endothelial cell numbers in the original study were low, correctly identifying and distinguishing these cell types without a larger healthy reference is challenging. Thus, by serving as a comprehensive healthy control, the HLCA prevents misinterpretation of limitations in sampling and experimental design as meaningful differences between healthy and diseased tissue. In addition, the HLCA can provide context to the results of large-scale genetic studies of disease. Genome-wide association studies (GWASs) link disease with specific genomic variants that may confer an increased risk of disease. Previous studies have linked such variants to cell type-specific mechanistic hypotheses, which are often lacking in the initial association study. Yet, these studies fail to include all known lung cell types in their cell type reference. To demonstrate the value of the HLCA core in contextualizing genetic data, we mapped association results from four GWASs of lung function or disease to the HLCA core cell types, by testing significant enrichment of both weakly and strongly disease-associated variants in regions of genes that characterize each cell type (Fig. 5d, Supplementary Fig. 7 and Methods). We show that genomic variants linked to lung function (forced vital capacity) are associated with smooth muscle (adjusted P value (Padj) = 0.07), alveolar fibroblasts (Padj = 0.07), peribronchial fibroblasts (Padj = 0.07) and myofibroblasts (Padj = 0.07), suggesting that these fibroblast subtypes play a causative role in inherited differences in lung function. We further find a significant association of lung T cells with asthma-associated single-nucleotide polymorphisms (SNPs) (Padj = 0.005). Lung adenocarcinoma-associated variants trend towards AT2 cells (Padj = 0.18) and myofibroblasts are significantly associated with chronic obstructive pulmonary disease (COPD) GWAS SNPs (Padj = 0.04). Thus, by linking genetic predispositions to lung cell types, the HLCA core serves as a valuable resource with which to improve our understanding of lung function and disease. Finally, the HLCA can be used as a reference for cell type deconvolution of bulk RNA expression samples, which have been shown to reflect cell type proportions more accurately than scRNA-seq datasets. Inferring cell type proportions from bulk RNA samples from nasal brushings and bronchial biopsies using the HLCA core (Supplementary Table 10, Supplementary Fig. 8a and Methods) revealed no significant cell type compositional changes associated with corticosteroid inhalation or asthma, respectively (Supplementary Fig. 8b,c and Supplementary Table 11). In contrast, we find that the proportion of capillary endothelial cells in lung resection tissue from the Lung Tissue Database is higher in samples from patients with severe COPD (GOLD stage 3 or 4) than in those from non-COPD controls matched for age and smoking history (Padj = 0.0004). Conversely, alveolar and interstitial macrophages, AT2 cells and dendritic cells decrease in proportion (Fig. 5e, Supplementary Fig. 8d and Supplementary Table 11; Padj = 0.0007, 0.0003, 0.005 and 3.21 × 10, respectively). Finally, smooth muscle shows the largest shift in proportion, increasing significantly in patients with severe COPD (P = 1.85 × 10) in line with previous work. As deconvolution of bulk samples using the HLCA can reveal disease-specific changes in cell type composition, we provide publicly available preprocessed cell type signature matrices based on the HLCA core (https://github.com/LungCellAtlas/HLCA). As knowledge of cell types in the lung expands, and the sizes of newly generated datasets increase, annotations in the HLCA core will need to be further refined. The HLCA and its annotations can be updated by learning from new data projected onto the reference. We simulated such an HLCA update using the previously projected healthy lung dataset, specifically focusing on the cell identities that were distinguished based on their tissue location in matched spatial transcriptomic data (spatially annotated cell types). These cell identities were present at very low frequencies (median: 0.005% of all cells; Supplementary Fig. 9a). Both spatially annotated mesenchymal cell types with more than 40 cells (immune-recruiting fibroblasts and chondrocytes) and two rare cell types (myelinating Schwann cells and perineurial nerve-associated fibroblasts) were recovered in distinct clusters (spatially annotated clusters), and three of these (all except chondrocytes) also contained cells from the HLCA core, thereby enabling a refinement of existing HLCA core annotations using the spatial context from the projected dataset (Fig. 5f and Supplementary Fig. 9b,c). In this manner the HLCA core and its annotations can be refined by mapping new datasets to the atlas and incorporating annotations from these new datasets into the reference. To extend the atlas and include samples from lung disease, we mapped 1,797,714 cells from 380 healthy and diseased individuals from 37 datasets (four unpublished and 33 published) to the HLCA core using scArches, bringing the HLCA to a total of 2.4 million cells from 486 individuals (Fig. 6a and Supplementary Table 1). Label transfer from the HLCA core to the newly mapped datasets enabled detailed cell type annotation across datasets even for rare cells, including 2,048 migratory dendritic cells identified across 28 datasets with label transfer, whereas this cell type was originally labeled in only two of 12 labeled datasets (Extended Data Fig. 8).Fig. 6The extended HLCA enables the identification of disease-associated cell states.a, UMAP of the extended HLCA colored by coarse annotation (HLCA core) or in gray (cells mapped to the core). b, Uncertainty of label transfer from the HLCA core to newly mapped datasets, categorized by several experimental or biological features. Categories with fewer than two instances are not shown. The numbers of datasets per category were as follows: 30 cells, 7 nuclei, 23 healthy, 5 IPF, 3 CF, 3 carcinoma, 4 ILD, 8 surgical resection, 7 donor lung, 12 lung explant, 6 bronchoalveolar lavage fluid, 4 autopsy, 9 10x 5′, 31 10x 3′, 4 Drop-Seq and 3 Seq-Well. c, Bottom, mean label transfer uncertainty per mapped healthy lung sample in the HLCA extension, grouped into age bins and colored by study. The numbers of mapped samples per age bin were as follows: 43 for 0–10 years, 33 for 10–20 years, 31 for 20–30 years, 23 for 30–40 years, 19 for 40–50 years, 12 for 50–60 years, 9 for 60–70 years, 8 for 70–80 years and 2 for 80–90 years. Top, bar plot showing the number of donors per age group in the HLCA core. d, Violin plot of label transfer uncertainty per transferred cell type label for a single mapped IPF dataset, split into cells from healthy donors (blue) and donors with IPF (orange). e, Uncertainty-based disease signature scores among alveolar fibroblasts and alveolar macrophages, split into cells from control donors (n = 10,453 and 1,812, respectively), and low-uncertainty cells (n = 1,419 and 200, respectively) and high-uncertainty cells (n = 1,172 and 162, respectively) from donors with IPF. f, UMAP embedding of alveolar fibroblasts (labeled with manual annotation (core) or label transfer (five IPF datasets)) colored by Leiden cluster. g, Composition of the clusters shown in f by study, with cells from control samples colored in gray. h, Expression of marker genes for IPF-enriched cluster 0 per alveolar fibroblast cluster. Cluster 5 was excluded as 96% of its cells were from a single donor. i, UMAP of all MDMs in the HLCA, colored by Leiden cluster. j, Composition of the MDM clusters from i by disease. k, Expression of cluster marker genes among all MDM clusters excluding donor-specific clusters 5 and 6. For h and k, mean counts were normalized such that the highest group mean was set to 1 for each gene. For b, c and e, the boxes show the median and interquartile range. Data points more than 1.5 times the interquartile range outside the low and high quartile are considered outliers. Whiskers extend to the furthest nonoutlier point. BALF, bronchoalveolar lavage fluid; CF, cystic fibrosis; Drop-Seq, droplet sequencing; ILD, interstitial lung disease; Mph, macrophages; SM, smooth muscle; uncert., uncertainty. a, UMAP of the extended HLCA colored by coarse annotation (HLCA core) or in gray (cells mapped to the core). b, Uncertainty of label transfer from the HLCA core to newly mapped datasets, categorized by several experimental or biological features. Categories with fewer than two instances are not shown. The numbers of datasets per category were as follows: 30 cells, 7 nuclei, 23 healthy, 5 IPF, 3 CF, 3 carcinoma, 4 ILD, 8 surgical resection, 7 donor lung, 12 lung explant, 6 bronchoalveolar lavage fluid, 4 autopsy, 9 10x 5′, 31 10x 3′, 4 Drop-Seq and 3 Seq-Well. c, Bottom, mean label transfer uncertainty per mapped healthy lung sample in the HLCA extension, grouped into age bins and colored by study. The numbers of mapped samples per age bin were as follows: 43 for 0–10 years, 33 for 10–20 years, 31 for 20–30 years, 23 for 30–40 years, 19 for 40–50 years, 12 for 50–60 years, 9 for 60–70 years, 8 for 70–80 years and 2 for 80–90 years. Top, bar plot showing the number of donors per age group in the HLCA core. d, Violin plot of label transfer uncertainty per transferred cell type label for a single mapped IPF dataset, split into cells from healthy donors (blue) and donors with IPF (orange). e, Uncertainty-based disease signature scores among alveolar fibroblasts and alveolar macrophages, split into cells from control donors (n = 10,453 and 1,812, respectively), and low-uncertainty cells (n = 1,419 and 200, respectively) and high-uncertainty cells (n = 1,172 and 162, respectively) from donors with IPF. f, UMAP embedding of alveolar fibroblasts (labeled with manual annotation (core) or label transfer (five IPF datasets)) colored by Leiden cluster. g, Composition of the clusters shown in f by study, with cells from control samples colored in gray. h, Expression of marker genes for IPF-enriched cluster 0 per alveolar fibroblast cluster. Cluster 5 was excluded as 96% of its cells were from a single donor. i, UMAP of all MDMs in the HLCA, colored by Leiden cluster. j, Composition of the MDM clusters from i by disease. k, Expression of cluster marker genes among all MDM clusters excluding donor-specific clusters 5 and 6. For h and k, mean counts were normalized such that the highest group mean was set to 1 for each gene. For b, c and e, the boxes show the median and interquartile range. Data points more than 1.5 times the interquartile range outside the low and high quartile are considered outliers. Whiskers extend to the furthest nonoutlier point. BALF, bronchoalveolar lavage fluid; CF, cystic fibrosis; Drop-Seq, droplet sequencing; ILD, interstitial lung disease; Mph, macrophages; SM, smooth muscle; uncert., uncertainty. Out of 37 new datasets, 27 were observed to map well to the HLCA, as evaluated by the mean label transfer uncertainty score (Fig. 6b, Supplementary Fig. 10a and Methods). The remaining ten datasets were often from coronavirus disease 2019 (COVID-19) studies or, unlike the HLCA core, contained pediatric samples (Fig. 6b,c and Supplementary Fig. 10b). In these datasets, higher uncertainty values may be attributable to true biological differences between the mapped data and the HLCA core adult, healthy lung samples. Overall, the successfully mapped datasets include disease samples, as well as single-nucleus and single-cell data from multiple chemistries (Fig. 6b), demonstrating the potential of the HLCA core as a universal reference. Pulmonary diseases are characterized by the emergence of unique disease-associated transcriptional phenotypes. We observed higher levels of label transfer uncertainty in datasets from diseased lungs (Fig. 6b, condition), possibly flagging cell types changed in response to disease. Specifically, labels of alveolar fibroblasts and alveolar macrophages, which interact to form a dysregulated cellular circuit in idiopathic pulmonary fibrosis (IPFs), are transferred with higher uncertainty in IPF samples than in samples from healthy controls from the same dataset (Fig. 6d and Extended Data Fig. 9a,b). Furthermore, uncertainty scores separate cells—derived from donors with IPF—within these cell types into more and less affected subsets: the genes more highly expressed in the high-uncertainty subset are also lowly expressed in healthy samples (Fig. 6e). Genes downregulated in high-uncertainty IPF macrophages are associated with homeostatic functions of tissue-resident alveolar macrophages and lipid metabolism (PPARG, FABP4 and others), while upregulated genes are associated with extracellular matrix remodeling and scar formation in the context of lung fibrosis (SPP1, PLA2G7 and CCL2; Supplementary Tables 12 and 13 and Extended Data Fig. 9b,c). Thus, the HLCA core can be used to annotate new data, identify previously unreported populations, and—using label transfer uncertainty scores—help to detect disease-affected cell states and corresponding gene expression programs. This vastly speeds up analysis and interpretation of new data, automatically prioritizing the most relevant populations. Automated mapping of new data to the HLCA core can be done by any user via an interactive web portal (https://github.com/LungCellAtlas/HLCA) or using code tutorials as provided online. Similar to healthy cellular states, the HLCA can provide insight into disease-specific states that are consistent across demographics and experimental protocols. To demonstrate this, we determined which cell types are consistently affected by IPF across datasets, extending the previous IPF analysis to five independent datasets. We found that cells labeled as alveolar fibroblasts consistently show high uncertainty levels in IPF samples compared with controls across all mapped IPF datasets that include controls (Extended Data Fig. 10a). Clustering of alveolar fibroblasts from the HLCA core and all IPF datasets shows that cells from patients with IPF predominantly cluster together in a single cluster (Fig. 6f,g and Extended Data Fig. 10b) characterized by high expression of genes previously associated with IPF (CCL2, COL1A1, CTHRC1 and MMP19), as well as further fibrosis-associated markers (SERPINE1, an inhibitor of extracellular matrix breakdown, and HIF1A, a chronic hypoxia response gene; Fig. 6h and Supplementary Table 14). These marker genes are consistently expressed across datasets (Extended Data Fig. 10c), confirming that the identification of this IPF-specific alveolar fibroblast state is reproducible. The HLCA contains data across more than ten lung diseases, providing the unique opportunity to discover cellular states shared across diseases. Discovering such common diseased cellular states could improve our understanding of lung diseases and accelerate the identification of effective treatments. For example, profibrotic SPP1 monocyte-derived macrophages (MDMs) have previously been reported in COVID-19, IPF and cancer. To test whether similar cross-disease MDM states could be discovered in the HLCA, we performed clustering of all MDMs from the HLCA (Fig. 6i). We identified four main MDM subtypes (Methods and Supplementary Table 15), each showing distinct gene expression and disease enrichment patterns, and representing different stages of monocyte-to-MDM differentiation and adaptation to the disease microenvironment. First, an early and inflammatory MDM state was observed that was high in the expression of CCL2, a gene involved in the recruitment of immune cells. This cluster predominantly contained cells from bronchoalveolar lavage fluid samples collected early during the course of COVID-19 pneumonia (cluster 2; IL1RN and S100A12; Fig. 6i–k and Extended Data Fig. 10d–h). We further observed an MDM subset expressing inflammation and phagocytosis-associated genes (cluster 4; CCL18, IL18, C1QA and TREM2) and enriched for samples from patients with COVID-19 pneumonia, as well as samples from patients with lung carcinoma (Fig. 6i–k and Extended Data Fig. 10d–h). A third MDM subset represented a more differentiated MDM phenotype, as indicated by the expression of MARCO and MCEMP1, dominated by cells from nondiseased samples (cluster 3; Fig. 6i–k and Extended Data Fig. 10d,f). The final MDM subset was dominated by IPF samples. Interestingly, this cluster was also enriched for cells from patients who died late in the course of COVID-19 and developed post-COVID-19 lung fibrosis, as well as cells from patients with lung carcinoma (cluster 0; Fig. 6i–k and Extended Data Fig. 10g–i). This multidisease cluster is marked by high expression of SPP1, LPL and CHIT1—markers that have been shown to play a causal role in the development of lung fibrosis (Fig. 6k), one of which (CHIT1) is currently being investigated as a therapeutic target for IPF. The expression of these markers is consistent across diseases and studies (Extended Data Fig. 10f), suggesting that also in cancer and late-stage COVID-19 samples a subset of MDMs adopt a fibrosis-associated phenotype. Together, this analysis shows that the HLCA enables a better understanding of cellular states shared between diseases and thereby has the potential to accelerate the discovery of effective disease treatments. In this study, we built the HLCA: an integrated reference atlas of the human respiratory system. While previous studies have described the cellular heterogeneity within the human lung, study-specific biases due to experimental design and a limited number of sampled individuals constrain their capacity to capture population variation and serve as a universal reference. The HLCA integrates data from 49 datasets to produce such a reference of 2.4 million cells, covering all major lung scRNA-seq studies published to date. The core of this atlas consists of a fully integrated healthy reference of 14 datasets with 61 cell identities, including rare and novel cell types, representing a data-derived consensus annotation of the cellular landscape of the human lung. We leveraged the unprecedented complexity of the HLCA to recover cell type-specific gene modules associated with covariates such as lung anatomical location, age, sex, BMI and smoking status. By projecting data onto the HLCA, we showed that the HLCA enables a fast and detailed annotation of new datasets, as well as the identification of unique, disease-associated cell states and cell states common to multiple diseases. The HLCA is publicly available as a resource for the community, together with an online platform for automated mapping of new data. Taken together, the HLCA is a universal reference for single-cell lung research that promises to accelerate future studies into pulmonary health and disease. The ultimate goal of a human lung cell atlas reference is to provide a comprehensive overview of all cells in the healthy human lung, as well as their variation from individual to individual. Despite its overall diversity, the HLCA is limited by the biological, demographic and experimental diversity in the foundational single-cell studies. For example, 65% of the HLCA core data are from individuals of European harmonized ethnicity, highlighting the urgent need for diversification of the population sampled in lung studies. Moreover, ethnicity metadata were based on self-reports and harmonized across datasets, which is an imperfect approach to representing the diversity of the atlas. SNP-based inference of genetic ancestry constitutes a more objective and therefore preferable approach to the grouping of individuals based on genetic background and would aid in better assessing the genetic diversity captured in the atlas. Overall, more diverse samples will enrich the atlas, diversify captured cell identities and improve the quality of the HLCA as a reference for new datasets. Such a reference will also enable comparison with model systems such as mice, cell lines or organoids, although further method development may be required to map across diverse in vitro and clinical datasets. The constituent datasets of the HLCA vary widely in experimental design, such as the sample handling protocol or single-cell platform used, causing dataset-specific batch effects. The quality of the HLCA hinges on the choice of data integration method, with methods such as Seurat’s RPCA and Harmony failing to correctly group rare cell identities into separate clusters. Nevertheless, also in the HLCA, certain subsets of T cells (regulatory T cells and γδ T cells) could not be identified as separate clusters, showing the limitations of the current HLCA in capturing cellular heterogeneity for a subset of immune cell types. Mapping additional datasets with high-resolution annotations (for example, derived from multimodal data) could provide the power to detect these cell identities in the atlas. Indeed, the HLCA must be viewed as a live resource that requires continuous updates. While we showed that mapping new, spatially annotated data to the HLCA core can refine HLCA annotations, this new knowledge must be consolidated by regular updates of the HLCA with new datasets (including epigenomic, spatial and imaging data) and refinements of HLCA annotations based on additional expert opinions. Thereby, the HLCA can serve as a community- and data-driven platform for open discussion on lung cell identities as the respiratory community progresses in charting the cellular landscape of the lung. In this process, we envision that the HLCA will be completed in two phases: first on the level of cellular variation (when no new consensus cell types can be found) and then in the description of individual variation (when population diversity is fully represented). Taken together, the HLCA provides a central single-cell reference of unprecedented size. It offers a model framework for building integrated, consensus-based, population-scale atlases for other organs within the Human Cell Atlas. The HLCA is publicly available, and combined with an open-access platform to map new datasets to the atlas, this resource paves the way toward a better and more complete understanding of both health and disease in the human lung. Ethics approval information per study was as follows. For the pooled data from refs. , approval was given by the Vanderbilt Institutional Review Board (IRB) (numbers 060165 and 171657) and Western IRB (number 20181836). All samples were collected from declined organ donors who were also consented for research. For ref. , the study was approved by the Comité de Protection des Personnes Sud Est IV (approval number 17/081). Informed written consent was obtained from all participants involved. For Jain_Misharin_2021 (A.V.M., M.J. and N.S.M., newly generated dataset), the protocol was approved by the Northwestern University IRB (STU00214826). Written informed consent was obtained from all study participants. For ref. , patient tissues were obtained under a protocol approved by Stanford University’s Human Subjects Research Compliance Office (IRB 15166). Informed consent was obtained from each patient before surgery. For ref. , healthy control lungs were obtained under a protocol approved by the University of Pittsburgh Committee for Oversight of Research and Clinical Training Involving Decedents (CORID protocol 718) and following rejection as candidate donors for transplant (IRB STUDY 19100326). For ref. , tissue samples were obtained from the Cambridge Biorepository for Translational Medicine (CBTM) with approval from the National Research Ethics Services (NRES) Committee of East of England—Cambridge South (15/EE/0152). Tissue samples were obtained with informed consent from the donor families. For ref. , the protocol was approved by the Northwestern University IRB (STU00212120). Written informed consent was obtained from all individuals in the study. For the pooled data from ref. and associated unpublished data, the protocol was approved by the IRB (Algemeen Beoordelings- en Registratieformulier number NL69765.042.19). Patients gave informed consent. For ref. , the National Jewish Health IRB approved the research under IRB protocols HS-3209 and HS-2240. Informed consent was obtained from authorized family members of all donors. For ref. , approval was given by the NRES Committee of East of England—Cambridge South (Research Ethics Committee (REC) reference: 15/EE/0152). Informed consent for use of the tissue was obtained from the donors’ families. For Barbry_unpubl (P.B., L.-E.Z., M.J.A., A.C., C.B. et al., newly generated dataset), the protocol was approved by the Centre Hospitalier Universitaire de Nice. Nasal and tracheobronchial samples were collected from patients with IPF after obtaining their informed consent. For ref. , approved was given by the IRB of Northwestern University (STU00212120, STU00213177, STU00212511 and STU00212579). For inclusion in this study, patients or their designated medical power of attorney provided informed consent. For Duong_lungMAP_unpubl (T.E.D., K.Z., X.S., J.S.H. and G.P., newly generated dataset), all postmortem human donor lung samples were obtained from the Biorepository for Investigation of Neonatal Diseases of the Lung (BRINDL), supported by the National Heart, Lung, and Blood Institute (NHLBI) LungMAP Human Tissue Core housed at the University of Rochester. Consent can be found on the repository’s website (brindl.urmc.rochester.edu/). For ref. , the study was conducted in accordance with the Declaration of Helsinki and Department of Health and Human Services Belmont Report. The use of biomaterial and data for this study was approved by the local ethics committee of the Medical Faculty Heidelberg (S-270/2001 and S-538/2012). All individuals gave informed consent for inclusion before they participated in the study. For ref. , human lung tissues were procured under each institution’s approved IRB protocol (numbers 00035396 (Cedars-Sinai Medical Center), 03-1396 (University of North Carolina at Chapel Hill), 1172286 (Cystic Fibrosis Foundation and WIRB-Copernicus Group Western IRB) and 16-000742 (University of California, Los Angeles)). Informed consent was obtained from lung donors or their authorized representatives. For ref. , the study was approved and monitored by the National Jewish Health IRB (FWA00000778). Written informed consent was obtained from all participants. For ref. , the study protocol was approved by the Partners Healthcare IRB (protocol 2011P002419). For ref. , lung tissue was obtained under a protocol approved by the University of Pittsburgh IRB (IRB STUDY 19100326) during transplantation surgery. For ref. , the study was conducted according to the principles expressed in the Declaration of Helsinki. Ethical approval was obtained from Ethics Committee Research UZ/KU Leuven (S63881). All participants provided written informed consent for sample collection and subsequent analyses. For ref. , approval was given by the NRES Committee of East of England—Cambridge South (15/EE/0152). The CBTM operates in accordance with UK Human Tissue Authority guidelines. Samples were obtained from deceased transplant organ donors by the CBTM with informed consent from the donor families. For ref. , ethical approval was given through the Living Airway Biobank, administered through the University College London Great Ormond Street Institute of Child Health (REC reference: 19/NW/0171; Integrated Research Application System (IRAS) project ID: 261511; North West Liverpool East REC), REC reference 18/SC/0514 (IRAS project ID: 245471; South Central Hampshire B REC; administered through the University College London Hospitals NHS Foundation Trust), REC reference 18/EE/0150 (IRAS project ID: 236570; East of England—Cambridge Central REC; administered through Great Ormond Street Hospital NHS Foundation Trust) and REC reference 08/H0308/267 (administered through the Cambridge University Hospitals NHS Foundation Trust), as well as by the local R&D departments at all hospitals. All of the study participants or their surrogates provided informed consent. For ref. , all protocols were reviewed and approved by the IRB at the Memorial Sloan Kettering Cancer Center (IRB protocol 14-091). Noninvolved lung, tumor tissues and metastatic lesions were obtained from patients with lung adenocarcinoma undergoing resection surgery at the Memorial Sloan Kettering Cancer Center after obtaining informed consent. For ref. , samples underwent IRB review and approval at the institutions where they were originally collected. Specifically, the Dana-Farber Cancer Institute approved protocol 13-416, the partners Massachusetts General Hospital and Brigham and Women’s Hospital approved protocols 2020P000804, 2020P000849 and 2015P002215, the Beth Israel Deaconess Medical Center approved protocols 2020P000406 and 2020P000418 and New York Presbyterian Hospital/Columbia University Irving Medical Center approved protocols IRB-AAAT0785, IRB-AAAB2667 and IRB-AAAS7370. Secondary analysis of samples at the Broad Institute was covered under Massachusetts Institute of Technology IRB protocols 1603505962 and 1612793224, or the Not Human Subjects Research protocol ORSP-3635. Donor identities were encoded at the hospitals before shipping to or sharing with the Broad Institute for sample processing or data analysis, respectively. For ref. , the study was approved by the local ethics committee of the Ludwig Maximilian University of Munich (EK 333-10 and 382-10). Written informed consent was obtained from all patients. For Schiller_2021 (H.B.S., J.G.-S., C.H.M., B.H.K., M.A. et al., newly generated dataset), the study was approved by the local ethics committee of the Ludwig Maximilian University of Munich (EK 333-10 and 382-10). Written informed consent was obtained from all patients. For Schultze_unpubl (J.L.S., C.S.F., T.S.K. and E.C., newly generated dataset), human lung tissue was available for research purposes following ethical approval from Hannover Medical School (ethical vote of the German Centre for Lung Research (DZL) number 7414, 2017). All patients in this study provided written informed consent for sample collection and data analysis. For ref. , samples were obtained under the Cells and Mediators IRB protocol (2003P002088). All individuals provided written informed consent. For ref. , the studies described were conducted according to the principles of the Declaration of Helsinki. The study was approved by the University of California, San Francisco IRB. Written informed consent was obtained from all individuals. For ref. , peripheral blood was obtained from healthy consenting adult volunteers by venipuncture through a protocol approved by the Columbia University IRB. All relevant ethical regulations for work with human participants were complied with. For ref. , donor lung samples were provided through the federal United Network for Organ Sharing via the National Disease Research Interchange and International Institute for the Advancement of Medicine and entered into the NHLBI LungMAP BRINDL at the University of Rochester Medical Center, overseen by the IRB as RSRB00047606. For for the pooled data from ref. and associated unpublished data, human lung tissue collection was approved by the Duke University IRB (Pro00082379) and University of North Carolina Biomedical IRB (03-1396) under exempt protocols. Consent was obtained to use human tissues for research purposes. For ref. , the study was approved by the local ethics committee at University Hospitals Leuven (B322201422081) and all of the relevant ethical regulations were complied with. Only patients with untreated, primary, nonmetastatic lung tumors who underwent lung lobe resection with curative intent and who provided informed consent were included in this study. For ref. , all of the research involving human participants was approved by the Northwestern University IRB. Samples from patients with COVID-19, viral pneumonia and other pneumonia, as well as controls without pneumonia, were collected from participants enrolled in the Successful Clinical Response in Pneumonia Therapy study STU00204868. All study participants or their surrogates provided informed consent. For ref. , the IRB of the University of Cincinnati College of Medicine approved all human-relevant studies. For ref. , the study was conducted according to the principles expressed in the Declaration of Helsinki. Ethical approval was obtained from the REC of Shenzhen Third People’s Hospital (2020-112). All participants provided written informed consent for sample collection and subsequent analyses. Further study details can be found in Supplementary Table 1. Several previously unpublished datasets were used for the HLCA and generated as follows. Participants recruited by the Pneumology Unit of Nice University Hospital were sampled between 1 and 15 December 2020. The full procedure, including patient inclusion criteria, is detailed at https://www.clinicaltrials.gov/ct2/show/NCT04529993. Nasal and tracheobronchial samples were collected from patients with IPF after obtaining their informed consent, following a protocol approved by the Centre Hospitalier Universitaire de Nice. The data were derived from the clinical trial registered at ClinicalTrials.gov under reference NCT04529993. This study was described as an interventional study instead of an observational study because the participants were volunteers and all assigned to a specific bronchoscopy not related to routine medical care. Participants were prospectively assigned to a procedure (bronchoscopy) according to a specific protocol to assess our ability to sample the airway. No other procedures were included in this study. Metadata of the donors’ sex was based on self-report. The libraries were prepared as described in Deprez et al. and yielded an average of 61,000 ± 11,000 cells per sample, with a viability above 95%. The single-cell suspension was used to generate single-cell libraries following the v3.1 protocol for 3′ chemistry from 10x Genomics (CG000204). Sequencing was performed on a NextSeq 500/550 sequencer (Illumina). Raw sequencing data were processed using the Cell Ranger 6.0.0 pipeline, with the reference genome GRCh38 and annotation using Ensembl98. For each sample, cells with fewer than 200 transcripts or more than 40,000 transcripts were filtered out, as well as genes expressed in fewer than three cells. Normalization and log transformation were done using the standard Scanpy pipeline. Principal component analysis (PCA) was performed on 1,000 highly variable genes (HVGs) to compute 50 principal components, and the Louvain algorithm was used for clustering. These clusters were then annotated by hand for each sample. Raw counts and the thus obtained cell annotations were used as input for the HLCA. Tumor-free, uninvolved lung samples (peritumor tissues) were obtained during tumor resections at the lung specialist clinic Asklepios Fachkliniken München-Gauting and accessed through the bioArchive of the Comprehensive Pneumology Center in Munich. The study was approved by the local ethics committee of the Ludwig Maximilian University of Munich (EK 333-10 and 382-10), and written informed consent was obtained from all patients. All fresh tissues from patients in a given time frame without any specific selection criteria were included, and only patients with obvious chronic lung disease as comorbidity based on their lung function parameters before tumor resection were excluded. Metadata of the donors’ sex were based on self-report. Single-cell suspensions for scRNA-seq were generated as previously described. In brief, lung tissue samples were cut into smaller pieces, washed with phosphate-buffered saline (PBS) and enzymatically digested using an enzyme mix composed of dispase, collagenase, elastase and DNAse for 45 min at 37 °C while shaking. After inactivating the enzymatic activity with 10% fetal calf serum (FCS)/PBS, dissociated cells were passed through a 70 µm cell strainer, pelleted by centrifugation (300g; 5 min) and subjected to red blood cell lysis. After stopping the lysis with 10% FCS/PBS, the cell suspension was passed through a 30 µm strainer and pelleted. Cells were resuspended in 10% FCS/PBS, assessed for viability and counted using a Neubauer hematocytometer. The cell concentration was adjusted to 1,000 cells per µl and ~16,000 cells were loaded on a 10x Genomics Chip G with Chromium Single Cell 3′ v3.1 gel beads and reagents (3′ GEX v3.1; 10x Genomics). Libraries were prepared according to the manufacturer’s protocol (CG000204_RevD; 10× Genomics). After a quality check, scRNA-seq libraries were pooled and sequenced on a NovaSeq 6000 instrument. The generation of count matrices was performed using the Cell Ranger computational pipeline (v3.1.0; STAR v2.5.3a). The reads were aligned to the GRCh38 human reference genome (GRCh38; Ensembl99). Downstream analysis was performed using the Scanpy package (version 1.8.0). We assessed the quality of our libraries and excluded barcodes with fewer than 300 genes detected, while retaining those with a number of transcripts between 500 and 30,000. Furthermore, cells with a high proportion (>15%) of transcript counts derived from mitochondrial-encoded genes were removed. Genes were considered if they were expressed in at least five cells. Raw counts of cells that passed filtering were used as input for the HLCA. All postmortem human donor lung samples were obtained from BRINDL, supported by the NHLBI LungMAP Human Tissue Core housed at the University of Rochester. Consent, tissue acquisition and storage protocols can be found on the repository’s website (brindl.urmc.rochester.edu/). Data were collected as part of the Human Biomolecular Atlas Program (HuBMAP). Metadata of the donor’s sex were based on self-report. For isolation of single nuclei, ten cryosections (40 µm thickness) from O.C.T.-embedded tissue blocks stored at −80 °C were shipped on dry ice and processed according to a published protocol. Single-nucleus RNA-seq was completed using 10x Chromium Single Cell 3’ Reagent Kits v3, according to a published protocol. Raw sequencing data were processed using the 10x Cell Ranger v3 pipeline and the GRCh38 reference genome. For downstream analysis, mitochondrial transcripts and doublets identified by DoubletDetection version 2.4.0 were removed. Samples were then combined and cell barcodes were filtered based on the genes detected (>200 and <7,500) and the gene unique molecular identifier (UMI) ratio (gene.vs.molecule.cell.filter function) using Pagoda2 (github.com/hms-dbmi/pagoda2). Also using Pagoda2 for clustering, counts were normalized to total counts per nucleus. For batch correction, gene expression was scaled to dataset average expression. After variance normalization, all significantly variant genes (n = 4,519) were used for PCA. Clustering was done at different k values (50, 100 or 200) using the top 50 principal components and the infomap community detection algorithm. Then, principal component and cluster annotations were imported into Seurat version 4.0.0. Differentially expressed genes for all clusters were generated for each k resolution using Seurat FindAllMarkers (only.pos = TRUE, max.cells.per.ident = 1000, logfc.threshold = 0.25, min.pct = 0.25). Clusters were manually annotated based on distinct differentially expressed marker genes. Raw counts and the thus obtained cell annotations were used as input for the HLCA. These data were a combination of published and unpublished data. In both cases, healthy volunteers were recruited for bronchoscopy at the University Medical Center in Groningen after giving informed consent and according to the protocol approved by the IRB (ABR number NL69765.042.19). Inclusion criteria and tissue processing were performed as previously described. In short, all donors were 20–65 years old and had a history of smoking <10 pack-years. Metadata of the donors’ sex were based on self-report. To exclude respiratory disease, the following criteria were used: absent history of asthma or COPD; no use of asthma- or COPD-related medication; a negative provocation test (concentration of methacholine that provokes a 20% decrease in the forced expiratory volume in 1 s (FEV1) > 8 mg ml); no airflow obstruction (FEV1/forced vital capacity ≥ 70%); and an absence of lung function impairment (that is, FEV1 ≥ 80% predicted). All donors underwent a bronchoscopy under sedation using a standardized protocol. Nasal brushes were obtained from the lateral inferior turbinate in a subset of the volunteers immediately before bronchoscopy using a Cyto-Pak CytoSoft nasal brush (Medical Packaging Corporation). Six macroscopically adequate endobronchial biopsies were collected for this study, located between the third and sixth generation of the right lower and middle lobe. Bronchial brushes were obtained from a different airway at similar anatomical locations using a Cellebrity bronchial brush (Boston Scientific). Extracted biopsies and bronchial and nasal brushes were processed directly, with a maximum of 1 h delay. Bronchial biopsies were chopped biopsies using a single-edge razor blade. A single-cell solution was obtained by tissue digestion using 1 mg ml collagenase D and 0.1 mg ml DNase I (Roche) in Hanks’ Balanced Salt Solution (Lonza) at 37 °C for 1 h with gentle agitation for both nasal brushes and bronchial biopsies. Single-cell suspensions were filtered and forced using a 70 µm nylon cell strainer (Falcon), followed by centrifugation at 550g and 4 °C for 5 min and one wash with PBS containing 1% bovine serum albumin (BSA; Sigma–Aldrich). The single-cell suspensions used for 10x Genomics scRNA-seq analysis were cleared of red blood cells using a red blood cell lysis buffer (eBioscience) followed by live cell counting and loading of 10,000 cells per lane. We used 10x Genomics Chromium Single Cell 3′ Reagent Kits v2 and v3 according to the manufacturers’ instructions. Raw sequencing data were processed using the Cell Ranger 3.1.0-based HLCA pipeline, with the reference genome GRCh38 and annotation using Ensembl98. Ambient RNA correction was performed with FastCAR, using an empty library cutoff of 100 UMI and a maximum allowed contamination chance of 0.05, ignoring the mitochondrial RNA. Data were merged and processed using Seurat, filtering to libraries with >500 UMIs and >200 genes and to the libraries containing the lowest 95% of mitochondrial RNA per sample and <25% mitochondrial RNA, normalized using sctransform while regressing out variation correlating with the percentage of mitochondrial RNA per cell. In general, 15 principal components were used for the clustering, at a resolution of 0.5 to facilitate manual annotation of the dataset. Clusters in the final object that were driven by single donors were removed. Raw counts and cell annotations were used as input for the HLCA. Nasal epithelial samples were collected from healthy volunteers who provided informed consent at Northwestern Medicine in Chicago. The protocol was approved by the Northwestern University IRB (STU00214826). Healthy volunteers were recruited to match a cohort of patients with cystic fibrosis for the ongoing study at Northwestern University (with M.J. as the principal investigator). In both studies, A.V.M. did not influence participant recruitment and did not introduce biases in sample selection. Metadata of the donors’ sex were based on self-report. Briefly, donors were seated and asked to extend their neck. A nasal curette (Rhino-Pro; VWR) was inserted into either nare and gently slid in the direction of posterior to anterior ~1 cm along the lateral inferior turbinate. Five curettes were obtained per participant. The curette tip was then cut and placed in 2 ml hypothermosol and stored at 4 C until processing. A single-cell suspension was generated using the cold-active dispase protocol reported by Deprez et al. and Zaragosi and Barbry with slight modification. Specifically, ethylenediaminetetraacetic acid (EDTA) was omitted and cells were dispersed by pipetting 20 times every 5 min using a 1 ml tip instead of tritration using a 21/23 G needle. The final concentration of protease from Bacillus licheniformis was 10 mg ml. The total digestion time was 30 min. Following the wash in 4 ml 0.5% BSA in PBS and centrifugation at 400g for 10 min, cells were resuspended in 0.5% BSA in PBS and counted using a Nexcelom K2 Cellometer with acridine orange/propidium iodide reagent. This protocol typically yields ~300–500,000 cells with a viability of >95%. The resulting single-cell suspension was then used to generate single-cell libraries following the protocol for 5′ V1 (CG000086 Rev M; 10x Genomics) or V2 chemistry (CG000331 Rev A; 10x Genomics). Excess cells from two of the samples were pooled together to generate one additional single-cell library. After a quality check, the libraries were pooled and sequenced on a NovaSeq 6000 instrument. Raw sequencing data were processed using the Cell Ranger 3.1.0 pipeline, with the reference genome GRCh38 and annotation using Ensembl98. To assign sample information to cells in the single-cell library prepared from two samples, we ran souporcell version 2.0 for that library and two libraries that were prepared from these samples separately. We used common genetic variants prepared by the souporcell authors to separate cells into two groups by genotype for each library, and Pearson correlation between the identified genotypes across libraries to establish correspondence between genotype and sample. Cell annotations were assigned to cell clusters based on expert interpretation of marker genes for each cluster. Cell clusters were derived with the Seurat version 3.2 workflow in which samples were normalized with sctransform, 3,000 HVGs were selected and integrated and clusters were derived from 30 principal components using the Louvain algorithm with default parameters. Clusters with a low number of UMIs and high expression of ribosomal or mitochondrial genes were excluded as low quality. Raw counts and the thus obtained cell annotations were used as input for the HLCA. Human lung tissue wabus available for research purposes following ethical approval from Hannover Medical School (Nr. 7414, 2017). All patients in this study provided written informed consent for sample collection and data analyses. At Hannover Medical School, patients with lung cancer were recruited in the course of their operation (that is, surgical tumor resection was performed according to the ethical vote of the German Centre for Lung Research, ethical vote 7414 and data safety guidelines). There was no bias in patient recruitment since the samples were collected as fresh native tissue following surgical tumor resection and according to the availability of surplus adjacent nonmalignant lung tissue, which was resected in parallel to the tumor tissue. Metadata of the donors’ sex were based on self-report or reported by medical professionals during consenting. Fresh adjacent normal tumor-free lung tissues from patients with non-small cell lung cancer tumors were obtained by the Lung Research group (D. Jonigk, Pathology, Hannover Medical School) and processed for single-cell isolation immediately. Lung tissue was chopped with a scalpel and scissors and digested using BD Tumor Dissociation Reagent (BD Biosciences) for 30 min in a 37 °C water bath. The digestion was stopped with 1% FCS and 2 mM EDTA in PBS without Ca/Mg and cells were filtered over a 70 µm cell strainer (BD Falcon). Erythrocytes were removed using a human MACSxpress Erythrocyte Depletion Kit (Miltenyi Biotec) and cells were filtered using a 40 µm cell strainer (BD Falcon). The viability of the cells was assessed microscopically and by flow cytometry using a LIVE/DEAD Fixable Yellow Dead Cell Stain Kit (Invitrogen) and was ~84%. The single-cell suspension was processed for scRNA-seq and library preparation with the Seq-Well protocol. Library pools with fewer than 100 cells were discarded and merged into one object. The Seurat v3.2 pipeline was used to further analyze the data. Genes in fewer than five cells in the dataset, as well as the mitochondrial genes MT-RNR1 and MT-RNR2, were removed. Cells with fewer than 200 genes were discarded, whereas cells with <5% mitochondrial genes or <30% intronic reads were kept for further analysis. The data were log normalized and 2,000 variable genes were calculated for each sample for integration with Seurat’s Canonical Correlation Analysis algorithm. The data were scaled, 50 principle components were selected and the data were clustered with 0.6 resolution. Cluster annotation revealed a low-quality cluster that was subsequently removed and the process (the calculation of variable genes, calculation of 30 principal components, clustering with 0.4 resolution) was repeated. Raw counts of the cells that passed all filtering were provided as input for the HLCA. To accommodate data protection legislation, scRNA-seq datasets of healthy lung tissue were shared by dataset generators as raw count matrices, thereby obviating the need to share genetic information. Count matrices were generated using varying software (Supplementary Table 1). Previously published scRNA-seq data were partly realigned by the dataset generators: the raw sequencing data of four previously published studies were realigned to GRCh38 using Ensembl84 for the HLCA. For two of these studies, the Cell Ranger 3.1.0-based HLCA pipeline was used for realignment. For the remaining two, data were processed as previously described, but with the reference genome and genome annotation adapted to the HLCA (GRCh38; Ensembl84). All other datasets in the HLCA core were originally already aligned to GRCh38 (Ensembl84) except data from ref. (GRCh38; Ensembl93) (Supplementary Table 1). For ref. , the count matrices provided had ambient RNA removed, as described previously. For all of the datasets from the HLCA core, a preformatted sample metadata form was filled out by the dataset providers for every sample, containing metadata such as the ID of the donor from whom the sample came, the donor’s self-reported ethnicity, the type of sample, the sequencing platform and so on (Supplementary Table 2). Ethnicity metadata were based on self-reported ethnicity for live donors or retrieved from medical records or assigned by the organ procurement team in the case of organ donors, as collected in the individual studies. For donor ethnicity, the following categories of self-reported ethnicity were used during metadata collection: Black, white, Latino, Asian, Pacific Islander and mixed. To conform to pre-existing 1,000 Genomes ancestry superpopulations, these self-reported ethnicity categories were then harmonized with the superpopulation categories as follows: Black was categorized as African, white as European and Latino as admixed American, while keeping the category Asian (merging the superpopulations East Asians and South Asians as this granularity was missing from the collected self-reported ethnicity data) and keeping Pacific Islander, as this category did not correspond to any of the superpopulations but does constitute a separate population in HANCESTRO. We refer to the resulting categories as harmonized ethnicity. Both self-reported ethnicity (as collected) and harmonized ethnicity per donor are detailed in Supplementary Table 2. Cell type annotations from dataset providers were included in all datasets. For tissue dissociation protocols, protocols were grouped based on: (1) enzyme(s) used for tissue dissociation; and (2) the digestion time in cases where large differences were observed between protocols (that is, cold protease protocols were split into two groups: 30–60 min versus overnight). Patients with lung conditions affecting larger parts of the lung, such as asthma or pulmonary fibrosis, were excluded from the HLCA core and later added to the extended atlas. For the HLCA core, all matrices were gene filtered based on Cell Ranger Ensembl84 gene-type filtering (resulting in 33,694 gene IDs). Cells with fewer than 200 genes detected were removed (removing 2,335 cells and 21 extra erythrocytes with close to 200 genes expressed as these hampered SCRAN normalization; see below), along with genes expressed in fewer than ten cells (removing 5,167 out of 33,694 genes). To normalize for differences in total UMI counts per cell, we performed SCRAN normalization. Since SCRAN assumes that at least half of the genes in the data being normalized are not differentially expressed between subgroups of cells, we performed SCRAN normalization within clusters. To this end, we first performed total count normalization, by dividing each count by its cell’s total count and multiplying by 10,000. We then performed a log transformation using natural log and pseudocount 1. A PCA was subsequently performed. Using the first 50 principal components, a neighborhood graph was calculated with the number of neighbors set to k = 15. Data were subsequently clustered with Louvain clustering at a resolution of r = 0.5. SCRAN normalization was then performed on the raw counts, using the Louvain clusters as input clusters and with the minimum mean (library size adjusted) average count of genes to be used for normalization set to 0.1. The resulting size factors were used for normalization. For the final HLCA (and not the benchmarking subset), cells with abnormally low size factors (<0.01) or abnormally high total counts after normalization (>10 × 10) were removed from the data (267 cells in total). To harmonize cell type labels from different datasets in the HLCA core, a common reference was created to which original cell type labels were mapped (Supplementary Table 4). To accommodate labels at different levels of detail, the cell type reference was made hierarchical, with level 1 containing the coarsest possible labels (immune, epithelial and so on) and level 5 containing the finest possible labels (for example, naive CD4 T cells). Levels were created in a data-driven fashion, recursively breaking up coarser-level labels into finer ones where finer labels were available. To map anatomical location to a 1D CCF score that could be used for modeling, a distinction was made between upper and lower airways. First, an anatomical coordinate score was applied to the upper airways, starting at 0 and increasing linearly (with a value of 0.5) between each of the following anatomical locations: inferior turbinate, nasopharynx, oropharnyx, vesibula and larynx. The trachea received the next anatomical coordinate score using the same linear increment as in the upper airways (a score of 2.5). In the lower airways, the coordinate score within the bronchial tree was based on the generation airway, with the trachea being the first generation and the total number of generations assumed to be 23 (ref. ). The alveolar sac was assigned the coordinate score of the 23rd generation airway. The coordinate score of each generation airway was calculated by taking the log2 value of the generation and adding it to the score of the trachea. Using this methodology, the alveolus received an anatomical coordinate score of 7.02. To calculate the final CCF coordinate, the coordinate scores (ranging from 0 to 7.02) were scaled to a value between 0 (inferior turbinate) and 1 (alveolus). Samples were then mapped to this coordinate system using the best approximation of the sampling location for each of the samples of the core HLCA (Supplementary Table 3). For computational efficiency, benchmarking was performed on a subset of the total atlas, including data from ten studies split into 13 datasets (ref. was split into 10xv1 and 10xv2 data; ref. was split into 10xv2 and 10xv3 data; and the pooled data from ref. and associated unpublished data were split into two based on the processing site). The data came from 72 donors, 124 samples and 372,111 cells. Preprocessing of the benchmarking data included the filtering of cells (minimum number of total UMI counts: 500) and genes (minimum number of cells expressing the gene: 5). For integration benchmarking, the scIB benchmarking framework was used with default integration parameter settings unless otherwise specified. All benchmarked methods except scGen (that is, BBKNN, ComBat, Conos, fas tMNN, Harmony, Scanorama, scANVI, scVI and Seurat RPCA) were run at least twice: on the 2,000 most HVGs; and on the 6,000 most HVGs. Of these methods, all that did not require raw counts as input were run twice on each gene set: once with gene counts scaled to have a mean of 0 and standard deviation of 1; and once with unscaled gene counts. scVI and scANVI, which require raw counts as input, were not run on scaled gene counts. scGen was only tested on 2,000 unscaled HVGs. For HVG selection, first, HVGs were calculated per dataset using Cell Ranger-based HVG selection (default parameter settings: min_disp=0.5, min_mean=0.0125, max_mean=3, span=0.3, n_bins=20). Then, genes that were highly variable in all datasets were considered overall highly variable, followed by genes highly variable in all datasets but one, in all datasets but two and so on until a predetermined number of genes were selected (2,000 or 6,000 genes). For scANVI and scVI, genes were subset to the HVG set and the resulting raw count matrix was used as input. For all other methods, SCRAN-normalized (as described above) data were used. Genes were then subset to the precalculated HVG sets. For integration of gene-scaled data, all genes were scaled to have mean of 0 and standard deviation of 1. Two integration methods allowed for input of cell type labels to guide the integration: scGen and scANVI. As labels, level 3 harmonized cell type labels were used (Supplementary Table 4), except for blood vessel endothelial, fibroblast lineage, mesothelial and smooth muscle cells, for which we used level 2 labels. Since scGen does not accept unlabeled cells, cells that did not have annotations available at these levels (that is, cells annotated as cycling, epithelial, stromal or lymphoid cells with no further annotations; 4,499 cells in total) were left out of the benchmarking data. The dataset rather than the donor of the sample was used as the batch parameter. The maximum memory usage was set to 376 Gb and all methods requiring more memory were excluded from the analysis. The quality of each of the integrations was scored using 12 metrics, with four metrics measuring the batch correction quality and eight metrics quantifying the conservation of biological signal after integration (Supplementary Fig. 1; metrics previously described). Overall scores were computed by taking a 0.4:0.6 weighted mean of batch effect removal to biological variation conservation (bioconservation), respectively. Methods were ranked based on overall score (Supplementary Fig. 1). For integration of the data into the HLCA core, we first determined for which cases studies had to be split into separate datasets (which were treated as batches during integration). Reasons for possible splitting were: (1) different 10x versions used within a study (for example, 10xv2 versus 10xv3); or (2) the processing of samples at different institutes within a study. To determine whether these covariates caused batch effects within a study, we performed principal component regression. To this end, we preprocessed single studies separately (total count normalization to median total counts across cells and subsequent PCA with 50 principal components). For each study, we then calculated the fraction of the variance in the first 50 principal components that could be explained (PCexpl) by the covariate of interest (that is, 10x version or processing institute):[12pt] $$}_}} = _^ }( }} )}}_^ }( }_i} )}}$$=∑i=150varcov∑i=150varPCiwhere var(PCi|cov) is the variance in scores for the ith principal component across cells that can be explained by the covariate under consideration, based on a linear regression. Then, 10x version or processing institute assignments were randomly shuffled between samples and PCexpl was calculated for the randomized covariate. This was repeated over ten random shufflings and the mean and standard deviation of PCexpl were then calculated for the covariate. If the nonrandomized PCexpl was more than 1.5 standard deviations above the randomized PCexpl, we considered the covariate a source of batch effect and split the study into separate datasets. Thus, both Jain_Misharin_2021 and ref. were split into 10xv1 and 10xv2; ref. was split into 10xv2 and 10xv3; and ref. and its pooled unpublished data were not split based on 10x version nor on processing location. For integration of the datasets into the HLCA core, coarse cell type labels were used as described for integration benchmarking (AT1, AT2, arterial endothelial cell, B cell lineage, basal, bronchial vessel 1, bronchial vessel 2, capillary, multiciliated, dendritic, fibroblast lineage, KRT5KRT17 epithelial, lymphatic endothelial cell, macrophages, mast cells, megakaryocytes, mesothelium, monocytes, neutrophils, natural killer/natural killer T cells, proliferating cells, rare, secretory, smooth muscle, squamous, submucosal secretory, T cell lineage, venous and unlabeled), except cells with lacking annotations were set to unlabeled instead of being removed. scANVI was run on the raw counts of the 2,000 most HVGs (calculated as described above), using datasets as the batch variable to enable the conservation of interindividual variation. The following parameter settings were used: number of layers: 2; number of latent dimensions: 30; encode covariates: True; deeply inject covariates: False; use layer norm: both; use batch norm: none; gene likelihood: nb; n epochs unsupervised: 500; n epochs semi-supervised: 200; and frequency: 1. For the unsupervised training, the following early-stopping parameters were used: early stopping metric: elbo; save best state metric: elbo; patience: 10; threshold: 0; reduce lr on plateau: True; lr patience: 8; and lr_factor: 0.1. For the semisupervised training, the following early-stopping parameter settings were used: early stopping metric: accuracy; save best state metric: accuracy; on: full dataset; patience: 10; threshold: 0.001; reduce lr on plateau: True; lr_patience: 8; and lr_factor: 0.1. The integrated latent embedding generated by scANVI was used for downstream analysis (clustering and visualization). For gene-level analyses (differential expression and covariate effect modeling), uncorrected counts were used. To cluster the cells in the HLCA core, a nearest neighbor graph was calculated based on the 30 latent dimensions that were obtained from the scANVI output, with the number of neighbors set to k = 30. This choice of k, while improving clustering robustness, could impair the detection of very rare cell types. Coarse Leiden clustering was done on the graph with a resolution of r = 0.01. For each of the resulting level 1 clusters, a new neighbor graph was calculated using scANVIs 30 latent dimensions, with the number of neighbors again set to k = 30. Based on the new neighbor graph, each cluster was clustered into smaller level 2 clusters with Leiden clustering at a resolution of r = 0.2. The same was done for levels 3 and 4 and (where needed) 5, with k set to 15, 10 and 10, respectively, and the resolution set to 0.2. Clusters were named based on their parent clusters and sister clusters (for example, cluster 1.2 is the third biggest subcluster (starting at 0) of cluster 1). For visualization, a 2D UMAP of the atlas was generated based on the 30 nearest neighbors graph. To quantify cluster cell type label disagreement for a specific level of annotation, the label Shannon entropy was calculated on the cell type label distribution per cluster as[12pt] $$- _^k p( )}[ )} ],$$−∑i=1kpxilogpxi,where x1…xk are the set of labels at that annotation level and p(xi) is the fraction of cells in the cluster that was labeled as xi. Cells without a label at the level under consideration were not included in the entropy calculation. If <20% of cells were labeled at the level under consideration, the entropy was set to not available for the figures. The entropy of donors per cluster (that is, diversity of donors in a cluster) was calculated in the same way. To set a threshold for high label entropy, we calculated the label entropy of a hypothetical cluster with 75% of cells given one label and 25% of cells given another label, as a cluster with <75% of cells with the same label suggests substantial disagreement in terms of cell type labeling. Clusters with a label entropy above that level (0.56) were considered to have high label entropy. Six small clusters with high label entropy even at the coarsest level of annotation highlighted doublet populations (identified via simultaneous expression of lineage-specific marker genes; for example, expression of both epithelial (AT2) and stromal (smooth muscle) marker genes) not labeled as such in the original datasets. These clusters were removed from the HLCA core, bringing the total number of clusters to 94. To set a threshold for low donor entropy, we calculated the label entropy for a hypothetical cluster with 95% of cells from one donor and the remaining 5% of cells distributed over all other donors, as clusters with >95% of the cells from the same cluster could be considered single-donor clusters, possibly caused by remaining batch effects or by donor-specific biology that is difficult to interpret. Clusters with a donor entropy below that level (0.43) were considered clusters with low donor entropy. To determine how well rare cell types (ionocytes, neuroendocrine cells and tuft cells) were clustered together and separate from other cell types after integration, we calculated recall (the percentage of all cells annotated as a specific rare cell type that were grouped into the cluster) and precision (the percentage of cells from the cluster that were annotated as a specific rare cell type) for all level 3 clusters. Nested clustering on Harmony and Seurat’s RPCA output was done based on PCA of the corrected gene counts, recalculating the principal components for every parent cluster when performing clustering into smaller children clusters, with clustering performed as described above under ‘UMAP embedding and clustering’. The level 3 clusters with the highest sensitivity for each cell type are included in Supplementary Fig. 3b. Re-annotation of cells in the HLCA core was done by six investigators with expertise in lung biology (E.M., M.C.N., A.V.M., L.-E.Z., N.E.B. and J.A.K.) based on original annotations and differentially expressed genes of the HLCA core clusters. Annotation was done per cluster, using finer clusters where these represented specific known cell types or states rather than donor-specific variation. Annotations of cell identities were hierarchical (as was the harmonized cell type reference) and each cluster was annotated at the finest known level, whereafter coarser levels could automatically be inferred (for example, a cell annotated as a CD8 T cell was then automatically annotated as of T cell lineage at level 3, lymphoid cell lineage at level 2 and immune cell lineage at level 1). The number of cells per cell type is shown for all levels in Supplementary Table 5. Mislabeling of original cells was identified by comparing final annotations with original harmonized labels and checking whether these were contradictory (and not only done at different levels of detail). Out of 61 final cell types, 18 included mostly mislabeled cells, 12 of which were previously known cell types. Despite consisting of mostly mislabeled cells from the original datasets, individual experts agreed on the annotation of these cell types: for all previously known cell types with a high proportion of mislabeled cells, the majority of experts agreed on the final annotation for the atlas, or only differed in the granularity of annotation. Marker genes were calculated based on per-sample, per-cell-type pseudo-bulks, calculating the mean (normalized and log-transformed) expression per pseudo-bulk for every gene. Pseudo-bulks were only calculated for a sample if it had at least ten cells of the cell type under consideration. An exception was made for cell types with fewer than 100 cells in total, for which the minimum number of cells per sample was set to 3. Pseudo-bulks rather than cell-level counts were used to ensure equal weighing of every sample in the differential expression test, thus statistically testing cell type-specific changes in expression that were significant across samples rather than cells. As pseudo-bulks represent the mean of a repeated draw from a single distribution, based on the central limit theorem, we expect pseudo-bulk gene counts to be normally distributed, and a t-test was therefore used to test differential gene expression, comparing a single cell type with all other cell types in the atlas (marker iteration 1). To further filter out differentially expressed genes that were not consistently expressed across samples, we applied a filtering step to remove genes expressed in <80% of the pseudo-bulks, or genes expressed in <50% of cells per pseudo-bulk (with the filtering based on the mean across pseudo-bulks). Similarly, to ensure specificity of gene expression, additional filtering was done to remove genes expressed in >20% of other pseudo-bulks. For many cell types, marker genes unique to a single cell type across the entire atlas could not be found. To nonetheless collect a robust and unique set of marker genes for every cell type, a hierarchical approach was taken, subsetting the atlas to four compartments (epithelial, endothelial, immune and stromal, for each of which a marker set was calculated) before calculating cell type-specific marker genes and filtering on uniqueness only within the compartment (marker iteration 2). When necessary, a second subsetting step was done, now subsetting to the next coarsest cell type set within the compartment (for example, lymphatic endothelial cells) and repeating the same procedure (marker iteration 3). Finally, filtering criteria were loosened for the remaining cell types for which no unique markers could be found in any of the iterations (marker iterations 4 and 5). Exact filtering parameters per iteration can be found in Supplementary Table 16. For lymphatic endothelial cell subtypes, one subtype contained sufficient cells for only a single sample, hampering a pseudo-bulk-based approach. Therefore, lymphatic endothelial cell subset markers (mature, differentiating and proliferating) were chosen based on known literature, after checking consistency with expression patterns observed in the HLCA lymphatic endothelial cells. To quantify the extent to which different technical and biological covariates correlated with interindividual variation in the atlas, we calculated the variance explained by each covariate for each cell type. We first split the data in the HLCA core by cell type annotation, merging substates of a single cell type into one (Supplementary Table 5; includes the number of cells per cell type). For every cell type, we excluded samples that had fewer than ten cells of the sample. We then summarized covariate values per sample for every cell type depending on the variable, taking the mean across cells from a sample for scANVI latent components (integration results), UMI counts per cell and fractions of mitochondrial UMIs, while for all other covariates (for example, BMI and tissue sampling method) each sample had only one value; therefore, these values were used. Next, we performed principal component regression on every covariate, as described previously (see the section ‘Splitting of studies into datasets’), but now using scANVI latent component scores instead of principal component scores for the regression, yielding a fraction of latent component variance explained per covariate. Samples that did not have a value for a given covariate (for example, where the BMI was not recorded for the donor) were excluded from the regression. Categorical covariates were converted to dummy variables. Cell type–covariate pairs for which only one value was observed for the covariate were excluded from the analysis. Quantification of the correlation or dependence between variables within a cell type and identification of the minimum number of samples needed to control for spurious correlation are described below. To check the extent to which covariates correlated with each other, thereby possibly acting as confounders in the principal component regression scores, we determined dependence between all covariate pairs for every cell type. If at least one covariate was continuous, we calculated the fraction of variance in the continuous covariate that could be explained by the other covariate (dummying categorical covariates) and took the square root (equal to Pearson’s r for two continuous covariates). For two categorical covariates, if both covariates had more than two unique values, we calculated normalized mutual information between the covariates using scikit-learn, since a linear regression between these two covariates is not possible. To control for spurious correlations between interindividual cell type variation and covariates due to low sample numbers, we assessed the relationship between sample number and mean variance explained across all covariates for every cell type. We found that for cell types sampled in fewer than 40 samples the mean variance explained across all covariates showed a high negative correlation with the number of samples (Supplementary Fig. 4a). We reasoned that for these cell types correlations between interindividual variation and our covariates were inflated due to undersampling. Moreover, we note that at lower sample numbers technical and biological covariates often strongly correlate with each other across donors (Supplementary Fig. 4c). This might lead to the attribution of true biological variation to technical covariates, and vice versa, complicating the interpretation of observed interindividual cell type variation. Consequently, we consider 40 a recommended minimum number of samples to avoid spurious correlations between observed interindividual variation and tested covariates, and excluded results from cell types with fewer samples. To select cell types for which covariate effects could be confidently modeled at the gene level, we followed the same procedure for every cell type: we filtered out all genes that were expressed in fewer than 50 cells and all samples that had fewer than ten cells of the cell type. We furthermore filtered out datasets with fewer than two donors and refrained from modeling categories in covariates that had fewer than three donors in their category for that cell type. We encoded smoking status as a continuous covariate, setting never to 0, former to 0.5 and current to 1. Anatomical region was encoded into anatomical region CCF scores as described earlier. As we noted that changes from the nose to the rest of the airways and lungs were often independent from continuous changes observed in the lungs only, we encoded nasal as a separate covariate, setting samples from the nose to 1 and all others to 0. BMI and age were rescaled, such that the 10th percentile of observed values across the atlas was set to 0 and the 90th percentile was set to 1 (25 and 64 years for age, respectively, and 21.32 and 36,86 for BMI). To determine whether covariance between covariates was low enough for modeling, we calculated the variance inflation factor (VIF) between covariates at the donor level. The VIF quantifies multicollinearity among covariates of an ordinary least squares regression and a high VIF indicates strong linear dependence between variables. If the VIF was higher than 5 for any covariate for a specific cell type, we concluded that covariance was too high and excluded that cell type from the modeling. As many cell types lacked sufficient representation of harmonized ethnicities other than European, including harmonized ethnicity in the analysis simultaneously decreased the samples that could be included in the analysis to only those with ethnicity annotations; hence, we excluded harmonized ethnicity from the modeling. To model the effects of demographic and anatomical covariates (sex, age, BMI, harmonized ethnicity, smoking status and anatomical location of the sample) on gene expression, we employed a generalized linear mixed model. We used sample-level pseudo-bulks (split by cell type), since the covariates modeled also varied at the sample or donor level and not at the cell level. Modeling these covariates at the cell level (that is, treating single cells as independent samples even when they come from the same sample) has been shown to inflate P values. First, we split the lung cell atlas by cell type annotation, pooling detailed annotations into one subtype (for example, grouping all lymphatic endothelial cell subtypes into one) (Supplementary Table 5; includes the number of cells per cell type). Subsequent filtering, covariate encoding and exclusion of cell types due to covariate dependence are described above. Gene counts were summed across cells for every sample, within cell type. Sample-wise sums (that is, pseudo-bulks) were normalized using edgeR’s calcNormFactors function, using default parameter settings. We then used voom, a method designed for bulk RNA-seq that estimates observation-specific gene variances and incorporates these into the modeling. Specifically, we used a voom extension (differential expression testing with linear mixed models) that allows for mixed-effects modeling and modeled gene expression as:[12pt] $$}[ }} ] 1 + } + } + } + } + } + }\, } \\+ ( }} )$$~1+age+sex+BMI+smoking+nose+CCFscore+1∣datasetwhere dataset is treated as a random effect to correct for dataset-specific changes in expression and all other effects are modeled as fixed effects. Resulting P values were corrected for multiple testing within every covariate using the Benjamini–Hochberg procedure. To identify more systematic patterns across genes and changes happening at the gene set level, a gene set enrichment analysis was performed using correlation-adjusted mean-rank gene set tests. The analysis was performed in R using the cameraPR function in the limma package, with the differential expression test statistic. Gene Ontology biological process terms were tested separately for each comparison. These sets were obtained from MSigDB (version 7.1), as provided by the Walter and Eliza Hall Institute (https://bioinf.wehi.edu.au/MSigDB/index.html). To stratify GWAS results from several lung diseases by lung cell type, we applied stratified linkage disequilibrium score regression in single cells (sc-LDSC). sc-LDSC can link GWAS results to cell types by calculating, for each cell type, whether disease-associated variants are enriched in genomic regions of cell-type specific genes (i.e. the region of each gene and its surrounding base pairs), while taking into account the genetic signal of proximal linkage disequilibrium-associated regions. Here cell-type specific genes are defined as genes differentially expressed in the cell type of interest. In contrast with simple enrichment testing of only significantly disease-associated genes from a GWAS among genes differentially expressed in a cell type, this method takes into account all SNPs included in the GWAS. Thus, consistent enrichment of weakly disease-associated genes (that would not individually pass significance tests) in a cell type could still lead to a significant association between the disease and the cell type. In this way, sc-LDSC provides more statistical power to detect associations between cell types and heritable phenotypes such as lung diseases. To perform sc-LDSC on the HLCA, first a differential gene expression test was performed for every grouped cell type (Supplementary Table 5) in the HLCA using a Wilcoxon rank-sum test, testing against the rest of the atlas. The top 1,000 most significant genes with positive fold changes were stored as genes characterizing that cell type (cell type genes) and used as input for LDSC. Gene coordinates of cell type genes were obtained based on the GRCh37.13 genome annotation. For SNP data (names, locations and linkage-related information), the 1000 Genomes European reference (GRCh37) was used, as previously described. Only SNPs from the HapMap 3 project were included in the analysis. For identification of SNPs in the vicinity of cell type genes, we used a window size of 100,000 base pairs. Genes from X and Y chromosomes, as well as human leukocyte antigen genes, were excluded because of their unusual genetic architecture and linkage patterns. For linkage disequilibrium score calculation, a 1 cM window was used. Significance of the link between a phenotype and a cell type was calculated using LDSC. P values yielded by LDSC were corrected for multiple testing for every disease tested using the Benjamini–Hochberg correction procedure. As a negative control, the analysis was performed with a GWAS of depression and no cell types were found to be significant (Supplementary Fig. 7). The numbers of cases and controls per GWAS study were as follows: n = 2,668 cases and 8,591 controls for IPF; n = 35,735 cases and 222,076 controls for COPD; n = 11,273 cases and 55,483 controls for lung adenocarcinoma; n = 321,047 individuals for lung function; n = 88,486 cases and 447,859 controls for asthma; and n = 113,769 cases and 208,811 controls for depression (used as negative control). To enable deconvolution of bulk expression data on the basis of the HLCA, HLCA cell type signature matrices were generated. One generic-purpose signature matrix was created per sublocation of the respiratory system (that is, one parenchyma, one airway and one nose tissue matrix; Supplementary Table 10). Additionally, a script to generate custom reference sets from the HLCA data is provided together with the HLCA code on GitHub (https://github.com/LungCellAtlas/HLCA) to tailor the deconvolution signature matrix to any specific research question. Cell types were included in the bulk deconvolution signature matrix on the basis of cell proportions (constituting >2% of cells within samples of the corresponding tissue in the HLCA core). In addition, cell types were merged when they were deemed too transcriptionally similar. For each included cell type, 200 cells were randomly sampled from the HLCA core, while all cells were included for cell types with fewer than 200 cells present in the HLCA core. Cells were sampled from the matching anatomical location (for example, nose T cells rather than parenchymal T cells were used for the nose signature matrix). Signature matrices were constructed using CIBERSORTx (version 1.0) according to default settings, and no cross-platform batch correction was applied. The reference data were optimized by deconvolution of pseudo-bulk samples constructed from the HLCA core data, assessing deconvolution performance per included cell type based on the correlation of predicted proportions with ground truth composition (Supplementary Fig. 8a). The following cell types were included in the deconvolution: endothelial cell arterial, endothelial cell capillary, lymphatic endothelial cell, basal and secretory (merged), multiciliated lineage, AT2, B cell lineage, innate lymphoid cell (ILC) natural killer and T cell lineage (merged), dendritic cells, alveolar macrophages, interstitial macrophages, mast cells, fibroblast lineage, smooth muscle, endothelial cell venous and monocytes (for the parenchyma); basal resting and suprabasal (merged), multiciliated lineage, club, goblet, dendritic cells, hillock like and T cell lineage (for the nose); and endothelial cell venous, CD4 T cells, fibroblasts, smooth muscle, basal and secretory (merged), multiciliated lineage, endothelial cell capillary, interstitial macrophages, B cell lineage, natural killer cells, CD8 T cells, dendritic cells, alveolar macrophages, mast cells and monocytes (for the airway). Capillary endothelial cells and interstitial macrophages (airway) were excluded from statistical testing due to poor performance in the benchmark. Venous endothelial cells and monocytes (parenchyma), hillock-like cells and T cell lineage cells (nose) and B cell lineage cells, natural killer cells, CD8 T cells, dendritic cells, alveolar macrophages, mast cells and monocytes (airways) were excluded from statistical testing due to >60% zero proportions. The parenchymal signature matrix was used to deconvolve RNA expression data of samples from the Lung Tissue Database (GEO accession number GSE23546) using only lung tissue samples from patients with COPD GOLD stages 3 and 4 (n = 27 and 56, respectively) and matched controls (n = 281). The Lung Tissue Database dataset was run on the Rosetta/Merck Human RSTA Custom Affymetrix 2.0 microarray platform (HuRSTA-2a520709; GPL10379). As this platform has multiple probe sets for each gene, we focused on the probe sets that were derived from curated RefSeq records (with NM_ accession prefixes) when present to maximize the accuracy of the deconvolution. Where genes did not have probe sets based on curated RefSeq records or had multiple probe sets mapping to curated RefSeq records, the probe set with the highest average microarray intensity across samples was selected. Quantile normalization of the data and subsequent deconvolution were performed using CIBERSORTx. A Wilcoxon rank-sum test between control and GOLD stage 3/4 samples was performed to identify statistically significant compositional changes in advanced-stage COPD compared with control tissue. GOLD 3/4 and control samples were matched for age and smoking history. Cell types with >60% of samples predicted to have a proportion of zero were excluded from the Wilcoxon test, as the high number of tied ranks (zeros in both groups) would result in inflated P values. P values were multiple testing corrected using the Benjamini–Hochberg procedure. The same procedure was followed for a dataset of nasal brush bulk RNA-seq samples from asthmatic donors pre- and postinhalation of corticosteroids (n = 54 and 26, respectively) and a dataset of airway biopsy bulk RNA-seq samples from asthmatic donors and controls (n = 95 and 38, respectively). As these consisted of RNA-seq data, no quantile normalization was applied. To map unseen scRNA-seq and single-nucleus RNA-seq data to the HLCA, we used scArches, our transfer learning-based method that enables mapping of new data to an existing reference atlas. scArches trains an adaptor added to a reference embedding model, thereby enabling it to generate a common embedding of the new data and the reference, allowing reanalysis and de novo clustering of the joint data. The data to map were subsetted to the same 2,000 HVGs that were used for HLCA integration and embedding, and HVGs that were absent in the new data were set to 0 counts for all cells. Raw counts were used as input for scArches, except for the ref. dataset, for which ambient RNA removal was run previously on the raw counts. Healthy lung data were split into two datasets: 3′ and 5′ based. Lung cancer data were also split into two datasets: 10xv1 and 10xv2. The model that was learned previously for HLCA integration using scANVI was used as the basis for the scArches mapping. scArches was then run to train adaptor weights that allowed for mapping of new data into the existing HLCA embedding, using the following parameter settings: freeze-dropout: true; surgery_epochs: 500; train base model: false; metrics to monitor: accuracy and elbo; weight-decay: 0; and frequency: 1. The following early-stopping criteria were used: early stopping metric: elbo; save best state metric: elbo; on: full dataset; patience: 10; threshold: 0.001; reduce lr on plateau: True; lr patience: 8l and lr_factor: 0.1. To enable cross-dataset gene-level analysis, harmonization of gene names from different datasets (using different reference genome builds and genome annotations; Supplementary Table 1) was necessary. Both annotation sources (for example, Ensembl or RefSeq) and annotation versions (for example, Ensembl release 84 or Ensembl release 91) contribute to the variation between different gene naming schemes. Therefore, both annotation sources and versions, including outdated ones, need to be taken into account to enable the mapping of all gene names to a single naming scheme. For the harmonization of gene names, we aimed to map all original gene names to the target scheme HUGO Gene Nomenclature Committee gene name, corresponding to the Ensembl release 107 publication. To find the most likely match between each original gene name and a target gene name, we first downloaded Ensembl releases 79 to 107, which included for each release: (1) all Ensembl gene IDs from the downloaded release (for example, ENSG00000081237.21); (2) corresponding Ensembl transcript and protein IDs (for example, ENST00000442510.8 and ENSP00000411355.3); (3) matching Ensembl IDs from the previous release; (4) matching gene IDs from other genome annotation sources (for example, RefSeq); and (5) matching gene, transcript and protein identifiers from various external resources, such as UniProt, the HUGO Gene Nomenclature Committee and the Consensus Coding Sequence Project. We then constructed a graph, with each Ensembl ID, other genome annotation ID and external resource identifier represented by a single node per release. Nodes were then connected based on the matching (points 2–5) provided by Ensembl, weighing edges based on Ensembl similarity scores where available. For each original gene name from the HLCA datasets, the path with the lowest mean edge weight from that gene name to a gene name from the target names (Ensembl release 107) was selected to find the most likely matching gene name from the target (Supplementary Table 17). Genes for which no target could be found were excluded from downstream analysis. When multiple genes were matched with the same target gene name, counts from the original genes were summed. To identify the genes most commonly exhibiting batch-specific expression, the HLCA was split by cell type and a differential expression analysis was performed (based on a Wilcoxon rank-sum test) in each cell type, comparing cells from one dataset (batch) with those from all other datasets and repeating this for all datasets. Datasets that had fewer than ten cells of the cell type or fewer than three samples with cells of the cell type were excluded from the test. For each test, genes were filtered such that only genes that were significantly upregulated were retained. Next, the fraction of included datasets in which a gene was significantly upregulated in the cell type (affected dataset fraction) was calculated for all genes. To find genes that were often batch effect associated across many cell types, the mean of the affected dataset fractions was calculated across cell types for each gene. To perform label transfer from the HLCA core to the mapped datasets from the extended HLCA, we used the scArches k nearest neighbor-based label transfer algorithm. Briefly, a k nearest neighbor graph was generated from the joint embedding of the HLCA core and the new, mapped dataset, setting the number of neighbors to k = 50. Based on the abundance and proximity in a cell’s neighborhood of reference cells of different types, the most likely cell type label for that cell was selected. Furthermore, a matching uncertainty score was calculated based on the consistency of reference annotations among the k nearest neighbors of the cell of interest[12pt] $$u_}_c} = 1 - p( }_}\! .$$,y,Nc=1−pY=y∣X=c,Ncwhere uc,y,Nc is the uncertainty score for a query cell c with transferred label y; Nc is its set of k nearest neighbors; and p(Y = y|X = c, Nc) is the weighted (by edge weights) proportion of Nc with label y, as previously described. Thus, high consistency in HLCA core annotations leads to low uncertainty scores and low consistency (that is, mixing of distinct reference annotations) leads to high uncertainty scores. For label transfer to lung cancer and healthy, spatially annotated projected data (Fig. 5b and Extended Data Fig. 7g), cells with an uncertainty score above 0.3 were set to unknown. Disagreement between original labels and transferred annotations (that is, transferred annotations with high certainty but not matching the original label) in the data from ref. highlighted three different cases: annotations not included in the mapped data (for example, preterminal bronchiole secretory cells, which were labeled as club and goblet in the mapped data; these may not be incorrect label transfers but cannot be verified by label comparison alone); cell types that are part of a continuum, with cutoffs between cell types chosen differently in the reference than in the projected data (for example, macrophage subtypes); and cell types missing in the HLCA core that have high transcriptional similarity to other cell types that are present in the HLCA, which was observed for several finely annotated immune cell identities. For example, γδ T cells, ILCs, megakaryocytes, natural killer T cells and regulatory T cells were not annotated in the HLCA core, as these cell types could not be distinguished with confidence in the integrated object and were often lacking in the constituent datasets. Indeed, cell types from the T cell/ILC/natural killer lineages are known to be particularly difficult to annotate using transcriptomic data only. Therefore, cells annotated with these labels in the projected dataset were largely incorrectly annotated as CD4 T cells, CD8 T cells and natural killer cells through label transfer (Fig. 5b and Extended Data Fig. 6e) For the extended atlas, we calibrated the uncertainty score cutoff by determining which uncertainty levels indicate possible failure of label transfer. To determine the uncertainty score at which technical variability from residual batch effects impairs correct label transfer, we evaluated how label transfer performed at the level of datasets, as these predominantly differ in experimental design. To determine an uncertainty threshold indicative of possible failure of label transfer, we harmonized original labels for 12 projected datasets (one unpublished: Duong_lungMAP_unpubl) and assessed the correspondence between original labels with the transferred annotations. Only cells with level 3 or 4 original annotations were considered, as these levels represent informative annotations while not representing the finest detail. Level 5 annotations will often display high uncertainty levels due to high annotation granularity rather than remaining batch effects. To assess the optimal uncertainty cutoff for labeling a new cell as unknown, we used these results to generate a receiver operating characteristic curve. We chose a cutoff around the elbow point, keeping the false positive rate below 0.5 (uncertainty cutoff = 0.2; true positive rate = 0.879; false positive rate = 0.495) to best distinguish correct from incorrect label transfers (Supplementary Fig. 10a). False positives were either due to incorrect label transfer or incorrect annotations in the original datasets. Cells with an uncertainty higher than 0.2 were set to unknown. The ref. study of healthy lung included cell type annotations based on matched spatial transcriptomic data. Many of these annotations were not present in the HLCA core. To determine whether these spatial cell types could still be recovered after mapping to the HLCA core, we looked for clusters specifically grouping these cells. We focused on six spatial cell types: perineurial nerve-associated fibroblasts; endoneurial nerve-associated fibroblasts; immune-recruiting fibroblasts; chondrocytes; myelinating Schwann cells; and nonmyelinating Schwann cells. As these cell types were often present at very small frequencies, we performed clustering at different resolutions to determine whether these cells were clustered separately at any of these resolutions. We clustered at resolutions of 0.1, 0.2, 0.5, 1, 2, 3, 5, 10, 15, 20, 25, 30, 50, 80 and 100, with the number of neighbors set to k = 30 for resolutions under 25 and k = 15 for resolutions of 25 or higher, to enable the detection of smaller clusters. Minimum recall (the percentage of cells with the spatial cell type annotation captured in the cluster) and minimum precision (the percentage of cells from ref. in the cluster that had the spatial cell type annotation) were both set to 25%. The cluster with the highest recall was selected for every spatial cell type (unless this cluster decreased precision by >33% compared with the cluster with the second highest recall). If the precision of the next best cluster was doubled compared with the cluster with the highest recall and recall did not decrease by >20%, this cluster was selected. To learn disease-specific signatures based on label transfer uncertainty scores, cells from the mapped data with the same transferred label (either alveolar fibroblasts or alveolar macrophages) were split into low-uncertainty cells (<0.2) and high-uncertainty cells (>0.4), excluding cells between these extremes (for alveolar fibroblasts, n = 11,119 (<0.2) and n = 2,863 (>0.4); for alveolar macrophages, n = 1,770 (<0.2) and n = 577 (>0.4)). We then performed a differential expression analysis on SCRAN-normalized counts using a Wilcoxon rank-sum test with default parameters, comparing high- and low-uncertainty cells. The 20 most upregulated genes based on log-fold changes were selected after filtering out genes with a false discovery rate-corrected P value (using the Benjamini–Hochberg procedure) above 0.05 and genes with a mean expression below 0.1 in the high-uncertainty group. To calculate the score of a cell for the given set of genes, the average expression of the set of genes was calculated, after which the average expression of a reference set of genes was subtracted from the original average, as described previously. The reference set consists of a randomly sampled set of genes for each binned expression value. The resulting score was considered the cell’s disease signature score. To uncover the cell identities affected in IPF, label transfer uncertainty was analyzed for three mapped datasets from the extended HLCA that included both IPF and healthy samples. Cell types of interest were determined based on the largest increase in mean label transfer uncertainty in IPF compared with healthy samples, while checking for consistency in increments across the three datasets. This highlighted alveolar fibroblasts as the main cell type of interest. To find IPF-specific alveolar fibroblast states, all alveolar fibroblasts from the abovementioned datasets and two more IPF datasets (for which no healthy data were mapped, as these were already in the core) were clustered together with the alveolar fibroblasts from the HLCA core. For clustering, a k nearest neighbor graph was calculated on the joint scArches-derived 30-dimensional embedding space setting k = 30, after which the cells were clustered using the Leiden algorithm with a resolution of 0.3. The resolution was chosen such that datasets were not isolated in single clusters, thus avoiding clustering driven solely by dataset-specific batch effects. One cluster (cluster 5) was small (n = 460 cells) and displayed low donor entropy (0.17), indicating that it almost exclusively came from a single donor (96% of cells from HLCA core donor 390C). It was therefore excluded from further analysis. To perform differential gene expression analysis, gene counts were normalized to a total of 7,666 counts (the median number of counts across the HLCA) and then log transformed with a pseudocount of 1. Finally, a Wilcoxon rank-sum test was used on the normalized data to detect differentially expressed genes for cluster 0 (n = 6,765 cells versus a total of n = 14,731). The results were filtered such that genes expressed in <30% of cells of the cluster of interest were excluded, as well as genes that were expressed in >20% of cells outside of the cluster and genes with a multiple testing-corrected P value (using the Benjamini–Hochberg procedure) above 0.05 (Supplementary Table 14). To investigate whether the HLCA can be used to identify disease-associated cell states shared across multiple diseases, MDMs from the HLCA core, together with all cells from the mapped datasets labeled as MDMs based on label transfer, were jointly analyzed. Datasets and diseases with fewer than 50 MDMs were excluded from the analysis. The cells were subsequently clustered as described above for the cross-dataset IPF analysis. Finally, a Wilcoxon rank-sum test was used on the normalized data to detect differentially expressed genes per cluster (number of cells per cluster: n = 64,915 (cluster 0), 47,539 (cluster 1), 32,027 (cluster 2), 31,097 (cluster 3), 25,267 (cluster 4), 1,998 (cluster 5) and 307 (cluster 6)). The results were filtered as described above (Supplementary Table 15). The following tools and versions were used: R (version 4.1.1 for covariate modeling and version 4.0.3 for GSEA); edgeR (version 3.28.1); lme4 (version 1.1-27.1); LDSC (version 1.0.1); Limma (version 3.46.0); Scanpy (version 1.9.1); scArches (version 0.3.5); scIB (version 0.1.1); scikit-learn (version 0.24.1); and scvi-tools (scANVI; version 0.8.1). Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41591-023-02327-2.
|
PMC1315352
|
Development of antibodies to human embryonic stem cell antigens
|
Using antibodies to specific protein antigens is the method of choice to assign and identify cell lineage through simultaneous analysis of surface molecules and intracellular markers. Embryonic stem cell research can be benefited from using antibodies specific to transcriptional factors/markers that contribute to the "stemness" phenotype or critical for cell lineage. In this report, we have developed and validated antibodies (either monoclonal or polyclonal) specific to human embryonic stem cell antigens and early differentiation transcriptional factors/markers that are critical for cell differentiation into definite lineage. These antibodies enable stem cell biologists to conveniently identify stem cell characteristics and to quantitatively assess differentiation.Although the stem cell concept was introduced decades ago, to date, stem cells can only be defined functionally, not morphologically or phenotypically. Two functions define stem cells. Firstly, they are self-renewing, thus able to propagate to generate additional stem cells. Secondly they can differentiate into various progenitor cells, which commit to further maturation along a specific lineage. While stem cells can be best defined functionally, a good number of molecular markers have been used to prospectively identify various stem cell populations. Although the functional importance of many of these antigens remains unknown, their unique expression pattern and timing of expression provide a useful tool for scientists to identify as well as isolate stem cells. Embryonic stem cells (ESC), derived from the inner cell mass of pre-implantation embryos, have been recognized as the earliest stem cell population . This pluripotent population can differentiate into all somatic tissue including germ cells. In the case of human ESC, they can differentiate into some extra-embryonic derivatives as well. Like mouse ESC, human ES cells can be maintained and propagated on mouse fibroblast feeders for extended periods in media containing basic fibroblast growth factor (bFGF) . Gene expression of undifferentiated human ES cells has been investigated among several ES cell lines by a variety of techniques. They include comparison with databases, reverse transcriptase-polymerase chain reaction, focused cDNA microarrays, and immunocytochemistry. A list of molecules comprised of known ES-specific or -highly expressed genes and candidates that can serve as markers for human ESCs and may also contribute to the "stemness" phenotype has been established [3-11]. For example, pluripotent ESC can be characterized by high level expression of Oct3/4 (POU domain, class 5, transcription factor 1, Pou5f1) and Nanog, which are a member of POU domain and homeobox transcription factors respectively. A critical amount of Oct3/4 and Nanog expression is required to sustain stem-cell pluripotency and both of these markers are downregulated as cells differentiate in vitro and in vivo [4-9]. Antibodies to Oct3/4 which cross react with human Oct 3/4 have been widely used to monitor the presence of undifferentiated ESC. No single marker however is sufficient or unique for identifying ESCs. Oct3/4 for example is expressed by germ cells and may be expressed by specific populations later in development. Likewise, Nanog has been shown to express in other tissues. We and other have noted however, that while no single marker is sufficient a constellation of positive and negative markers can in concert unambiguously allow one to define the state of ESC cultures and that surface markers in combination can be used to prospectively sort for ESC. Based on published data at the level of gene expression, we have cloned a number of candidate marker genes. We have also expressed the recombinant protein and generated a panel of monoclonal or polyclonal antibodies to these proteins. Using these antibodies we have confirmed the specificity and selectivity of these antibodies on several ESC lines and established their utility as stem cells markers. Our results confirm the expression pattern and timing of these cell markers at the protein level, whereas previous data reported at the level of gene expression. All monoclonal antibodies were initially selected for their abilities to recognize recombinant proteins in direct ELISAs. A subset were also tested by Western Blot analysis using recombinant proteins and cell lysate to confirm binding to a single epitope. The best clone was later screened for its applications for immunocytochemistry and flow cytometry using various cell lines. Human peripheral blood platelets were used for screening mouse anti-human CD9 antibody. MCF-7 cells were used for screening mouse anti-human E-Cadherin and PODXL (podocalyxin-like) antibodies. MG-63 cells were used for screening mouse anti-human GATA1 (GATA binding protein 1) antibody. Beta-TC6 cells were used for screening for mouse anti-human/mouse PDX-1 (pancreatic duodenal homeobox-1) antibody. NTERA-2 cells were used for screening mouse anti-human Oct3/4 and SOX2 (sex-determining region Y-box 2) antibodies. All polyclonal antibodies were affinity-purified using recombinant proteins and validated by direct ELISAs and Western. Caco-2 cells were used for validation of goat anti-human GATA6 antibody and NTERA-2 cells were used for validation of goat anti-human Nanog and anti-human Oct3/4 antibodies (Summarized in Table 1). Summary list of antibody verification by western blot. N/A: 1. DPPA5 is still being subcloned. Only Elisa verification is available. 2. The clone for GATA-1 (MAB1779) does not work for Western blot application but is useful for IHC, The clone picked for Western blot analysis does not work for IHC (MAB17791, see data in ). After antibodies were validated in direct ELISAs, Western blot or cell lines (Fig. 1 and data not shown), they were used to examine the expression of individual molecules in undifferentiated human ES cells and differentiated EBs. When examined by immunohistochemistry, high level of expressions of Oct3/4, SOX2, E-Cadherin, PODXL and Nanog were observed in undifferentiated human ES cells (Fig. 2A, 2B and 2C). DPPA5 (developmental pluripotency associated 5) expression was also observed in undifferentiated human ES cells (data not shown). We noted that a subset of the proteins used were membrane bound proteins. To test if any of the antibodies generated could recognize an extracellular epitope and thus be used for live cell sorting, we repeated staining of live cells as previously described. The CD9, E-Cadherin and PODXL antibodies recognized an extracellular epitope and their ability to select cells by FACS was confirmed (Fig. 3). Minimal or no expressions of Oct3/4, E-Cadherin, PODXL and Nanog were detected in the differentiated EBs (Fig. 2D, 2E and 2F). However, SOX2 expression, which is observed in neural progenitor cells, is persistent in subsets of EBs. Western blot analysis for Gt × hOct3/4 (A), Gt × hNanog (B) and Ms × hSOX2 (C) in NTERA-2 cell lysate, Ms × hE-Cadherin (D) in MCF-7 cell lysate, Ms × hCD9 (E) in PBMC lysate and Ms × hPDX-1(F) in β-TC-6 cell lysate. Numbers indicate the positions of molecular weight markers. Undifferentiated human ES cells (A, B, and C) and differentiated EBs (D, E and F) were analyzed using antibodies to indicated molecular markers. Immunostaining with goat anti-human Oct3/4 (Red in A and D), mouse anti-human SOX2 (Green in A and D), goat anti-human E-Cadherin (Red in B and E), mouse anti-human PODXL (Green in B and E), and goat anti-human Nanog (Red in C and F), are contrasted with DAPI nuclear staining (Blue in C-F). Note the dramatic downregulation of ESC specific markers (Oct3/4, E-Cadherin, PODXL, and Nanog) in EBs. However, SOX2 expression is persistent in subsets of EB cells. Scale bars = 100 μm. Human embryonic stem cells stained with anti-CD9 (A), anti-E-Cadherin (B), and anti-PODXL (C) and antigen expression detected by a flow cytometer. The specific staining is indicated by green histogram and corresponding isotype control is indicated by black histogram. Suspension culture with FGF withdrawal is known to induce differentiation of ES cells to all three germ layer precursors . The differentiation status of the EB used here was detected to contain all germ cell markers by RT-PCR (Fig. 4). In order to examine how more antibodies can be used for characterization of early differentiation events from human ES cells, we examined the expressions of endodermal markers, SOX17, GATA6 and PDX-1, and mesodermal markers, Brachyury and GATA1, in the undifferentiated human ES cells and differentiated EBs. Expressions of SOX17, GATA6, PDX-1, Brachyury and GATA1 were not detected in undifferentiated human ES cells (data not shown). In contrast to the undifferentiated ES cells, subpopulations of SOX17-, GATA6-, Brachyury- and GATA1-positive cells were observed (Fig 4). These results suggest that both endodermal and mesodermal precursors exist in EBs with FGF withdrawal for 8 days. However, no PDX-1-positive cells were seen in EBs differentiated with the same treatment (data not shown). Differentiated EBs were analyzed by either immunocytochemistry or RT-PCR to the indicated molecular markers. (A) Immunostaining with goat anti-human SOX17 (Red), is contrasted with Fluoro Nissl nuclear staining (Green). (B) Immunostaining with goat anti-human GATA6 (Red), is contrasted with DAPI nuclear staining (Blue). (C) Immunostaining with goat anti-human brachyury (Red), is contrasted with DAPI nuclear staining (Blue). (D) Immunostaining with mouse anti-human GATA1 (Red). Note that each antibody recognizes subsets of EB cells. Scale bars = 100 μm. (E) The differentiation status of EB is detected by RT-PCR using different germ layer cell markers. Selected endoderm markers AFP, FoxA2; mesoderm markers Hand1, MSX1 and ectoderm marker Msl1 were all highly expressed in the EB samples while their expression was either undetectable or at low level in the ES samples. G3PDH was a positive control showing similar amount of RNA samples were used for analysis. We have also examined the cross-reactivities of these antibodies to mouse ES cells using mouse D3 ES cell line and mouse fetal endodermal tissue. Cross-reactivity to mouse of goat anti-Oct3/4, goat anti-PDX-1, goat anti-SOX17 and mouse anti-SOX2 was detected. Minimal cross-reactivity to mouse, measured by 10% intensity to human by higher than control cells, was observed in mouse anti-CD9 and mouse anti-E-cadherin antibodies. Goat anti-Nanog and mouse anti-PODXL antibodies appear to be human-specific as well (data not shown). The subtypes of monoclonal antibodies were also identified in the best clones. These results are summarized in Table 2. Summary of antibodies detection in ES and EB samples. *NT, Not tested; ND, Not determined. The expression patterns detected using antibodies developed in our facility are consistent with data reported using reverse transcriptase-polymerase chain reaction or cDNA microarrays. Moreover several of the monoclonal antibodies have differing heavy chain subunits allowing double labeling using subtype specific markers to be performed. In summary, we have developed a useful collection of antibodies that would be useful for identification of stem cell characteristics and assessment of differentiation. Several additional antibodies to the molecules that have been identified as potential cell lineage markers are currently under development using the same approach. Brachyury (aa. 1–202), DPPA5 (a.a. 1–116), GATA1 (a.a. 1–413), GATA6 (aa. 1–449), Nanog (aa. 153–305), Oct3/4 (aa. 1–265), PDX-1 (aa. 1–283), SOX2 (aa. 135–317) and SOX17 (aa. 177–414) were expressed in E. Coli and extracellular domains of CD9, E-Cadherin, PODXL were expressed in mouse NSO cells. All proteins were purified and sequenced before they were used as antigens for immunizations and as substrate for antibody screening and subcloning. All monoclonal antibodies were derived from fusions of mouse myeloma with B cells obtained from BALB/c mice which had been immunized with purified antigen. The IgG fraction of the culture supernatant was purified by Protein G affinity chromatography (Sigma). Each panel of antibodies was screened and selected for their abilities to detect purified recombinant antigen in direct ELISA and Western blot. All polyclonal antibodies were derived from sera of goats which had been immunized and boost it with purified antigen. Antibody was purified from the sera by an antigen-affinity chromatography. Human Caco-2, MG-63, MCF-7, NTERA-2 and mouse D3 cells were purchased from American Type Culture Collection (ATCC). Cells were cultured according to the ATCC instructions. Information regarding human ES cell line HSF-6 (NIH code UC06) can be obtained at the website . Undifferentiated human ES cells were cultured according to the protocol provided by the University of California, San Francisco in human ES culture medium [DMEM supplemented with 20% KnockOut Serum Replacement (Invitrogen) and 5 ng/mL of bFGF (R&D Systems)]. To induce formation of embryoid bodies (EBs), ES colonies were harvested, separated from the MEF feeder cells by gravity, gently resuspended in ES culture medium and transferred to non-adherent suspension culture dishes (Corning). Unless otherwise noted, EBs derived from human ES cell aggregates were cultured for 8 days in ES culture medium deprived of bFGF and used for analysis by immunohistochemistry as described. Cells are solubilized in hot 2× SDS gel sample buffer (20 mM dithiothreitol, 6% SDS, 0.25 M Tris, pH 6.8, 10% glycerol, 10 mM NaF and bromophenyl blue) at 2 × 10per mL. The extracts are heated in a boiling water bath for 5 minutes and sonicated with a probe sonicator with 3–4 bursts of 5–10 seconds each. Samples are diluted with 1× SDS sample buffer to the desired loading of 1–5 × 10per lane. Lysates were resolved by SDS-PAGE, transferred to Immobilon-P membrane, and immunoblotted with 0.5 μg/mL primary Abs as described in R&D Systems Website . Antibodies were used with the appropriate secondary reagents at a concentration of 5 to 10 μg/ml. Cells or sections of EBs were fixed with 4% paraformaldehyde in PBS at room temperature for 20 min, then blocked and permeabilized with 0.1% Triton X-100, 1% BSA, 10% normal donkey serum in PBS at room temperature for 45 min. After blocking, cells were incubated with diluted primary antibody overnight at 4°C followed by coupled anti-mouse or anti-goat IgG (Molecular Probes) at room temperature in the dark for an hour. Between each step cells were washed with PBS with 0.1% BSA. Total RNA was extracted from EBs using Trizol LS (Invitrogen). cDNA was synthesized by using Superscript II reverse transcriptase (Invitrogen) according to the manufacturer's recommendations. The PCR primers are available upon request. Antibodies were prepared at the concentration of 0.1 mg/mL. 10 μL of the stock solution was added to 1 – 2.5 × 10cells in a total reaction volume not exceeding 200 μL. The sample was then incubated for 20 min at 2–8 °C. Following incubation, excess antibody was removed by washing cells twice with FACS buffer (2% FCS and 0.1% sodium azide in Hank's buffer). After wash, cells were resuspend in 200 μL of FACS buffer and the binding of unlabeled monoclonal antibodies was visualized by adding 10 μL of a 25 μg/mL stock solution of a secondary developing reagent such as goat anti-mouse IgG conjugated to a fluorochrome for 20 min at 2–8°C. Following incubation, cells were washed once with FACS buffer, once with PBS. After wash, cells were resuspend in 400 μL of PBS and analyzed on a FACScant flow cytometer (Becton-Dickinson, Mountain View, CA). Five thousand events were collected and analyzed using CELL Quest software. Dr. Cai contributed significantly in validating antibodies in human ES cells and human EBs. Ms. Olson performed initial screening of antibodies in various cell lines. Dr. Rao initiated the project, supervised Dr. Cai, and participated in all discussions for this report. Ms. Stanley and Ms. Taylor performed Western blot analysis. Dr. Ni coordinated collaborative work between two labs, monitored the generation of the antibodies, and directed the project at R&D Systems.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.