Exploiting the diphtheria toxin internalization receptor enhances delivery of proteins to lysosomes for enzyme replacement therapy – Science Advances

Abstract

Enzyme replacement therapy, in which a functional copy of an enzyme is injected either systemically or directly into the brain of affected individuals, has proven to be an effective strategy for treating certain lysosomal storage diseases. The inefficient uptake of recombinant enzymes via the mannose-6-phosphate receptor, however, prohibits the broad utility of replacement therapy. Here, to improve the efficiency and efficacy of lysosomal enzyme uptake, we exploited the strategy used by diphtheria toxin to enter into the endolysosomal network of cells by creating a chimera between the receptor-binding fragment of diphtheria toxin and the lysosomal hydrolase TPP1. We show that chimeric TPP1 binds with high affinity to target cells and is efficiently delivered into lysosomes. Further, we show superior uptake of chimeric TPP1 over TPP1 alone in brain tissue following intracerebroventricular injection in mice lacking TPP1, demonstrating the potential of this strategy for enhancing lysosomal storage disease therapy.

Lysosomal storage diseases (LSDs) are a group of more than 70 inherited childhood diseases characterized by an accumulation of cellular metabolites arising from deficiencies in a specific protein, typically a lysosomal hydrolase. Although each individual disease is considered rare, LSDs have a combined incidence of between 1/5000 and 1/8000 live births, and together, they account for a substantial proportion of the neurodegenerative diseases in children (1). The particular age of onset for a given LSD varies depending on the affected protein and the percentage of enzymatic activity still present; however, in most cases, symptoms manifest early in life and progress insidiously, affecting multiple tissues and organs (2). In all but the mildest of cases, disease progression results in severe physical disability, possible intellectual disability, and a shortened life expectancy, with death occurring in late childhood or early adolescence.

As they are monogenic diseases, reintroducing a functional form of the defective enzyme into lysosomes is in principle a viable strategy for treating LSDs. Enzyme replacement therapy (ERT) is now approved for the treatment of seven LSDs, and clinical trials are ongoing for five others (3). However, delivering curative doses of recombinant lysosomal enzymes into lysosomes remains a major challenge in practice. ERT typically takes advantage of a specific N-glycan posttranslational modification, mannose-6-phosphorylation (M6P), which controls trafficking of endogenous lysosomal enzymes, as well as exogenous uptake of lysosomal enzymes from circulation by cells having the cation-independent M6P receptor (CIMPR) (4). Hence, a combination of factors including (i) the abundance of the M6P receptor in the liver, (ii) poor levels of CIMPR expression in several key target tissue types such as bone and skeletal muscle, (iii) incomplete and unpredictable M6P labeling of recombinant enzymes, and (iv) the highly variable affinity of recombinant lysosomal enzymes for CIMPR [viz., Kds (dissociation constants) ranging from low to mid micromolar (5, 6)] all contribute to diminishing the overall effectiveness of therapies using CIMPR for cell entry (3).

To improve the delivery of therapeutic lysosomal enzymes, we drew inspiration from bacterial toxins, which, as part of their mechanism, hijack specific host cellsurface receptors to gain entry into the endolysosomal pathway. While we and others have explored exploiting this pathway to deliver cargo into the cytosol (7, 8), here we asked whether this same approach could be used to enhance the delivery of lysosomal enzymes into lysosomes. We choose the diphtheria toxin (DT)diphtheria toxin receptor (DTR) system owing to the ubiquitous nature of the DTR, in particular its high expression levels on neurons.

Corynebacterium diphtheriae secretes DT exotoxin, which is spread to distant organs by the circulatory system, where it affects the lungs, heart, liver, kidneys, and the nervous system (9). It is estimated that 75% of individuals with acute disease also develop some form of peripheral or cranial neuropathy. This multiorgan targeting results from the fact that the DTR, heparin-binding EGF (epidermal growth factor)like growth factor (HBEGF), is ubiquitously expressed. The extent to which DT specifically targets difficult-to-access tissues such as muscle and bone, however, is not currently known.

DT is a three-domain protein that consists of an N-terminal ADP (adenosine diphosphate)ribosyl transferase enzyme (DTC), a central translocation domain (DTT), and a C-terminal receptorbinding domain (DTR). The latter is responsible for both binding cell surface HBEGF with high affinity [viz., Kd = 27 nM (10)] and triggering endocytosis into early endosomes (Fig. 1A). Within endosomes, DTT forms membrane-spanning pores that serve as conduits for DTC to enter the cytosol where it inactivates the host protein synthesis machinery. The remaining portions of the toxin remain in the endosomes and continue to lysosomes where they are degraded (11, 12). We hypothesized that the receptor-binding domain, lacking any means to escape endosomes, would proceed with any attached cargo to lysosomes and, thus, serve as a means to deliver cargo specifically into lysosomes following high-affinity binding to HBEGF.

(A) DT intoxication pathway (left), DT domain architecture, and LTM structure (right). (B and C) DTK51E/E148K, LTM, mCherry-LTM, and LTM-mCherry compete with wild-type DT for binding and inhibit its activity in a dose-dependent manner with IC50 (median inhibitory concentration) values of 46.9, 10.1, 52.7, and 76.1 nM, respectively (means SD; n = 3). (D and E) C-terminal and N-terminal fusions of LTM to mCherry were immunostained (red) and observed to colocalize with the lysosomal marker LAMP1 (39). (F) Fractional co-occurrence of the red channel with the green channel (Manders coefficient M2) were calculated for mCherry-LTM and LTM-mCherry and were found to be 0.61 0.10 and 0.52 0.11, respectively (means SD; n = 6).

In this study, we generated a series of chimeric proteins containing the DTR-binding domain, DTR, with the goal of demonstrating the feasibility of delivering therapeutic enzymes into lysosomes through the DT-HBEGF internalization pathway. We showed that DTR serves as a highly effective and versatile lysosome-targeting moiety (LTM). It can be placed at either the N or C terminus of the cargo, where it retains its high-affinity binding to HBEGF and the ability to promote trafficking into lysosomes both in vitro and in vivo. On the basis of its advantages, over M6P-mediated mechanisms, we further investigated the utility of LTM for the lysosomal delivery of human tripeptidyl peptidase-1 (TPP1) with the long-term goal of treating Batten disease.

To evaluate whether the DTR-binding fragment could function autonomously to traffic cargo into lysosomes, we first asked whether the isolated 17-kDa DTR fragment could be expressed independently from DT holotoxin and retain its affinity for HBEGF. We cloned, expressed, and purified the receptor-binding fragment and evaluated its ability to compete with full-length DT for the DTR, HBEGF. Before treating cells with a fixed dose of wild-type DT that completely inhibits protein synthesis, cells were incubated with a range of concentrations of LTM or a full-length, nontoxic mutant of DT (DTK51E/E148K). LTM-mediated inhibition of wild-type DT-mediated toxicity was equivalent to nontoxic DT (Fig. 1B), demonstrating that the receptor-binding fragment can be isolated from the holotoxin without affecting its ability to fold and bind cell surface HBEGF. Next, we evaluated whether LTM had a positional bias (i.e., was able to bind HBEGF with a fusion partner when positioned at either terminus). To this end, we generated N- and C-terminal fusions of LTM to the model fluorescent protein mCherry (i.e., mCherry-LTM and LTM-mCherry). To determine binding of each chimera to HBEGF, we quantified the ability of each chimera to compete with wild-type DT on cells in the intoxication assay. Both constructs competed with wild-type DT to the same extent as LTM alone and DTK51E/E148K (Fig. 1C), demonstrating that LTM is versatile and autonomously folds in different contexts.

To evaluate intracellular trafficking, HeLa cells were treated with either LTM-mCherry or mCherry-LTM and then fixed and stained 4 hours later with an antibody against the lysosomal marker LAMP1. In both cases, we observed significant uptake of the fusion protein (Fig. 1, D and E). We calculated Manders coefficients (M2) to quantify the extent to which signal in the red channel (LTM-mCherry and mCherry-LTM) was localizing with signal in the green channel (LAMP1). The fraction of red/green co-occurrence was calculated to be 0.61 for mCherry-LTM and 0.52 for LTM-mCherry, indicating trafficking to the lysosomal compartments of the cells and no significant difference (P = 0.196) between the two orientations of chimera (Fig. 1F). Together, these results confirm that the LTM is capable of binding HBEGF and trafficking associated cargo into cells and that the LTM can function in this manner at either terminus of a fusion construct.

With minimal positional bias observed in the mCherry fusion proteins, we next screened LTM fusions to TPP1 to identify a design that maximizes expression, stability, activity, and, ultimately, delivery. TPP1 is a 60-kDa lysosomal serine peptidase encoded by the CLN2 gene, implicated in neuronal ceroid lipofuscinosis type 2 or Batten disease. Loss of function results in the accumulation of lipofuscin, a proteinaceous, autofluorescent storage material (13). Exposure to the low-pH environment of the lysosome triggers autoproteolytic activation of TPP1 and release of a 20-kDa propeptide that occludes its active site. From a design perspective, we favored an orientation in which the LTM was N terminal to TPP1, as autoprocessing of TPP1 would result in the release of the upstream LTM-TPP1 propeptide, liberating active, mature TPP1 enzyme in the lysosome (Fig. 2A). Given the need for mammalian expression of lysosomal enzymes, we generated synthetic genetic fusions of the LTM to TPP1, in which we converted the codons from bacterially derived DT into the corresponding mammalian codons. Human embryonic kidney (HEK) 293F suspension cells stably expressing recombinant TPP1 (rTPP1) and TPP1 with an N-terminal LTM fusion (LTM-TPP1) were generated using the piggyBac transposon system (14). A C-terminal construct (TPP1-LTM) was also produced; however, expression of this chimera was poor in comparison with rTPP1 and LTM-TPP1 (~0.4 mg/liter, cf. 10 to 15 mg/liter).

(A) Design of LTM-TPP1 fusion protein and delivery schematic. (B) Enzyme kinetics of rTPP1 and LTM-TPP1 against the synthetic substrate AAF-AMC are indistinguishable. Michaelis-Menten plots were generated by varying [AAF-AMC] at a constant concentration of 10 nM enzyme (means SD; n = 3). Plots and kinetic parameters were calculated with GraphPad Prism 7.04. (C) Maturation of TPP1 is unaffected by the N-terminal fusion of LTM. (D) LTM-TPP1 inhibits wild-type DT activity in a dose-dependent manner (IC50 of 17.2 nM), while rTPP1 has no effect on protein synthesis inhibition by DT (means SD; n = 3). (E) LTM and DTR-TPP1 bind HBEGF with apparent Kds of 13.3 and 19.1 nM, respectively. (F) LTM-TPP1 (39) colocalizes with LAMP1 staining (red).

The activity of rTPP1 and LTM-TPP1 against the tripeptide substrate Ala-Ala-Phe-AMC (AAF-AMC) was assessed to determine any effects of the LTM on TPP1 activity. The enzyme activities of rTPP1 and LTM-TPP1 were determined to be equivalent, as evidenced through measurements of their catalytic efficiency (Fig. 2B), demonstrating that there is no inference by LTM on the peptidase activity of TPP1. Maturation of LTM-TPP1 through autocatalytic cleavage of the N-terminal propeptide was analyzed by SDSpolyacrylamide gel electrophoresis (PAGE) (Fig. 2C). Complete processing of the zymogen at pH 3.5 and 37C occurred between 5 and 10 min, which is consistent with what has been observed for the native recombinant enzyme (15).

The ability of LTM-TPP1 to compete with DT for binding to extracellular HBEGF was first assessed with the protein synthesis competition assay. Similar to LTM, mCherry-LTM, and LTM-mCherry, LTM-TPP1 prevents protein synthesis inhibition by 10 pM DT with an IC50 (median inhibitory concentration) of 17.2 nM (Fig. 2D). As expected, rTPP1 alone was unable to inhibit DT-mediated entry and cytotoxicity. To further characterize this interaction, we measured the interaction between LTM and LTM-TPP1 and recombinant HBEGF using surface plasmon resonance (SPR) binding analysis (Fig. 2E). By SPR, LTM and LTM-TPP1 were calculated to have apparent Kds of 13.3 and 19.1 nM, respectively, values closely corresponding to the IC50 values obtained from the competition experiments (10.1 and 17.2 nM, respectively). Consistent with these results, LTM-TPP1 colocalizes with LAMP1 by immunofluorescence (Fig. 2F).

To study uptake of chimeric fusion proteins in cell culture, we generated a cell line deficient in TPP1 activity. A CRISPR RNA (crRNA) was designed to target the signal peptide region of TPP1 in exon 2 of CLN2. Human HeLa Kyoto cells were reverse transfected with a Cas9 ribonucleoprotein complex and then seeded at low density into a 10-cm dish. Single cells were expanded to colonies, which were picked and screened for TPP1 activity. A single clone deficient in TPP1 activity was isolated and expanded, which was determined to have ~4% TPP1 activity relative to wild-type HeLa Kyoto cells plated at the same density (Fig. 3A). The small residual activity observed is likely the result of another cellular enzyme processing the AAFAMC (7-amido-4-methlycoumarin) substrate used in this assay, as there is no apparent TPP1 protein being produced (Fig. 3B). Sanger sequencing of the individual alleles confirmed complete disruption of the CLN2 gene (fig. S1). In total, three unique mutations were identified within exon 2 of CLN2: a single base insertion resulting in a frameshift mutation and two deletions of 24 and 33 base pairs (bp), respectively.

(A) CLN2 knockout cells exhibit ~4% TPP1 activity relative to wild-type HeLa Kyoto cells (means SD; n = 3). (B) Western blotting against TPP1 reveals no detectable protein in the knockout cells. (C) (Left) In vitro maturation of pro-rTPP1 and LTM-TPP1 (16 ng) was analyzed by Western blot. (Right) TPP1 present in wild-type (WT) and TPP1/ cells, and TPP1/ cells treated with 100 nM rTPP1 and LTM-TPP1. (D) Uptake of rTPP1 and LTM-TPP1 into HeLa Kyoto TPP1/ cells was monitored by TPP1 activity (means SD; n = 4). (E) TPP1 activity present in HeLa Kyoto TPP1/ cells following a single treatment with 50 nM LTM-TPP1 (means SD; n = 3).

Next, we compared the delivery and activation of rTPP1 and LTM-TPP1 into lysosomes by treating TPP1/ cells with a fixed concentration of the enzymes (100 nM) and by analyzing entry and processing by Western blot (Fig. 3C). In both cases, most enzymes were present in the mature form, indicating successful delivery to the lysosome; however, the uptake of LTM-TPP1 greatly exceeded the uptake of rTPP1. As both rTPP1 and LTM-TPP1 receive the same M6P posttranslational modifications promoting their uptake by CIMPR, differences in their respective uptake should be directly attributable to uptake by HBEGF. To quantify the difference in uptake and lysosomal delivery, cells were treated overnight with varying amounts of each enzyme, washed, lysed, and assayed for TPP1 activity. The activity assays were performed without a preactivation step, so signal represents protein that has been activated in the lysosome. For both constructs, we observed a dose-dependent increase in delivery of TPP1 to the lysosome (Fig. 3D). Delivery of LTM-TPP1 was significantly enhanced compared with TPP1 alone at all doses, further demonstrating that uptake by HBEGF is more efficient than that by CIMPR alone. TPP1 activity in cells treated with LTM-TPP1 was consistently ~10 greater than that of cells treated with rTPP1, with the relative difference increasing at the highest concentrations tested. This may speak to differences in abundance, replenishment, and/or recycling of HBEGF versus CIMPR, in addition to differences in receptor-ligand affinity. Uptake of LTM-TPP1 and rTPP1 into several other cell types yielded similar results (fig. S2). To assess the lifetime of the delivered enzyme, cells were treated with LTM-TPP1 (50 nM) and incubated overnight. Cells were washed and incubated with fresh media, and TPP1 activity was assayed over the course of several days. Cells treated with LTM-TPP1 still retained measurable TPP1 activity at 1 week after treatment (Fig. 3E).

While the DT competition experiment demonstrated that HBEGF is involved in the uptake of LTM-TPP1 but not rTPP1 (Fig. 2D), it does not account for the contribution of CIMPR to uptake. Endoglycosidase H (EndoH) cleaves between the core N-acetylglucosamine residues of high-mannose N-linked glycans, leaving behind only the asparagine-linked N-acetylglucosamine moiety. Both rTPP1 and LTM-TPP1 were treated with EndoH to remove any M6P moieties, and delivery into Hela TPP1/ was subsequently assessed. While rTPP1 uptake is completely abrogated by treatment with EndoH, LTM-TPP1 uptake is only partially decreased (Fig. 4), indicating that while HBEGF-mediated endocytosis is the principal means by which LTM-TPP1 is taken up into cells, uptake via CIMPR still occurs. The fact that CIMPR uptake is still possible in the LTM-TPP1 fusion means that the fusion is targeted to two receptors simultaneously, increasing its total uptake and, potentially, its biodistribution.

Uptake of LTM-TPP1 via the combination of HBEGF and CIMPR was shown to be 3 to 20 more efficient than CIMPR alone in cellulo (fig. S2). To interrogate this effect in vivo, TPP1-deficient mice (TPP1tm1pLob or TPP1/) were obtained as a gift from P. Lobel at Rutgers University. Targeted disruption of the CLN2 gene was achieved by insertion of a neo cassette into intron 11 in combination with a point mutation (R446H), rendering these mice TPP1 null by both Western blot and enzyme activity assay (16). Prior studies have demonstrated that direct administration of rTPP1 into the cerebrospinal fluid (CSF) via intracerebroventricular or intrathecal injection results in amelioration of disease phenotype (17) and even extension of life span in the disease mouse (18). To compare the uptake of LTM-TPP1 and rTPP1 in vivo, the enzymes were injected into the left ventricle of 6-week-old TPP1/ mice. Mice were euthanized 24 hours after injection, and brain homogenates of wild-type littermates, untreated, and treated mice were assayed for TPP1 activity (Fig. 5A). Assays were performed without preactivation, and therefore, the results report on enzyme that has been taken up into cells, trafficked to the lysosome, and processed to the mature form.

(A) Assay schematic. (B) TPP1 activity in brain homogenates of 6-week-old mice injected with two doses (5 and 25 g) of either rTPP1 or LTM-TPP1 (5 g, P = 0.01; 25 g, P = 0.002). (C) TPP1 activity in brain homogenates following a single 25-g dose of LTM-TPP1, 1, 7, and 14 days postinjection. Data are presented as box and whisker plots, with whiskers representing minimum and maximum values from n 4 mice per group. Statistical significance was calculated using paired t tests with GraphPad Prism 7.04.

While both enzymes resulted in a dose-dependent increase in TPP1 activity, low (5 g) and high (25 g) doses of rTPP1 resulted in only modest increases of activity, representing ~6 and ~26% of the wild-type levels of activity, respectively (Fig. 5B). At the same doses, LTM-TPP1 restored ~31 and ~103% of the wild-type activity. To assess the lifetime of enzyme in the brain, mice were injected intracerebroventricularly with 25 g of LTM-TPP1 and euthanized either 1 or 2 weeks postinjection. Remarkably, at 1 week postinjection, ~68% of TPP1 activity was retained (compared with 1 day postinjection), and after 2 weeks, activity was reduced to ~31% (Fig. 5C).

ERT is a lifesaving therapy that is a principal method of treatment in non-neurological LSDs. Uptake of M6P-labeled enzymes by CIMPR is relatively ineffective due to variable receptor affinity (5, 6), heterogeneous expression of the receptor, and incomplete labeling of recombinantly produced enzymes (19). Despite its inefficiencies and high cost (~200,000 USD per patient per year) (20), it remains the standard of care for several LSDs, as alternative treatment modalities (substrate reduction therapy, gene therapy, and hematopoietic stem cell transplantation) are not effective, not as well developed, or inherently riskier (2125). Improving the efficiency and distribution of recombinant enzyme uptake may help address some of the current shortcomings in traditional ERT.

Several strategies have been used to increase the extent of M6P labeling on recombinantly produced lysosomal enzymes: engineering mammalian and yeast cell lines to produce more specific/uniform N-glycan modification (19, 26, 27), chemical or enzymatic modification of N-glycans posttranslationally (28), and covalent coupling of M6P (29). M6P-independent uptake of a lysosomal hydrolase by CIMPR has been demonstrated for both -glucuronidase (28) and acid -glucosidase (30, 31). In the latter work, a peptide tag (GILT) targeting insulin-like growth factor II receptor (IGF2R) was fused to recombinant alpha glucosidase, which enabled receptor-mediated entry into cells. CIMPR is a ~300-kDa, 15-domain membrane protein with 3 M6P-binding domains and 1 IGF2R domain. By targeting the IGF2R domain with a high-affinity (low nanomolar) peptide rather than the low-affinity M6P-binding domain, the authors were able to demonstrate a >20-fold increase in the uptake of a GAA-peptide fusion protein in cell culture and a ~5-fold increase in the ability to clear built-up muscle glycogen in GAA-deficient mice.

In this study, we have demonstrated efficient uptake and lysosomal trafficking of a model lysosomal enzyme, TPP1, via a CIMPR-independent route, using the receptor-binding domain of a bacterial toxin. HBEGF is a member of the EGF family of growth factors, and DT is its only known ligand. Notably, it plays roles in cardiac development, wound healing, muscle contraction, and neurogenesis; however, it does not act as a receptor in any of these physiological processes (32). Intracellular intoxication by DT is the only known process in which HBEGF acts as a receptor, making it an excellent candidate receptor for ERT, as there is no natural ligand with which to compete. Upon binding, DT is internalized via clathrin-mediated endocytosis and then trafficked toward lysosomes for degradation (33, 34). Acidification of endosomal vesicles by vacuolar ATPases (adenosine triphosphatases) promotes insertion of DTT into the endosomal membrane and subsequent translocation of the catalytic DTC domain into the cytosol. In the absence of an escape mechanism, the majority of internalized LTM should be trafficked to the lysosome, as we have demonstrated with our chimera (Figs. 2F and 3C). Uptake of LTM-TPP1 in vitro is robustly relative to rTPP1 (Fig. 3D and fig. S2), and TPP1 activity is sustained in the lysosome for a substantial length of time (Fig. 3E). We have also demonstrated that the increase in uptake efficiency that we observed in cell culture persists in vivo. TPP1 activity in the brains of CLN2-null mice was significantly greater in animals treated with intracerebroventricularly injected LTM-TPP1, as compared with those treated with TPP1 at two different doses (Fig. 5B), and, remarkably, this activity persists with an apparent half-life of ~8 days (Fig. 5C).

An important consideration for further development of the LTM platform for clinical development is the potential immunogenicity of using a bacterial fragment in this context. Previously, we demonstrated that the receptor-binding fragment of DT could be replaced with a human scFv (single-chain fragment variable) targeting HBEGF (8). With our demonstration of the potential for targeting HBEGF for LSDs, future efforts will focus on increasing the affinity and specificity of these first-generation humanized LTMs to develop high-affinity chimeras with greatly reduced immunogenicity for further development.

While the ability of LTM-TPP1 to affect disease progression has yet to be determined, recent positive clinical trial results (35) and the subsequent approval of rTPP1 (cerliponase alfa) for treatment of neuronal ceroid lipofuscinosis 2 (NCL2) provide support for this approach. In that clinical trial, 300 mg of rTPP1 was administered by biweekly intracerebroventricular injection to 24 affected children, and this was able to prevent disease progression. While this dose is of the same order of magnitude as other approved ERTs (<1 to 40 mg/kg) (36, 37), it represents a substantial dose, especially considering that it was delivered to a single organ. Improving the efficiency of uptake by targeting an additional receptor as we have done here, is expected to greatly decrease the dose required to improve symptoms, while at the same time decreasing costs and the chances of dose-dependent side effects.

DTK51E/E148K, LTM, LTM-mCherry, mCherry-LTM, and HBEGF constructs were cloned using the In-Fusion HD cloning kit (Clontech) into the Champion pET SUMO expression system (Invitrogen). Recombinant proteins were expressed as 6His-SUMO fusion proteins in Escherichia coli BL21(DE3)pLysS cells. Cultures were grown at 37C until an OD600 (optical density at 600 nm) of 0.5, induced with 1 mM IPTG (isopropyl--d-thiogalactopyranoside) for 4 hours at 25C. Cell pellets harvested by centrifugation were resuspended in lysis buffer [20 mM tris (pH 8.0), 160 mM NaCl, 10 mM imidazole, lysozyme, benzonase, and protease inhibitor cocktail] and lysed by three passages through an EmulsiFlex C3 microfluidizer (Avestin). Following clarification by centrifugation at 18,000g for 20 min and syringe filtration (0.2 m), soluble lysate was loaded over a 5-ml His-trap FF column (GE Healthcare) using an AKTA FPLC. Bound protein was washed and eluted over an imidazole gradient (20 to 150 mM). Fractions were assessed for purity by SDS-PAGE, pooled, concentrated, and frozen on dry ice in 25% glycerol for storage at 80C.

TPP1 cDNA was obtained from the SPARC BioCentre (The Hospital for Sick Children) and cloned into the piggyBac plasmid pB-T-PAF (J.M.R., University of Toronto) using Not I and Asc I restriction sites to generate two expression constructs (pB-T-PAF-ProteinA-TEV-LTM-TPP1 and pB-T-PAF-ProteinA-TEV-TPP1). Stably transformed expression cell lines (HEK293F) were then generated using the piggyBac transposon system, as described (14). Protein expression was induced with doxycycline, and secreted fusion protein was separated from expression media using immunoglobulin G (IgG) Sepharose 6 fast flow resin (GE Healthcare) in a 10-ml Poly-Prep chromatography column (Bio-Rad). Resin was washed with 50 column volumes of wash buffer [10 mM tris (pH 7.5) and 150 mM NaCl] and then incubated overnight at 4C with TEV (Tobacco Etch Virus) protease to release the recombinant enzyme from the Protein A tag. Purified protein was then concentrated and frozen on dry ice in 50% glycerol for storage at 80C.

Cellular intoxication by DT was measured using a nanoluciferase reporter strain of Vero cells (Vero NlucP), as described previously (8). Briefly, Vero NlucP cells were treated with a fixed dose of DT at EC99 (10 pM) and a serial dilution of LTM, LTM-mCherry, mCherry-LTM, DTK51E/E148K, LTM-TPP1, or rTPP1 and incubated overnight (17 hours) at 37C. Cell media was then replaced with a 1:1 mixture of fresh media and Nano-Glo luciferase reagent (Promega), and luminescence was measured using a SpectraMax M5e (Molecular Devices). Results were analyzed with GraphPad Prism 7.04.

SPR analysis was performed on a Biacore X100 system (GE Healthcare) using a CM5 sensor chip. Recombinant HBEGF was immobilized to the chip using standard amine coupling at a concentration of 25 g/ml in 10 mM sodium acetate (pH 6.0) with a final response of 1000 to 2500 resonance units (RU). LTM and LTM-TPP1 were diluted in running buffer [200 mM NaCl, 0.02% Tween 20, and 20 mM tris (pH 7.5)] at concentrations of 6.25 to 100 nM and injected in the multicycle analysis mode with a contact time of 180 s and a dissociation time of 600 s. The chip was regenerated between cycles with 10 mM glycine (pH 1.8). Experiments were performed in duplicate using two different chips. Binding data were analyzed with Biacore X100 Evaluation Software version 2.0.2, with apparent dissociation constants calculated using the 1:1 steady-state affinity model.

HeLa cells were incubated with LTM-mCherry (0.5 M), mCherry-LTM (0.5 M), or LTM-TPP1 (2 M) for 2 hours. Cells were washed with ice-cold phosphate-buffered saline (PBS), fixed with 4% paraformaldehyde, and permeabilized with 0.5% Triton X-100. mCherry constructs were visualized with a rabbit polyclonal antibody against mCherry (Abcam, ab16745) and anti-rabbit Alexa Fluor 568 (Thermo Fisher Scientific). LAMP1 was stained with a mouse primary antibody (DSHB 1D4B) and anti-mouse Alexa Fluor 488 (Thermo Fisher Scientific).

Colocalization was quantified using the Volocity (PerkinElmer) software package to measure Manders coefficients of mCherry signal with LAMP1 signal. The minimal threshold for the 488- and 568-nm channels was adjusted to correct the background signal. The same threshold for both channels was used for all the cells examined.

CLN2/ fibroblast 19494 were incubated with LTM-TPP1 (2 M) for 2 hours. Cells were washed with ice-cold PBS, fixed with 4% paraformaldehyde, and permeabilized with 0.5% Triton X-100. LTM-TPP1 was visualized with a mouse monoclonal against TPP1 (Abcam, ab54685) and anti-mouse Alexa Fluor 488 (Thermo Fisher Scientific). LAMP1 was stained with rabbit anti-LAMP1 and anti-rabbit Alexa Fluor 568 (Thermo Fisher Scientific).

TPP1 protease activity was measured using the synthetic substrate AAF-AMC using a protocol adapted from Vines and Warburton (38). Briefly, enzyme was preactivated in 25 l of activation buffer [50 mM NaOAc (pH 3.5) and 100 mM NaCl] for 1 hour at 37C. Assay buffer [50 mM NaOAc (pH 5.0) and 100 mM NaCl] and substrate (200 M AAF-AMC) were then added to a final volume of 100 l. Fluorescence (380 nm excitation/460 nm emission) arising from the release of AMC was monitored in real time using a SpectraMax M5e (Molecular Devices). TPP1 activity in cellulo was measured similarly, without the activation step. Cells in a 96-well plate were incubated with 25 l of 0.5% Triton X-100 in PBS, which was then transferred to a black 96-well plate containing 75 l of assay buffer with substrate in each well.

crRNA targeting the signal peptide sequence in exon 2 of CLN2 was designed using the Integrated DNA Technologies (www.idtdna.com) design tool. The gRNA:Cas9 ribonucleoprotein complex was assembled according to the manufacturers protocol (Integrated DNA Technologies) and reverse transfected using Lipofectamine RNAiMAX (Thermo Fisher Scientific) into HeLa Kyoto cells (40,000 cells in a 96-well plate). Following 48 hours of incubation, 5000 cells were seeded into a 10-cm dish. Clonal colonies were picked after 14 days and transferred to a 96-well plate. Clones were screened for successful CLN2 knockout by assaying TPP1 activity and confirmed by Sanger sequencing and Western blot against TPP1 antibody (Abcam, ab54385).

The pro-form of TPP1 was matured in vitro to the active form in 50 mM NaOAc (pH 3.5) and 100 mM NaCl for 1 to 30 min at 37C. The autoactivation reaction was halted by the addition of 2 Laemmli SDS sample buffer containing 10% 2-mercaptoethanol and boiled for 5 min. Pro and mature TPP1 were separated by SDS-PAGE and imaged on a ChemiDoc gel imaging system (Bio-Rad).

Proteins or cellular lysate were separated by 4 to 20% gradient SDS-PAGE before being transferred to a nitrocellulose membrane using the iBlot (Invitrogen) dry transfer system. Membranes were then blocked for 1 hour with a 5% milktris-buffered saline (TBS) solution and incubated overnight at room temperature with a 1:100 dilution of mouse monoclonal antibody against TPP1 (Abcam, ab54685) in 5% milk-TBS. Membranes were washed 3 5 min with 0.1% Tween 20 (Sigma-Aldrich) in TBS before a 1-hour incubation with a 1:5000 dilution of sheep anti-mouse IgG horseradish peroxidase secondary antibody (GE Healthcare) in 5% milk-TBS. Chemiluminescent signal was developed with Clarity Western ECL substrate (Bio-Rad) and visualized on a ChemiDoc gel imaging system (Bio-Rad).

rTTP1 and LTM-TPP1 were treated with EndoH (New England Biolabs) to remove N-glycan modifications. Enzymes were incubated at 1 mg/ml with 2500 U of EndoH for 48 hours at room temperature in 20 mM tris (pH 8.0) and 150 mM NaCl in a total reaction volume of 20 l. Cleavage of N-glycans was assessed by SDS-PAGE, and concentrations were normalized to native enzyme-specific activities.

Cryopreserved TPP1+/ embryos were obtained from P. Lobel at Rutgers University and rederived in a C57/BL6 background at The Centre for Phenogenomics in Toronto. Animal maintenance and all procedures were approved by The Centre for Phenogenomics Animal Care Committee and are in compliance with the CCAC (Canadian Council on Animal Care) guidelines and the OMAFRA (Ontario Ministry of Agriculture, Food, and Rural Affairs) Animals for Research Act.

TPP1/ mice (60 days old) were anesthetized with isoflurane (inhaled) and injected subcutaneously with sterile saline (1 ml) and meloxicam (2 mg/kg). Mice were secured to a stereotactic system, a small area of the head was shaved, and a single incision was made to expose the skull. A high-speed burr was used to drill a hole at stereotaxic coordinates: anteroposterior (A/P), 1.0 mm; mediolateral (M/L), 0.3 mm; and dorsoventral (D/V), 3.0 mm relative to the bregma, and a 33-gauge needle attached to a 10-l Hamilton syringe was used to perform the intracerebroventricular injection into the left ventricle. Animals received either 1 or 5 l of enzyme (5 g/l), injected at a constant rate. Isoflurane-anesthetized animals were euthanized by transcardial perfusion with PBS. Brains were harvested and frozen immediately, then thawed and homogenized in lysis buffer [500 mM NaCl, 0.5% Triton X-100, 0.1% SDS, and 50 mM Tris (pH 8.0)] using 5-mm stainless steel beads in TissueLyser II (Qiagen). In vitro TPP1 assay was performed, as described, minus the activation step.

Acknowledgments: We thank P. Lobel at Rutgers University for providing the TPP1-deficient mice. Funding: We are grateful to the Canadian Institutes of Health Research for funding. Author contributions: S.N.S.-M. devised and performed experiments and drafted the initial manuscript. G.L.B. provided materials and assisted in conceptualization and experimental design. X.Z., D.Z., and R.H. contributed to the experimental design and performed experiments. P.K.K. and B.A.M. contributed to the experimental design. J.M.R. contributed to the experimental design and revised the manuscript. R.A.M. assisted in conceptualization, contributed to the experimental design, and assisted in writing the manuscript. Competing interests: B.A.M. is a chief medical advisor at Taysha Gene Therapies. The authors declare that they have no other competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

Read more:
Exploiting the diphtheria toxin internalization receptor enhances delivery of proteins to lysosomes for enzyme replacement therapy - Science Advances

Haywards Heath woman’s bid to fund stem cell treatment to combat MS – Mid Sussex Times

Joceline Colvert was diagnosed with relapsing remitting Multiple Sclerosis in her early 20s and says she spent the first eight years researching and managing her condition while trying to mention it as little as possible to others and completing her Sound Production degree.

I spent most of my late 20s and early 30s finding ways to manage relapses, the symptoms of which have included whole body numbness, loss of the use of both hands, right eye blindness, vertigo and double vision, she said. Thankfully these symptoms did resolve however left scarring on my nerves. This results in reduced vision in my formerly blind eye and hands that dont function very well with repetitive tasks.

This semi-denial worked for me until about 2010 when I started to become a bit limpy which I did my best to hide. After a couple of memorable falls and fractures I decided to face up to being slightly rickety and got a hiking pole that I used occasionally in public. Since then Ive needed to get used to being visibly disabled, and switch between two hiking poles for very short distances and a wheelchair everywhere else.

Joceline, who lives with her husband and her five beloved cats and dogs, says she is not eligible for Haematopoietic Stem Cell Transplantation (HSCT), on the NHS which is the first treatment I have ever got excited about and believe could work. It could be truly life-changing.

As a result she is trying to raise money to fund the treatment herself.

HSCT is a procedure that aims to reset the faulty immune system which, in my case, is attacking my nervous system from within, Joceline said. Stem cells will be taken from my bone marrow or blood before my immune system is wiped out with chemotherapy. My cells are then reintroduced into my blood, where they grow a new immune system which will hopefully no longer attack my nerves or have any memory of MS.

The aim of HSCT is to completely halt progression, putting MS into remission with no requirement for immunosuppressant drug therapy. The success rate for relapsing remitting MS is 80% - 90% which is absolutely phenomenal compared to the limited available drug treatments, which only aim to slow down disability.

HSCT is available on the NHS, however there is a very strict criteria for which I do not qualify. The expense of the treatment and the increased pressures on the public purse mean the NHS will only treat patients who have been diagnosed for fewer than 15 years.

I have been diagnosed for 18 years.

I had prepared myself for this possibility and, for the last year, have been researching treatment with The National Pirogov Medical Centre Russia (Moscow). Russia has been pioneering in their use of HSCT to treat MS and are world renowned for their expertise and care. Im excited to have a treatment date in March 2021 which fills me with hope for a future free from progression. I need your help to get there.

Joceline, who loves making stop-motion animation puppets and props and playing musical instruments, says the treatment will cost 40,800, and the flights 800.

She has launched a Go Fund Me page at https://gf.me/u/y538k2 which has already seen donations of more than 26,000.

I am incredibly grateful for any help you can give towards enabling me to access this life-changing treatment, she said.

After almost two decades of managing MS flare-ups and their consequences, its hard to put into words just what a future without them would mean to me.

Thank you for reading this and for any help you can put towards this goal.

Read the original:
Haywards Heath woman's bid to fund stem cell treatment to combat MS - Mid Sussex Times

Family of sick girl whose stem cell donor pulled out at the last minute find a replacement – The Sun

THE campaigning family of a sick girl whose stem cell transplant match pulled out at the last minute have found a new one.

Evie Hodgson, eight, has a deadly blood disorder and was preparing for the operation when her donor cancelled without explanation.

4

4

Her family launched an appeal to find another, telling their story on ITVs This Morning, which saw 25,000 people sign up and found a second match.

Evies mum Tina, 37, said: This is the best Christmas present we could ever wish for.

Evie, of Whitby, North Yorks, was diagnosed with aplastic anaemia in May.

Medics were delighted when their global search for a bone marrow donor found a match only for disaster to strike.

RAF worker Tina, married to chief executive Andy, 49, said: We got the call about the new match while Evie was being treated for an infection.

4

4

Exclusive

Latest

"She said, Mummy, youre my hero. It was so emotional.

Evie, who has a brother, is to have the op at the Great North Childrens Hospital in Newcastle in January.

She said: Thank you so much to everyone. Youve saved my life.

GOT a story? RING The Sun on 0207 782 4104 or WHATSAPP on 07423720250 or EMAILexclusive@the-sun.co.uk

Go here to see the original:
Family of sick girl whose stem cell donor pulled out at the last minute find a replacement - The Sun

Integration of intra-sample contextual error modeling for improved detection of somatic mutations from deep sequencing – Science Advances

INTRODUCTION

The process of single-nucleotide variant (SNV) accumulation is an important universal element of cancer initiation and progression. While the genetic landscape of the most common malignancies has been broadly described (13), accurate identification of driver mutations in specimens with low cancer DNA purity continues to be of great importance yet presents substantial challenges. Hybrid-capturebased next-generation sequencing (NGS) is one of the most common techniques being used for circulating tumor DNA profiling (4, 5), detection of therapy-resistant clones (6, 7) and preleukemia (8, 9), and monitoring disease burden during therapy (10). Nevertheless, in all of these settings, the relevant genomic alterations typically exist at low relative abundance.

Several different methods have been developed in recent years to address the barrier of identifying the minute fraction of DNA molecules harboring an alteration against the high background of NGS-associated errors. Among the various methods, state-of-the-art techniques for error suppression typically can be categorized into two groups: (i) those that incorporate unique molecular identifiers (UMIs) to suppress library amplification errors by the assembly of consensus sequences (1113) and (ii) those that use probabilistic models to estimate background sequencing noise. The latter group can be further segregated into those that generate models that estimate error rates by the analysis of data from a single sample (i.e., single sample/tumor-only mode) (1416), data from a single control sample (1618), or data from multiple control samples (e.g., cohort of healthy controls) (1921). In the case of paired patients tumor and matched normal sample, Bayesian statistics models are commonly used to identify tumor-specific somatic variants that are distinguishable from the background and the germline variants detected within the matched normal sample (22, 23). Some techniques rely on a ploidy assumption to calculate genotype probabilities (24), while others have adapted statistical models to analyze allele frequencies directly (16), thus allowing the identification of rare subclones in existing, complex cancer genomes. Since a single control sample cannot fully account for the stochastic nature of NGS errors, other algorithms have been developed to generate site-specific error estimations using a larger cohort of controls (1921). This approach could be problematic as proper control samples are not always available. When control samples are completely lacking, stringent preprocessing steps can be applied to prioritize high-confidence mutations, for instance, thresholds on base quality scores, supporting read counts, and variant allele frequencies.

Despite advances enabled by the diverse approaches mentioned above, each is associated with inherent disadvantages that can lead to increased assay complexity, elevated sequencing costs, and/or suboptimal exchange between sensitivity and specificity (fig. S1, table S1, and Supplementary Note). To overcome these limitations, we characterized the contextual patterns of high-frequency errors observed during targeted hybrid-capture NGS in >1000 samples, divided across multiple technically diverse and clinically relevant human cohorts. On the basis of these patterns, we developed Espresso, a novel UMI-independent method that optimizes the suppression of artifacts from deep NGS for accurate SNV mutation calling.

To demonstrate the challenges associated with lowvariant allele fraction (VAF) mutation calling from hybrid-capturetargeted NGS, we interrogated multiple benchmarking datasets that differ by their library preparation techniques, captured genomic loci, number of samples, and sequencing depths (Fig. 1A, table S2, and Materials and Methods). Briefly, these datasets include the following: (i) CB: a human cord blood dataset; (ii) CL: a cell line dilution series using genomic DNA from the acute myeloid leukemia (AML) cell line MOLM13 and the colon cancer cell line SW48; (iii and iv) pre-AML1 and pre-AML2: peripheral blood DNA from two separate cohorts, each composed of pre-AML cases (that is, blood was drawn before clinical diagnosis of AML) and age- and sex-matched controls (9); and (v) AML-MRD: a cohort composed of peripheral blood DNA samples obtained from patients with AML during the course of treatment.

(A) Raw, SSCS, and duplex average sequencing depths across all the samples included in this study. Different colors represent different datasets, and these are consistent across all of the figure panels. (B) Sample-wide error abundance in the diverse NGS cohorts. The fraction of genomic positions being observed with at least one nonreference allele supporting read in each sample is indicated. Error burden is significantly different among the investigated datasets (Mann-Whitney test: P < 1.2 1053 for the indicated comparisons). (C) Inverse correlation between the abundance of genomic positions with nonreference allele and their corresponding allele frequencies is demonstrated (Spearmans rank order correlation: r = 0.95; *P < 2.2 1016). Each dot represents a single sample. (D) Panel-wide error abundance in the diverse NGS cohorts as determined by the inclusion of positions with a minimum of one nonreference supporting read in at least one sample. NA, not applicable.

Three different target panels were used to sequence these cohorts, resulting in 83,000 to 1.2 million interrogated bases (table S2). Investigating these genomic loci revealed that the percentage of positions with nonreference alleles per sample varied widely among the different datasets and, in some cases, among samples within a particular dataset (Fig. 1B). Samples with a lower percentage of positions with nonreference alleles displayed higher average error rates (Fig. 1C). Furthermore, almost all genomic positions sequenced harbored a nonreference allele in at least one sample in each dataset (Fig. 1D). Overall, these observations reveal the magnitude of the challenge presented by potential false-positive variants produced by hybrid-capture NGS. Since such a large number of technical artifacts may mask clinically relevant variants, we conducted an unbiased exploration of multiple strategies aiming to specifically suppress NGS errors while maintaining high sensitivity in identifying real mutations.

To evaluate the contextual dependencies of errors in the datasets described above, we investigated how error rates differ with respect to the substitution type, and its 5 and 3 one-base flanking genomic sequence. We found that error rates are highly heterogeneous across the 192 distinct trinucleotide sequence contexts (Fig. 2A, top, and fig. S2) and are highly variable between samples within the same experimental cohort (Fig. 2A, bottom). High error rates were frequently observed at C>A and C>T substitutions (Fig. 2, A and B, and fig. S2). C>T error rates were particularly high when they occurred at a CpG context (Fig. 2, A and C, and fig. S2). Initiated by spontaneous deamination of 5-methylcytosine, real mutations in this context accumulate during aging (25), are frequent in germline cells (Supplementary Note), and are also highly prevalent in cancer genomes (26), emphasizing the importance of evaluating error rates in relation to their associated genomic contexts.

(A) Nonreference average error rates at the 192 distinct trinucleotide contexts are shown using the AML-MRD dataset. Vertical lines in each box represent individual samples. Samples order is kept among distinct contexts. Arrows represent a group of samples with high error rates across multiple contexts. The bottom panels exemplified variation among contextual error rates (*Wilcoxon signed-rank test: P < 1.8 1017) and samples (Mann-Whitney test, samples with the highest and lowest error rates. C[G>T]C: P < 7.7 1041, T[A>C]C: P < 3.6 106). (B). C>T and C>A substitutions are more frequent (Wilcoxon signed-rank test, P < 1.4 10252 for all the comparisons with the other substitution types). (C) High error rates at CpG sites (Wilcoxon signed-rank test, P < 1.1 1064 for all comparisons). (D) Error rates vary between error contexts and their reciprocals (Wilcoxon rank sum test, P < 0.05; #significance was not reached). (E) Average sequencing depths. Arrows represent a group of samples with low sequencing depths across multiple contexts. (F) Reduced sequencing depth at contexts that include reference cytosine and an increasing number of guanine (Pearson correlation: r = 0.35; P = 2.3 10264) and at contexts that include reference guanine with an increasing number of cytosine (r = 0.29; P = 8.6 10179). (G) Low sequencing depth at contexts with C>G or G>C base substitutions (Wilcoxon signed-rank test: P = 1.7 10217). (H) Inverse correlation between depth and error rates (black dashed line, log-log scaled Pearson correlation: r = 0.27; P = 9.7 10308). Correlation strengths differ among different error contexts (colored dashed lines). (I) The number of nonreference supporting reads at the 192 distinct trinucleotide contexts is shown. The samples order is identical across (A), (E), and (I).

While contextual error patterns were generally similar between their complementary counterparts, they did not always mirror each other perfectly within any particular sample (fig. S3A). Small yet statistically significant asymmetric error rates were consistently observed among the majority of error contexts in each of the cohorts (Fig. 2D and fig. S3B). For instance, we measured asymmetric error rates involving G>T/C>A, in line with prior observations (27). Error rate asymmetries were markedly directional and consistently elevated in specific contexts as compared with their matched reciprocals in all of the investigated datasets. As an example, each of the 16 trinucleotide contexts containing A>T substitutions demonstrated elevated error rates as compared with their corresponding reciprocal contexts containing T>A substitutions. Together, these results indicate that 192, rather than 96, contextual error types would need to be considered to accurately model error rates.

Next, we investigated how sequencing depth may influence error frequencies. As with error rates, sequencing depth differed between distinct contextual error types (Fig. 2E). We noticed a marked inverse correlation between sequencing depth and guanine or cytosine content within specific trinucleotide contexts, a possible reflection of the systemic under-coverage in GC-rich regions reported in NGS (Fig. 2F) (28, 29). Sequencing depth was also lower within trinucleotide contexts that included C>G and G>C substitutions as compared with those that included nucleotide substitutions that reduce GC content (Fig. 2G). These data illustrate how sequencing depth can be influenced by both the trinucleotide context and the nonreference allele.

Overall, a modest, statistically significant inverse correlation was observed between sequencing depth and error rates (Fig. 2H). Correlation strengths were not equal among distinct contextual error types. Further supporting this trend, individual samples with lower average sequencing depth displayed high error rates in multiple contextual error types (see arrows in Fig. 2, A and E). In contrast to the error rates, the absolute number of nonreference supporting reads at the distinct contextual error types showed reduced inter-sample differences in those samples; however, the differences between distinct contextual errors were preserved (Fig. 2I). Collectively, the results obtained here suggest that integration of intra-sample contextual error modeling of nonreference supporting reads at each of the 192 contexts may be a promising strategy for accurate suppression of errors produced by hybrid-capture NGS.

As described above, errors varied across samples yet were highly stereotypical according to sequence context and sequencing depth. We reasoned that intra-sample contextual error patterns could be leveraged for in silico error suppression. Such an approach could have several inherent advantages over existing error suppression methods that rely on UMIs, apply thresholds based on intra-samplewide error rates, or use control samples to train error rate models. Therefore, we devised a computational approach, called Espresso, to model within a sample of interest the nonreference allele counts at each of the 192 distinct contextual error types. Espresso incorporates three distinct features that make it robust to different sequencing datasets (Supplementary Note): (i) pragmatic pre-filters that prepare the dataset for error modeling (fig. S4), (ii) automatic selection of the most appropriate probabilistic distribution for error modeling at a particular contextual error type (fig. S5), and (iii) utilization of nonreference supporting reads as opposed to VAF for error modeling (fig. S6). Unlike applying fixed and arbitrary cutoffs (e.g., minimum VAF, coverage, and number of supporting reads), nonreference alleles would not be indiscriminately eliminated by such an approach; rather, mutations would only be called if they reached statistical significance when compared to their corresponding error distributions (Fig. 3, A to E, and Materials and Methods).

Flowchart illustrating the error modeling technique that is implemented by Espresso. (A) Following the summarization of the sequencing data to include the dominant alleles at each investigated genomic position, their corresponding read counts, and the average mapping read qualities in each sample of interest, a set of filters is being applied, aiming to deplete potential somatic SNVs and common polymorphism from being included in the error models. (B) On the basis of the distribution of the nonreference supporting reads in the enriched error list, Espresso selects between either the exponential or the Weibull probabilistic approaches. (C) The nonreference supporting read (SR) counts in each sample are being grouped based on the genomic sequence context to generate 192 context-specific distribution models. (D) The models are being reapplied to the entire samples data for outlier identification. True positives are being determined if they reach statistical significance when compared to their corresponding error distribution. (E) The cumulative distribution function graph displays the empirical data (black dots) and the theoretical data (blue line) generated by the 192 models in all the samples included in the CB dataset (top, exponential models) and the AML-MRD dataset (bottom, Weibull models). (F) Panel-wide error rates defined as the number of nonreference alleles supporting reads following error suppression, divided by all the reads from the same category (i.e., raw, SSCS, and duplex reads) across the entire 1,264,830-bp panel and (G) percentage of error-free positions in the 10 cord blood samples are illustrated. For error suppression, a cutoff P value 0.05 (Bonferroni-adjusted) was used. SSCS and duplex cutoffs are 1 nonreference supporting read unless indicated otherwise. * indicates Wilcoxon signed-rank test: P < 0.002.

To evaluate the performance of Espresso, we first applied it to the CB dataset. We reasoned that CB would have a minimal burden of somatic mutations, allowing for a more precise estimation of true error rates. We also tested in parallel other common error suppression techniques for unbiased comparative performance assessment (Materials and Methods). The techniques selected for comparison were representative of the spectrum of previously published tools. Specifically, we used two UMI-based methods, namely, single-strand consensus sequences (SSCSs) and duplex sequences (12), and two statistical methods for error correction that model background error distributions differently. Among the two statistical methods used, one relies on a training cohort to estimate error rates at the allele level (termed AL here) (20), and the other estimates error rates at the sample level (termed SL here) (14) without consideration for distinct sequence contexts.

Panel-wide error rates were highly similar among the 10 CB samples but varied significantly among the different error suppression methods (Fig. 3F). As compared with the various statistical approaches (i.e., SL, AL, and Espresso), the UMI-based methods demonstrated inferior error suppression capabilities. A minimum of nine nonreference supporting SSCS reads or three nonreference supporting duplex reads were required to achieve panel-wide error rates comparable to that of SL and Espresso in the CB dataset. We observed similar relative performance among the methods to maximize the number of error-free positions across the entire target panel (Fig. 3G). Considering the highest panel-wide error rate obtained by Espresso (2.74 106) and the lowest of the panel-wide error rate observed without error suppression (0.025) across the CB samples, Espresso achieved an error rate reduction of more than 9000-fold.

To evaluate the sensitivity and specificity exchange delivered by Espresso, we analyzed the sequencing data from the CL dataset, which consisted of a dilution series using two cancer cell lines, MOLM13 and SW48. For sensitivity measurements, we assessed the ability of the different methods to detect 119 MOLM13-specific germline variants at the different dilutions (table S3). To evaluate specificity, we assessed the miscalling of 186 AML-related somatic hotspot mutations that are covered by the target panel but are absent from both cell lines (table S3). Espresso outperformed all the other methods in distinguishing between true and false variants (Fig. 4A). In contrast, duplex sequencing achieved the smallest area under the receiver operator curve (AUC), highlighting the low diagnostic accuracy of this method and, consequently, its limited clinical utility in detecting variants across large hybrid-capture panels.

(A) Espresso demonstrates improved sensitivity versus specificity and (B) preferable precision-recall trade-offs as compared with the various indicated methods. The ability of each method to differentiate between 119 positive alleles and 186 negative control variants in a set of serially diluted cell line DNA samples was tested. (C and D) No substantial benefit of using UMIs to augment Espressos performance could be determined. Sensitivities and specificities were measured at all the possible combinations of the unique P values outputted by Espresso and the unique numbers of SSCS or duplex nonreference supporting reads that were observed in the dataset. The maximum sensitivities at each calculated value of specificity are illustrated. (E to H) Sensitivity versus specificity trade-offs derived by the reduced and extended contextual error modeling approaches are illustrated in comparison with Espresso. Ninety-five percent confidence intervals (shaded colors) and average values were derived by three random subsets of the data for each one of the indicated in silico decreased panel sizes. (I) Heatmap illustrating the percentage of contextual models that can be generated by Espresso when data are being restricted by either panel size reduction or sequencing depth reduction, or both. Data removal was controlled for both the reference and nonreference supporting reads, thus keeping the variant allele frequencies of the nonreference alleles similar to those in the original samples. The red line illustrates such combinations, of which 90% or more of the distinct contextual models could have been generated in every sample in the CL dataset. With datasets that fall below this line, the 12-model contextual error modeling approach can be used in addition to Espresso.

The use of hybrid-capture NGS panels allows for the detection of mutations at thousands of genomic positions. However, their use also creates unique challenges for true variant identification across so many bases. In addition to high sensitivity and specificity, positive predictive value (PPV) must be prioritized to maximize utility. We assessed PPV in conjunction with sensitivity (i.e., precision-recall analysis). We focused on variants with expected VAF 0.2%, since accurate variant detection below this threshold is clinically important yet has proven to be a great challenge for existing hybrid-capture NGS platforms (5, 30). Espresso provided a sensitivity of 19.9%, thus achieving the highest number of true-positive, low-VAF alleles at 100% PPV among the tested methods (Fig. 4B). This corresponds to a 6.8-fold improvement as compared to AL, which was the next best-performing method to detect low-VAF alleles without scarifying PPV. Notably, SL performed far worse in this analysis than the other methods due to a high number of false-positive calls across various sensitivity thresholds. This result highlights the limited power of noncontextual, sample-level error modeling in detecting mutations with very low read support despite its ability to achieve an extremely high level of error suppression (Fig. 3, F and G). Further supporting this, we compared the false-positive and true-positive calls obtained by Espresso with that of Mutect2 (16) at tumor-only mode. Once more, Espresso demonstrated superior results (table S4).

Previously, the suppression of errors through statistical error modeling was shown to be enhanced by combination with UMI-based approaches (20). However, integrating UMI information with Espresso did not confer significant performance improvements (Fig. 4, C and D), suggesting that accurate detection of low-frequency variants can be achieved with Espresso alone. Collectively, the comparative analysis using the CL dataset indicates that the bioinformatic strategy applied here outperformed other methods in the reliable distinction of low-frequency errors from real SNVs.

To characterize pragmatic constraints of our method, we compared Espresso with alternative sequence context-based error models. Specifically, we included (i) a simplified 12-model design that accounts only for the 12 possible distinct substitution types without consideration of flanking bases and (ii) an expanded 3072-model design that accounts for the substitution type and for two additional 3 and 5 flanking bases. We evaluated the impact of panel size (i.e., number of interrogated bases) and sequencing depth on the performance of Espresso and the alternative sequence context-based models using the CL dataset.

This comparative analysis exposed critical factors affecting the performance of the alternative models. On the one hand, the performance of the 3072-model approach suffered with reduced panel size (Fig. 4, E to H, and fig. S7A). This is an expected observation that is attributed to the reduction in the number of nonreference alleles being used to populate a relatively high number of models, thus resulting in either model generation failure or an inadequate estimation of the background error noise. In contrast, performance of the 12-model approach was less dependent on panel size since the relatively small number of models was easily populated with nonreference alleles (Fig. 4, E to H, and fig. S7B); however, Espresso consistently outperformed the 12-model approach, presumably because the 12 models were insufficient to account for errors arising within distinct sequence contexts. Moreover, the 12-model approach performed poorly on the largest panel size, possibly as a result of model overfitting from high-VAF errors that escape the initial filtering steps (Materials and Methods). The performance of Espresso was relatively consistent across a broad range of panel sizes from ~1 Mb down to ~50 kb (Fig. 4, E to H, and fig. S7C).

Next, we serially downsampled the CL dataset to simulate various practical scenarios of panel sizes (1 Mb to 32.5 kb) and sequencing depths (4500 to 1000). At each simulated panel-depth combination, we determined the percentage of trinucleotide contexts that could be modeled directly by Espresso (Fig. 4I). Notably, low represented nonreference alleles that cannot be modeled directly by Espresso would still be analyzed automatically by alternative techniques that are included in the software package (see Data and materials availability). Overall, these results illustrate the performance dependencies of Espresso and related sequence contextbased models to assist with their implementation in a wide range of sequencing settings.

Having demonstrated Espressos high analytical performance in the CB and CL datasets, we next sought to evaluate its clinical utility. The presence of persistent AML clones that carry genetic abnormalities during or after treatment has been shown to carry crucial prognostic information (31). Therefore, we assembled a cohort of 42 patients with AML (AML-MRD; table S5) whose mutations were previously determined at diagnosis (table S3). Forty of the 42 patients had serial samples analyzed by ultra-deep hybrid-capture NGS at two time points during therapy; for the other two patients, single follow-up samples were available.

Since minimal/measurable residual disease (MRD) monitoring may guide clinical decisions (3234), in addition to true positives, both false positives and false negatives could have tremendous implications for patient care. We therefore evaluated F1 scores, which represent the harmonic mean of PPV and sensitivity. For comparative performance evaluation, mutations reported at diagnosis were considered as true positives if they were detected in the follow-up samples of the same patient or as false positives if they were detected in other patients. We first applied a cutoff of 0.05 (Bonferroni-adjusted) for the probabilistic methods SL, AL, and Espresso and a heuristic threshold of 1 nonreference supporting reads for the UMI-based methods SSCS and duplex. Tested on the subset of samples obtained at either the first time point (T1, closer to diagnosis) or the second time point (T2, further into treatment), Espresso delivered the highest F1 scores (0.71 at T1 and 0.74 at T2) followed by AL and duplex (Fig. 5A). We next applied the optimized SSCS and duplex cutoffs used in the CB analysis (i.e., 9 and 3 nonreference supporting reads, respectively). Although F1 scores improved with these parameters, they still fell short due to an increased number of false positives for SSCS 9 and an increased number of false negatives for duplex 3 in both the T1 and the T2 data subsets as compared with Espresso (Fig. 5B).

(A) Espresso provides a preferred balance between precision (PPV) and recall (sensitivity), as determined by the inspection of 78 SNVs reported across 35 of 42 patients at the time of AML diagnosis. Mutations were called in the patients sample at 21 different iterations. In each iteration, 6 random patients of the 42 were excluded. Median F1 scores and 1 SD are shown for the various methods tested at two time points during the course of treatment (T1 and T2, Wilcoxon signed-rank test: P 6.4 105 for all the comparisons with Espresso). (B) The variation in the mutations being called by Espresso ( 0.05, Bonferroni-adjusted), SSCS (9 nonreference supporting reads), and duplex (3 nonreference supporting reads) is illustrated. Red color indicates called mutations, while blue color indicates that mutations were not detected. FP, false positives; FN, false negatives. (C) Sensitivity versus specificity as determined by the different tested methods. (D) Enrichment of clones, carriers of TP53, and DNMT3A mutations is observed in patients with AML following therapy. The y axis represents the number of mutations detected, classified by the affected genes.

Despite the technical differences between the CL and AML-MRD datasets, Espresso once again produced the most preferred balance between sensitivity and specificity (Fig. 5C). We compared Espresso with additional algorithms and saw consistent outcomes. Espresso outperformed Mutect2 (16) in both the tumor-only mode and the panel of normals mode when samples obtained from 14 healthy adults were used (table S4). Espresso also outperformed deepSNV (18), a statistical algorithm that was developed specifically for the accurate detection of SNVs from deep targeted sequencing experiments. The comparison with deepSNV extrapolates beyond the probabilistic approaches being used and illustrates the benefits of other features implemented in our bioinformatic pipeline for the reduction of false-positive calls (fig. S8).

Having established Espresso as the preferred methodology to maximize the accuracy of SNV detection from peripheral blood, we next sought to implement it for the characterization of clonal dynamics in patients with AML. Since the competitive balance among different hematopoietic clones is likely to change during multiple rounds of chemotherapy, we hypothesized that Espresso would enable the identification of resistant clones that were not reported at diagnosis. We therefore extended our analysis to include an additional 147 highly recurrent AML SNVs that are covered by the AML-MRD hybrid-capture panel (table S3). Across all the samples, Espresso identified 92 mutations ( 0.05, Bonferroni-adjusted) with the lowest being reported at VAF = 0.0135% (table S6 and fig. S9). These correspond to 59 distinct mutations, out of which 47 (~80%) were present in at least two samples of the same patient (that is, reported at diagnosis and detected in at least one additional time point by Espresso or detected in the two follow-up samples by Espresso). Such a high percentage of validated mutations is an indicator of Espressos reliable mutation calling. Among these, Espresso has enabled the detection of 22 new putative driver SNVs not reported at diagnosis in 15 patients, including in 3 of the 7 patients (~43%) with no SNVs in the diagnostic report (table S6). Further supporting the validity of the mutations called by Espresso, most of these newly identified mutations were in genes that commonly contribute to positive clonal selection following cytotoxic chemotherapy (3537), including TP53 and DNMT3A (Fig. 5D).

Together, our results demonstrate substantial advantages of Espresso over other methods for SNV detection from peripheral blood of patients with AML during the course of therapy. Encouraged by a recent consensus document release from the European LeukemiaNet MRD Working Party (38), many studies are now underway to evaluate the prognostic and predictive significance of clonal dynamics in AML and the proposed role of MRD detection as a surrogate endpoint for clinical trials (39). Implementation of Espresso in these contexts has the potential for significant clinical utility.

Age-related clonal hematopoiesis (ARCH) is a common phenomenon evident by the presence of somatic mutations in hematopoietic stem cells of otherwise healthy individuals that cause a clonal expansion of the stem cells and their progeny (40). Recently, our group reported several hundred ARCH-associated mutations spread across 27 genes with various contributions to the risk of AML transformation (9). Our study provided a proof of concept for risk prediction of AML. Nevertheless, large population screens using broad sequencing panels remain socioeconomically unattractive because of high costs, the relatively low incidence of AML, and the relatively high incidence of ARCH in the general population.

To address these challenges, we reasoned that interrogating a small number of highly recurrent AML mutations would be a more tractable approach than broad hybrid-capture sequencing. This approach could theoretically result in improved segregation between pre-AML and controls while reducing sequencing costs. The success of this approach relies on the accurate identification of preleukemic mutations in asymptomatic individuals.

We first compiled datasets that would allow comparisons among the distinct methods used in our previous analyses. For this reason, we focused initially on the pre-AML1 dataset, which contains UMIs in the sequencing reads, and the CB dataset, which could be used as a training set for error rate estimation at the AL. Putative driver SNVs (that is, mutations in coding sequences other than synonymous SNVs and mutations at splice sites) identified by each method at the recurrently mutated genomic loci were used to derive random forest classifiers that were trained and tested on their corresponding methods mutation calls (table S7). For the probabilistic methods, 0.05 (Bonferroni-adjusted) was used, and for the UMI-dependent methods, we applied either a threshold of one supporting consensus read or SSCS 9 and duplex 3. The Espresso-derived classifier exhibited the highest level of performance for discriminating pre-AML from controls (AUC: 0.74) and reported the highest sensitivity (46.8%) at 100% specificity (Fig. 6A). A reduction in specificity down to 96.3 or 93.7% was needed to achieve the same sensitivity with the SL-derived and SSCS-derived classifiers, respectively. The SSCS-derived model also underperformed the Espresso-derived classifier when the SSCS 1 cutoff was applied (AUC: 0.66, Fig. 6A, dashed line). The duplex 3 derived classifier had the poorest performance (AUC: 0.42), owing to poor duplex consensus efficiency (fig. S1B), low duplex coverage (Fig. 1A), and subsequent dropout of mutations not meeting the required cutoff. On the contrary, with a threshold of one supporting duplex read, a large number of putatively false-positive SNVs were called, resulting in poor classification accuracy (AUC: 0.65, Fig. 6A, dashed line). The AL-derived classifier also performed poorly due to a high number of false-positive SNVs (AUC: 0.62).

Classification performance evaluation of pre-AML and control, mutated samples. (A) Each classifier was trained and tested on the mutations that were obtained from the classifiers corresponding method. (B) Comparison between the Espresso and the SL-derived classifiers. In this iteration, each classifier was trained using its corresponding methods mutation calls and was tested in its accuracy to classify pre-AML cases and controls, including mutated samples identified by the other method as well. (C) Comparative performance validation between the Espresso and the SL-derived classifiers to differentiate between pre-AML and control samples obtained from an additional validation dataset (8). Information regarding the study participants age, specific mutations, and their VAFs was obtained directly from the main text. (D) Performance estimation using the validation dataset and simulated controls. (E) Precision-recall trade-offs are calculated at the individual level (that is, serial samples are accounted for single individuals and individuals without any mutations are also included in the performance measurements). The red dot indicates AMLs incidence rate. This is equivalent to a situation where no screen is being conducted at all [PPV = incidence rate = 0.006% (44), SN = 100%]. The green dot indicates the model performance using an additional published dataset consisting of 11,262 individuals when the model was set to achieve 100% specificity in the training set. Horizontal color bars represent PPV ranges determined for screening mammography for breast cancer (54) and fecal immunochemical test for advanced adenomas and colorectal cancer (CRC) (55). Comparison with the genetic risk model performance shows the extent to which sensitivity must be compromised to achieve PPV comparable with these widely applied early detection tests.

There is a low cumulative risk of ARCH progression to hematologic neoplasms (41). For this reason, the implementation of a population-based pre-AML genomic screening test would need to achieve exceedingly high specificity and low false-positive rate. We therefore prioritized the Espresso- and SL-derived classifiers for subsequent performance evaluation. Additional mutations that were found by Espresso and SL in the pre-AML2 dataset were included in the analysis (table S7). Each classifier was trained on the mutations found by its corresponding method in both the datasets (pre-AML1 and pre-AML2) and tested on the data that include all the mutations detected by either of the two methods. The Espresso-derived classifier once more provided a better overall sensitivity-specificity balance and a greater sensitivity at 100% specificity (Fig. 6B). Similar trends were observed when both the classifiers were applied to an external validation set consisting of mutations called in 188 pre-AMLs and 181 controls (8), with the Espresso-derived classifier again displaying higher discriminatory accuracy (Fig. 6C). Together, the superior classifier performance using mutations called by Espresso illustrates that accurate mutation calling is imperative when designing genetic risk prediction models.

To estimate how well the winning classifier would perform as a population-wide screening test, we spiked the validation set into >4 million in silico simulated controls (prevalence ~0.005%; Materials and Methods). Despite the small genomic footprint (table S8), the Espresso-derived classifier resulted in accurate identification of the mutated pre-AML samples (AUC: 0.84; Fig. 6D). As an example, when the model was tuned to minimize false-positive calls based on the pre-AML1/pre-AML2 merged training dataset, a sensitivity of 29.3% and a specificity of 99.8% were obtained. Precision-recall analysis revealed the extent to which the Espresso-derived classifier may enrich for individuals at high risk of developing AML as compared with current practice (no screening, i.e., AML incidence rate) (Fig. 6E). Sensitivity was 4.8% at 100% PPV; this small subset detected with no false positives was enriched for highly penetrant SRSF2/IDH2 double-positive individuals with the highest risk for AML development (table S9). Last, we estimated the model performance in an additional published cohort of 11,262 individuals (42). In this cohort, when the model was tuned to minimize false positives within the training dataset, a sensitivity of 14.3% and a PPV of 4.8% were obtained (Fig. 6E and table S9).

In this study, we described the rationale, technical performance characteristics, and potential clinical utility for Espresso, a novel method to improve hybrid-capture sequencingbased SNV detection. Unlike many other NGS error suppression methods, including the representative published UMI-based and probabilistic modelbased approaches tested here, Espresso does not rely on UMIs or a training set of controls for error rate estimations; therefore, Espresso improves practicality by reducing library preparation complexity, assay costs, and analysis time. We observed additional notable advantages of Espresso over alternative methods, and these were consistent across diverse datasets. Specifically, Espresso produced superior error suppression and an improved trade-off between sensitivity and specificity for detection of low-VAF alleles.

These advantages of Espresso were the result of several key features. First, Espresso applies a set of pre-filters to prepare the data for error modeling. Second, Espresso automatically selects between two statistical models to estimate the number of alternative supporting reads rather than the VAFs; thus, in addition to selecting the more appropriate error distribution model, it better accounts for error rate bias resulting from variation in sequencing depth within hybrid-capture NGS datasets. Third, Espresso markedly reduces false-positive calls by considering only the dominant nonreference allele at each interrogated genomic position. Fourth, Espresso leverages a large number of errors that share the same trinucleotide sequence context within the investigated sample; thus, it reduces the potential for misrepresentation of real error rates by relatively small control cohorts.

To explore its potential use in clinical settings, we tested the performance of Espresso to detect SNVs in serial peripheral blood samples from 42 patients with AML who achieved clinical remission. Consistent with the performance in the other investigated datasets, Espresso outperformed all the other tested methods in this setting. Using Espresso, we found resistant subclones enriched for TP53 and DNMT3A mutations that were genetically distinct from the AML clones present at diagnosis. In the future, more extensive cohort studies are needed to determine whether the selection and enrichment of such clones following induction therapy may affect patient outcomes in a nonautonomous fashion, similar to the observations in solid malignancies (43). Furthermore, combining accurate detection of persistent mutations together with other independent prognostic markers will be necessary to build clinically relevant models for accurate determination of the risk of relapse.

Our results emphasize the importance of accurate mutation detection for the derivation of classification models in the setting of early detection of AML. Using Espresso, we derived a risk prediction model that is focused on a minimal yet highly informative set of genomic loci that are recurrently mutated in patients with AML. With only 1594 genomic bases being interrogated, our results imply that up to 29.3% of de novo AML cases can be predicted years in advance with a specificity of 99.8%. Although sensitivity may greatly suffer with elevated PPV, considering the incidence rates of AML in the general population (~6:100,000) (44), our approach would still provide meaningful patient enrichment. Modest sensitivity may be acceptable when screening the general population as long as specificity and PPV remain high. Further prospective validation studies are required to assess the feasibility, utility, and cost-effectiveness of this targeted approach. Our findings should also be extended to incorporate additional predictive biomarkers. As AML is a blood-borne disease, we envision that epigenetic and metabolomic perturbations within leukocytes may further improve prediction accuracy, thus making AML predictions more clinically useful. Our results indicate that certain biomarker-enriched populations may be at an exceedingly high risk of developing AML. In time, novel therapeutic developments and targeted therapies against blood cells with high-risk mutations may provide the minimal side effects necessary to deliver a favorable risk-benefit ratio that justifies the initiation of early intervention clinical studies.

In summary, we have described, benchmarked, and validated a new practical NGS error suppression technique. We have demonstrated the superiority of Espresso in detecting somatic SNVs as compared with existing state-of-the-art approaches and defined its limitations with respect to sequencing depths and hybrid-capture panel sizes. We used Espresso to derive new biological insights, augmenting our understanding of the genetic mutations that define high-risk malignant transformation and therapy resistance clones in patients with AML. We envision that Espresso will prove useful in guiding clinical decisions and scientific research alike.

CB dataset: This dataset is composed of 10 human umbilical cord blood genomic DNA samples obtained from Trillium Hospital (Mississauga, Ontario, Canada) with informed consent in accordance with guidelines approved by the University Health Network Research Ethics Board. Cord blood was processed 24 to 48 hours after delivery. Mononuclear cells were enriched using Ficoll-Paque followed by red blood lysis by ammonium chloride and CD34+ selection before DNA extraction. CL dataset: MOLM13 cell line DNA was mixed with SW48 cell line DNA at relative concentrations of 100, 5, 1, 0.2, 0.04, and 0% and was sequenced in duplicate. Pre-AML1 and pre-AML2 datasets: Detailed information regarding these cohorts is described elsewhere (9). Briefly, the pre-AML1 dataset contains peripheral blood genomic DNA samples obtained from a total of 509 individuals upon enrollment into the European Prospective Investigation into Cancer and Nutrition (EPIC) study (45) between 1993 and 1998. Together, 414 control individuals who did not develop any hematological disorders during the extended follow-up period and 95 individuals who developed AML were included in this study. The pre-AML2 dataset contains peripheral blood genomic DNA samples obtained from individuals enrolled in the EPIC-Norfolk longitudinal cohort study between 1994 and 2010. Samples were available from 37 patients with AML and 262 age- and sex-matched controls without a history of cancer or any hematological conditions. Samples taken at multiple time points were available for a fraction of the participants in this cohort. Notably, samples from eight pre-AML patients in the pre-AML2 cohort were separately sequenced in the pre-AML1 dataset (by independent investigators using a different methodology). To avoid statistical misrepresentation of AML predictions, we removed those samples from the pre-AML2 dataset before the derivation of the described genetic risk models. AML-MRD dataset: This dataset is composed of peripheral blood genomic DNA from 42 patients with AML treated at the Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada. All 42 patients achieved morphologic leukemia-free state (MLFS) on chemotherapy. Complete count recovery occurred when absolute neutrophil count recovered to 1 109/liter and platelet count recovered to 100 109/liter up to 7 days following the bone marrow assessment that confirmed MLFS status. All patients were deidentified with patient IDs. Their demographic and clinical features were captured (table S5). All the samples in this study, including healthy individuals and patients with cancer, were collected with informed consent for research use and were approved by Institutional Review Boards in accordance with the Declaration of Helsinki. Protocols were approved by the following ethics committees: (i) International Agency for Research on Cancer Ethics Committee approval #14-31, (ii) East of EnglandCambridgeshire and Hertfordshire Research Ethics Committee reference number 98CN01, and (iii) University Health Network Research Ethics Board # 01-0573.24.

Library construction and sequencing were done as previously described (9). Briefly, for each sample in the CB, CL, and pre-AML1 datasets, 100 ng of genomic DNA was sheared to 250base pair (bp) fragments before library construction (KAPA HyperPrep Kit KK8504, Kapa Biosystems) with a Covaris E220 instrument using the recommended settings. After end repair and A-tailing, ligation of UMI-containing adaptors was performed with 100-fold molar excess. Agencourt AMPure XP beads (Beckman Coulter) were used for library cleanup following eight cycles of fragment amplification with 0.5 M Illumina universal and indexing primers. Targeted hybrid-capture was carried out on pools of three indexed libraries. Five microliters of Cot-I DNA (1 mg ml1; Invitrogen) and 1 nmol each of xGen Universal Blocking Oligo, TS-p5, and xGen Universal Blocking Oligo, TS-p7 (8 nucleotides) were added to each pool of adaptor-ligated DNA. The mixture was dried using a SpeedVac and then was resuspended in 1.1 l of water, 3.4 l of NimbleGen hybridization component A, and 8.5 l of NimbleGen 2 hybridization buffer. The mixture was heat-denatured at 95C for 10 min following the addition of 4 l of xGen Lockdown Probes (3 pmol; xGen AML Cancer Panel v.1.0). Hybridization was conducted at 47C for 72 hours. Washing and recovery of the captured DNA were initiated with 100 l of clean streptavidin beads that were added to each capture. Following separation of the libraries and the supernatant using a magnet, 200 l of 1 Stringent Wash Buffer was added, and the reaction was incubated for 5 min at 65C. The supernatant containing unbound DNA was removed before repeating the high stringency wash for the second time. The bound DNA was then washed one time with 200 l of each of the following: 1 Wash Buffer, 1 Wash Buffer II, and 1 Wash Buffer III. The washed DNA on beads was resuspended in 40 l of nuclease-free water, and this volume was divided into two polymerase chain reaction (PCR) tubes that were subjected to 10 cycles of post-capture amplification (Kapa Biosystems, recommended conditions). Libraries were spiked with 2% PhiX before sequencing. The procedure used for the pre-AML2 dataset is described elsewhere (referred to as the validation cohort) (9). For each sample in the AML-MRD dataset, peripheral blood samples were collected during remission in PAXgene Blood DNA Tubes (PreAnalytiX, Hombrechtikon, Switzerland). DNA was extracted according to the manufacturers instructions. Illumina-compatible libraries were constructed from 100 ng of sheared genomic DNA using the Covaris M220 sonicator (Covaris, Woburn, MA, USA) and the KAPA HyperPrep Kit (#KK8504, Kapa Biosystems, Wilmington, MA, USA). Following end repair and A-tailing, adapter ligation was performed for 16 hours at 4C using 100-fold molar excess of adapters. Agencourt AMPure XP beads (Beckman Coulter) were used for library cleanup, and ligated fragments were amplified by PCR for 6 cycles using 0.5 M universal and indexed primers. Following hybrid-capture at 47C for 72 hours, the captured DNA fragments were enriched with 12 cycles of PCR. Paired-end 2 125-bp sequencing was performed on an Illumina HiSeq 2500 instrument with eight libraries multiplexed into each lane.

Paired-end sequencing data from the Illumina platform were converted to FASTQ format. When included, the unique molecular barcode information at each read of the pair was trimmed and was added to the read header. The Burrows-Wheeler aligner (BWA-mem) (46) was used for the alignment of the processed FASTQ files to the reference hg19 genome. To eliminate the chance of ambiguous short indel alignment on neighboring SNV miscalls, we removed reads with indels. We further cleaned the data from short and hard clipped reads and any nonunique read alignments. We found that, together, these preprocessing steps can improve SNV detection (fig. S8). Consensus read assembly into read families was done in a similar way to previous reports (47, 48). Specifically, reads that share the same molecular barcode sequence, the genomic position of where each read of the pair maps to the reference, and the CIGAR string were grouped. Families that consisted of at least two reads were used to generate SSCS, and a consensus base was called when there was full agreement. When a consensus base was called, it was assigned with the maximum base quality score observed in its corresponding precollapsed reads. Similarly, when two SSCSs with corresponding UMIs on the reciprocal strand were observed, duplex reads were generated. After converting the raw-, SSCS-, and duplex-containing sam files into coordinate-sorted bam files, we used samtools (49) version 1.2 and Varscan2 (14) version 2.2.8 to summarize the data. The following parameters were used: (i) mpileup parameters: -s -x -BQ0 -q1 -d100000 and (ii) pileup2cns parameters: --min-coverage 10 --min-reads2 1 --min-avg-qual 30 --min-var-freq 0.0001 --p-value 1 --strand-filter 0. These are rather permissive parameters allowing the output of all the dominant alleles in each one of the investigated genomic positions. To allow unbiased performance comparisons, we used this format as an input for all the probabilistic methods (SL, AL, and Espresso) and the UMI-based methods (SSCS and duplex).

With Espresso, we deployed a novel approach to model errors based on their association with either one of the 192 contextual contexts (Fig. 3, A to E). These correspond to 12 base substitution types, four alternative 5 bases, and four alternative 3 bases. To mitigate the impact of outliers and real mutations on overfitting, a set of filters is applied to exclude specific variants from the contextual error models (Supplementary Note and fig. S4). These include the removal of alleles (i) that are observed as germline variants in the general population (50, 51) with minor allele frequency 0.1%, (ii) with VAF/error rates 5%, (iii) that have MapQual<59 and MapQual!=0 [for additional information, please refer to the manual of Varscan2 (14)], (iv) that describe recurrent cancer mutations, and (v) that disproportionally persist across multiple samples in the dataset (see the Flagged alleles section; Materials and Methods). Notably, to prevent performance comparison bias, we used these filters together with all the probabilistic methods (SL, AL, and Espresso) and the UMI-based methods (SSCS and duplex) tested.

To determine the more appropriate distribution type for error modeling, Espresso first investigates the overall distribution of nonreference supporting reads in a context-independent manner, in the samples filtered, error-enriched list. On the basis of the observed peak occurrence, either exponential or Weibull distribution models are selected to generate all the contextual models. If the peak corresponds to a single nonreference supporting read, exponential distribution will be used to represent the data; otherwise, if this value is larger than 1, Weibull distribution will be used. Either the pexp or pweibull R functions are then being used together with the modeled parameters from the fitdistrplus package (either rate or shape and scale) to determine how high any nonreference allele of interest is being represented above its corresponding contextual background. A Bonferroni-corrected P value 0.05 was used to determine whether any nonreference allele received significantly more supported reads.

For comparative performance analysis, error rate models at the AL were constructed as previously described (20). Briefly, if the total number of nonzero allele frequencies seen in the training set used for error modeling was 5, we used Gaussian distribution; otherwise, we fit a Weibull distribution to the allele frequencies observed in the training set. Specifically, the pnorm or pweibull R functions were used together with the modeled parameters (either mean and SD or shape and scale) to estimate the likelihood that any allele frequency value of interest is above the corresponding modeled distribution derived for the same interrogated position in the corresponding training set. The yielded P values were adjusted by incorporating the fraction of nonzero allele frequencies into the final models [for additional information, please refer to iDES (20)]. Training datasets were constructed as follows: (i) The pre-AML1 dataset was used for the CB analysis (Fig. 3) and the CL analysis (Fig. 4). (ii) A training set composed of peripheral blood genomic DNA samples from 14 healthy individuals was sequenced and used in the analysis of the AML-MRD data (Fig. 5). (iii) The CB dataset was used as a training set for the derivation of the AL-based model for AML risk prediction (Fig. 6). To evaluate allele mutated status at the SL, we used Varscan2 (14) that computes statistical significance in single samples by Fishers exact test.

While parameters such as specific genomic context, the presence of a repetitive region, and low base or read mapping quality may explain the basis of some errors, these do not always capture artifacts that may persist across multiple samples. We therefore derived a statistical approach to flag recurrently specious alleles. To flag potentially low-frequency artifactual alleles that escaped conventional filtering, we iterated between the 99 and 99.9% nonreference allele frequency quantiles in the entire investigated cohort in increments of 0.1% (user-defined parameters). The 10 derived VAF values were used consecutively to apply Fishers exact tests, determining whether errors with VAF above the quantile-derived cutoff distribute proportionately among all the observed nonreference alleles in the dataset or being clustered in a low number of alleles across many samples in an unbalanced fashion. Then, if included, we removed recurrent Catalogue of Somatic Mutations in Cancer (COSMIC) (52) mutations (that is, SNVs with classification other than synonymous with at least three case reports of hematopoietic and lymphoid tissues; COSMIC version 80) to derive a final list of dataset-specific flagged alleles to be excluded from contextual error modeling.

To derive with a list of mutations that are highly associated with leukemic transformation for AML risk prediction model derivation, we interrogated the COSMIC database (52) and ranked variants according to their evidence for functional relevance in AML. All the SNVs with classification other than synonymous with at least 10 case reports of hematopoietic and lymphoid tissues were considered hotspot variants. For the future implementation of our findings, we reasoned that any hybrid-capture probe design and short sequencing reads would efficiently encompass at least several genomic bases surrounding these hotspots. Therefore, we extended the variant calls to capture mutations with a putative deleterious effect that are within fiveamino acid distance surrounding each hotspot variant. Genomic loci that were found to be mutated in the training cohort (pre-AML1 and pre-AML2) were used for the final model derivation (table S8). Notably, we discarded genomic loci with mutations in KIT, KRAS, and PHF6 as these were found solely in the training sets controls. Such enrichment surely does not correlate with real-life evidence and can bias classification. We then used a random forest algorithm via the R package randomForest. Mutations were grouped by genes, and their VAFs were used to train the model together with the age of the individuals at sampling and the number of the mutations that they carry. If more than one mutation was detected in the same gene, the highest VAF was used. The number of features used for each one of the 5000 generated trees was two.

To simulate a large population screen, we used the mutations detected by Espresso in the controls from the pre-AML1 and pre-AML2 (termed merged dataset here). We first calculated the frequency of controls that carry at least one mutation at the following age groups: 20 to 49, 50 to 64, 65 to 74, and >75 years old. For these age groups, we obtained the incidence rates of AML through the Surveillance, Epidemiology, and End Results Program (53). By assuming similar age distribution for the validation cohort (8) and the individuals interrogated in the merged dataset and knowing the number of pre-AML cases interrogated in the validation cohort (n = 188), we were able to estimate the number of simulated controls needed to mimic real incidence rates for each age group. Overall, 4,033,904 controls were simulated.

The frequency of ARCH and the number of mutations that each individual carries within each control age group from the merged dataset helped us to estimate how many of the simulated individuals are expected to carry mutations in the relevant genomic loci (table S8). Overall, 5.05, 7.69, 10.70, and 19.09% of the individuals within the age range of 20 to 49, for 50 to 64, for 65 to 74, and 75 years, respectively, were simulated to have ARCH. A total of 285,629 individuals (~7%) were simulated to carry one mutation, 934 with two mutations (~0.02%), and 156 with three mutations (~0.004%). We next assigned the specific mutations to the simulated individuals based on their association with each age group. For example, for the 149,423 simulated mutated controls with a simulated age of 50 to 64, we populated a list of 149,423 specific mutations that were detected in control individuals in the same age group or in younger age groups in the merged dataset. We also allowed 10% of the mutations detected in the merged dataset in one age group older to be randomly included. Last, we aimed to assign VAF to the simulated mutations. We observed that the VAF of the detected mutations in the merged dataset did not significantly correlate with age [R(Pearson) = 0.20; P = 0.07] and that a lognormal distribution accurately captures the VAF distribution among all the detected mutations. We therefore used the rlnorm R function to simulate VAFs. This resulted with a median VAF of 1.45% and a mean VAF of 2.45% for the simulated controls; 37.46% of the simulated VAFs received a value of VAF 2%. As intended, these values are highly comparable with those of the mutations found in the merge datasets controls (table S7).

Read more from the original source:
Integration of intra-sample contextual error modeling for improved detection of somatic mutations from deep sequencing - Science Advances

MPN Driver Mutations Can Be Acquired as Early as in Utero, Study Shows – Targeted Oncology

A study presented during the 2020 ASH Annual Meeting has suggested that certain driver mutations for myeloproliferative neoplasms (MPNs) can be traced back to when they were acquired as early as in utero.

If you can not only detect clones early but then calculate their rates of growth with a repeat sample you can then plot the growth trajectory of these clones and estimate the latency to a potential clinical manifestation, thus offering opportunities for early preventive strategies, said Jyoti Nangalia, MBBChir, senior study author, and cancer research UK clinician scientist at Sanger Institute, in a virtual presentation of the data.

The study comprised a cohort of 10 patients with essential thrombocythemia (ET), polycythemia vera (PV), and myelofibrosis (MF), with a median age of 48.5 (range, 20-76).

Each patients peripheral blood and bone marrow samples were grown into single cellderived hematopoietic colonies. Each colony underwent whole-genome sequencing. A total of 952 whole-genome sequences were produced, each reflecting that of the single cell from which the colony was derived.

Right from the start of life, as our cells are dividing, mutations are being acquired, and theyre being passed down from generation to generation such that at any one time, the mutations within individual cells represent natural bar codes that can be used to trace back the ancestry of those cells right to the start of life, and so we did this in MPNs, said Nangalia.

Phylogenetic trees were drawn based on the somatic mutations that had been identified. Driver mutations were subsequently assigned to the tree and evaluated for appearance patterns across each colony, reflecting the relative development of the driver mutations in each patient.

Because the total number of somatic mutations in an individual colony was shown to correlate with age, investigators converted the relative development of mutations to absolute development to more accurately understand clonal evolution.

Our blood stem cells require mutations throughout life, roughly 18 mutations across the genome per year. Therefore, by applying patient-specific mutation rates and clone-specific mutation rates, we were able to put a start time and an end time to each individual branch across the phylogenetic trees in our cohort, said Nangalia.

In the first patient who had been diagnosed with ET at the age of 21, the JAK2 mutation was acquired early, timing between 6.2 weeks post-conception and 1.3 years of age. In the phylogenetic tree, the branching downstream of JAK2 demonstrated how the single stem cell that acquired JAK2 expanded into a clone of stem cells in rapid succession.

Similarly, in the second patient diagnosed with PV at the age of 31, the JAK2 mutation was acquired early, timing between 4.2 weeks post-conception and 8.6 years of age. Clonal expansion also demonstrated a DNMT3A mutation, which is the most common mutation in age-related clonal hematopoiesis, said Nangalia. However, in this patient, the mutation was also acquired early, timing between 4.6 weeks post-conception and 7.8 years of age, growing slowly before reaching a detectable clonal fraction.

In some patients, the JAK2 and DNMT3A mutations were acquired very early, including in utero. In one patient diagnosed with PV at age 33, the JAK2 mutation was acquired between 9.1 weeks post-conception and 4.1 months after birth, and the DNMT3A mutation was acquired between 19.4 weeks and 22.2 weeks post-conception. Another patient with PV showed that the DNMT3A mutation was acquired as early as between 1.2 weeks and 7.9 weeks post-conception.

In the patient with PV diagnosed at age 33, evidence of clonal evolution in the MPN clone was shown over 3 decades, cascading from a JAK2 mutation 4.1 months post-birth to homozygous JAK2 at 17.8 years to 1q amplification at 33 years.

In other patients, JAK2 was acquired as the second driver mutation in an already expanded DNMT3A-mutated clone.

Notably, a patient diagnosed with ET at age 54 had acquired DNMT3A R882H by 2 years of age, which is one of the most common mutations found in acute myeloid leukemia, said Nangalia. However, in another patient who had been diagnosed with ET at 76 years of age, there was a driverless clonal expansion.

We know that clonal hematopoiesis can often lack driver mutations and is thought to be the case in up to 50% of patients with clonal hematopoiesis. Here, we showed that that also has a single-cell origin. What is driving clonal expansion in this individual in that particular clone, we do not know, said Nangalia.

Across the cohort, recurrent or similar genetic aberrations were found in individual patients. A patient diagnosed with PV at age 53, who acquired a JAK2 exon 12 mutation in childhood, developed 3 independent DNMT3A mutations and 4 independent homozygous acquisitions stemming from their JAK2 exon 12 mutation.

Other patients demonstrated evidence of independent acquisitions of 1q amplifications, leading to myelofibrotic transformation. Another patient demonstrated evidence of multiple independent acquisitions of JAK2 V617F, suggesting that factors other than driver mutations, such as the patients germline or microenvironment within the bone marrow, also influence clonal evolution.

In the second portion of the study, investigators revisited the patients for whom phylogenetic trees had been drawn and re-sequenced the mutations that had been identified in the phylogenetic trees in the whole blood.

Combining the mutant clonal fractions in blood with the pattern of branching in the trees, investigators calculated the rate at which the clones had been growing over each patients lifespan.

Clone rates varied significantly. For example, in a patient with 3 mutant clones that had been acquired in uteroDNMT3A, JAK2, and JAK2/TET2the rates of annual growth were 9% (95% CI, 5%-25%), 67% (95% CI, 6%-246%), and 233% (95% CI, 143%-360%), respectively, the latter of which translates to a doubling-in-size time of every 7 months.

Regarding clones that had the same genetic makeup, such as clones consisting solely of the JAK2 V617F driver mutation, the annual growth rate varied among individual patients, ranging from 18% (95% CI, 13%-23%) to 68% (95% CI, 41%-95%).

This again suggests that there are factors other than JAK2 that determine the consequences of acquiring it in individual patients, said Nangalia.

Additional results revealed that the rate of growth was associated with the time of diagnosis. In patients with slow growth rates of less than 50%, over 50 years had gone by prior to diagnosis, whereas in patients with growth rates over 100%, it took less than 10 years before a diagnosis was made.

In retroactively calculating what the clonal fractions would have been leading up to diagnosis, the slow growing JAK2 clones could have been detected with sensitive assays 40 years prior to diagnosis and up to 10 years before diagnosis for faster growing clones.

Providing additional perspective during a press briefing, Robert Brodsky, MD, moderator, and director of the Division of Hematology at Johns Hopkin Medicine stated, These results suggest that there may be untapped opportunities to detect these conditions much earlier and potentially intervene and prevent disease development.

Reference

Williams N, Lee J, Moore L, et al. Driver mutation acquisition in utero and childhood followed by lifelong clonal evolution underlie myeloproliferative neoplasms. Presented at: 2020 ASH Annual Meeting & Exposition; December 5-8, 2020; virtual. Abstract LBA-1.

See the original post here:
MPN Driver Mutations Can Be Acquired as Early as in Utero, Study Shows - Targeted Oncology

Bid to fund stem cell treatment in Mexico for woman with MS – Bournemouth Echo

A FORMER hairdresser from Bournemouth is appealing for people to help raise money to have life-saving surgery in Mexico to get rid of her Multiple Sclerosis once and for all.

Having been admitted to Royal Bournemouth Hospital for a suspected stroke or brain tumour in March 2017, at the age of 47, Kirsten Hannibal was found to have multiple lesions on her brain and was diagnosed with CIS which later progressed to MS.

During lockdown, Kirsten has researched into different ways to stop Multiple Sclerosis dead in its tracks, one of them being Hematopoietic Stem Cell Transplantation.

Although the procedure, which involves the transplantation of multipotent hematopoietic stem cells, usually derived from bone marrow, is not widely accessible in the UK, it is available in Mexico, considered a world class hub for HSCT.

However she must raise over 40,000 to cover flights to Mexico as well as the cost of the procedure.

Vicky Dixon has set up a crowdfunding page to raise money for Kirstens medical procedure.

In a statement written on her crowdfunding page, she said: Our family are joining forces to raise the money needed to send our Kirsten to Mexico for Hematopoietic Stem Cell Transplantation treatment that is not universally available on the NHS, but will hopefully give Kirsten a chance of a future; a life free of pain, disability and heart breaking challenges.

We hope that Kirsten can follow the footsteps of other British MS sufferers and go to Mexico, a world class centre for HSCT, and cheaper than the UK, at the cost of 43,500.

The first large, randomised control trial, and several meta-analyses of HSCT, have confirmed that HSCT is a very effective therapy. This is now tipping the scales for HSCT becoming a mainstream treatment for MS in Britain.

However, the treatment has to take place before the MS becomes too advanced, and as it will be years before HSCT might be offered more widely, Kirsten would by then be swallowed up by the MS and not a suitable candidate for treatment.

Kirsten is on the brink of becoming too disabled for this treatment, hence the urgency of our appeal.

Sadly, the 46-year-old is now travelling a path similar to one her family have walked before.

In 1984 her mother at the age of 32 was diagnosed with lymphoblastic leukaemia and the Echo covered the story.

Her mother underwent aggressive chemotherapy and was the receiver of a ground-breaking treatment with a bone marrow transplant.

She was the first patient to receive this treatment in the south and, whilst the treatment was deemed a success, sadly her mother died.

Lynda Smiths legacy lives on because her bravery in allowing this treatment to take place is now the lifeline to many children and adults alike who survive leukaemia.

The treatment Kirsten is looking to have is similar to her mothers treatment, except it would be her own bone marrow that would be harvested. She will then be given chemotherapy and then the day Kirsten longs for, freedom from the disease.

The new birthday she dreams of is a stem cell birthday celebrated when the bone marrow is put back into her body giving her the chance of stopping Multiple Sclerosis.

So far, Kirstens fundraising appeal has raised 4,535, just over 10 per cent of her target.

To donate, visit https://www.gofundme.com/f/multiple-sclerosis-and-an-urgent-bid-for-freedom?utm_source=customer&utm_medium=email&utm_campaign=p_cp+sharesheet.

Originally posted here:
Bid to fund stem cell treatment in Mexico for woman with MS - Bournemouth Echo

CollPlant to Supply rhCollagen to STEMCELL Technologies for Use in a Broad Range of Cell Culture Applications – PRNewswire

REHOVOT, Israel and VANCOUVER, BC, Dec. 10, 2020 /PRNewswire/ -- CollPlant (NASDAQ: CLGN), a regenerative medicine company, and STEMCELL Technologies, Canada's largest privately owned biotechnology company, which develops cell culture media, cell separation systems, instruments, and other reagents for life sciences research, today jointly announced they have entered into aproduct manufacturing and supply agreement. CollPlant will sell its proprietary recombinant human Type I collagen (rhCollagen), the world's first plant-based rhCollagen, to STEMCELL Technologies, which will incorporate CollPlant's product into cell culture media kits.

The recently signed agreement follows the companies' established business relationship, which started in 2014 when STEMCELL began purchasing and incorporating CollPlant's rhCollagen into some of its cell culture expansion and differentiation media kits. To date, hundreds of companies, as well as research and academic institutes, have used these kits for research and development projects. STEMCELL will distribute the kits globally for use in the regenerative medicine research market.

"Incorporation of rhCollagen into STEMCELL's cell culture applications sold to researchers worldwide is designed to help advance the science in a broad range of dynamic fields including stem cells, immunology, cancer, regenerative medicine, and cellular therapy. We are happy to have entered into this agreement with STEMCELL, which, as Canada's largest biotechnology company, is very well positioned to make rhCollagen-containing cell culture kits widely available in the market," stated Yehiel Tal, Chief Executive Officer of CollPlant. "The cell culture market is just one example of the vast potential of our rhCollagen platform technology in life science applications. We continuously evaluate new fields in which CollPlant's products and technologies have the potential to enable breakthroughs that improve patients' lives."

Dr. Sharon Louis, STEMCELL's Senior Vice President of Research and Development noted that "STEMCELL is pleased to utilize CollPlant's animal component free rhCollagen to promote cell attachment in several products that support the culture of diverse human progenitor cell types. The quality and animal component-free composition of CollPlant's rhCollagen is what first brought this product to STEMCELL's attention, and the robust performance rhCollagen provides with a variety of STEMCELL media is what we want to be able to provide to our customers. Upon entering into this agreement, STEMCELL and CollPlant will together provide high-quality reagents that will be used to further our understanding in life sciences and potentiate regenerative medicine research."

About STEMCELL Technologies

STEMCELL Technologies is Canada's largest biotechnology company. Based in Vancouver, STEMCELL supports life sciences research around the world with more than 2,500 specialized reagents, tools, and services. STEMCELL offers high-quality cell culture media, cell separation technologies, instruments, accessory products, and educational resources that are used by scientists advancing the stem cell, immunology, cancer, regenerative medicine, microbiology, and cellular therapy fields.

Find more information at http://www.stemcell.com

About CollPlant Biotechnologies

CollPlant is a regenerative and aesthetic medicine company focused on 3D bioprinting of tissues and organs, and medical aesthetics. Our products are based on our rhCollagen (recombinant human collagen) that is produced with CollPlant's proprietary plant based genetic engineering technology.

Our products address indications for the diverse fields of tissue repair, aesthetics and organ manufacturing, and, we believe, are ushering in a new era in regenerative and aesthetic medicine.

Our flagship rhCollagen BioInk product line is ideal for 3D bioprinting of tissues and organs. In October 2018, we entered into a licensing agreement with United Therapeutics, whereby United Therapeutics is using CollPlant's BioInks in the manufacture of 3D bioprinted lungs for transplant in humans.Recently, the parties announced the expansion of the collaboration with the exercise by United Therapeutics of its option to cover a second lifesaving organ, human kidneys.

Safe Harbor for Forward-Looking Statements

This press release may include forward-looking statements. Forward-looking statements may include, but are not limited to, statements relating to CollPlant's objectives, plans and strategies, as well as statements, other than historical facts, that address activities, events or developments that CollPlant intends, expects, projects, believes or anticipates will or may occur in the future. These statements are often characterized by terminology such as "believes," "hopes," "may," "anticipates," "should," "intends," "plans," "will," "expects," "estimates," "projects," "positioned," "strategy" and similar expressions and are based on assumptions and assessments made in light of management's experience and perception of historical trends, current conditions, expected future developments and other factors believed to be appropriate. Forward-looking statements are not guarantees of future performance and are subject to risks and uncertainties that could cause actual results to differ materially from those expressed or implied in such statements. Many factors could cause CollPlant's actual activities or results to differ materially from the activities and results anticipated in forward-looking statements, including, but not limited to, the following: the CollPlant's history of significant losses and its need to raise additional capital and its inability to obtain additional capital on acceptable terms, or at all; CollPlant's expectations regarding the timing and cost of commencing clinical trials; regulatory action with respect to rhCollagen-based products, including but not limited to acceptance of an application for marketing authorization, review and approval of such application, and, if approved, the scope of the approved indication and labeling; commercial success and market acceptance of the CollPlant's rhCollagen-based BioInk; CollPlant's ability to establish sales and marketing capabilities or enter into agreements with third parties and its reliance on third-party distributors and resellers; CollPlant's reliance on third parties to conduct some aspects of its product manufacturing; the scope of protection CollPlant is able to establish and maintain for intellectual property rights and the company's ability to operate its business without infringing the intellectual property rights of others; the overall global economic environment; the impact of competition and new technologies; general market, political, and economic conditions in the countries in which the company operates; projected capital expenditures and liquidity; changes in the company's strategy; and litigation and regulatory proceedings. More detailed information about the risks and uncertainties affecting CollPlant is contained under the heading "Risk Factors" included in CollPlant's most recent annual report on Form 20-F, filed with the SEC, and in other filings that CollPlant has made. The forward-looking statements contained in this press release are made as of the date of this press release and reflect CollPlant's current views with respect to future events, and CollPlant does not undertake, and specifically disclaims, any obligation to update or revise any forward-looking statements, whether as a result of new information, future events or otherwise.

Contact atCollPlant:

Eran Rotem Deputy CEO & CFO Tel: + 972-73-2325600 [emailprotected]

Contact at STEMCELL: Luba Metlitskaia Vice President, Business Development & Licensing [emailprotected]

SOURCE CollPlant

Read more here:
CollPlant to Supply rhCollagen to STEMCELL Technologies for Use in a Broad Range of Cell Culture Applications - PRNewswire

NASA-partnered Pluristem crashes to Earth as it axes leading therapy – FierceBiotech

Israeli biotech Pluristem is canning its experimental phase 3 critical limb ischemia therapy after an outside review said it was no good.

Haifa, Israel-based Pluristems R&D operation is built upon placenta-derived adherent stromal cells, which the biotech has designed for use in patients of all human leukocyte antigen types. This approach is made possible by the low immunogenicity of the cells. Once inside the body, Pluristem hopes the cells will drive the healing of injured tissue.

But one of its leading contenders using this approach has been judged a failure in phase 3: An independent data monitoring committee (DMC) took a look at the ongoing data for its pivotal phase 3 in patients with critical limb ischemia (CLI), a severe obstruction of the arteries which markedly reduces blood flow to the extremities and can lead to amputation.

During this webinar, attendees will learn about Thermo Fisher Scientifics new 5KL bioreactor and how it benefits clients who outsource late-phase biologics drug substance manufacturing. An introduction of the 5KL bioreactor will be provided, as well as application data around performance and scalability, process economy comparison with traditional stainless steel bioreactors, and decision criteria that could be helpful in choosing between different cell culture strategies. Register Today!

The DMC said the test was unlikely to meet the primary endpoint, and that the CLI study population has experienced a substantial low number of events (major amputation of the index leg or death), different from what is known in clinical medicine for the rate of these events in this patient population. The lower than anticipated event rate in the placebo group reduced the statistical power of the study to meet its primary endpoint.

The biotech is now tossing out the therapy and will instead focus on other pipeline areas, including a long-shot stem cell attempt at treating COVID-19. The biotechs shares fell nearly 40% on the news.

We are deeply disappointed by the outcome of the CLI interim analysis. In light of the DMCs recommendation, we decided that it would be in the best interests of the company and its shareholders to terminate the CLI study and focus our resources and efforts on our other lead indications, said Pluristem CEO and President Yaky Yanay.

We expect to present topline clinical results during calendar year 2021, including our phase 3 study in muscle regeneration following hip fracture, phase 2 studies in Acute Respiratory Distress Syndrome associated with COVID-19 and our phase 1 study in incomplete hematopoietic recovery following hematopoietic cell transplantation. Pluristem is well positioned to advance and support future development of these indications.

Last year, Pluristem penned a deal with NASA to assess its cell therapies against the health problems caused by spending time in space, teaming up with NASAs Ames Research Center for the project, which focuses on using its PLX placenta-derived cell therapies to try to prevent or treat medical conditions that can occur during and after space missions, including conditions that affect the blood, bone, muscle, brain and heart.

Read the original post:
NASA-partnered Pluristem crashes to Earth as it axes leading therapy - FierceBiotech

Global Animal Stem Cell Therapy Market To Reach A New Threshold of Growth By 2026 – The Courier

There are millions of them around the globe waiting for clutching on to some of the latest vital information circulating across the globe. The up-to-the-minute Animal Stem Cell Therapy market report based on the growth and the development of theAnimal Stem Cell Therapy marketis systematically listed down. The Animal Stem Cell Therapy market report comprises statistically verified facts such the unique essence including topological investigations, worldwide market share, government stringent norms, applications, current trends, futuristic plans, market bifurcations, and so on mentioned in a crystal clear pattern.

The statistical plus scientific Animal Stem Cell Therapy market report has all the important market aspects penciled down in a layman language format so that the data based on the markets productivity or future strategy can be easily extrapolated from the reports. The Animal Stem Cell Therapy market report has the dominant market players ANIMAL CELL THERAPIES, Celavet, Animacel, VETSTEM BIOPHARMA, Cell Therapy Sciences, Magellan Stem Cells, Cells Power Japan, Animal Care Stem, Aratana Therapeutics, VetCell Therapeutics, MediVet Biologic, U.S. Stem Cell, J-ARM explained in detail.

Sample of global Animal Stem Cell Therapy Market Report at::https://www.marketresearchstore.com/report/global-animal-stem-cell-therapy-market-report-2020-751651#RequestSample

The essential futuristic segments such as {Dogs, Horses, Others}; {Veterinary Hospitals, Research Organizations} have also been detailed out in the Animal Stem Cell Therapy market report for the clients convenience and more of vital data embracing capability. The forecast trends along with the current market status can better understand the Animal Stem Cell Therapy market development on a global basis. The intricate industrial strategies and the supply-demand chain are also discussed in the contextual report.

Key points of the global Animal Stem Cell Therapy market

Theoretical analysis of the global Animal Stem Cell Therapy market stimulators, products, and other vital facets Recent, historical, and future trends in terms of revenue and market dynamics are reported Pin-point analysis of the competitive market dynamics and investment structure is predicted to grow Future market trends, latest innovations, and various business strategies are reported Market dynamics include growth influencers, opportunities, threats, challenges, and other crucial facets

The Animal Stem Cell Therapy market report has the imperative data mentioned in a systematic way only after comprehensive inspection and vigilant referencing. The diagrammatic representations such as a pie chart of the Animal Stem Cell Therapy market are also drawn out so as to attract the punters and make it easy for them to comprehend the entire Animal Stem Cell Therapy market from in and out. The most eye-catching format of the Animal Stem Cell Therapy market report is its market bifurcation based on the product type, application, geography, end-users, and more as per the particular market. The geographical segments Europe, North America, Latin America, Asia Pacific, and Middle East & Africa are further exhaustively mentioned.

Read Detailed Index of full Research Study at::https://www.marketresearchstore.com/report/global-animal-stem-cell-therapy-market-report-2020-751651

Questions answered in the report include

1. What is the expected market size by the end of the forecast period? 2. What are the major factors initiating the global Animal Stem Cell Therapy market growth? 3. What are the latest developments and trending market strategies that are influencing the growth of the Animal Stem Cell Therapy market? 4. What are the key outcomes of the Animal Stem Cell Therapy market developments? 5. Who are the key players in the market? 6. What are the opportunities and challenges faced by the key players?

The Animal Stem Cell Therapy market report endows the global market dominance, market segmentation, growth factors, and others reported such that the clients can have a total tour of the market without any much off efforts needed.

Key Reasons for Purchasing Global Animal Stem Cell Therapy Market Report

New approaches and latest development trend that describe the structure of the market Advanced market breakdown structure Historical data and future market scope In-depth market analysis based on statistics, growth stimulators, and market developments Statistical data representation through figurative, numerical, and theoretical elaboration Report provides insight of the business and sales activities

Ask Any Kind of Query Before Purchasing the Report Here:https://www.marketresearchstore.com/report/global-animal-stem-cell-therapy-market-report-2020-751651#InquiryForBuying

Visit link:
Global Animal Stem Cell Therapy Market To Reach A New Threshold of Growth By 2026 - The Courier

Cryo-shocked cancer cells for targeted drug delivery and vaccination – Science Advances

Abstract

Live cells have been vastly engineered into drug delivery vehicles to leverage their targeting capability and cargo release behavior. Here, we describe a simple method to obtain therapeutics-containing dead cells by shocking live cancer cells in liquid nitrogen to eliminate pathogenicity while preserving their major structure and chemotaxis toward the lesion site. In an acute myeloid leukemia (AML) mouse model, we demonstrated that the liquid nitrogentreated AML cells (LNT cells) can augment targeted delivery of doxorubicin (DOX) toward the bone marrow. Moreover, LNT cells serve as a cancer vaccine and promote antitumor immune responses that prolong the survival of tumor-bearing mice. Preimmunization with LNT cells along with an adjuvant also protected healthy mice from AML cell challenge.

Acute myeloid leukemia (AML) is a hematological malignancy with a dismal prognosis and 5-year survival of only 30% (1, 2). The standard-of-care cytoreductive chemotherapy induces AML remission (35), but disease relapse frequently occurs (6, 7). Hematopoietic stem cell transplantation (HSCT) in patients who achieve remission after chemotherapy represents the only curative approach so far (8, 9). However, HSCT is associated with either the lack of suitable hematopoietic stem cell donors or the high risk of transplantation-related mortality (10). Hence, there is an urgent need to find further strategies for AML treatment.

AML originates in the bone marrow, and bone marrow creates leukemia-niches that promote leukemia survival (11). Furthermore, biodistribution of chemotherapeutics to the bone marrow is frequently poor (12, 13), and higher doses of chemotherapy required to ablate leukemia are toxic to normal tissues. Thus, developing targeting drug delivery to the bone marrow may not only enhance the therapeutic index of chemotherapy but also reduce its toxicity to nonhematopoietic tissues. Nevertheless, it is still challenging to engineer bone marrowtargeting moieties and bypass the bloodbone marrow barriers (14). Leveraging cells intrinsic properties offers solutions to overcome these limitations (1517). Because AML cells naturally exhibit bone marrow homing capabilities (1820), we developed an approach to use AML cells as drug carriers while eliminating their intrinsic pathogenicity.

Here, we used a liquid nitrogenbased cryo-shocking method to obtain therapeutic dead cells. These cells maintained the intact structure allowing for drug encapsulation, but lost their proliferation ability and pathogenicity. Specifically, cryo-shocked AML cells kept their bone marrow homing capability and served as a drug delivery vehicle of doxorubicin (DOX), which is a critical drug used in the induction chemotherapy in AML. Cryo-shocked AML cells stimulated an immune response that was in conjunction with chemotherapy to eradicate leukemia in tumor-bearing mice. Preimmunization with LNT cells together with an adjuvant protected healthy mice from AML cell challenge. We thus proposed a dead cellbased delivery vehicle that can be rapidly manufactured for clinical use compared with other live cellbased drug delivery systems (21).

To obtain the liquid nitrogentreated (LNT) cells, AML cells were suspended in the cell cryopreservation medium and immersed in liquid nitrogen for 12 hours. LNT cells were then thawed at 37C and washed with phosphate-buffered saline (PBS) (Fig. 1A). When analyzed by confocal microscopy, LNT cells showed the same cellular structure as untreated live cells when assessed by nucleus and cytoskeleton staining (Fig. 1B). A slight decrease in cellular size was observed (Fig. 1C), with an average size of 11 m for LNT cells and 12 m for untreated live cells. The forward scatter (FSC) values measured by flow cytometry corroborated the cell size reduction of LNT cells, and similar side scatter (SSC) values suggested that the internal structure of LNT cells was maintained (Fig. 1D). Scanning electron microscopy (SEM) images revealed the sphere-like structure of LNT cells and the rougher cellular surface as compared with control live cells (Fig. 1E and fig. S1).

(A) Schematic of the procedure to prepare LNT cells. (B) Cellular structure of live and LNT C1498 cells. Cell nucleus was stained by Hoechst 33342 (blue), and cytoplasm F-actin was stained by AF488 phalloidin (green). Scale bars, 10 m. (C) Cellular sizes of live and LNT C1498 cells. The cells were imaged by confocal microscopy, and cellular size was measured by the software Nano Measurer (cell numbers = 200). (D) Flow cytometry analysis of live and LNT C1498 cells under same voltages. FSC, forward scatter; SSC, side scatter. (E) SEM images of live and LNT cells. Scale bars, 1 m. (F) Cell viability analysis of live and LNT cells by LIVE/DEAD viability kit. Calcein AM: live cells; EthD-1: dead cells. Scale bar, 10 m. (G) Cell viability analysis of live and LNT cells by CCK8 assay (n = 6). a.u., arbitrary unit. (H) In vivo proliferation of 2 106 luciferase tagged live and LNT C1498 cells indicated by the bioluminescence signal (n = 5). (I) Survival of mice after challenge with 2 106 live and LNT tumor cells (n = 5). Typical flow cytometry images (J) and DsRed intensities (K) of peripheral blood 20 days after challenge with live and LNT DsRed tagged C1498 cells (n = 3). MFI, mean fluorescence intensity. Data are presented as means SD (G and K). Statistical significance was calculated via the log-rank (Mantel-Cox) test (I) and ordinary one-way analysis of variance (ANOVA) (K). *P < 0.05, **P < 0.01. NS, not significant.

Next, we evaluated the cell viability of LNT cells. As shown in Fig. 1F, nearly all the LNT cells were labeled with EthD-1 (indicating dead cells) and did not show intact fluorescence signal of calcein AM (indicating live cells). In addition, LNT cells did not show proliferative activity as compared with live cancer cells as measured with cell counting kit-8 (CCK8) assay (Fig. 1G). Furthermore, we confirmed the necrosis-dependent cell death of LNT cells by annexin-Vpropidium iodide (PI) staining (fig. S2). We further verified the absence of pathogenicity of LNT cells in vivo. As shown in Fig. 1H, live C1498 AML cells quickly proliferated in mice and caused 100% death in 31 days, while mice receiving C1498 LNT cells exhibited no detectable bioluminescence signal, and all mice survived for at least 180 days (Fig. 1, H and I). Moreover, we quantitatively analyzed cancer cells in the peripheral blood at day 20 after injection. A notably higher DsRed signal was observed in mice injected with live C1498 cells, indicating a high portion of leukemia cells circulating in the blood, while the DsRed intensity for the mice challenged with LNT cells was similar to that of healthy mice (Fig. 1, J and K).

Leukemia cells exhibit bone marrow homing and resident capabilities, which are at least in part associated with the expression of CXCR4 and CD44 chemokine, two typical adhesion receptors that interact with bone marrow (18, 22, 23). SDSpolyacrylamide gel electrophoresis (PAGE) showed that most of the proteins expressed by live C1498 cells were retained in LNT cells (fig. S3A). CXCR4 and CD44 were detected in both live and LNT cells as assessed by confocal imaging and flow cytometry (Fig. 2, A and B, and fig. S3, B and C). Despite some reduction in expression levels, Western blotting analysis indicated that CXCR4 and CD44 expression were 39 and 60%, respectively, in LNT cells compared with live cells (fig. S3, D and E). The bone marrow homing capacity of LNT cells was also evaluated. Upon intravenous infusion, LNT cells exhibited similar accumulation efficiency in bone barrow compared with live C1498 cells (Fig. 2, C and D, and fig. S4A). Cell signal was notably higher compared with paraformaldehyde-fixed cells, which reflects the loss of bioactivities upon paraformaldehyde fixation (Fig. 2, C and D). LNT cells also distributed in the liver, kidney, and spleen (fig. S4B), and were cleared from the bloodstream within 24 hours (fig. S5).

CXCR4 (A) and CD44 (B) expression of live and LNT C1498 cells analyzed by confocal microscopy (top) and flow cytometry (bottom). Scale bars, 10 m. (C) Fluorescence images of bone isolated 6 hours after injection of cy5.5-labeled live C1498 cells, LNT C1498 cells, and paraformaldehyde-fixed C1498 cells. (D) Fluorescence intensities of the bone of indicated groups (n = 6). (E) Typical confocal image of DOX-loaded LNT cells. Scale bar, 10 m. (F) Cumulative release profile of DOX from LNT cell/DOX (n = 3). (G) Plasma DOX concentration after intravenous injection of free DOX and LNT cell/DOX with DOX dose of 2.5 mg/kg (n = 4). (H) Bone marrow DOX content 3 hours after administration of the drug (n = 3). Data are presented as means SD (D and F to H). Statistical significance was calculated via ordinary one-way ANOVA (D) and Students t test (G and H). *P < 0.05, **P < 0.01, ***P < 0.001.

Because nuclear and cytoplasmic cellular structures are preserved in LNT cells (Fig. 1B), we assessed if these cells can be payload with DOX, via DNA intercalation and the electrostatic interactions between DOX and cytoplasm proteins (2426), and deliver DOX to bone marrow. Briefly, DOX could be loaded into LNT cells via mixing and incubation with a loading capacity of 65 16 g per 1 107 LNT cells (Fig. 2E and fig. S6A). DOX was released from the drug-loaded LNT cells (LNT cell/DOX) in a sustained manner, and 81% of DOX was released within 10 hours (Fig. 2F). We then studied the in vitro cytotoxicity against C1498 cells of free DOX and LNT cell/DOX. The IC50 (median inhibitory concentration) values were 0.32 and 1.05 g/ml, respectively (fig. S6B). Even though free DOX exhibited higher cytotoxicity against C1498 cells in vitro, LNT cell/DOX allowed longer detection of DOX in the blood and higher DOX accumulation within the bone marrow (Fig. 2, G and H). We used murine AML models to evaluate the therapeutic efficacy of LNT cell/DOX. In tumor-bearing C57BL/6J mice, tumor growth was monitored by bioluminescence signals upon treatment (fig. S7, A to C). In this leukemia model, although LNT cells alone exhibited no antitumor effects, LNT cell/DOX treatment reduced the tumor growth compared with control treatments (fig. S7, D to H).

Tumor cell lysates can function as cancer vaccines and initiate tumor-specific immune responses (27, 28). We hypothesized that LNT cells can enhance the antigen uptake and maturation of antigen-presenting cells (APCs). LNT cells cocultured with dendritic cells (DCs) caused their maturation as assessed by up-regulation of CD40, CD80, CD86, and major histocompatibility complex II (MHC-II) (fig. S8A). Moreover, CD4+ T cells and CD8+ T cells increased in the peripheral blood of the mice receiving LNT cells and the adjuvant of monophosphoryl lipid A (MPLA) (fig. S8B). DC maturation and T cell activationrelated cytokines, including interferon- (IFN-), tumor necrosis factor (TNF-), and interleukin-6 (IL-6), were also detected in mice treated with LNT cell and adjuvant (fig. S8C). We next evaluated the antitumor efficacy of LNT cell/DOX with adjuvant in leukemia-bearing mice. As demonstrated in Fig. 3 (A and B), bioluminescence of AML cancer cells increased rapidly in untreated mice, while AML had been partially inhibited after DOX or LNT cell and adjuvant treatment. AML cells were almost completely eliminated in mice treated with LNT cell/DOX and adjuvant up to 21 days after tumor inoculation (Fig. 3B). Quantitative analysis of tumor bioluminescence and survival analysis also demonstrated superior therapeutic activity of LNT cell/DOX combined with adjuvant (Fig. 3, C to E). Increased serum levels of IFN- and TNF- (Fig. 3, F and G), as well as elevation of CD3+ T cell and CD8+ T cells, supported the occurrence of boosted immunity in the mice receiving LNT Cell/DOX and adjuvant treatment (Fig. 3, H and I).

(A) Schematic of the treatment model. (B) AML progression in vivo as indicated by bioluminescence signal expressed by luciferase tagged C1498 cells during different treatments (G1, saline; G2, DOX; G3, LNT cell + adjuvant; G4, LNT cell/DOX + adjuvant). (C) Quantified bioluminescence of different treatment groups. (D) Bioluminescence intensity of treated mice on day 21 (n = 6). (E) Survival of the mice of different treatment groups (n = 6). Serum cytokine levels of IFN- (F), TNF- (G), and proportion of peripheral CD3+ T cells (H) and CD8+ T cells (I) on day 13 (n = 6). Data are presented as means SD. (D and F to I). Statistical significance was calculated via ordinary one-way ANOVA (D and F to I) and log-rank (Mantel-Cox) test (E). *P < 0.05, **P < 0.01, ***P < 0.001.

We further evaluated the efficacy of LNT cells as a prophylactic cancer vaccine. Mice were first immunized at 21, 14, and 7 days before challenge with live C1498 cells. The onset of AML in mice was prevented in mice preimmunized with LNT cells and adjuvant (Fig. 4, A to C). Quantitative data also revealed that the tumor bioluminescence intensity of the group of LNT cells with adjuvant was substantially lower than control groups (Fig. 4D). Moreover, 71% of the mice treated with LNT cells and adjuvant were tumor free 90 days after tumor challenge, while all control mice died by day 34 (Fig. 4E). Serum levels of IFN-, TNF-, IL-12, and IL-6 were significantly increased in mice treated with LNT cells and adjuvant (Fig. 4F), indicating that a prompt immune response was triggered upon tumor cell inoculation. In addition, CD3+ T cells and CD8+ T cells were significantly increased in the peripheral blood of mice vaccinated with LNT cells and adjuvant (Fig. 4, G and H, and fig. S9).

(A) Schematic of the treatment model. Bioluminescence images (B) and quantified bioluminescence (C) of the mice preimmunized with different treatment formulations (G1, saline; G2, adjuvant; G3, LNT cell + adjuvant). (D) Bioluminescence intensity of treated mice on day 47 (n = 5 for G1 and G2 for one mice died before day 47; n = 7 for G3). (E) Survival of the mice after tumor challenge (n = 6 for G1 and G2; n = 7 for G3). (F) Serum cytokine levels 3 days after challenge of live C1498 cells (n = 6 for G1 and G2; n = 7 for G3). (G) Representative flow cytometry images of CD3+ T cells (left) and proportion of peripheral CD3+ T cells (right) on day 24 (n = 6 for G1 and G2; n = 7 for G3). (H) Representative flow cytometry images of CD8+ T cells (left) and corresponding proportion of peripheral CD8+ T cell gating on CD3+ T cells (right) on day 24 (n = 6 for G1 and G2; n = 7 for G3). Data are presented as means SD. (D and F to H). Statistical significance was calculated via ordinary one-way ANOVA (D and F to H) and log-rank (Mantel-Cox) test (E), *P < 0.05, **P < 0.01, ***P < 0.001.

In this study, we demonstrated the feasibility, efficacy, and safety of tumor dead cells used as a drug-targeting carrier and tumor vaccine for cancer therapy. Compared with the synthetic materialmediated delivery vehicles, cell-based carriers show unique targeting capacities and can bypass biological barriers (15, 29). AML cells originate in the bone marrow and naturally exhibit similar bone marrow homing capabilities as HSCs (22, 30, 31), rendering them suitable to be used as cellular drug carriers for AML therapy. However, it remains essential to develop strategies allowing the elimination of AML tumorigenicity while transiently preserving cellular integrity to deliver the payload at the tumor site. We therefore proposed to use the dead but functional AML cells as the drug carrier.

Usually, the structure of the live cells can disintegrate upon dying with the loss of proteins and cytokines (32). In addition, external stimuli that could induce cell death, such as heat or radiation, will deactivate proteins as well (33, 34). Our data support the concept that cryo-shocked tumor cells obtained by rapid immersion of live cells in liquid nitrogen lose tumorigenicity while preserving transiently the integrity of the cell structure, which is critical for the drug loading and cargo release. Furthermore, certain critical functional proteins that include CD44 and CXCR4 were retained in LNT cells. CD44 can interact with hyaluronic acid that is highly expressed in the endosteum of bone marrow (18). CXCR4 enables cells to migrate toward the chemokine stromal cellderived factor 1 (SDF-1) that is constitutively produced by the osteoblasts and stromal cells (23). CD44 and CXCR4 are two important adhesion receptors mediating AML cells homing toward bone marrow (35). Retention of both CD44 and CXCR4 in LNT cells, even if at reduced levels compared with live cells, is likely critical to promote their bone marrow homing. The proposed LNT-based strategy is simple and straightforward from a manufacturing point of view. Tumor cells in the case of liquid tumors can be readily collected in large quantity, for example, by leukapheresis. Similarly, for solid tumors, multiple devises are currently available to generate single cell suspension from resected tumors or tumor biopsies. The process of cell shocking in liquid nitrogen is also feasible to standardize in good manufacturing practice conditions.

We evaluated the proliferation and tumorigenicity of LNT tumor cells both in vitro and in vivo. The data that all mice treated with LNT cells exhibited no obvious side effects and no leukemia growth was recorded for 6 months after inoculation of LNT C1498 cells support at least in our mouse model the safety of the proposed strategy. After exposure to liquid nitrogen, the cellular membrane of LNT cells becomes permeable. While live cells require treatment with cell membrane detergent to obtain intracellular staining, LNT cells do not require this treatment, indicating the loss of long-term integrity of the cell membrane, which is essential to cell survival. However, our experiments demonstrate that liquid nitrogen treatment does not impair the capacity of LNT cells to function as drug carrier and tumor vaccine. The cryo-shocking technique could be a platform technology in cell bioengineering and could be applicable to various cell types. Here, we have further tested feasibility in 4T1 tumor cells (fig. S10). Regarding the potential impact in clinical use, the safety of LNT tumor cells, besides C1498 cells adopted in this work, should be evaluated thoroughly in other experimental animal models. In addition, the application of LNT cells to serve as drug carriers of other therapeutics, such as immune checkpoint inhibitors (36), is worth investigating.

In summary, we engineered LNT tumor cells to serve simultaneously as a drug delivery carrier and cancer vaccine. The simple liquid nitrogen treating process abrogates the tumorigenicity of tumor cells but preserves the integrity of their cellular structure. This in turn allows the possibility to load LNT cells with chemotherapy drugs and preserves the homing capacity of these cells to the tumor site. LNT cells in combination with adjuvant could elicit both therapeutic and protective immune antitumor responses and may avoid the complex quality control associated with isolated cells and synthesized materialbased vehicles and enable large-scale production for clinical use.

The aim of this study was to use the cryo-shocked tumor cells as a kind of drug-targeting carrier and tumor vaccine for chemo-immunotherapy in the treatment of AML. After treating the live cells in liquid nitrogen, the cellular structure of the cryo-shocked cells was observed. The proliferation behavior, in vivo tumorigenicity, and targeting capability toward the bone marrow of the cryo-shocked cells were assessed. In vivo antitumor efficacy was analyzed in an AML model by intravenously injecting C1498 cells in C57BL/6J mice. Mice were randomly assigned to groups based on body weights. After different treatments, the mice were captured by in vivo imaging system (IVIS) to evaluate in vivo tumor progression. Survival curves, immune cell proportions, and cytokine levels were determined according to previous experimental experience. Specific information about treatment groups, sample numbers, and data analysis was denoted in the figure captions.

Doxorubicin hydrochloride was purchased from Fisher Scientific Co. (D4193; purity, >95%). Noncontrolled-rate cell cryopreservation medium was bought from Cyagen Co. (NCRC-10001-50). AML cell line C1498 was purchased from the American Type Culture Collection (ATCC). Luciferase and DsRed tagged C1498 cell line was provided by B. Blazar of the University of Minnesota. The cells were cultured in 90% Dulbeccos modified Eagles medium (Gibco) and 10% fetal bovine serum (Gibco) with penicillin (200 U ml1) and streptomycin (200 U ml1) (Gibco). The cells were passaged every 1 to 2 days. C57BL/6J mice (4 to 6 weeks, female) were purchased from the Jackson laboratory. All animal tests complied with the animal protocol approved by the Institutional Animal Care and Use Committee of the University of California, Los Angeles.

C1498 cells were centrifuged at 250g for 3 min and suspended in noncontrolled-rate cell cryopreservation medium at a cell density of 1 106 to 1 107 ml1. The cell-containing medium was immersed in liquid nitrogen for 12 hours. Before use, the medium was thawed at 37C and LNT cells were pelleted at 500g for 3 min. After washing with PBS solution (pH 7.4), LNT cells were suspended in PBS and kept at 4C. For preparation of DOX-loaded LNT cells, the LNT cells were suspended in DOX containing PBS. After incubation for 2 hours, the medium was centrifuged at 500g for 5 min and the pellets were DOX-loaded LNT cells.

The AML model was established by intravenous injection of 5 106 C1498 cells on day 0. On day 8 and day 15, saline, LNT cell + adjuvant, free DOX, and LNT cell/DOX + adjuvant were administrated intravenously with DOX dose of 5 mg/kg and adjuvant (MPLA) 20 g per mouse. Specifically, MPLA was intravenously injected 10 hours after injection of LNT cell or LNT cell/DOX. The bioluminescence images of mice were captured every 3 days. The exposure time was 2 min. On day 13, 400 l of blood was collected via the orbital vein. Blood (200 l) was treated with ammonium-chloride-potassium (ACK) buffer and centrifuged at 800g for 8 min to obtain pellets of white blood cells. After staining with BV421-CD3, PE-CD4, and APC-CD8, the samples were analyzed by flow cytometry. Another 200 l of blood in blood serum collection tubes (BD Microtainer 365967) was centrifuged at 3000 rpm for 10 min. The upper serum was detected with the following enzyme-linked immunosorbent assay kits: IFN- (BioLegend 430804) and TNF- (BioLegend 430904).

The results were presented as means SD or mean standard error of the mean (means SEM) as indicated. The data were compared by Students t test between two groups and ordinary one-way analysis of variance (ANOVA) for three or more groups. The survival curves were analyzed via the log-rank (Mantel-Cox) test. All statistical analyses were conducted by the GraphPad Prism software. The threshold of a statistically significant difference was defined as P < 0.05.

Acknowledgments: We acknowledge B. Blazar at the University of Minnesota for providing the luciferase and DsRed tagged C1498 cell line. Funding: This work was supported by the NIH (R01 CA234343-01A1) and grants from the start-up packages of UCLA. Author contributions: Z.G. and T.C. proposed the conception of the project. T.C. and H.L. performed all the experiments and collected the data with the help of G.C., Z.W., J.W., P.A., and Y.T. All authors analyzed the data and contributed to the writing of the manuscript, discussed the results and implications, and edited the manuscript at all stages. Competing interests: Z.G. and T.C. are inventors on a U.S. patent application related to this work filed by University of California, Los Angeles (no. 63/094,034, filed [Oct 20th, 2020]). Z.G. is a scientific cofounder of ZenCapsule Inc. The authors declare that they have no other competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

See the rest here:
Cryo-shocked cancer cells for targeted drug delivery and vaccination - Science Advances