Cancer is characterized by aberrant heme metabolism

Compared with normal cells, cancer cells have enhanced metabolic dependencies (Levine & Puzio-Kuter, 2010; Luengo et al, 2017; Martinez-Outschoorn et al, 2017). Our initial data analyses of genome-scale CRISPR/Cas9 gene loss of function from the publicly available project DepMap screens of cell lines derived from a set of cancer cell lines derived from diverse tissue origins (Meyers et al, 2017) indicated that cancer cells depend on heme synthesis (Fig S1) (Table S1). Significantly, these cancer cell lines developed dependencies on heme metabolism–related proteins, such as uroporphyrinogen III decarboxylase (UROD), the enzyme that catalyzes the fifth step in the heme biosynthetic pathway. These cancers also depend on a set of hemoproteins, such as cytochrome c-1 (CYC1), succinate dehydrogenase complex subunit C (SDHC) for cellular respiration, and DGCR8 for microRNA biogenesis (Nguyen et al, 2015), each of which uses heme as a co-factor. Unexpectedly, the ferrochelatase (FECH) gene encoding FECH that catalyzes the final critical step of heme biosynthesis is dispensable in several cancers, suggesting that cancer cell lines are capable of bypassing endogenous biosynthesis of heme in vitro, yet are still dependent on genes encoding enzymes that mediate intermediate steps of heme biosynthesis. This initial analysis of essentiality data indicated a greater dependency of these cancer cell lines on the mid-steps of biosynthesis genes, as opposed to the last step of the pathways. This observation suggests that the cancer pathways accumulate intermediates more than the end product, which is indicative of an “inefficient” pathway because of imbalanced synthesis enzyme steps. This stands in contrast to normal heme biosynthesis, which is well balanced, with no intermediate accumulation, and efficient conversion of substrates into products (Fig 1A).

Next, we investigated the genes that encode proteins acting as gatekeepers for heme biosynthesis. The first committed and key regulatory step in mammalian heme biosynthesis is catalyzed by 5-aminolevulinic acid (ALA) synthase (ALAS; EC 2.3.1.37) (Ponka et al, 2014). Two chromosomally distinct genes, ALAS1 and ALAS2, encode the housekeeping and erythroid-specific ALAS isoforms, respectively (Ponka et al, 2014). Although ALAS1 is expressed in every cell, the expression of ALAS2 is restricted to developing erythrocytes (Ponka et al, 2014). We examined the data generated from genome-wide CRISPR-based screens of diverse cancer cell lines, which were designed to provide a complete gene essentiality dataset for these cell lines (Meyers et al, 2017; Tsherniak et al, 2017). In contrast to ALAS2, the essentiality of ALAS1 is a feature of diverse cancer cells independent of cancer type (Fig S2A). Consistent with these forward genetic results, the gene expression patterns in ∼10,000 patient tumors from the Genotype-Tissue Expression (GTEx) (Lonsdale et al, 2013; The GTEx Consortium et al, 2015) and The Cancer Genome Atlas (TCGA) (Cancer Genome Atlas Research Network et al, 2013) datasets show that ALAS2 expression is absent in most tumors except myeloid leukemias (Fig S2B). These results show that despite ALAS2 being responsible for most heme production (∼85%) in humans, many cancer cell lines, including erythroleukemia of RBC lineages (e.g., HEL), require ALAS1 (Fig S2A–C).

Porphyrin overdrive operates in diverse cancers and early embryos

To better understand the molecular and genetic mechanisms underlying the cancer dependency on heme metabolism, we determined gene essentiality from genome-scale CRISPR/Cas9 loss-of-function screens in over 300 human cancer cell lines covering different cell lineages and estimated gene dependency (Meyers et al, 2017; Tsherniak et al, 2017). Then, we focused on the gene dependencies associated with the eight enzymatic steps of the heme biosynthetic pathway by computing the respective pan-cancer essentiality scores, similar to the lethality scores used in the development of a cancer dependency map (DepMap) (Dempster et al, 2019; Pacini et al, 2021). Gene essentiality score values lower than zero indicate decreased cell growth/viability with loss of gene X, and hence, the lower the X gene essentiality score, the more dependent are cells on the X gene. Consistent with recent cross-platform studies (Dempster et al, 2019; Pacini et al, 2021), over 2,000 genes were defined as pan-cancer–essential, based on the distribution of their essentiality scores in over 300 cancer lines derived from 27 of the most prevalent cancer types. The gene essentiality scores were confirmed by using the same method on the total DepMap collection (DepMap release 21Q3) of over 1,000 cell lines (Pearson’s R > 0.9, P < 0.001). Strikingly, instead of finding the expected complete and balanced heme biosynthetic pathway characteristic of normal cells, we found that the survival of cancer cells depends on the various steps of the heme biosynthetic pathway in a “non-balanced” manner. In 100 different types of cancers, the tumor cells particularly rely on the intermediate enzymatic steps of heme biosynthesis, as assessed by loss of cellular viability upon CRISPR/Cas9-mediated gene ablation in whole-genome loss-of-function screens (Fig S3) (Tables S2 and S3). Specifically, as shown in Figs 1B and S3, the pan-cancer gene essentiality is the highest for the genes encoding the enzymes responsible for the fifth and sixth steps of the pathway, UROD and coproporphyrinogen III oxidase (CPOX), respectively. This result was notably surprising, as the genes for the ALAS isoforms, which catalyze the initial and rate-limiting step in the pathway, and the gene for FECH, the final enzyme in the pathway that catalyzes heme production, had lower essentiality values and were dispensable in several cancers (Figs 1B and S3). Though the heme biosynthesis genetic dependences of 27 major forms of human cancer vary, the estimated patterns of essentiality across the cancer lines clearly revealed a range of heme and heme precursor requirements. These results uncovered the “unbalanced” nature of heme biosynthesis in cancer, with many cancers surviving without functional first and terminal enzymatic steps but depending on the intermediate steps (Fig S3). The gene essentiality analysis also highlighted the importance of heme trafficking (e.g., heme importer FLVCR2) and key hemoproteins (e.g., CYC1) in cancer cell survival (Figs 1B and S3).

To examine cancer heme metabolism in vivo, we analyzed the in vivo mouse CRISPR/Cas9 loss-of-function and metabolic essentiality data from pancreatic and lung cancer models (Zhu et al, 2021). The genetic screen (Zhu et al, 2021) indicates that most metabolic gene essentialities were similar between the in vitro and in vivo settings, except for the heightened requirement of the intermediate steps of heme biosynthesis in vivo. As reported in the previous study, of the ∼2,900 studied metabolic genes (Zhu et al, 2021), the genetic dependencies of both murine pancreatic and lung cancers were only increased for the enzymes responsible for the intermediate steps of the heme biosynthetic pathway (Fig S4A and B) (Table S4). Strikingly, our careful examination of each step of the heme biosynthetic pathway shows that the essentiality extent of either ALAS or FECH, which both code for the first and terminal enzymes of the heme biosynthetic pathway, respectively, did not differ significantly between the in vivo murine cancer models. Although the Zhu study interprets these results as enhanced heme production in vivo, our close analysis of the linear heme biosynthetic pathway reveals that, in fact, these CRISPR KO results show the significance of heme intermediates and the imbalanced pathway’s role in intermediate accumulation within an in vivo setting (Fig S4).

To assess heme biosynthesis in patient tumor samples, we analyzed data available from the GTEx project (Lonsdale et al, 2013; The GTEx Consortium et al, 2015) and TCGA program (Cancer Genome Atlas Research Network et al, 2013). We used methods designed for the pairwise comparative gene expression of GTEx and TCGA datasets (Tang et al, 2017). In over 80% of all tumors, the genes for hydroxymethylbilane synthase (HMBS), the third enzyme of the heme biosynthetic pathway, and the heme exporter FLVCR1 were among the most commonly up-regulated across diverse tumor types (Figs S5A and B and S6) (Table S5). By the same criteria, the onco-signaling genes AURKA, KRAS, and MYC were up-regulated in over 90%, 60%, and 50% of all tumor types, respectively. In contrast, the expression of the genes encoding the second and terminal enzymes of the heme biosynthetic pathway, aminolevulinate dehydratase (a.k.a. porphobilinogen synthase) and FECH, respectively, is down-regulated, suggesting the presence of a buildup of intermediate heme precursors in tumors. HPX, the gene for the heme-binding and scavenger hemopexin (Tolosano et al, 2010), was also down-regulated in most of the tumors (Fig S5C). This finding, which corroborates the previously reported role of hemopexin as a key player for the checkpoint in cancer growth and metastases (Canesin et al, 2020), leads, compared with the gene expression patterns in ∼10,000 tumors versus those in normal tissues, us to suggest that the heme content is high in the tumor microenvironment. ALAS2 belongs to the 10% of genes that were not expressed in most of the large tumor collections, except myeloid leukemias (<2% of total tumors). Overall, our comparative analysis of the differential gene expression in tumor tissues, spanning 31 cancer types, indicated that only a subset (e.g., third step) of the nine genes encoding the enzymes of the heme biosynthetic pathway were up-regulated in cancer cells, whereas the second-step or the last-step genes in fact are often down-regulated in tumors.

Based on the gene essentiality results from both in vitro (Figs 1B and S3) and in vivo (Fig S4) settings, the pan-cancer gene expression patterns (Figs S5 and S6), the detected porphyrin accumulation in a wide range of tumors (Fiorito et al, 2020), and the reported elevated heme flux in diverse cancer cells (Hanna et al, 2016), we defined the salient porphyrin overdrive–associated features. They are as follows: (1) heme biosynthetic enzymes with aberrantly increased (HMBS, UROS, and UROD) and suppressed (aminolevulinate dehydratase and FECH) activities, (2) accumulation of heme precursors (porphyrins), and (3) enhanced heme flux (Fig 1A and B). This cellular status contrasts with the tightly regulated steady state or homeostasis, defined by a flawless enzyme-catalyzed channeling of substrates to products along the heme biosynthetic pathway and circumventing the toxicity and instability of the pathway intermediates, while ensuring that the cell heme requirements are met (Ferreira, 2013). With porphyrin overdrive, the production of heme precursor molecules exceeds that of heme, porphyrin intermediates accumulate, and “imbalanced” heme biosynthesis of increased porphyrin intermediates and decreased end product arises. Likely, to compensate for the reduced amount of heme synthesized, heme flux is increased (Fiorito et al, 2020). Cancer cells appear to have adopted a dependence on this seemingly “inefficient” metabolic pathway designed to produce aberrant levels of heme intermediates with unknown biological functions to date.

Early embryonic stem cells represent a special metabolic state that bears similarities with that of cancer cells (Smith & Sturmey, 2013). Therefore, we also studied the expression of the heme biosynthetic enzyme–encoding genes in human preimplantation embryos and embryonic stem cells. Specifically, we explored a high-resolution dataset generated upon single-cell RNA sequencing in human early embryonic cells and embryonic stem cells (Yan et al, 2013; Li et al, 2017). In the human zygote and early preimplantation embryos, the level of ALAS1 expression is 100-fold higher than that of FECH (Fig 1C) (adjusted P < 0.001). This large difference in gene expression for two key enzymes in heme biosynthesis was observed in all 17 early human embryos examined (Table S6). The large difference in expression levels is restricted to very early embryos because it rapidly narrows, via an increase in FECH expression and a decrease in ALAS1 expression after initial rounds of embryonic cell division. ALAS2 expression is not detected in any of the early human embryos, which is congruent with the non-essentiality of ALAS2 in all examined cancers. As a control, the genes AURKA and SOX4 show expected expression patterns during early embryogenesis (Penzo-Méndez et al, 2007; Sasai et al, 2008). The similarity of these gene expression data to those observed in cancer cells supports the possibility that a form of porphyrin overdrive operates in preimplantation human embryos.

Porphyrin overdrive is absent in normal cells

To evaluate whether porphyrin overdrive is specific to cancer cells (and possibly preimplantation embryos), we investigated and compared heme biosynthesis in diverse types of non-cancerous cells, that is, normal differentiated cells, replicating fibroblasts, human primary hepatocytes, and human primary hematopoietic stem cells. We looked for evidence of porphyrin overdrive, defined by the non-homeostatic, imbalanced heme biosynthetic pathways and heightened accumulation of heme pathway intermediates. As existing literature predominantly identifies PPIX as the major accumulated form in both in vitro and in vivo settings (Szeimies et al, 1995; Moesta et al, 2001; Bellini et al, 2008; Kaneko & Kaneko, 2016), in this study, we refer to PPIX as the representative porphyrin species. Future studies will conduct a more detailed biochemical analysis to identify specific heme intermediate forms in different cancer cell types.

Using published datasets, we found that porphyrin overdrive is absent in normal differentiated cells and somatic stem cells (Fig 2A). First and consistent with this finding, normal human PBMCs do not accumulate heme intermediates even upon ALA induction (Oka et al, 2018). Second, colon cancer cells incubated with ALA to induce protoporphyrin IX (PPIX) production to accumulate PPIX, whereas no significant PPIX accumulation is observed in replicating human colon–derived fibroblasts (stromal cells) previously incubated with ALA (Krieg et al, 2002). Furthermore, the increased HMBS activity in colon cancer cells relative to stromal cells (Krieg et al, 2002) is consistent with a buildup of heme intermediates including PPIX. Third, using our previously described primary human hepatocyte system (Maher et al, 2020), we demonstrate that normal, metabolically active, primary hepatocytes do not accumulate porphyrins as indicated by the absence of the PPIX fluorescence (Fig 2B). In contrast, >99% of liver cancer cells produced and accrued PPIX upon induction with ALA, demonstrating a canonical feature of porphyrin overdrive.

To uncover the role of heme metabolism in normal human stem cells, we analyzed the erythropoiesis data generated by RNAi-based knockdown of gene expression in human hematopoietic progenitor cells (Egan et al, 2015). By mapping normal stem cell gene essentiality following the methods described in Egan et al (2015), we show that erythropoiesis is heme-dependent in both early progenitor (undifferentiated) cells and differentiated cells from normal human bone marrow–derived hematopoietic stem cells. In contrast to cancer cells, where ALAS2 is dispensable, normal human hematopoietic stem cells depend on ALAS2 for survival (Fig 2C) (Fig S7). The differential gene essentiality profiles between cancer and stem cells are consistent with the molecular basis for specific ex vivo purging of cancer hematopoietic progenitor cells, via ALA induction/photodynamic therapy (PDT) as a way to eradicate malignant cells while sparing hematopoietic stem cells (de Lima & Shpall, 2004). Taken together, the distinct genetic essentialities and the markedly different porphyrin accumulation modes between cancer and non-cancerous cells indicate that porphyrin overdrive is absent in somatic stem cells.

CRISPR/Cas9-mediated KO of ALAS2 and FECH validates abnormal heme metabolism underlying cell proliferation and oncogenesis

We modified chronic myeloid leukemia K562 cells with CRISPR/Cas9-mediated KO of the genes encoding FECH and ALAS2 (Fig 2D). Under a heme homeostasis model, a cell line without FECH should not survive, because FECH is a single-copy gene in humans and with no functional substitute (Ponka et al, 2014). Indeed, knockdown of FECH in human stem cells led to cell death (Egan et al, 2015). However, under porphyrin overdrive conditions, we predicted that cancer cells rely on heme trafficking and are addicted to the porphyrin intermediates, and thus do not depend on the final step of heme biosynthesis for survival. As hypothesized, K562 cells with the FECH edited out (K562-FECH KO) (Fig 2D) had normal growth (Fig 2E) under standard culture conditions. The unimpeded growth of the K562-FECH KO cells presumably indicates that at least some cancer cell lines can metabolically function as heme auxotrophs.

As predicted, K562 cells with the ALAS2 deleted (K562-ALAS2 KO) (Fig 2D) lost their differentiation capacity although their growth was not hindered (Fig 2E and F). Given the proposal that oxidation stress, with the involvement of reactive oxygen species (ROS), is the first step in erythroid differentiation of K562 cells (Chenais et al, 2000), we used the organic peroxide tert-butyl hydroperoxide (t-BHP) to trigger ROS-mediated erythroid differentiation. As early as 30 min after t-BHP–mediated ROS induction, the parental K562 cells started synthesizing heme, an early step in cellular differentiation (Chenais et al, 2000), whereas the K562-ALAS2 KO cells failed to initiate this metabolic step, indicated by the lack of heme production (Fig 2F) (Chenais et al, 2000). Furthermore, at 48 h, the parental K562 cells readily differentiated into erythroid cells, but the K562-ALAS2 KO cells continued to proliferate and failed to differentiate (Fig S8), as expected because ALAS2 is required for erythroid differentiation (Fig S7). Conceivably, in the absence of endogenous production of heme destined to hemoglobin, the erythroid lineage commitment is blocked, forcing the K562-ALAS2 KO cells to an obligatory porphyrin overdrive path and into a proliferative/non-differentiable, or cancerous, state (Fig S8).

Single-cell transcriptomics reveals porphyrin overdrive in acute myeloid leukemia (AML) patient biopsies

To study porphyrin overdrive in primary cancer cells, we performed single-cell RNAseq on biopsies from AML patients (Tables S7 and S8). For each patient, we obtained around 2,000 single-cell transcriptomes, and we analyzed cells of various differentiation stages and cell types. Interestingly, in two patients, only a minority of cells (5–10% of total cells) were identified as cancer progenitor cells with hyperproliferation features (Fig 3A), and they exhibited hybrid features of mature erythroid cells (e.g., expressing the genes HBB and GYPA for hemoglobin β and glycophorin A, respectively) and stem cells (e.g., expressing SOX4). The cancer progenitor cells (Fig 3B) had both mature cell features, for example, HBB expression, and early progenitor properties, for example, proliferative phenotype. Specifically, they were characterized by elevated iron and heme metabolic gene expression, cell proliferation markers, KRAS, and anabolism process genes, yet they had reduced normal stem cell progenitor marker gene expression, for example, SOX4 and JUN (Fig 3C) (Merryweather-Clarke et al, 2016). In addition, cancer/AML progenitor cells exhibited imbalanced heme biosynthesis, inferred and exemplified by elevated gene expression for enzymes that catalyze the intermediate steps of heme biosynthesis, for example, HMBS and UROD (Fig 3D and E). The cancer progenitor cells also had a higher expression level of genes related to heme trafficking such as FLVCR1 and DNM2 (Fig 3C). Our results show that these cancer progenitor cells exhibit features of porphyrin overdrive, consistent with the abnormal heme anabolism and endogenous PPIX accumulation upon ALA administration in leukemia cells compared with normal PBMCs (Oka et al, 2018). Notably, in two solid tumors (Fig 3F) with independently generated single-cell transcriptomes (Tirosh et al, 2016; Wu et al, 2020), the genes encoding the enzymes for the intermediate steps of heme biosynthesis were similarly overexpressed in the cancer progenitor cells (Fig 3F).

From our single-cell transcriptome results, we searched for clues of specialized oncogenic microenvironments computationally that could support enhanced substrate trafficking and molecular interactions. Interestingly, only the cancer progenitor cells highly express cell-to-cell interaction-related genes, for example, ICAM4, MAEA, ITGA4 (Fig S9A), which are critical in establishing the stem cell niche for erythropoiesis (Manwani & Bieker, 2008). We infer that the specific expression of these niche interaction genes suggests that the cancer progenitor cells exist in a specialized niche with intimate contact with neighboring cells and extracellular matrix, and establish evidence for future characterization of this unknown microenvironment.

Expression was remarkably enhanced for genes encoding proteins involved in metabolic substrate trafficking processes in the cancer progenitor cells, including genes for key protein players in lipid import purine and pyrimidine salvage pathways (Fig S9B). These specific metabolic flux–related genes, such as DNM2 (endocytosis), APOC1 (lipid transport), MRI1 (L-methionine salvage), MTAP (NAD salvage), and SLC29A1 (nucleoside import), indicate that the cancer progenitor cells represent hubs of metabolic trafficking activities. In particular, the lipid import gene APOC1 is expressed in cancer progenitor cells, but not in the other cell populations, including normal progenitor cells, from the same patient samples (Fig S10A and B). The metastatin S100 family members linked to tumor niche construction (Lukanidin & Sleeman, 2012) are also enriched in the progenitor cells. These potential dynamic metabolic trafficking patterns inferred from the abundant expression of metabolite import genes in cancer early progenitors (Fig S9C) are reminiscent of “metabolic parasitism” (Icard & Lincet, 2013), a bioenergetically favorable process that uses energy-rich and preexisting macromolecules to fuel robust cell proliferation.

Targeting cancer porphyrin overdrive with a bait-and-kill strategy

Two aspects of cancer porphyrin overdrive offer opportunities for therapeutic intervention. First, an imbalanced heme biosynthetic pathway (i.e., with aberrantly increased and suppressed enzyme activities) is absent in normal cells. Second, porphyrin overdrive is inducible 100-fold to 1,000-fold in cancer cells, but not in non-cancerous cells. In fact, PDT, which uses ALA to induce biosynthesis and accumulation of photosensitizing porphyrins, and exemplifies the inducible principle of porphyrin overdrive, has been used for cancer imaging and treatment for decades (Krammer & Plaetzer, 2008). Thus, porphyrin overdrive may be a unique metabolic reprogramming feature of cancer cells that can be therapeutically exploited. Selective cancer cell death could possibly be achieved by activating PPIX-induced cytotoxicity with a two-step “bait-and-kill” strategy.

Knowing that cancer cells readily take up the heme precursor ALA and bypass the first and rate-limiting step of porphyrin biosynthesis catalyzed by ALAS, we set to assess our engineered two-step process for cancer metabolic death. This two-step process involves, first, to “bait” cancer cells with ALA to induce PPIX accumulation, and, second, to “kill” the cancer cells with a compound that exploits the metabolic stress associated with porphyrin accumulation. As expected, ALA addition to the culture medium did enhance PPIX buildup in K562 (Fig 4A) and HEL (Fig 4B) cells, as demonstrated by the few 100-fold to 1,000-fold PPIX concentration increase in relation to respective cells cultured in the absence of ALA (Fig 4A and B). Furthermore, HEL cell viability was not affected by supplementing the culture medium with ALA (1 mM final concentration), or ALA and DMSO (1 mM and 0.1% final concentrations, respectively) (Fig 4C). These results validated that the experimental criteria had been met to proceed with the second step of the “bait-and-kill” strategy.

To exploit the metabolic stress associated with porphyrin accumulation (e.g., oxidative stress, lipid peroxidation), we treated the human erythroleukemia HEL cell line with a range of concentrations of RSL3, an inhibitor of the cells’ endogenous antioxidant defense system that protects cells against membrane lipid peroxidation (Cheff et al, 2023). As such, RSL3 induces ferroptosis, an iron-dependent form of programmed cell death highly relevant to iron and heme metabolism. Pretreatment of HEL cells with ALA increased their sensitivity toward RSL3 at least 1,000-fold (Fig 4D). Even at a RSL3 concentration of 100 mM, the viability of HEL cells remained similar to that of cells cultured in the absence of the RSL3 inhibitor. However, the viability of HEL cells grown in the presence of ALA for 4 h diminished with RSL3, reaching non-viability at a concentration of 17 mM. The decline was yet more acute when the incubation time of the HEL cells with ALA was extended to 24 h, and the cells became non-viable with 16 nM RSL3. Similar strong synergistic cancer-killing results, using ALA as “bait,” and RSL3 and/or artemisinin as “kill,” were found in murine BaF2_Jak2mt leukemic cells and human liver cancer cell line HC04 (Fig 4E and F). In contrast, ALA pretreatment did not sensitize cells to ganetespib, a heat shock protein inhibitor with antineoplastic activity (Alexandrova et al, 2015), suggesting that ALA is not a general inducer or sensitizer of cell death. Specifically, at ganetespib concentrations greater than 33 mM, HEL cells persist viably regardless of being grown in the presence or absence of ALA (Fig 4G). Notably, a healthy human primary lung fibroblast does not show PPIX accumulation (not shown), nor have treatment toxicity under the same conditions to eradicate cancer cells (Fig 4H and I).

In conclusion, we report that the essentiality pattern of the genes for the enzymes of the heme biosynthetic pathway varies significantly between human healthy and cancer cells. With the genes for the third, fourth, and fifth enzymes being particularly critical for cancer cell proliferation, accumulation of porphyrin intermediates exceeds heme production. We propose that cancer cells, in an attempt to counteract unbalanced heme biosynthesis, scavenge heme from their environment and deploy vigorous heme trafficking strategies. We designate the combined non-homeostatic heme synthesis and enhanced heme trafficking as porphyrin overdrive. We suggest that porphyrin accumulation because of porphyrin overdrive and the resulting oxidative stress present a vulnerability of cancer cells that may be amenable to novel therapeutic strategies to specifically destroy cancer cells, as exemplified by our proof-of-principle results (Fig 4A–I).

Cell culture

The human chronic myelogenous leukemia cell line K562 (Cat No. CCL-243; ATCC) and human erythroleukemia cell line HEL 92.1.7 (Cat No. TIB-180; ATCC) were obtained from the American Type Culture Collection. Both the cells were grown in RPMI 1640 medium (Gibco) supplemented with 10% FBS (Sigma-Aldrich), gentamicin (1,000x, Cat No. 15-710-072; Thermo Fisher Scientific), Pen-Strep-Neo solution (100x, Cat No. 15640-055; Thermo Fisher Scientific), and 2 mM L-glutamine at 37°C in a humidified 5% CO₂ atmosphere. 5-Aminolevulinic acid hydrochloride (ALA), purchased from Alfa Aesar (Cat No. A16942ME), was dissolved in distilled water to yield a stock concentration of 1.0 M, and stored at −20°C. Glycine, purchased from Thermo Fisher Scientific (Cat No. BP381-500), was dissolved in phenol red–free culture medium purchased from Gibco (Cat No. 11835055) to give a stock concentration of 1 M.

Cell growth and quantification

K562 or HEL cells were plated at 1–2 × 10⁵/ml, and cells were counted by trypan blue (Cat No. 25-900-CI; Corning) exclusion over time, with passage back to the starting cell density as needed. Total cell numbers were determined based on the passage dilution at each time point, on a disposable hemocytometer (Cat No. DHC-N01; INCYTO).

Generation of KO cell lines by CRISPR

K562 cells containing deletions in the ALAS2 or FECH genes were created using a multi-guide strategy via nucleofection (Lonza) of Cas9/RNP complexes (Gene KO Kit v2; Synthego) following the manufacturer’s instructions. The guide RNAs used were as follows: ALAS2: UGAAGGCUUUCAAGACAGGU, CAAUCUUGCUCUUCCCAUCC, and AGAUUCUCCAUCUUGGGCGA; FECH: UUAGACUCAUACCUCUUCUG, CUGGGCUGUUUCUGUGGUGA, and CUGACAGACCCUCCAGCUGC. CRISPR/Cas9 deletions were confirmed after amplification of the genomic region of interest, Sanger sequencing of the amplified region, and Inference of CRISPR Edits analyses (Synthego). Cells were lysed in RIPA buffer with protease and phosphatase inhibitors and immunoblotted with antibodies that recognize FECH (SC-377377) and vinculin (SC-73614) (Santa Cruz Biotechnology). PCR amplification of the first-strand cDNA from K562 and K562-ALAS2 KO was used to confirm the presence of the homozygous 107-bp out-of-frame deletion that was detected in the Inference of CRISPR Edits analyses of the edited genomic region in K562-ALAS2 KO cells.

In vitro erythropoiesis protocols

In vitro erythropoiesis was monitored by following cellular differentiation of the cell lines (K562 and K562-ALAS2 KO) with benzidine staining. The benzidine solution was prepared by mixing 5 ml of 30% hydrogen peroxide (H₂O₂) with 1 ml of 0.2% benzidine dihydrochloride in 0.5 M acetic acid. Around one million cells were washed thrice with Dulbecco’s phosphate-buffered saline (DPBS, 1X, Cat No. 21-031-CV; Corning) before being resuspended in 250 μl of 1x DPBS mixed with 250 μl benzidine solution and incubated at room temperature for 10 min. The cells stained brown-blue were visually recognized and counted as positives, on a disposable hemocytometer (Cat No. DHC-N01; INCYTO). The experiments were conducted in triplicates, and multiple microscopic fields were counted.

AML patient single-cell RNAseq and heme biosynthetic pathway–related gene expression analysis

Cells were obtained from human donors with over 60% blast expansions in marrow biopsies. Blast cells were washed and isolated according to the standard tissue-banking protocols. Cells were carefully washed in DPBS (1X, Cat No. 21-031-CV; Corning) and resuspended at 10⁶ cells/ml to avoid cell aggregates. Cells were processed using the 10x Genomics Chromium controller and the Chromium Single Cell 3′ Library and Gel Bead kit (10x Genomics, Cat No. PN-1000075) following the standard manufacturer’s protocols. First, gel beads-in-emulsion (GEMs) were generated by combining barcoded single-cell 3′ v3 gel beads, a master mix containing cells, and partitioning oil onto Chromium Chip B. To achieve single-cell resolution, cells were delivered at a limiting dilution, such that most (∼90–99%) of the generated GEMs contain no cells, whereas the remainder contain predominantly a single cell. Between 2,000 and 21,000 live cells were loaded onto the Chromium controller to recover between 1,500 and 15,000 cells for library preparation and sequencing. Immediately after GEM generation, the gel beads were dissolved, primers were released, and any co-partitioned cells were lysed. An Illumina TruSeq Read 1 (read 1 sequencing primer), 16-nt 10x barcode, 12-nt unique molecular identifier, and 30-nt poly(dT) sequence were mixed with the cell lysate and a master mix containing reverse transcription (RT) reagents. Incubation of the GEMs produces barcoded, full-length cDNAs from polyadenylated mRNAs. Next, GEMs were broken, and cDNA was amplified and quantified using Agilent High Sensitivity DNA ScreenTape (Cat No. 5067-5592; Agilent Technologies). SPRIselect magnetic beads (Cat No. B23317; Beckman Coulter) were used to purify the first-strand cDNA from the post–GEM-RT reaction mixture, which included leftover biochemical reagents and primers. The barcoded, full-length cDNA was amplified via PCR to generate sufficient mass for library construction. Enzymatic fragmentation and size selection were used to optimize the cDNA amplicon size. TruSeq Read 1 (read 1 primer sequence) was added to the cDNAs during GEM incubation. P5, P7, a sample index, and TruSeq Read 2 (read 2 primer sequence) were added via end repair, A-tailing, adaptor ligation, and PCR. The final library quality was assessed using an Agilent Bioanalyzer high-sensitivity chip. Samples were then sequenced on the Illumina NextSeq 550 with a target of 150,000 reads/cell.

The Cell Ranger Single-Cell Software Suite (10x Genomics) was used for data processing, sample demultiplexing, and gene expression quantification. For data analysis, genes with more than one unique molecular identifier count were used. The top 1,000 most variably expressed genes were used for further clustering, and t-distributed stochastic neighbor embedding analysis was performed to reduce the data dimension to a two-dimensional space, and k-means clustering was used to identify cell populations. The mean expression of genes in all cells in a given cluster was calculated, and the expression of each gene was compared with that of the same gene in all the other clusters. For cell classification, the mean expression profiles of all cells were first calculated, and then, each cell was assigned to a subpopulation by the highest Spearman correlation.

Heme biosynthetic pathway–related gene expression analysis

The gene expression patterns related to heme biosynthesis from patient-derived tumor samples were analyzed using the publicly available resources to study tissue-specific gene expression, the GTEx project (Lonsdale et al, 2013; The GTEx Consortium et al, 2015), and TCGA program (Cancer Genome Atlas Research Network et al, 2013). The gene expression patterns of the heme biosynthetic pathway were investigated in a comprehensive analysis involving 10,000 tumors and matched normal tissues. This analysis covered a diverse range of major tumor types sourced from TCGA and GTEx projects. The examined tumor types included acute myeloid leukemia (LAML), adrenocortical carcinoma, bladder urothelial carcinoma, brain lower grade glioma, breast invasive carcinoma, cervical squamous cell carcinoma and endocervical adenocarcinoma, cholangiocarcinoma, chronic myelogenous leukemia, colon adenocarcinoma, controls, esophageal carcinoma, FFPE Pilot Phase II, glioblastoma multiforme, head and neck squamous cell carcinoma, kidney chromophobe, kidney renal clear cell carcinoma, kidney renal papillary cell carcinoma, liver hepatocellular carcinoma, lung adenocarcinoma, lung squamous cell carcinoma, lymphoid neoplasm diffuse large B-cell lymphoma, mesothelioma, miscellaneous, ovarian serous cystadenocarcinoma, pancreatic adenocarcinoma, pheochromocytoma and paraganglioma, prostate adenocarcinoma, rectum adenocarcinoma, sarcoma, skin cutaneous melanoma, stomach adenocarcinoma, testicular germ cell tumors, thymoma, thyroid carcinoma, uterine carcinosarcoma, uterine corpus endometrial carcinoma, and uveal melanoma. An index, reflecting the genes predominantly up-regulated in cancers, was computed by calculating the paired differences between tumor and normal tissues within each major tumor type. Mann–Whitney’s tests were performed, and P-values were adjusted for multiple testing using the Benjamini–Hochberg correction.

CRISPR pan-cancer gene essentiality analysis

The data from the DepMap Portal were used to calculate the essentiality scores in a similar manner to published methods (Aguirre et al, 2016; Meyers et al, 2017; Kim & Hart, 2021). The whole-genome CRISPR/Cas9 datasets were used to identify significantly depleted mutant cells bearing a specific gene knockout in a pooled experiment. Gene essentiality was estimated from a given gene dependency inferred from CRISPR/Cas9 gRNA gene KO. The essential score was used to evaluate the cell growth fitness. The lower the essentiality score value, the larger the gene loss effect on cell viability. Thus, a score of 0, < 0, and > 0 indicates no fitness change, fitness loss, and fitness gain (i.e., possible growth advantage for the cell line) under the assay conditions, respectively. The method as described in Aguirre et al (2016); Kim & Hart (2021) was employed to correct the copy-number bias in whole-genome CRISPR/Cas9 screens by computing the mean of sgRNAs versus the control plasmid library. Commonly essential genes were required for the fitness of most cell lines across cancer types (Dempster et al, 2019; Pacini et al, 2021). For in vivo gene essentiality analysis in pancreatic and lung cancer models, the published data in Zhu et al (2021) were used to specifically examine the genes encoding the heme biosynthetic pathway enzymes and heme transporters. We conducted Mann–Whitney’s tests comparing the essentiality of the mid-step gene UROD with that of the first- and last-step genes. This analysis showed significant differences (P < 0.001) in essentiality patterns among different genes in the same biosynthetic pathways.

In vitro primary human hepatocyte culture, quantification, and imaging

Sterilized 384-well plates (Cat No. 781091; Greiner) were unpackaged in a class II biosafety cabinet and placed in a secondary container (i.e., plates were placed in large assay pans) to serve as a lid and control evaporation. The day before hepatocyte seeding, wells were collagen-coated with 40 μl of 15 μg μl⁻¹ rat tail collagen I (Cat No. 354236; Corning) in sterile-filtered 0.02 M acetic acid (Cat No. AC456850050; Thermo Fisher Scientific), and kept at 37°C overnight. Immediately before seeding, the wells were washed thrice with sterile PBS and then filled with 20 μl in vitro GRO CP plate medium (Cat No. Z99029; BioIVT) supplemented with 1x Pen-Strep-Neo solution (100x, Cat No. 15640-055; Thermo Fisher Scientific) and 20 μM gentamicin (1,000x, Cat No. 15-710-072; Thermo Fisher Scientific). Vials of cryopreserved (male) primary human hepatocytes (Cat No. M00995-P; BioIVT) were thawed by immersion in a 37°C water bath for 2 min and sterilized with 70% ethanol in a sterile field, and the contents were added directly to 4 ml plate medium. Live and dead cells were quantified by trypan blue exclusion on a Neubauer-improved hemocytometer. The hepatocyte density was set at 1 × 10³ live cells μl⁻¹, and 18 μl cell suspension was added to each well. Medium was exchanged with the GRO CP plating medium, described above, thrice weekly. The cells were incubated with 1.0 mM ALA at 37°C for 4 h. Both ALA-treated and non-treated cells were handled under very low light conditions. During the last 45 min of incubation, a staining solution diluted in phenol-free, serum-free RPMI (Cat No. 11835055; Gibco) containing Hoechst 33342 (Cat No. H3570; Life Technologies) at a final concentration of 10 μM was added to the cells. Live-cell imaging was performed on CellInsight CX7 High-Content Screening Platform (Thermo Fisher Scientific), and each plate well was counted for hepatocyte nucleus staining.

In vitro hepatocyte cell line HC-04 culture, quantification, and imaging

Cryopreserved HC-04 hepatocyte cells were thawed, suspended into a previously prepared hepatocyte culture medium, and transferred to a T75 flask coated with collagen (Cat No. 354236; Corning) at 5 μg/cm². The previously prepared hepatocyte cell line culture medium consisted of a mixture of F12 base medium (Cat No. 11765-054; Invitrogen) and MEM base medium (Cat No. A10490-01; Invitrogen) on 1:1 (vol/vol) ratio, containing 10% FBS (Cat No. SH30070; HyClone), 1.0 M Hepes (Cat No. 15630-080; Invitrogen), and 200 mM glutamine (Cat No. 25030-081; Invitrogen) (Sattabongkot et al, 2006). Cells were allowed to grow until reaching 70% confluence, and the medium was changed every other day. Then, the cells were trypsin-hydrolyzed with TrypLE Express Enzyme (1X) (Cat No. 12605028; Gibco) and washed with a hepatocyte culture medium. The HC-04 cells were seeded at a density of 6,000 cells/well and cultured in 384-well plates (Cat No. 781091; Greiner) in 20 μl of the above medium/well. Cells were incubated either in the absence or in the presence of 1.0 mM ALA at 37°C for 4 h. Both ALA-treated and non-treated cells were handled under very low light conditions. During the last 45 min of incubation, a staining solution diluted in phenol-free, serum-free RPMI (Cat No. 11835055; Gibco) containing Hoechst 33342 (Cat No. H3570; Life Technologies) at a final concentration of 10 μM was added to the cells. Live-cell imaging was performed on CellInsight CX7 High-Content Screening Platform (Thermo Fisher Scientific), and each plate well covering 15 fields at 20x was counted for hepatocyte nucleus staining.

Cellular PPIX quantification

Protoporphyrin (PPIX) accumulation was determined using FACS as previously described (Fratz et al, 2014). K562 and K562-ALAS2 KO cells were independently seeded in 24-well plates for suspension culture (Cat No. 662102; Greiner) overnight at a density of 1.2 × 10⁵ cells/well. The K562 and K562-ALAS2 KO cells were incubated in a medium (described in the Cell culture section) containing 1.0 mM ALA, at 37°C for 4 h. Preparation of either 5-ALA–treated cells or control (non–ALA-treated) cells was under very low light conditions. After incubation, cells were washed thrice with DPBS (1X, Ca²⁺- and Mg²⁺-free, Cat No. 21-031-CV; Corning) and resuspended in 250 μl of 1x DPBS. Briefly, cells were washed once with serum-free medium (Cat No. 11835055; Gibco) and incubated in six-well plates containing the medium (described in the Cell culture section) in either the absence or the presence of 100 mM glycine or ALA (0.1, 0.25, 0.5, and 1.0 mM) at 37°C. Intracellular PPIX concentration was measured 18 h later by FACS. The cell suspension was passed through a 40-μm Flowmi cell strainer to eliminate clumps and debris before transferring to BD Falcon tubes under very low light conditions to minimize phototoxicity caused by PPIX accumulation. FACS analyses were performed using BD LSR II Analyzer (Becton, Dickinson and Company) and FACSDiva version 6.1.3 software. To eliminate any background red fluorescence, the 633-nm red laser was blocked during the collection of the PPIX emission data. PPIX emission was determined in the 619-nm and 641-nm range (630/22 BP filter) upon excitation of the cells with the 405-nm laser. Forward-scatter versus side-scatter dot plots were used to gate the whole cells and thus remove the contribution of the cell debris from the population being examined. A minimum of 10,000 of the gated cells were then depicted in dot plots of side-scatter versus PPIX fluorescence; the gate was defined based on wild-type K562 cells without any perturbation as negative controls.

Cellular ROS detection and cell viability assays

Both K562 and K562-ALAS2 KO cultures were seeded in 24-well plates for suspension culture (Cat No. 662102; Greiner) at a density of 1.2 × 10⁵ cells/well in the medium described in the Cell culture section, at 37°C overnight. After the incubation, cells were washed thrice with DPBS (1X, Ca²⁺- and Mg²⁺-free, Cat No. 21-031-CV; Corning), resuspended in 250 μl of 1x DPBS, and stained for 30 min at 37°C with 500 nM CellROX Green Reagent (Cat No. C10444; Life Technologies). During the last 15 min of staining, 1 μl of the 5 μM SYTOX Blue Dead Cell stain (Cat No. S34857; Life Technologies) was added to distinguish live from dead cells. Cells were immediately analyzed by FACS using BD LSR II Analyzer (Becton, Dickinson and Company) and FACSDiva Version 6.1.3 software. Emission of the oxidized fluorogenic CellROX Green was measured at 525 nm (530/30 BP filter) upon excitation of the cells using the 488-nm laser. K562 cells treated with 250 μM t-BHP (Cat No. 180340050; Thermo Fisher Scientific) for 15 min were used as a positive control. K562 cells were treated with 1 mM ALA for 4 h as a “PPIX fluorescence positive control.” Both ALA-treated and non-treated cells are handled under very low light conditions.

Drug (“bait-and-kill”) assays

The compounds RSL3 (Cat No. HY-100218A; MedChemExpress) and ganetespib (Cat No. HY-15205; MedChemExpress) were dissolved in 100% DMSO (Cat No. 4-X; ATCC) to yield 10-mM stocks, which were stored at −80°C until further use. HEL cells were seeded in 38-well plates (Cat No. 781091; Greiner) at a density of 6,000 cells/well and in 20 μl of medium (described in the Cell culture section) per well. The cells were allowed to proliferate for the next 48 h. Both ALA-treated and non-treated cells were handled under very low light conditions throughout the assay. The cells were treated with 1 mM ALA in individual plates for 4 and 24 h. Both ALA-treated and non-treated plates were tested with the compounds RSL3 and ganetespib in triplicate wells using an 18-point concentration format with twofold dilutions (final concentrations of 132 μM to 1 nM) bringing the total volume to 25.6 μl per well. Cell proliferation was measured using the CellTiter-Glo 2.0 reagent (Cat No. G9243; Promega) to quantify cellular ATP according to the manufacturer’s instructions by adding 25.6 μl of the reagent per well. Luminescence was measured with CLARIOstar Plus Microplate Reader (BMG Labtech). For each assay plate, a DMSO control (0.1%), a positive control, a negative control, and blanks were added, and a minimum of 12 wells per plate were analyzed. Data were reported as arbitrary luminometric units.