Abstract

   Limited oxygen (hypoxia) in solid tumors poses a challenge to
   successful immunotherapy with natural killer (NK) cells. NK cells have
   impaired cytotoxicity when cultured in hypoxia (1% oxygen) but not
   physiologic (>5%) or atmospheric oxygen (20%). We found that changes to
   cytotoxicity were regulated at the transcriptional level and
   accompanied by metabolic dysregulation. Dosing with interleukin-15
   (IL-15) enhanced NK cell cytotoxicity in hypoxia, but preactivation
   with feeder cells bearing IL-21 and 4-1BBL was even better.
   Preactivation resulted in less perturbed metabolism in hypoxia; greater
   resistance to oxidative stress; and no hypoxia-induced loss of
   transcription factors (T-bet and Eomes), activating receptors, adhesion
   molecules (CD2), and cytotoxic proteins (TRAIL and FasL). There
   remained a deficit in CD122/IL-2Rβ when exposed to hypoxia, which
   affected IL-15 signaling. However, tri-specific killer engager
   molecules that deliver IL-15 in the context of anti-CD16/FcγRIII were
   able to bypass this deficit, enhancing cytotoxicity of both fresh and
   preactivated NK cells in hypoxia.
     __________________________________________________________________

   Activation strategies alter the metabolic profile of human NK cells to
   overcome defects driven by hypoxic tumor microenvironments.

INTRODUCTION

   Natural killer (NK) cells are innate cytotoxic cells that can eliminate
   cancer cells without prior sensitization. They have great potential for
   cancer immunotherapy because they recognize tumor cells through an
   array of germline encoded receptors, including Fc receptor CD16, which
   drives antibody-dependent cellular cytotoxicity (ADCC); and they
   activate broad immune responses through secretion of cytokines ([68]1,
   [69]2). Notably, NK cell–based therapies have limited side effects
   ([70]3–[71]5). There are two main areas of NK cell–based immunotherapy:
   treatments that enhance endogenous NK cell function, such as the
   addition of cytokines, monoclonal antibodies, checkpoint inhibitors and
   NK cell engagers ([72]6–[73]8); and adoptive transfer of primary NK
   cells or expanded NK cells to make hundreds of doses as an “off the
   shelf” therapeutic product ([74]9–[75]11). These immunotherapies have
   shown promise in hematological malignancies, but their efficacy in
   solid tumors is limited ([76]12). The solid tumor microenvironment
   (TME) has a wide variety of immunosuppressive mechanisms ([77]13), but
   a key challenge is the lack of oxygen, or “hypoxia,” driven by several
   tumor-related factors ([78]14). The severity of oxygen deprivation
   varies within and across tumor types, with glioblastoma, head and neck
   cancer, and pancreatic cancer often displaying severely hypoxic regions
   ([79]15, [80]16). While hypoxia may directly affect tumor cells,
   influencing stemness, metastasis, and progression across a range of
   cancer types ([81]17–[82]20), here, we focus on the direct effects of
   hypoxia on the NK cells themselves.

   Hypoxia within the TME can be acute or chronic ([83]21). Clinically, NK
   cells can be found circulating in patients after adoptive transfer for
   up to 3 weeks ([84]4, [85]11, [86]22). Cytokine administration and
   repeat dosing are designed to increase circulating NK cell numbers.
   Previous studies on the interaction of hypoxia and interleukin-2
   (IL-2)/IL-15 on NK cell function relied on high doses of cytokine that
   can alter metabolic requirements ([87]23–[88]25). Nevertheless, NK
   cells in hypoxia have consistently demonstrated reduced cytotoxicity
   when assays are performed during hypoxia ([89]23, [90]26), after
   overnight exposure ([91]23, [92]25), or longer intervals ([93]27,
   [94]28). This has been linked to reduced activating receptor expression
   and subtle defects in cytotoxic proteins ([95]23, [96]27). More
   recently, a failure to kill in hypoxia was linked to fragmented
   mitochondria ([97]28), suggesting that metabolism plays a key role in
   maintaining NK cell functionality. In mice, genetic modification of
   hypoxia-inducible factor 1-alpha (HIF1A), to improve NK cell activity
   in hypoxia, has shown a variety of effects, positive and negative,
   depending on the model system used ([98]29–[99]31).

   For T cells, hypoxia is a strong driver of exhaustion
   ([100]32–[101]36). Mitochondrial fragmentation, loss of mitochondrial
   biogenesis, an accompanying deficit in adenosine triphosphate (ATP)
   production, and the increased susceptibility to reactive oxygen species
   (ROS) all contribute to this impaired immunologic state. For these
   effector cells, activation within the hypoxic environment appears to
   further compound exhaustion ([102]34).

   Eukaryotic cells generate ATP from adenosine diphosphate (ADP) through
   glycolysis and oxidative phosphorylation. In oxidative phosphorylation,
   oxygen is required to accept electrons from the electron transport
   chain and generate a proton gradient. In doing so, ROS are generated.
   An imbalance in ROS can contribute to an impaired functional state on
   immune cells. Where the lack of availability of oxygen limits oxidative
   phosphorylation, cells must rely on other metabolic processes to
   generate ATP. How much NK cells rely on these two metabolic processes
   for cytotoxicity and cytokine production likely depends on their mode
   of activation.

   Stimulating NK cells with IL-2 or IL-15 ([103]37, [104]38),
   IL-12/IL-18/IL-15 ([105]39), or IL-21/41-BBL bearing feeder cells for
   adoptive transfer ([106]40) increases glycolysis and oxidative
   phosphorylation. Blocking oxidative phosphorylation in primary NK cells
   prevents the enhanced release of cytotoxic granules that normally
   follows cytokine stimulation ([107]37). Blocking oxidative
   phosphorylation in IL-21/4-1BBL–activated NK cells has been reported to
   have no impact on cytotoxicity ([108]41) or to strongly inhibit
   cytotoxicity if focusing on a subset of the NK cells that lack
   inhibitory receptors that bind self human leukocyte antigens ([109]40).
   Metabolic requirements can differ for interferon-γ (IFN-γ) production
   depending on whether the stimulus for release was a cytokine
   (IL-12/IL-18) or an activating receptor (NK1.1) ([110]42). In this
   study, we comprehensively examined the cytotoxic, proliferative,
   proteomic, transcriptomic, and metabolic state of NK cells over 7 days
   in hypoxia to determine which elements of these cells contribute to
   functional deficits to determine optimal therapeutic strategies to
   enhance NK cell immunotherapy efficacy to treat cancer.

RESULTS

Hypoxia diminishes NK cell proliferation

   Healthy donor NK cells were incubated in different oxygen (O[2]) and
   pressure conditions for up to 7 days to model levels found in arterial
   blood (12% O[2], 2 psi) or bone marrow (5% O[2], 0.6 psi) ([111]43) or
   within the TME (1% O[2]) at low or high pressure (0.3 or 2 psi)
   ([112]44–[113]46). Cells were then removed from the incubators, and
   assays were performed to assess NK cell proliferation, cytotoxicity,
   gene expression, and metabolism. NK cells incubated in a conventional
   5% CO[2] incubator (20% O[2]) served as a control (“normoxia”). Our
   initial focus was to identify conditions that would allow us to
   investigate NK cell function over time under conditions expected in the
   TME. We found that once the culture media was replaced on day 4 to
   avoid acidification in 1% O[2], there were no significant differences
   in NK cell viability across conditions as measured by annexin V and
   live/dead staining by flow cytometry (fig. S1A). Since oxygen and
   pressure both varied in this model, we asked whether pressure had an
   independent effect on NK cell proliferation and cytotoxicity in hypoxia
   at low or high pressure. There was essentially no impact of pressure on
   NK cell viability and proliferation or on cytolysis of tumor cells
   [discussed in Materials and Methods and in fig. S1 (B to D)]. Given
   these negative results, subsequent figures are presented with the focus
   on oxygen as the driving biological variable.

   To evaluate how oxygen levels affect proliferation, NK cells were
   CellTrace labeled, and dilution of the dye was evaluated at day 7.
   Oxygen had a titratable effect on proliferation and NK cell numbers,
   with the least proliferation and fewest NK cells obtained from the 1%
   O[2] condition ([114]Fig. 1, A and B). It should be noted that the
   severe hypoxia condition with higher pressure was selected to represent
   the TME going forward ([115]44–[116]46), since it best mimics the
   challenges to overcome in the hostile TME. To account for cell
   interactions, proliferation assays were performed in the presence or
   absence of other peripheral blood mononuclear cells (PBMCs), which
   might influence the response of NK cells to hypoxia over 7 days (fig.
   S2, A and B). There were greater numbers of NK cells achieving six or
   seven divisions when cultured with PBMCs compared to when enriched NK
   cells were isolated and placed alone in normoxia (fig. S2A). This is
   consistent with the capacity of monocytes and dendritic cells to
   enhance proliferation in part by the trans-presentation of IL-15
   through the IL-15 receptor alpha expressed on these cell types
   ([117]47). In contrast to normoxia, coculture with PBMCs did not
   enhance proliferation of NK cells in severe hypoxia (1% O[2]),
   suggesting that other immune cells in blood are not enough to rescue
   this dysfunction.

Fig. 1. Hypoxia diminishes NK cell proliferation and cytotoxicity but not
granule release.

   [118]Fig. 1.
   [119]Open in a new tab

   PBMCs (A and B) or pbNK cells (C to K) cells were cultured under 20%
   O[2], 12% O[2] + 2 psi, 5% O[2] + 0.6 psi, or 1% O[2] + 2 psi and
   examined on day 7. (A) Live, CD56^+ CD3^− NK cells and (B)
   proliferating NK cells (%) were analyzed by flow cytometry (n = 7) with
   two-way repeated measures (RM) Analysis of Variance (ANOVA) and post
   hoc Dunnett’s comparisons (horizontal bars). [(C) to (F)] Live cell
   imaging assessed [(C) and (D)] natural cytotoxicity (n = 10) and [(E)
   and (F)] ADCC (n = 7). Nuclight Red^+ tumor cell survival was
   normalized to growth of tumor cells without NK cells. [(C) and (E)]
   Representative single donors are shown alongside [(D) and (F)]
   quantification of multiple donors (two-way RM ANOVA, post hoc Dunnett’s
   comparisons comparing each condition within a time point to 20% O[2],
   indicated by horizontal bars if P ≤ 0.05). [(G) to (K)] The cytotoxic
   process was examined through (G) quantification of granule release (n =
   5, analyzed by two-way RM ANOVA) by flow cytometry after 5 hours and
   [(H) to (K)] polarization of the cytotoxic machinery to the synapse.
   [(H) to (K)] NK cells incubated with K562 for 5 min at 37°C, 20% O[2],
   were fixed and stained for LFA-1, the mTOC, perforin, and nuclei. An
   example of confocal assessment of synapse formation is shown in (H).
   Scale bar, 5 μm. Cells were scored for “a” enrichment of LFA-1 at the
   synapse compared to the back of the cell, “b” polarization of the mTOC
   toward the synapse, and “c” aggregation of perforin granules around the
   mTOC; quantified in (I) to (K), n = 2, minimum 14 cells per condition;
   [(I) and (J)] compared by Kruskal-Wallis test (horizontal bars); (K)
   analyzed by Fisher Exact test comparing each condition with 20% O[2],
   but P > 0.05 for all. Vertical bars show mean and SD [(A), (B), and
   (G)] or SEM [(C) to (F)] or median, range, and interquartile range
   (IQR) [(I) to (K)]. ns, not significant. **P ≤ 0.01, ***P ≤ 0.001, and
   ****P ≤ 0.0001.

   To determine whether T cells themselves might be influencing the NK
   cell proliferative response to severe hypoxia, T cells were enriched
   and added back into culture with autologous donor NK cells and then
   subjected to normoxia or severe hypoxia for 7 days (fig. S2, C and D).
   In normoxia and hypoxia, NK cells divided less when T cells were
   present, possibly competing for IL-15. This suggests that T cells
   themselves are not responsible for the enhanced NK cell proliferation
   induced by PBMCs in normoxia but could be partially responsible for
   limiting proliferation when PBMCs are present with NK cells in severe
   hypoxia.

Cytokine production is maintained in severe hypoxia

   NK cells can secrete cytokines to activate a wider immune response. We
   therefore asked whether secretion of the inflammatory cytokine IFN-γ
   was maintained in hypoxia over time. We performed these studies with
   both enriched NK cells and NK cells cultured with PBMCs (fig. S3). K562
   (leukemia cell line) were selected as targets for examining natural
   cytotoxicity by NK cells, and Raji (B cell malignancy cell line)
   combined with the therapeutic antibody rituximab were selected to
   investigate ADCC. For enriched NK cells alone (cultured for 3 or 7
   days), oxygen had no overall impact on production of IFN-γ when
   challenged with K562 or Raji with rituximab (fig. S3, A and C). When
   cultured with PBMCs, NK cell secretion of IFN-γ in response to these
   tumor cells was influenced by oxygen availability. After NK cells were
   cultured with PBMCs for 3 days, normoxic NK cells produced more IFN-γ
   than hypoxic NK cells when triggered through natural cytotoxicity, but
   not ADCC (fig. S3B). After NK cells were cultured with PBMCs for 7
   days, higher oxygen negatively affected cytokine production (fig. S3D).
   Cytokine production declined with cell division in all cases (fig.
   S3E), so in severe hypoxia the lack of cell division sustained cytokine
   production over time (fig. S3D). These studies show that oxygen content
   has a time-dependent effect on IFN-γ production influenced by
   interaction with other immune cells found in blood emphasizing the
   balance of many variables involved in cytokine production.

Hypoxia diminishes NK cell cytotoxicity

   We used a live cell imaging assay to follow cytolysis of tumor cells
   through a nuclear fluorescent reporter in K562 and Raji cells and an
   indicator of apoptosis (Caspase-3/7 Green). We determined that a ratio
   of 2:1 was sensitive to differences in natural cytotoxicity and ADCC
   imparted by oxygen content on NK cells (fig. S4). NK cells incubated at
   oxygen concentrations of 5% or higher controlled tumor cells to a
   similar extent, but after exposure to severe hypoxia, both natural
   cytotoxicity and ADCC were significantly impaired ([120]Fig. 1, C to
   F). Thus, killing was more sensitive to severe hypoxia than cytokine
   production. Considering this and the apparent importance of
   cytotoxicity for tumor control, we focused our investigations on the
   origin of the cytotoxic deficit.

Cytotoxic granule release is sustained, despite the deficit in killing

   There are several steps that lead to efficient NK cell cytotoxicity,
   including adhesion through integrins, aggregation of cytotoxic granules
   around the microtubule organizing center (mTOC), polarization of the
   mTOC to the synapse, followed by degranulation ([121]48). To determine
   how different oxygen levels affect NK cell degranulation, we measured
   the accumulation of granule protein CD107a at the NK cell surface when
   challenged with K562 or Raji cells with rituximab. Unexpectedly, there
   were no differences in the extent of degranulation after NK cells were
   incubated in the different oxygen conditions ([122]Fig. 1G). In line
   with this, we confirmed by imaging that there were no differences in
   the recruitment of integrin LFA-1 to the synapse, polarization of the
   mTOC to the synapse, or perforin-bearing granule aggregation around the
   mTOC ([123]Fig. 1, H to K). Nor were there any differences in the
   frequency with which NK cells formed conjugates with tumor cells (fig.
   S5). NK cell cytotoxicity started to decrease as early as 3 days under
   hypoxic conditions with lower natural cytotoxicity and ADCC (fig. S6).
   There was a decrease in degranulation after 3 days in 1% O[2] compared
   to 20% O[2] when triggered by ADCC (fig. S7A). Examining the impact of
   PBMCs on degranulation (fig. S7), there was a more substantial
   impairment in degranulation for both natural cytotoxicity and ADCC on
   day 3 of culture in hypoxia compared to normoxia, but these differences
   disappeared by day 7 (fig. S7, A to D), again highlighting the time
   dependence of this effect. Analyzing the impact of cell divisions on
   degranulation, considerable degranulation still occurred after more
   than one division (fig. S7E). Throughout all these assays,
   degranulation in the context of tumor cells was significantly higher
   than unstimulated controls, suggesting that whatever was causing the
   failure to kill tumors cells ([124]Fig. 1, C to F) was not a failure to
   release cytotoxic granules and the mechanistic defect must be at
   another level.

Cytotoxicity in hypoxia is impaired against solid tumor cell lines

   Since solid tumors experience the most severely hypoxic conditions, we
   examined whether the deficit in NK cell cytotoxicity was broadly
   applicable to a range of malignancies ([125]Fig. 2 and fig. S8). We
   selected cell lines of tumor types known to have TMEs characterized by
   severe hypoxia like head and neck cancer (Cal33), glioblastoma (LN229),
   and cancers that can have more variable oxygen availability like
   ovarian cancer (OVCAR8) and colorectal cancer (HT29). Cal33 is
   extremely resistant to natural cytotoxicity, making it a good model for
   understanding ADCC. We used cetuximab to mediate ADCC, since all these
   lines express epidermal growth factor receptor. In comparison to Cal33,
   OVCAR8 is more sensitive to natural cytotoxicity, so that the addition
   of cetuximab adds little to the overall control of OVCAR8 by NK cells.
   Normalizing to the growth of tumor cells alone, it was apparent for all
   tumor types and stimulations where robust cytotoxicity was observed in
   normoxia that cytotoxicity was significantly diminished in severe
   hypoxia. As with our blood cancer models of natural cytotoxicity and
   ADCC, there was no statistical difference in killing by NK cells
   cultured in 20, 12, or 5% oxygen. Overall, these data suggest that the
   models of natural cytotoxicity and ADCC reveal a marked deficit in NK
   cell biology in severe hypoxia that is applicable across a wide range
   of solid tumor types, adding to the definitiveness of this finding with
   multiple models.

Fig. 2. Severe hypoxia diminishes NK cell cytotoxicity of solid tumor lines.

   [126]Fig. 2.
   [127]Open in a new tab

   NK cells were enriched from blood and cultured (under 20% O[2] with
   atmospheric pressure, 12% O[2] + 2 psi, 5% O[2] + 0.6 psi, or 1% O[2] +
   2 psi) for 7 days and then examined for their ability to kill solid
   tumor lines by natural cytotoxicity or ADCC with cetuximab. Impedance
   was used as a measure of solid tumor cell survival, for natural
   cytotoxicity-sensitive ovarian cancer line, OVCAR8 (A and B), and
   natural cytotoxicity-resistant head and neck cancer line, Cal33 (C and
   D), normalized to the growth of tumor cells alone without cetuximab.
   The effector:target cell ratio (4:1) was selected for greatest
   sensitivity to differences between groups, and the dose of cetuximab
   was selected for each cell line to have the minimal direct impact on
   tumor cell growth alone but maximal ADCC. In representative graphs,
   conditions were measured in duplicate, and vertical bars show SD. In
   aggregate graphs, data from four independent donors are shown, and
   vertical bars show SEM, analyzed by two-way RM ANOVA. If O[2] failed to
   affect the result, then this is reported within the graph. If O[2]
   affected the result (P ≤ 0.05), then a post hoc Dunnett’s multiple
   comparisons test was applied, comparing all conditions to 20% O[2]
   within each time point, indicated by horizontal bars for any
   significant result, P ≤ 0.05. not significant, P > 0.05; *P ≤ 0.05; and
   **P ≤ 0.01.

Hypoxia reduces cytotoxic proteins, receptors, and transcription factors

   If the toxic payload of NK cell granules was diminished in hypoxia,
   that could explain the disconnect between successful granule release
   but impaired cytotoxicity. We examined the abundance of pore-forming
   protein, perforin; lytic protein, granzyme B; and death receptors,
   tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) and
   FasL; by time-of-flight mass cytometry (CyTOF). We included 38
   antibodies against activating and inhibitory receptors, cytokine
   receptors, transcription factors, and activation markers to examine
   important cellular processes. We applied viSNE, a t-distributed
   stochastic neighbor-embedding algorithm, to organize NK cell subsets
   into “islands” (CD56^bright immature cells, CD57^+ mature cells, and
   Ki67^+ proliferating cells) that can be seen in the key to fig. S9. The
   Ki67^+ subset was absent in 1% O[2], but for both mature and immature
   NK cells, perforin decreased when NK cells were incubated in decreasing
   oxygen conditions over time. Activating receptor NKG2D that recognizes
   stress ligands on tumor cells decreased with oxygen concentration but
   not necessarily with time (3 and 7 days in culture induced more NKG2D
   than 1 day in culture). For statistical analysis, we focused on severe
   hypoxia (1% O[2]) and normoxia (20% O[2]) after 3 or 7 days in culture
   ([128]Fig. 3). Across our panel, most were decreased in abundance in
   severe hypoxia compared to normoxia ([129]Fig. 3A, blue). There was a
   decrease in cytotoxic proteins ([130]Fig. 3B): granzyme B and TRAIL,
   with a similar trend seen for FasL and perforin, but the latter did not
   reach statistical significance. This suggests that the process of
   granule release was not impaired, but the contents of those NK cell
   granules were less cytotoxic in hypoxia. There was an accompanying
   decrease in the abundance of activating receptors (NKG2D, NKp44, and
   NKp46; [131]Fig. 3C), inhibitory receptors (NKG2A and CD94; P = 0.013
   for CD94; [132]Fig. 3, A and D), transcription factors (Tbet and Eomes;
   [133]Fig. 3E), and chemokine receptor CXCR3 ([134]Fig. 3F) and a
   decrease in proliferation marker, Ki67 ([135]Fig. 3G). NK cells require
   activating receptors to recognize tumor cells; they require inhibitory
   receptors to tune their cytotoxic function in a process known as
   “licensing” or “education” ([136]49–[137]51), and many of the cytotoxic
   processes are controlled by the transcription factors T-bet and Eomes
   ([138]52–[139]54), so the loss of these factors likely impairs NK cell
   cytotoxicity. Opposing this were the activation markers, CD69 and CD25
   ([140]Fig. 3H), that were reproducibly overexpressed in hypoxia. This
   observation suggests that hypoxia is inducing a broad stress response.
   Together with the defect in cytotoxic proteins, activating receptors,
   and key transcription factors, the stress response could be
   contributing to the impaired cytotoxicity seen in hypoxia.

Fig. 3. Severe hypoxia decreases NK cell receptors, cytotoxic proteins, and
transcription factors but increases activation markers.

   [141]Fig. 3.
   [142]Open in a new tab

   NK cells were exposed to 20% or 1% oxygen for 3 to 7 days and analyzed
   by CyTOF. (A to H) Differential expression analysis reveals the change
   in protein abundance on NK cell under severe hypoxia (1% O2 + 2 psi),
   relative to normoxia (20% O[2]), as quantified in Astrolabe Diagnostics
   software. Positive fold changes indicate up-regulation in hypoxia. In
   (A), each square is the average value for all donors after 3 days or 7
   days in culture; in (B) to (H), each line represents a donor after 7
   days in culture. Differential expression analysis is adjusted for
   multiple testing (n = 6). *P ≤ 0.05, **P ≤ 0.01, and P > 0.05 are
   listed in full. In (A), # indicates that the values for Ki67 are beyond
   the range of the listed scale (−2.66 on day 3; −2.03 on day 7).

   We next asked whether we could observe changes in specific NK cell
   subsets upon exposure to hypoxia considering stages 4A (CD56^bright), 5
   (CD56^dim), and 6 (CD57+) as the primary NK cells found in peripheral
   blood ([143]55, [144]56). In addition, stage 5 is often subdivided into
   immature (NKG2A^+ KIR^−) and mature cells (NKG2A^− KIR^+) ([145]57),
   and adaptive NK cells (NKG2C^+ CD57^+) are a subset of stage 6 cells
   that are found in high frequency in individuals who are serum positive
   for cytomegalovirus. Using the CyTOF panel, we evaluated the
   CD56^bright populations (CD62L^+, CD16^−, CD57^−, KIR^−), CD56^dim
   immature populations (CD62L^− CD57^− KIR^− NKG2A^+), CD56^dim mature
   populations (CD62L^− CD57^− KIR^+ NKG2A^−), late-stage cells (CD57^+
   NKG2C^−), and adaptive cells (CD57^+ NKG2C^+). Hypoxia had no
   significant effects on the proportions of NK cell populations (fig.
   S10A). Phenotypic comparison of NK cells in hypoxia compared to
   normoxia revealed no differences within NK cell subsets compared to the
   bulk NK cell population (fig. S11A). There were decreases in Ki67 and
   granzyme B and concomitant increases in CD25 and CD69 in all subsets.
   Overall, it was clear that there were no substantial changes in
   individual NK cell subsets upon culture in hypoxia, and the impact was
   global and not dependent on maturational stage.

Shifts in gene expression underpin the loss of function in hypoxia

   To assess the consequences of oxygen concentration on the
   transcriptomic and signaling responses of NK cells, we performed an RNA
   sequencing (RNA-seq) analysis of NK cells incubated under different
   oxygen conditions for 1, 3, or 7 days. The 50 most differentially
   altered genes across time points and oxygen concentrations revealed
   different patterns of activation with time. Many of the genes in this
   set were strongly up-regulated at 24 hours after exposure to hypoxia
   and continued to show differential expression 7 days into exposure
   ([146]Fig. 4A). We observed changes to pathways related to
   proliferation, DNA repair, and cellular metabolism. By day 7 under
   normoxia, there was a clear up-regulation of pathways related to DNA
   replication and cell cycle, which were absent in hypoxia ([147]Fig.
   4B). The mitotic checkpoint, BUB1, typified this expression pattern. It
   was up-regulated in normoxia by day 3, but not in hypoxia ([148]Fig.
   4A). In hypoxia, there was an up-regulation of pathways related to
   membrane and organelle recycling: endocytosis, mitophagy, autophagy;
   energy usage: carbon metabolism; and transcription factors: HIF-1α and
   forkhead box O (FOXO; [149]Fig. 4B and fig. S12). Endocytosis is a
   process that internalizes exogenous material by budding membrane. In
   doing so, it removes membrane proteins from the cell surface, which
   could contribute to the loss of NK cell surface receptors observed by
   CyTOF ([150]Fig. 3). Autophagy involves the recycling of cellular
   components within the lysosome, and mitophagy is the removal of
   mitochondria. These active processes suggest that NK cells are not
   passive in hypoxia but actively remodeling their cellular processes. In
   the presence of oxygen, HIF-1α is posttranslationally modified by
   oxygen-dependent proteases, rapidly targeting it for degradation. In
   the absence of oxygen, HIF-1α is stable and regulates the expression of
   genes that allow cells to adapt to low oxygen conditions ([151]19).
   Proteomic regulation of HIF-1α resulted in strong up-regulation of
   HIF-1α–regulated genes, such as the glycolytic enzyme PFKFB4 at 24
   hours ([152]Fig. 4A), and there is persistent up-regulation of the
   HIF-1α pathway after 7 days (fig. S12).

Fig. 4. Gene expression signatures steadily change with hypoxia and are found
in tumor-infiltrating NK cells from glioblastoma patients.

   [153]Fig. 4.
   [154]Open in a new tab

   (A to C) Enriched NK cells were cultured under 20% O[2], 12% O[2] + 2
   psi, 5% O[2] + 0.6 psi, or 1% O[2] + 2 psi and harvested after 1, 3, or
   7 days for RNA-seq analysis. (A) Heatmaps show the relative change in
   the 50 most differentially expressed genes (DEGs) across oxygen
   conditions and time points. Each column of small squares represents a
   donor. (B) Bar graphs plot the Kyoto Encyclopedia of Genes and Genomes
   (KEGG) pathways up-regulated with increasing oxygen concentration
   (blue) or up-regulated with decreasing oxygen concentration (orange) at
   day 7. (C) The number of DEGs in comparison to the 20% oxygen condition
   is highlighted for the different time and oxygen culture conditions. (D
   to F) Genes and pathways up-regulated in hypoxia identified by bulk
   RNA-seq are compared to signatures within the NK cell population of
   scRNA-seq datasets from patients with glioblastoma (GBM) or healthy
   human blood. (D) The probability of separating GBM from blood scRNA-seq
   data using all detected genes (>4000, positive control), a random
   selection of genes (six genes, negative control), or the genes shared
   between these ex vivo sets and the in vitro hypoxic dataset of 50 most
   DEGs (six genes) was calculated using a machine learning method using
   logistic classifier. Bars show the mean and SD. (E) The relative
   enrichment in the hypoxic (bulk RNA-seq) or GBM (scRNA-seq) datasets is
   plotted relative to normoxia (bulk RNA-seq) or blood (scRNA-seq) for
   three genes important in glycolysis or autophagy. (F) Gene enrichment
   within GBM samples relative to blood samples is plotted for genes
   within three KEGG pathways (endocytosis, mitophagy, and autophagy),
   identified as differentially regulated in our in vitro RNA-seq dataset
   (B).

   Chemokine receptors are an important factor for homing of NK cells. Our
   data indicate that at early time points (fig. S13A) some chemokine
   receptors are unregulated in hypoxia (CCR7 and CXCR4), but at later
   time points (fig. S13B), several chemokine receptors get down regulated
   (CCR1, CCR2, CCR5, and CXCR6). While these changes were strongest in
   severe hypoxia, decreases are also seen at 5% oxygen, indicating that
   migration could be affected in other compartments like the bone marrow
   at this intermediate oxygen level.

   When we looked at the total number of differentially regulated genes at
   day 7, we found that there was almost no difference between the
   standard incubator condition (20% O[2]) and the 12% O[2] condition (2
   total). There were few differences when NK cells were incubated under
   5% O[2] (<300 total), but the biggest differences were observed when NK
   cells were incubated under the 1% O[2] (>2000 total; [155]Fig. 4C). We
   had observed no effect of pressure on proliferation and cytotoxicity at
   1% O[2] (fig. S1), which was supported by a lack of difference in gene
   expression when NK cells were exposed to these conditions (<30 total).
   These results demonstrate two things with regard to oxygen content.
   First, there is a remodeling of the transcriptome when NK cells are
   exposed to hypoxia that is detectable at 24 hours and compounds over 7
   days. Second, the largest differences are between normoxia (20% O[2])
   and severe hypoxia (1% O[2]). Given the physiologic conditions
   throughout the body’s tissues, it is logical to conclude that these
   differences occur along an oxygen continuum to affect NK cell function
   in the microenvironment where they home.

Hypoxic gene expression signatures are found in glioblastoma-infiltrating NK
cells

   Having identified pathways that are regulated in NK cells by hypoxia,
   we set out to determine whether these pathways would be present in NK
   cells extracted from human tumors. Glioblastoma is known to be hypoxic
   compared to peripheral blood ([156]15, [157]58, [158]59). We mined
   previously published single-cell RNA-seq (scRNA-seq) datasets from
   peripheral blood of healthy donors ([159]GSE130430) ([160]60) and
   excised glioblastoma tissue ([161]GSE147275) ([162]61) for this
   purpose. NK cells were identified as previously described ([163]60,
   [164]62), and genes were ranked according to their expression level
   within each tissue. The published scRNA-seq datasets had a high level
   of gene dropout compared to our bulk RNA-seq datasets. When we examined
   the 50 most differentially regulated genes within our RNA-seq data
   ([165]Fig. 4A), only eight genes could be found within the published
   scRNA-seq datasets, and only six could be compared for fold change
   between glioblastoma and blood ([166]Fig. 4D). Nevertheless, a machine
   learning method using logistic classifier was able to separate blood NK
   cells from glioblastoma NK cells with a confidence of >0.85 (±0.01 SD)
   using these six genes, compared to a confidence of nearly 1 if using
   all detected genes (>4000; positive control) or a confidence of <0.6
   (±0.02 SD) when six genes were selected at random from this list
   (negative control). This confirms the clinical applicability of the
   hypoxic gene panel we generated in vitro. The relative enrichment of
   three of these genes, which relate to important pathways up-regulated
   in hypoxia according to our RNA-seq data, is highlighted for all
   datasets in [167]Fig. 4E. These include glycolytic enzyme enolase 2
   (ENO2), glucose transporter GLUT3 (SLC2A3)—both part of the glycolytic
   pathway—and Bcl2-interacting protein 3 (BNIP3)—a mitochondrial protein
   involved in mitophagy and prolonged NK cell responses ([168]63). When
   considering the top three pathways up-regulated in hypoxia in our
   RNA-seq dataset (endocytosis, mitophagy, and autophagy), genes in the
   top two pathways (endocytosis and mitophagy) are enriched within the
   glioblastoma dataset compared to blood ([169]Fig. 4F), but autophagy
   genes are normally distributed across both samples. These data suggest
   that pathways identified in hypoxia in vitro are relevant to NK cell
   function in tumors in vivo.

NK cells switch to glycolysis and have less available ATP under hypoxia

   Metabolism within a solid tumor is different from that of the
   surrounding tissue. In tumor cells, the result of HIF-1α activation in
   hypoxia is to shift energy production by increasing glycolysis and
   decreasing mitochondrial function ([170]64). Our RNA-seq data revealed
   that by 7 days, NK cells in hypoxia had up-regulated genes involved in
   glycolysis ([171]Fig. 5A). To confirm the increase in glycolysis, we
   conducted a glycolytic rate assay using a Seahorse Analyzer ([172]Fig.
   5B). NK cells incubated under hypoxia for 7 days had a higher rate of
   glycolysis at rest (basal; [173]Fig. 5C) compared to normoxic NK cells
   at rest. The maximal capacity for glycolysis was similar for normoxic
   and severely hypoxic cells when mitochondria were inhibited, forcing
   the cells to meet all energy demands through glycolysis (compensatory;
   [174]Fig. 5D). The final product of glycolysis is pyruvate. Under
   hypoxia, tumor cells do not use pyruvate in the mitochondria but
   convert it to lactate by lactate dehydrogenase (LDHA). To assess the
   role of LDHA in NK cells incubated under hypoxia, we added the LDHA
   inhibitor NCI-737 (1 μM) during the 7-day incubation. NK cell viability
   decreased more robustly when NK cells were treated with the LDHA
   inhibitor in severe hypoxia compared to normoxia ([175]Fig. 5E), and no
   impact was seen on NK cell proliferation ([176]Fig. 5F). These data are
   consistent with the notion that NK cells increase glycolysis under
   hypoxia. Unlike many cancer cell types, glycolysis supports NK cell
   survival but not proliferation.

Fig. 5. NK cells rely on glycolysis in hypoxia.

   [177]Fig. 5.
   [178]Open in a new tab

   (A) A heatmap shows the relative expression of genes in the glycolysis
   pathway (KEGG) after NK cells were cultured under 20% O[2], 12% O[2] +
   2 psi, 5% O[2] + 0.6 psi, or 1% O[2] + 2 psi for 7 days. Each column of
   squares represents a donor. (B) Extracellular acidification rate (ECAR)
   of NK cells, as an indicator of aerobic glycolysis, was measured after
   addition of rotenone/antimycin A and 2-deoxy-d-glucose (2-DG)
   (representative kinetic graph, measured in triplicate). The calculated
   (C) basal and (D) compensatory glycolysis (rate of glycolysis following
   addition of Rot/AA) are plotted for six donors (paired t test). (E and
   F) NK cells were treated with 1 μM NCI-737 or dimethyl sulfoxide (DMSO)
   during 7-day culture. The proportion of (E) live cells and (F)
   proliferating cells, as indicated by dilution of CellTrace dye, was
   analyzed by flow cytometry (live, single, CD56^+ CD3^− cells; n = 6;
   two-way RM ANOVA with Dunnett’s multiple comparisons; not significant,
   P > 0.05; *P ≤ 0.05; and **P ≤ 0.01). All bars show the mean and SD.
   PER, protein efflux rate; LDHAi, LDHA inhibitor.

   Our RNA-seq data also revealed an increase in expression of genes
   relating to mitophagy in NK cells incubated under hypoxia ([179]Fig.
   6A). Oxidative phosphorylation occurs across mitochondrial membranes,
   and so a loss of mitochondria will affect this metabolic process. We
   observed a decrease in oxidative phosphorylation in severe hypoxia
   compared to normoxia as measured by oxygen consumption rate ([180]Fig.
   6, B and C). Thus, the relative contributions of glycolysis and
   oxidative phosphorylation differed for NK cells incubated in severe
   hypoxia or normoxia, with more ATP generated by glycolysis in hypoxia
   and less generated by oxidative phosphorylation relative to normoxia
   ([181]Fig. 6D). Nevertheless, the overall rate of ATP production was
   not different for NK cells cultured in normoxia or hypoxia ([182]Fig.
   6D, sum height of bars). The available ATP within the cell was lower
   for NK cells cultured in hypoxia relative to normoxia ([183]Fig. 6E).
   This suggests that the energy demands of hypoxic NK cells at rest are
   greater than for normoxic NK cells, since they generate the same amount
   of energy, but severely hypoxic NK cells have lower ATP reserves
   compared to normoxic NK cells.

Fig. 6. NK cells reduce mitochondrial function in hypoxia, leading to less
ATP, but still considerable ROS.

   [184]Fig. 6.
   [185]Open in a new tab

   (A) A heatmap shows the relative expression of genes involved in
   mitophagy (KEGG) after enriched NK cells were cultured under 20% O[2],
   12% O[2] + 2 psi, 5% O[2] + 0.6 psi, or 1% O[2] + 2 psi. (B) Oxygen
   consumption rate (OCR) of NK cells, as an indicator of oxidative
   phosphorylation, was measured after addition of oligomycin, carbonyl
   cyanide p-trifluoromethoxyphenylhydrazone (FCCP), and
   rotenone/antimycin A (one representative kinetic graph of three,
   measured in triplicate, bars shows mean and SD). (C) The basal and
   maximal respiration (following addition of FCCP) are plotted for all
   donors (paired t test). (D) The rate of ATP generation via glycolysis
   (glycoATP) or oxidative phosphorylation (mitoATP) was measured after
   treatment with oligomycin and rotenone/antimycin A (n = 7, two-way RM
   ANOVA comparison of 20% O[2] versus 1% O[2] for glycoATP or mitoATP).
   (E) The amount of ATP in NK cells was measured using a luciferase assay
   (n = 6); Wilcoxon signed rank test. (F to H) After 7 days in 20 or 1%
   O[2], NK cells were stained for mitochondria (MitoTracker far red) and
   nuclei [4′,6-diamidino-2-phenylindole (DAPI)]. (F) Representative sum
   intensity z-projection images of mitochondria or mitochondria overlaid
   on nuclei are shown. Scale bar, 5 μm. Violin plots show quantification
   of mitochondrial surface area (G) and volume (H), as modeled in Imaris
   software. n = 2 donors, 20% O[2] (48 cells), 1% O[2] (40 cells);
   Mann-Whitney test. (I) ROS were measured by mitoSOX staining measured
   by flow cytometry (n = 6; Wilcoxon signed rank test). Bars show the
   mean and SD; *P ≤ 0.05, **P ≤ 0.01, and ***P ≤ 0.001.

   Severe hypoxia was associated with a significant decrease in NK cell
   mitochondrial surface area and volume as measured by microscopy
   ([186]Fig. 6, F to H). In addition to reducing the capacity for
   oxidative phosphorylation, loss of mitochondria can affect the
   generation of ROS. We observed a greater amount of ROS in NK cells
   cultured in severe hypoxia compared to normoxia at rest ([187]Fig. 6I)
   presumably arising from dysfunctional mitochondria in this altered
   metabolic state. Overall, these results show that NK cell metabolism is
   altered under hypoxia, and while some aspects are similar to the
   metabolic shift seen in tumor cells, there are substantive differences
   that affect NK cell function and survival in the hypoxic TME.
   Considering how these variables change with different therapeutic
   interventions is of translational importance to maximize antitumor
   activity in vivo.

Activation of NK cells with IL-15 or feeder cells enhances cytotoxicity under
hypoxia

   Having identified loss of granzyme B, greater reliance on glycolysis,
   lower ATP availability, and higher ROS at rest as features of severely
   hypoxic NK cells, we asked whether activation strategies being used
   therapeutically could overcome these limitations. A challenge for NK
   cell–based immunotherapy in solid tumors is cell numbers and
   persistence. IL-15 can enhance survival and proliferation, but we had
   selected a low dose of IL-15 (1 ng/ml) that induced minimal
   proliferation for our investigations of hypoxia. A higher pharmacologic
   dose of IL-15 (10 ng/ml) that induced robust proliferation in normoxia
   was selected to assess proliferation and cytotoxicity of peripheral
   blood NK (pbNK) cells in hypoxia. Another method of therapeutically
   enhancing NK cell function is by adoptive transfer, using
   feeder-expanded NK (eNK) cells to generate sufficient numbers. To
   replicate this process, NK cells were expanded in normoxia by culturing
   them with IL-2 and K562 feeder cells expressing membrane-bound IL-21
   and 4-1BBL. The resulting product was frozen, and we evaluated whether
   these eNK cells, thawed and incubated in normoxia or hypoxia for 7 days
   to mimic exposure to the TME posttransfer, would show improved
   cytotoxicity under hypoxia.

   We focused on the differences induced by 20 and 1% O[2] after 7 days,
   because these conditions impaired NK cell function and phenotype most
   profoundly. Neither treatment was capable of driving increases in cell
   proliferation and numbers within the TME ([188]Fig. 7, A to D). IL-15
   (10 ng/ml dose) induced almost no proliferation in hypoxia, despite
   inducing greater proliferation in normoxia ([189]Fig. 7, A and B). In
   contrast, eNK cells were capable of proliferating in hypoxia, but this
   came at a cost: Fewer cells survived of these proliferating cells
   ([190]Fig. 7, A to C) resulting in similar NK cell numbers at the end
   of 7 days in hypoxia ([191]Fig. 7D). In all contexts, we observed lower
   viability in severe hypoxia compared to normoxia, but since killing
   assays were adjusted for viable cells at the start of the assay, any
   deficits in killing would not arise solely from a decrease in viability
   under hypoxic conditions.

Fig. 7. Preactivation with IL-15 or IL-2 plus feeder cells overcomes
diminished cytotoxicity, but not diminished proliferation, in hypoxia.

   [192]Fig. 7.
   [193]Open in a new tab

   As a model of endogenous NK cell activation, pbNK cells were cultured
   in normoxia (20% O[2]) or hypoxia (1% O[2]) as before [IL-15 (1 ng/ml)]
   or with a higher dose of IL-15 (10 ng/ml) for 7 days. As a model of
   adoptive NK cell transfer, NK cells expanded with feeder cells (eNK)
   were thawed and incubated in 20 or 1% O[2] for 7 days. (A to D) For
   proliferation assays, enriched NK cells were labeled with CellTrace
   Violet before culture. (A) Representative histograms show CellTrace dye
   dilution within live single NK cells (live/dead^−, annexin V^−,
   singlets, CD56^+, CD3^−) analyzed by flow cytometry. (B) The proportion
   of NK cells that proliferated [marked in (A)]. (C) The proportion of NK
   cells that were alive (live/dead^− annexin V^−). (D) The relative
   number of NK cells per condition was quantified. For (A) to (D), n = 6,
   analyzed by two-way RM ANOVA with Tukey’s and Sidak’s multiple
   comparisons tests. (E to H) Graphs show cytotoxicity analysis of pbNK
   cells (n = 11) treated with 1 or 10 ng/ml or eNK cells (n = 5), killing
   K562 cells [(E) and (F)], or Raji cells with rituximab treatment [(G)
   and (H)] as in [194]Fig. 1. Killing by representative donors over time
   are displayed in (E) and (G) (bars show mean and SEM of triplicate
   measurements). Quantification of surviving tumor cells at 24 hours
   after the addition of NK cells are shown in (F) and (H) (bars show the
   mean and SD; analyzed by mixed effects analysis with Sidak’s multiple
   comparisons test; not significant, P > 0.05; *P ≤ 0.05; **P ≤ 0.01;
   ***P ≤ 0.001; and ****P ≤ 0.0001.

   Examining cytotoxicity of K562 or Raji and rituximab revealed pbNK
   cells treated with IL-15 (10 ng/ml), and eNK cells were more cytolytic
   in severe hypoxia than pbNK cells treated with IL-15 (1 ng/ml), with
   eNK cells being the most cytolytic overall. Compared to their
   counterparts in normoxia, pbNK cells treated with IL-15 (10 ng/ml)
   showed no effect of oxygen on natural cytotoxicity, but ADCC was worse
   in severe hypoxia. In comparison, eNK cells showed no effect of oxygen
   on natural cytotoxicity, but their ADCC was superior in severe hypoxia
   compared to normoxia ([195]Fig. 7, E to H). To determine the kinetics
   of these interventions within the setting of hypoxia, we evaluated
   cytolytic activity at three earlier time points, day 0 (baseline), day
   1, and day 3 (fig. S14, A to E). For both natural cytotoxicity and ADCC
   of solid tumor lines, our findings indicated that high-dose IL-15 and
   NK cell expansion improved cytolysis relative to less activated pbNK
   cells at all time points. Cytolysis was maximal in hypoxia on day 1,
   after initial priming with IL-15, and there was a decline in all
   responses in severe hypoxia from day 1 to day 3. eNK cells always had
   superior cytotoxicity to pbNK cells treated with IL-15 (10 ng/ml) in
   severe hypoxia, but this was not true of normoxia. Together, these data
   highlight the complex interactions of NK cell source and activation
   with oxygen over time. A consistent conclusion was the strong function
   of eNK cells in severe hypoxia.

   Modifying adoptively transferred cells to express or secrete intrinsic
   IL-15 is a clinical approach being taken with induced pluripotent stem
   cell–derived NK (iNK) cells ([196]65) and cord blood–derived NK cells
   ([197]66). To determine whether this intrinsic expression of IL-15
   would achieve a similar benefit to exogenous IL-15 (10 ng/ml), we made
   use of iNK cells with or without a fusion protein of IL-15 receptor α
   and IL-15, which is tethered to the iNK cell membrane. These research
   lines of iNK cells were unmodified apart from the IL-15 fusion protein
   (IL-15FP). Low-dose IL-15 was added to support survival of all iNK
   cells and maintain consistency between groups but failed to induce
   proliferation in normoxia and hypoxia. iNK with IL-15FP proliferated in
   normoxia, but proliferation was impaired in severe hypoxia (fig. S15, A
   and B). Upon examining natural cytotoxicity of iNK cells with or
   without IL-15FP (fig. S15, C and D), we found similar results to pbNK
   cells treated with a high dose of IL-15, with the IL-15FP inducing
   better iNK cytotoxicity in both normoxic and hypoxic conditions. These
   data show that both exogenous IL-15 and intrinsic IL-15 enhance NK cell
   killing in severe hypoxia, but severe hypoxia limits proliferation in
   both cases.

IL-15 prevents the loss of granzyme B, but proteomic deficits remain

   To examine how IL-15 (10 ng/ml) improved NK cell function in hypoxia,
   we compared these cells by CyTOF to the previously analyzed pbNK cells
   treated with IL-15 (1 ng/ml; fig. S16). Many of the key proteins that
   had been down-regulated in severe hypoxia at the lower dose of cytokine
   remained low in severe hypoxia with the higher cytokine dose, including
   cytotoxic proteins (TRAIL and FasLigand) and transcription factors
   (Eomes). The cytotoxic protein granzyme B and transcription factor
   T-bet were notable in that they had been suppressed in hypoxia with low
   dose of IL-15 but were no longer suppressed relative to normoxia when
   NK cells were stimulated with high dose of IL-15. A drop in
   CD122/IL-2Rβ suggests that a higher dose of IL-15 may be ineffective in
   stimulating NK cells in hypoxia, because signaling is compromised by
   the lack of this component within the IL-15 receptor complex.
   Activation markers CD69 and CD25 remained high with high dose of IL-15
   in hypoxia, suggesting that enhanced function of NK cells in hypoxia
   with IL-15 (10 ng/ml) was not a result of reversing this activated
   phenotype. Similar to what we saw previously with 1 ng/ml IL-15 (figs.
   S10A and S11A), separating these NK cells according to known
   developmental subsets revealed no change in subset frequencies when
   cultured in normoxia or hypoxia (fig. S10B) and similar patterns in
   protein abundance were seen across subsets compared to bulk NK cells
   (fig. S11B). There was maintenance of granzyme B in all subsets and
   relative increases in CD25 and CD69 in hypoxia in all subsets. An
   increased abundance of NKG2D and Eomes in hypoxia was present for
   adaptive pbNK cells treated with IL-15 (1 ng/ml). This was also
   apparent for pbNK cells treated with the higher dose of IL-15 for both
   the small adaptive population and larger CD57^+ population. This
   suggests that CD57^+ NK cells might not be as susceptible to
   suppression by hypoxia as less mature subsets. While stimulation of NK
   cells with high-dose IL-15 enhances cytotoxicity in severe hypoxia,
   there is still opportunity to enhance their function further.

Preactivation with feeder cells retains key proteins in hypoxia

   We asked what proteomic changes occurred in eNK cells when they were
   exposed to hypoxia, given that preactivation with feeder cells induced
   superior cytotoxicity in hypoxia, compared to IL-15 treatment. In
   contrast with pbNK cells, eNK cells had increased abundance of nearly
   all the key proteins in hypoxia relative to normoxia ([198]Fig. 8).
   This included transcription factors (T-bet and, to a lesser extent,
   Eomes), activating receptors (NKp30, NKp46, and others), adhesion and
   costimulatory protein (CD2), and cytotoxic proteins (TRAIL, FasL, but
   not granzyme B). The CD56^bright population and adaptive populations
   were absent from our subset analysis of eNK cells, but we were able to
   analyze a continuum of more mature to less mature NK cells of which the
   majority were immature NKG2A^+ KIR^− CD57^− cells (fig. S10C). There
   were no differences in the frequency of eNK cell subsets when cells
   were cultured in normoxia or hypoxia, and the subsets tended to follow
   the proteomic trends of the bulk population (fig. S11C). The IL-15
   receptor CD122 was down on the immature populations in hypoxia but
   increased on CD57^+ eNK cells in hypoxia. CD57^+ eNK cells were a rare
   subset in our expansion method, but other expansion protocols that
   focus on generating adaptive cells might produce cells that respond
   more favorably to hypoxic environments. Emerging as a consistent
   feature of severely hypoxic NK cells regardless of the NK cell source
   and activation, markers of activation (CD25 and CD69) remained high
   across all eNK subsets.

Fig. 8. Preactivation of NK cells with IL-2 and feeder cells increases
abundance of death receptors, activating receptors and transcription factors
in severe hypoxia.

   [199]Fig. 8.
   [200]Open in a new tab

   NK cells expanded with feeder cells were thawed and incubated in
   normoxia (20% O[2]) or severe hypoxia (1% O[2] + 2 psi) for 7 days. (A)
   A heatmap shows the relative change in protein abundance with 1%
   compared to 20% O[2] by CyTOF analysis in Astrolabe Diagnostics. (B to
   I) Box (median and IQR) and whisker plots (minimum to maximum) show the
   relative abundance of key proteins as quantified by Astrolabe
   Diagnostics software, normalized to pbNK cells cultured in normoxia
   with IL-15 (1 ng/ml). Kruskal-Wallis test with Dunn’s multiple
   comparisons to 1% O2 pbNK cells (eNK cells; n = 5; pbNK cells are
   reproduced in fig. S16 for reference, n = 7); not significant, P >
   0.05; *P ≤ 0.05; **P ≤ 0.01; and ***P ≤ 0.001

   Chemokine receptor expression is known to strongly differ between pbNK
   cells compared to eNK cells ([201]67). The majority of chemokine
   receptors are not shared. CXCR3 is unusual in that it is found on both
   pbNK cells and eNK cells and on a number of NK cell lines ([202]68).
   While CXCR3 was less abundant on pbNK cells in severe hypoxia compared
   to normoxia (fig. S16J), eNK cells had significantly more CXCR3 in
   severe hypoxia compared to pbNK, suggesting that they are more capable
   of trafficking in response to CXCR3 ligands in severe hypoxia.

Feeder cell activation, but not IL-15 activation, stabilizes NAD^+/NADH
ratios

   Having determined that eNK cells had superior cytotoxicity and protein
   abundance compared to pbNK cells in hypoxia, we asked whether the
   energy status and redox balance in these cells could be contributing to
   their enhanced function ([203]Fig. 9). In terms of energy usage, pbNK
   cells treated with IL-15 (1 or 10 ng/ml; fig. S17A) relied on
   glycolysis in hypoxia, decreasing oxidative phosphorylation. eNK cells,
   in comparison to pbNK cells (1 ng/ml IL-15), increased their rate of
   ATP production by glycolysis even more in severe hypoxia ([204]Fig. 9A)
   and similarly shut down oxidative phosphorylation ([205]Fig. 9B), which
   led to a slightly higher rate of ATP production overall for eNK cells
   relative to pbNK cells in hypoxia ([206]Fig. 9C). Both pbNK cells and
   eNK cells had less ATP in the cell under hypoxia than normoxia
   ([207]Fig. 9, D to F), suggesting that enhanced cytotoxicity of eNK
   cells did not result from enhanced ATP generation leading to enhanced
   ATP availability. A deeper dive into metabolites revealed unchanged
   levels of ADP but very low levels of adenosine monophosphate (AMP) in
   pbNK cells (suggesting catabolism), but hypoxic eNK cells maintained
   their levels of AMP. Lower AMP relative to ATP within the cell (the
   energy charge; fig. S17B) is generally considered an indicator that ATP
   availability is not limiting. The high energy charge in hypoxic pbNK
   cells thus suggests that ATP is not limiting in these cells.

Fig. 9. Preactivation of NK cells with feeders sustains AMP and NADH in
severe hypoxia.

   [208]Fig. 9.
   [209]Open in a new tab

   Freshly enriched NK cells (pbNK) or thawed eNK cells were cultured in
   normoxia (20% O[2]) or severe hypoxia (1% O[2] + 2 psi). After 7 days,
   the rate of ATP generation via glycolysis (glycoATP) (A) or oxidative
   phosphorylation (mitoATP) (B) was measured by Seahorse analysis; eNK (n
   = 4); pbNK (n = 5 of which five of seven were included in [210]Fig. 6D;
   shown here opaque for reference only); analyzed by one-way ANOVA with
   Dunnett’s multiple comparisons test. The total rate of ATP generation
   from these two processes is plotted in (C). Bars represent the mean and
   SD. In (D) to (I), the energy status of cultured NK cells (as before)
   after 7 days in normoxia (20% O[2]) or severe hypoxia (1% O2 + 2 psi)
   was determined by analytical chemistry (n = 5); dots represent donors,
   and bars show the mean and SD. Analyzed by two-way ANOVA with Sidak’s
   multiple comparisons test. (J) ROS (mitoSOX^+) were detected by flow
   cytometry within live NK cells (TOPRO-3^−, singlets, CD56+, CD3^−) when
   exposed to 0 or 10 μM hydrogen peroxide (H[2]O[2]); pbNK, n = 8; eNK, n
   = 7; bars show the mean and SD, analyzed by two-way ANOVA with Sidak’s
   multiple comparisons test. In all graphs, bars mark tests that were
   performed where P > 0.05 (not significant), *P ≤ 0.05, **P ≤ 0.01, ***P
   ≤ 0.001, and ****P ≤ 0.0001.

   In pbNK cells and eNK cells, there was more nicotinamide adenine
   dinucleotide (NAD^+) in hypoxic cells than normoxic cells ([211]Fig.
   9G). Glycolysis converts NAD^+ to NADH (reduced form of NAD^+), so
   NAD^+ must be regenerated for glycolysis to proceed. This suggests that
   NAD^+ levels are being actively regenerated in hypoxia-adapted cells.
   The lethality of LDHA inhibition in hypoxia also suggests that the
   regeneration of NAD^+ by LDHA is critical for NK cell survival. In pbNK
   cells, higher NAD^+ levels combined with lower availability of NADH led
   to a skewed NAD^+/NADH ratio, but in eNK cells, NADH levels were
   maintained, and the ratios did not differ for eNK in hypoxia and
   normoxia ([212]Fig. 9, G to I). These altered NADH levels were
   significantly higher for eNK cells compared to IL-15–treated pbNK cells
   (1 ng/ml) under hypoxia. pbNK cells treated with IL-15 followed a
   similar pattern regardless of treatment dose (fig. S17B). Other
   metabolites (acetyl coenzyme A and succinyl coenzyme A; fig. S17B)
   varied according to therapeutic intervention in hypoxia, suggesting
   that altered metabolic pathways in these cells extend beyond ATP
   availability and NAD^+/NADH ratios.

   It should be noted that the energy measurements described are based on
   the total amount of these molecules within the cells, but it is the
   amount of free NAD^+ and NADH that has regulatory roles. To address
   whether the altered NAD^+/NADH ratio correlated with oxidative stress,
   we challenged NK cells at the end of culture with increasing doses of
   peroxide ([213]Fig. 9J and fig. S17C). Titrating in the dose of
   peroxide revealed that pbNK cells treated with IL-15 (1 or 10 ng/ml)
   under hypoxia were very sensitive to oxidative stress, with more ROS
   detected by mitoSOX staining than in eNK cells that were very resistant
   to oxidative stress, highlighting another difference of NK cells based
   on their activation conditions. This aligns with the NAD^+/NADH balance
   being maintained in eNK cells but not pbNK cells. Thus, the redox
   balance within hypoxia-cultured eNK cells likely contributes to their
   improved function in the energetically demanding context of tumor cell
   cytotoxicity.

Knocking out HIF1A fails to enhance cytotoxicity of eNK cells in hypoxia

   Features that would improve the performance of adoptively transferred
   NK cells in solid tumors are of interest to the scientific community.
   Since HIF-1 signaling was up-regulated in hypoxia in pbNK cells over
   time (fig. S12C), we asked whether knocking out HIF1A would result in
   enhanced cytotoxicity of tumor cells in our experimental system.
   Translation of HIF1A is known to be driven by IL-15/IL-2 signaling in
   NK cells ([214]69–[215]71) so their history of activation with
   cytokines may affect their reliance on HIF-1α in hypoxic environments.
   We used a CRISPR-Cas9 system to knock out (KO) HIF1A on NK cells at an
   intermediate point of expansion (fig. S18A) and then proceeded to
   further expand the HIF1A KO cells. After 7 days in hypoxia, HIF-1α was
   detectable by flow cytometry in hypoxic wild-type (WT) cells, but the
   levels were significantly lower in KO cells, aligning with the staining
   for HIF-1α seen in normoxic cells (fig. S18, B and C). Harvesting the
   cells on day 7 revealed no differences in viability, proliferation, or
   yield in hypoxia between WT and KO cells (fig. S18, D to F), although
   there were differences in proliferation in normoxia, with KO cells
   proliferating less than WT cells (fig. S18E). HIF1A CRISPR KO did not
   induce substantial changes in either natural cytotoxicity or ADCC (fig.
   S18, G to J). Together, these data suggest that the expansion process
   primes eNK cells to function in hypoxia, not only removing their
   reliance on HIF-1α but also negating benefit of knocking out HIF1A in
   this setting.

CD16-targeted delivery of IL-15 further augments NK cell function under
hypoxia

   The NK cell activations tested (IL-15 and eNK) enhanced cytotoxicity in
   severe hypoxia, but there were opportunities to further enhance
   antitumor activity. The consistently low abundance of CD122/IL-2Rβ in
   severely hypoxic pbNK and eNK ([216]Fig. 10A) suggested that IL-15
   signaling may still be a limiting factor through this receptor. Our lab
   has previously described how targeting both CD16 and IL-15R signaling
   with tri-specific killer engager (TriKE) molecules drives robust
   proliferation of CD16^+ NK cells in preference to CD16^− T cells
   ([217]72). We hypothesized that CD16 targeting of the IL-15 moiety to
   hypoxic NK cells might overcome decreased abundance of CD122, enhancing
   signaling through the IL-15 receptor relative to IL-15. To test this,
   we measured the levels of phosphorylated signal transducer and
   activator of transcription 5 (STAT5) after an initial stimulation and
   upon restimulation with a TriKE or a dose of IL-15 (“equifunctional”)
   that drove equivalent activation/proliferation in CD16-null cells, as
   previously described ([218]73). After 4 days in culture, TriKE-treated
   cells had greater levels of phosphorylated STAT5 at baseline and upon
   restimulation than IL-15–treated cells, indicating better downstream
   signaling through IL-15 ([219]Fig. 10, B and C). In normoxia,
   restimulation with IL-15 or TriKE generated more phosphorylated STAT5
   than in severe hypoxia. This suggests that the deficit in CD122 was
   impairing STAT5 signaling relative to normoxia, even in the presence of
   TriKE (fig. S19). The antitumor element of the TriKE used here targeted
   mesothelin (SS1 TriKE) ([220]73) that can be found on the lung cancer
   cell line H460. To test whether this enhanced IL-15 signaling led to
   enhanced function, we incubated pbNK cells or eNK cells with the SS1
   TriKE or an equifunctional dose of IL-15 for 7 days in severe hypoxia,
   in the absence of any tumor target. When these cells were washed and
   challenged with H460 at the end of the 7-day incubation,
   TriKE-pretreated cells showed enhanced cytotoxic activity against these
   mesothelin-expressing targets, compared to IL-15 pretreatment alone
   ([221]Fig. 10, D and E). If TriKE was not included in the 7-day
   pretreatment, but instead added upon exposure to the H460 cells, then
   the differential effect was greatly diminished (fig. S20). This
   suggested that the benefit to the NK cells was provided by enhanced
   IL-15 signaling during hypoxia, not just because of the enhanced tumor
   targeting through CD16 that a TriKE provides. Treating with the SS1
   TriKE during exposure to hypoxia and upon tumor challenge maximized
   both IL-15 delivery and tumor-targeting mechanisms to provide the
   greatest benefit relative to IL-15 alone (fig. S20). Overall, these
   data show that pbNK and eNK cell responsiveness to IL-15 is limited by
   decreased abundance of CD122. The result of targeting CD16 and IL-15
   together is enhanced responsiveness to IL-15 signals and enhanced
   antitumor activity in hypoxia. Together, these data point to potent
   IL-15 signaling, effector protein abundance, and a stable redox balance
   as critical factors for optimal tumor control by NK cells in severe
   hypoxia.

Fig. 10. CD16-targeted delivery of IL-15 further improves function of primary
and preactivated NK cells in severe hypoxia.

   [222]Fig. 10.
   [223]Open in a new tab

   (A) Freshly enriched NK cells (pbNK) or thawed eNK cells were cultured
   in normoxia (20% O[2]) or severe hypoxia (1% O[2] + 2 psi) for 7 days
   and analyzed by CyTOF, as in [224]Fig. 7. A box (median and IQR) and
   whisker plot (minimum to maximum) show the relative abundance of
   CD122/IL-2Rβ, normalized to pbNK cells cultured in normoxia with IL-15
   (1 ng/ml), quantified by Astrolabe Diagnostics software. Mixed-effects
   analysis with Dunnett’s multiple comparisons test to 20% O[2] pbNK
   cells (eNK cells; n = 5; pbNK cells are reproduced in fig. S16 for
   reference, n = 7). (B) During 7-day culture in hypoxia, pbNK cells were
   incubated with no cytokines (empty histogram), 3 nM
   mesothelin-targeting TriKE (dark histogram), or an equifunctional dose
   of IL-15 (light histogram). Representative histograms are shown of
   pSTAT5 staining assessed by flow cytometry on day 4 of culture, before
   restimulation with drugs. (C) The median fluorescence intensity of the
   pSTAT5 signal on day 4, before (n = 4) or after (n = 6) restimulation
   with TriKE or IL-15 is quantified for independent donors. Mixed effects
   analysis with Holm-Sidak’s multiple comparisons test; bars shown the
   mean and SD. (D) Primary or (E) expanded NK cells were incubated with
   TriKE (black) or an equifunctional dose of IL-15 (gray) during 7-day
   culture in hypoxia. After washing away the drugs, cytotoxicity analysis
   of NCI-H460 was performed as in [225]Fig. 1, shown for one of three
   representative donors. In all graphs, *P ≤ 0.05 and ***P ≤ 0.001.

DISCUSSION

   We have demonstrated that oxygen concentrations commonly found in
   arterial blood (12% O[2]) do not affect NK cell cytotoxicity or
   transcriptional programming, relative to normoxic controls (20% O[2]),
   but more severe hypoxia has profound effects on NK cell function. This
   changes our interpretation of how to study NK cell biology under
   conditions of physiologic relevance. While conventional incubators are
   acceptable in modeling the oxygen concentrations found in arterial
   blood, they do a poor job of replicating the challenging conditions
   found in the TME, where 1% O[2] strongly inhibits NK cell function. In
   addition, there is a transcriptional imprint when NK cells are exposed
   to intermediate oxygen levels found in bone marrow (5% O[2]). This
   imprint does not result in altered killing by NK cells but may be
   relevant to NK cell therapy of myeloid leukemias ([226]74), now in
   clinical trial testing by many groups ([227]11, [228]22,
   [229]75–[230]77). In contrast to hematologic malignancies, solid tumors
   are more heterogeneous, with great variation in oxygen availability in
   different tumor types and locations within a tumor mass. These
   experiments have demonstrated that as NK cells penetrate into the TME
   and experience greater oxygen restriction, they reprogram themselves to
   shut down the cell cycle, increase mitophagy, autophagy and
   endocytosis, and switch over to rely on glycolysis. Proteomic and
   mitochondrial changes seen in our system resemble NK cells extracted
   from hypoxic solid tumors ([231]28, [232]62). Transcriptionally,
   reanalysis of a glioblastoma scRNA-seq dataset looking at
   tumor-infiltrating NK cells [[233]GSE147275 ([234]61)] compared to
   blood NK cells [[235]GSE130430 ([236]60)] confirmed up-regulation of
   key genes relating to glycolysis, autophagy, endocytosis, and
   mitophagy. These data suggest that our results are clinically relevant
   and that oxygen deprivation has the potential to be a major driver of
   NK cell function in vivo.

   We observed profound changes in NK cell metabolic regulation under
   hypoxic conditions. NK cells switch to rely on glycolysis with
   increased expression of LDHA. There was a decrease in mitochondrial
   metabolism, mitochondrial mass, and the net amount of ATP in cells,
   coupled with an increase in ROS. The rate of ATP production was similar
   in normoxic or severely hypoxic NK cells, suggesting that reduced ATP
   levels might result from greater energy demands in hypoxia. Among the
   pathways up-regulated in hypoxic pbNK cells, endocytosis is required in
   hypoxic epithelial cells to remove Na^+/K^+ adenosine triphosphatase
   from the plasma membrane to limit ATP expenditure ([237]78). Thus,
   endocytosis and autophagy may be a necessary step for pbNK to
   acclimatize to hypoxia by preserving ATP turnover in the cell. NK cells
   that switch to rely on glycolysis in oxygenated environments (aerobic
   glycolysis) use this metabolic pathway to proliferate, kill, and
   produce cytokines as knocking out LDHA prevents these functions
   ([238]79). In contrast, severely hypoxic NK cells rely on glycolysis
   but were only able to produce cytokines and died when LDHA was
   inhibited. Others have also shown that cytokine and chemokine
   production is independent of oxygen and glucose availability in
   short-term assays ([239]80). In mice, deletion of Cox10, a component of
   complex IV that is required for oxidative phosphorylation, did not
   impair cytokine production at baseline ([240]81), suggesting that
   cytokine production can be independent of oxygen. Nevertheless, the
   metabolic requirements for cytokine production vary with the kind of
   stimulation used ([241]42, [242]79, [243]81). Given the complexities of
   hypoxia on immune cells, we need to carefully interpret results between
   studies as precise conditions seem to determine ultimate functional
   outcomes, and understanding these mechanisms is important to develop
   effective therapeutic strategies.

   HIF-1α is regulated at many levels, but low oxygen is known to rescue
   HIF-1α protein from E3 ligase degradation. Downstream of IL-15/IL-2
   receptor activation, STAT3, and the phosphoinositide 3-kinase
   (PI3K)/mammalian target of rapamycin (mTOR) pathway increase
   transcription and translation of HIF1A in human NK cells ([244]69,
   [245]70). Murine NK cells with a HIF1A KO had less ATP, more ROS, were
   less viable, and produced less IFN-γ when examined ex vivo without
   IL-2/IL-15 activation ([246]31). In a different study, the same cells
   with strong IL-2 stimulation had better IFN-γ production than their WT
   counterparts ([247]29). Thus, cytokine and receptor activation–induced
   PI3K/mTOR stimulation can alter the cellular response to HIF-1α. Our
   results show that NK cells activated by a combination of 4-1BBL,
   membrane-bound IL-21, IL-2, and irradiated K562 feeders resulted in
   highly functional NK cells in severe hypoxia, independent of HIF-1α.
   Using a feeder-free method of NK cell expansion (IL-2, IL-18,
   anti-NKp46, and anti-CD16), others have shown that a HIF1A KO in human
   NK cells gave a subtle improvement in cytotoxicity in hypoxia, so the
   context of priming and evaluation are critical to any benefit observed
   ([248]82). While it is known that hypoxia response elements exist in
   the LDHA promoter ([249]83) and HIF-1α mediates much of its metabolic
   regulation via LDHA and pyruvate dehydrogenase kinase 1 up-regulation
   ([250]84), we conclude that the role of HIF-1α may be different for
   resting NK cells or those activated using various activation
   strategies.

   IL-15 (either intrinsic or extrinsic) and the expansion process
   enhanced cytotoxicity, and our data suggest distinct molecular
   mechanisms underlying these strategies. pbNK and eNK cells lacked CD122
   in hypoxia and pbNK cells phosphorylated STAT5 to a lesser extent in
   response to IL-15 after exposure to severe hypoxia. In IL-15–treated
   cells, this STAT5 signaling was presumably sufficient to prevent the
   observed loss of granzyme B, TRAIL, and T-bet, but several key factors
   were still decreased (Eomes and TRAIL). In contrast, eNK cells had
   higher amounts of a much wider range of critical factors (T-bet, Eomes,
   FasL, TRAIL, and activating receptors), and their ability to kill was
   not impaired. eNK responsiveness peaked early after thaw in both
   normoxia and hypoxia, but peak responsiveness for IL-15 varied by
   oxygen treatment, peaking early in hypoxia, but later in normoxia. This
   drop in responsiveness in hypoxia mirrors the temporal loss of CD122.
   While overexpression of CD122 might be expected to prevent this
   decline, it also risks exhausting NK cells through sustained IL-15
   stimulation ([251]85). If overexpression of CD122 was under the control
   of hypoxia-inducible factors, this might circumvent this issue.
   Metabolically, eNK cells were distinct from high-dose IL-15–treated NK
   cells, with a lower NAD^+/NADH ratio, better maintenance of AMP, and
   greater resistance to ROS. This aligned with a previous study of NK
   cells in liver cancer and other studies in T cells showing
   interventions that improve mitochondrial function and alleviate ROS
   enhance antitumor responses ([252]28, [253]35, [254]86). Overall, IL-15
   alone boosts cytolytic capacity in hypoxia, but greater metabolic
   stability of eNK cells renders them more adept at functioning in severe
   hypoxia, similar to what others have observed ([255]66).

   The use of expanded NK cells as immunotherapy has grown over the years
   to include overnight activated pbNK cells, NK cell lines, and cord
   blood–derived cells as well as genetically modified NK cells from
   different sources including induced pluripotent stem cells
   ([256]87–[257]89). Preclinical data using NK cells expanded with K562
   feeder cells have been published for a wide range of cancers including
   a few solid tumors ([258]77, [259]90). A phase 1 clinical trial of cord
   blood–derived, feeder-expanded CD19-chimeric antigen receptor NK cells
   infused into patients with relapsed/refractory lymphoid malignancies
   proved safe with promising clinical activity ([260]4). This ex vivo
   expansion system was initially developed to boost cell numbers for NK
   cell–based cellular products because the K562 feeder cells enable
   log-phase growth of NK cells for many weeks without inducing senescence
   ([261]91). The results from our study show that including an expansion
   step during the generation of NK cellular products prepares the cells
   to function better under hypoxia. Both 4-1BBL and IL-21 are known to
   contribute to this expansion process for therapeutic products ([262]91,
   [263]92) and have the potential to contribute to the metabolic benefit
   we observed. In T cells, short-term stimulation through 4-1BB can drive
   metabolic reprogramming to improve control of cancer in preclinical
   models ([264]93). IL-21 is a pleiotropic cytokine that could itself be
   contributing to this metabolic rewiring alone or through synergy with
   IL-2/IL-15 stimulation and activating receptor stimulation via K562.
   Regardless of the contribution of each individual factor, the combined
   effect of this expansion process—one that is well-established within
   clinical trials at present—is to prime adoptive NK cells for
   therapeutic function in hypoxia.

   While IL-15 and NK cell expansion mitigated the loss of function and
   improved NK cell killing under hypoxia, there are opportunities for
   further investigation. Many immunotherapies in development make use of
   different modalities of IL-15 ([265]7). Our study highlights the
   importance of efficient delivery of IL-15 since CD122/IL-2Rβ is
   decreased by severe hypoxia. Systemic treatment with an IL-15 receptor
   superagonist (nogapendekin alfa inbakicept), recently US Food and Drug
   Administration approved for the treatment of bladder cancer, can
   enhance rejection of allogeneic NK cells by enhancing cytolytic T cell
   responses when dosed frequently ([266]76). Therefore, targeted delivery
   of IL-15 to NK cells, cell-intrinsic IL-15 production, or careful
   dosing schema to avoid exhaustion or allogeneic depletion may be needed
   and require further investigation ([267]94).

   Our study highlights the importance of considering oxygen content of
   the environment when evaluating therapeutic potential. Immunotherapies
   have demonstrated precision and durability of responses in the clinic,
   but to broaden the applicability and success of NK cell therapies, we
   must build the suppressive elements of the TME into the design and
   testing of new immunotherapies. As NK cell immunotherapies are being
   tested in solid tumor settings, our data indicate that activation using
   targeted delivery of IL-15 and/or expansion of NK cells pose
   considerable advantages in terms of preserving cytolytic activity and
   mediating hypoxia resistance within inhospitable solid TMEs.

MATERIALS AND METHODS

   In depth descriptions of protocols are available as a collection at
   protocols.io: [268]dx.doi.org/10.17504/protocols.io.5qpvo3y9xv4o/v1.

Cell lines

   The cell lines NCI-H460, LN229, HT29, Raji, K562, and P815 were
   obtained from American Type Culture Collection, and OVCAR8 was received
   from the National Cancer Institute (NCI). Cal33 were a gift from R.
   Garcia-Escudero [Centro de Investigaciones Energéticas,
   Medioambientales y Tecnológicas (CIEMAT), Spain], and K562 and Raji
   expressing Nuclight Red were received from Fate Therapeutics,
   California. Cal33, K562, and Raji were all authenticated by short
   tandem repeat typing. The K562, Raji, H460, OVCAR8, and HT29 cells were
   cultured in R10 media: Roswell Park Memorial Institute 1640 (Gibco,
   catalog no. 2240-089) + 10% fetal bovine serum (Gibco, catalog no.
   26140079) + penicillin and streptomycin (100 U/ml; Gibco, catalog no.
   15140122). LN229, Cal33, and P815 were cultured in D10 media:
   Dulbecco’s modified Eagle’s medium (DMEM; Corning, catalog no.
   MT10013CV) + 10% fetal bovine serum (Gibco, catalog no. 26140079) +
   penicillin and streptomycin (100 U/ml; Gibco, catalog no. 15140122).
   Nuclight Red levels were maintained with puromycin (1 μg/ml;
   Sigma-Aldrich, catalog no. P9620-10ML). All cells were cultured in
   humidified incubators at 37°C, set to 5% CO[2]. All cells were
   routinely tested for mycoplasma.

Isolation and culture of immune cells

   Deidentified human blood products were obtained from Memorial Blood
   Bank (Minneapolis, MN). Their use was approved by the University of
   Minnesota and National Marrow Donor Program institutional review board
   (ID9709M00134). Blood was processed to obtain PBMCs using density
   gradient Ficoll-Paque (GE Healthcare, catalog no. GE17-5442-03). For
   experiments with enriched NK cells (pbNK), PBMCs were processed fresh
   using the EasySep Human NK Cell Enrichment Kit (STEMCELL Technologies,
   catalog no. 19055). For experiments studying the interaction of NK
   cells with T cells, the EasySep Human CD3 Positive Selection Kit II
   (STEMCELL Technologies, catalog no. 17851) was used to enrich T cells
   for addition into wells with NK cells (1:1). All these preparations of
   NK cells were cultured in R10 media supplemented with IL-15 (1 ng/ml;
   R&D Systems) in a standard humidified incubator (20% O[2], 5% CO[2],
   37°C, atmospheric pressure) or AVATAR System (Xcell Biosciences)
   incubator imitating blood (12% O[2], 5% CO[2], 37°C, + 2 psi), bone
   marrow (5% O[2], 5% CO[2], 37°C, + 0.6 psi), low pressure tumor (1%
   O[2], 5% CO[2], 37°C, + 0.3 psi), or high pressure tumor (1% O[2], 5%
   CO[2], 37°C, + 2 psi). We selected the AVATAR System, because, in our
   hands, it stabilized faster and gave more consistent results than other
   hypoxic chambers and because pressure was a variable factor in the
   experimental setup.

Assessment of oxygen and pressure as biological variables

   Two factors that can affect the severity of hypoxia are pressure and
   timing. Within tissues, oxygen availability is determined by partial
   pressure—a combination of the concentration of oxygen and the physical
   pressure the tissue experiences ([269]43). Tumors can have high
   interstitial pressure, which may affect their growth and response to
   therapy ([270]44–[271]46). Thus, we considered pressure a relevant
   variable to consider in evaluating the function of NK cells in hypoxic
   environments.

   Since oxygen and pressure were both varying in this model, we asked
   whether pressure had an effect on its own by comparing NK cell
   proliferation and cytotoxicity at 1% O[2] at low or high pressure.
   CellTrace dye was used to track NK cell proliferation under these
   conditions over 7 days. This revealed no impact of pressure on the
   number of NK cells recovered at the end of 7 days and no impact on the
   proportion of cells dividing (fig. S1B). There were measurable
   increases in degranulation at higher pressure, when NK cells were
   challenged with tumor cells to evaluate their capacity for natural
   cytotoxicity and ADCC (fig. S1C). This suggested that the lower partial
   pressure of oxygen at lower pressure might be impairing cytotoxic
   granules release. However, this did not result in improved lysis of
   tumor cells by natural cytotoxicity or ADCC under high pressure
   compared to low (fig. S1D) as measured by live cell imaging of
   fluorescent tumor cells in the presence of an indicator of apoptosis.
   [272]Figure 1 shows that, in these same assays, there were large
   differences in NK cell cytotoxicity and proliferation based on the
   different oxygen levels so this became the focus of how this research
   is presented.

   For 7-day incubations, the media was changed on day 4. This procedure
   was selected because it avoided acidification of the media in 1% O[2],
   and there were no significant differences in NK cell viability across
   oxygen conditions, as evaluated by flow cytometry annexin V^+ and
   live/dead^+ staining of NK cells among PBMCs on day 7 (fig. S1A). In
   comparison, under these same conditions, there were substantial
   differences in NK cell proliferation and cytotoxicity ([273]Fig. 1). In
   later assays with higher n numbers, measurable differences in viability
   did emerge between 20% O[2] and atmospheric pressure and 1% O[2] + 2
   psi ([274]Fig. 7). These are discussed when they appear in the text.
   However, the size of the effect was always small compared to the
   impacts on proliferation and cytotoxicity, and so data are focused on
   proliferation and cytotoxicity as key readouts that are both essential
   processes for cancer immunotherapy.

Models of therapeutic interventions—IL-15, expanded NK cells, and iNK cells

   Where enriched NK cells were cultured in IL-15 (10 ng/ml; R&D Systems)
   as a therapeutic intervention strategy and cultured as above for 7
   days, these assays were performed in parallel to the same donor treated
   with IL-15 (1 ng/ml), as the matched control. eNK cells were prepared
   according to ([275]95) with the following slight alterations. NK cells
   were enriched from PBMCs as above. K562 membrane–bound IL-21–4-1BBL
   feeder cells were a gift of Fate Therapeutics and were maintained as
   previously described ([276]96). Irradiated feeder cells and enriched NK
   cells were incubated at 2:1 ratio for the first week, and fresh
   irradiated feeders were added at a 1:1 ratio in the second week for a
   total of 14 days. The media was R10, supplemented with IL-2 (50 U/ml;
   Prometheus, catalog no. NDC 65483-116-07), which was refreshed every 2
   to 3 days. In the second week, cells were cultured in G-rex vessels
   (Wilson Wolf, catalog no. 80660M). After 14 days, cells were frozen at
   >10 million/ml in 90% fetal bovine serum and 10% dimethyl sulfoxide. To
   assess their function in hypoxia, they were thawed and placed
   immediately into culture, as above. The eNK donors were not matched to
   the pbNK donors.

   iNK cells were provided as frozen vials from Fate Therapeutics. An
   unmodified research line was provided alongside a research line with a
   single modification of the membrane-tethered fusion protein of IL-15
   and IL-15 receptor α (IL-15FP) that is one of their available edits for
   their multiplex edited products. After thaw, freezing media was removed
   by centrifugation, and cells were cultured with IL-15 (1 ng/ml; R&D
   Systems) in B0 media: 60% DMEM (Corning, catalog no. 10-017-CV), 30%
   modified Ham’s F-12 (Corning, catalog no. 10-080-CV), 10%
   heat-inactivated human AB serum (Valley Biomedical Inc., catalog no.
   HP1022), penicillin/streptomycin (100 U/ml; Gibco, catalog no.
   15140-122), 20 μM β2-mercaptoethanol (Sigma-Aldrich, catalog no.
   M7522), 50 μM ethanolamine (Sigma-Aldrich, catalog no. E0135), ascorbic
   acid (20 μg/ml; Sigma-Aldrich, catalog no. A4544), and sodium selenite
   (5 ng/ml; Sigma-Aldrich, catalog no. S5261) for up to 7 days with a
   media change on day 4.

Proliferation assays

   PBMCs ([277]Fig. 1) or enriched NK cells ([278]Fig. 7) were labeled
   with the CellTrace Violet Proliferation Kit (catalog no. [279]C34557,
   Thermo Fisher Scientific), cultured in oxygen-specific conditions as
   described above, harvested on day 7, and analyzed by flow cytometry on
   an LSR II (BD Biosciences). Relative NK cell counts on day 7, from
   300,000 cells (PBMCs or NK) on day 0, were determined by harvesting,
   staining, and resuspending the day 7 cells in a known volume (200 μl)
   that was run on the cytometer at constant speed for 60 s. Proliferation
   was assessed by dilution of the dye. NK cells were identified using
   LIVE/DEAD near infrared (NIR; catalog no. [280]L34976, Thermo Fisher
   Scientific), PE-CY7 conjugated anti-CD56 (RRID:AB_604107, HCD56,
   BioLegend), and PE-CF594–conjugated anti-CD3 (RRID:AB_604107, UCHT1, BD
   Biosciences). When viability was being directly quantified, fluorescein
   isothiocyanate (FITC)–conjugated annexin V (BD 556419, BD Pharmingen)
   was included, and cells were stained according to the manufacturer’s
   protocol. NK cells were identified as LiveDead^−/annexin
   V^−/CD56^+/CD3^−. Flow cytometric analysis was performed using FlowJo
   software (Tree Star Inc.).

Functional assays (degranulation and cytokine production)

   Following 7 days of culture in different oxygen conditions, assessment
   of NK cell degranulation by flow cytometry was carried out in those
   same oxygen conditions in a 5-hour assay. Effector cells (enriched NK
   cells) were plated in a 2:1 effector to target ratio with tumor cells:
   K562 without drug or Raji cells with rituximab (10 μg/ml; Genentech).
   Following addition of the treatments and targets, cells were stained
   with FITC-conjugated anti-CD107a (RRID:AB_1186036, clone H4A3,
   BioLegend). One hour after the addition of anti-CD107a, cells were
   given Golgi Stop and Golgi Plug (catalog nos. 554724 and 555029, BD
   Biosciences) and incubated for a further 4 hours. At the end of the
   incubation, cells were stained with the Live/Dead Fixable Aqua Staining
   Kit (catalog no. L-34966, Thermo Fisher Scientific), PE-CY7–conjugated
   anti-CD56 (RRID:AB_604107, HCD56, BioLegend), and PE-CF594–conjugated
   anti-CD3 (RRID:AB_11153674, UCHT1, BD Biosciences) and fixed [2%
   paraformaldehyde in phosphate-buffered saline (PBS)]. Cells were
   permeabilized with Triton X-100 and then stained with BV650-conjugated
   IFN-γ (RRID:AB_2563608, 4S.B3, BioLegend). The appearance of CD107a at
   the cell surface and build-up of IFN-γ within NK cells (live, CD56^+
   CD3^−) were evaluated by flow cytometry on an LSR II (BD Biosciences).

   A variant of this assay was used to examine degranulation and cytokine
   production, segregated by cell division. NK cells or PBMCs were labeled
   with the CellTrace Violet Proliferation Kit (catalog no. [281]C34557,
   Thermo Fisher Scientific) as for a proliferation assay and then
   cultured for 7 days. After 7 days, these cells were challenged in a
   redirected lysis assay where Fc receptor–bearing P815 cells cross-link
   antibodies directed against NK cell receptors (anti-CD16 clone 3G8, BD
   Biosciences, catalog no. 550383; anti-NKp30 clone P30-15 LEAF,
   BioLegend, catalog no. 325204) or isotype control antibody (mouse IgG1
   κ clone MOPC-21 LEAF, BioLegend, catalog no. 400124). These cells were
   stained, and responses were measured as above except with live/dead NIR
   (catalog no. [282]L34976, Thermo Fisher Scientific) in place of
   live/dead aqua.

Cytotoxicity by time-lapse microscopy (IncuCyte)

   K562 and Raji cells were labeled using the CellTrace Far Red
   Proliferation Kit (catalog no. [283]C34564, Thermo Fisher Scientific)
   or tracked through expression of Nuclight Red. Enriched NK cells were
   cultured in various oxygen conditions for 7 days (or 3 days where
   indicated) before challenging them with these fluorescent targets cells
   (K562 or Raji with rituximab at 10 μg/ml) in incubators capable of
   time-lapse microscopy (IncuCyte S3 and IncuCyte Zoom, Sartorius Inc.;
   humidified, 20% O[2] 5% CO[2], 37°C) for up to 48 hours. Death of tumor
   targets was tracked using Caspase-3/7 Green Apoptosis Assay reagent
   (catalog no. 4440, Essen BioScience). Images were taken every 30 min. A
   graph was created representing the percentage of live K562 and Raji
   targets (CellTrace Far Red^+, Caspase-3/7^−) normalized to the growth
   of target cells alone, and the number of live target cells in that well
   before NK cells was added (0 hours).

   For experiments involving the mesothelin-targeted TriKE (cam1615SS1),
   the TriKE was added either during the 7-day exposure to different
   oxygen conditions; at the end of the 7 days, during the live cell
   cytotoxicity assay performed in 20% O[2], 5% CO[2]; or both—as
   indicated. The TriKE was compared to an equifunctional dose of IL-15,
   calculated as that required to induce the same extent of proliferation
   in NK-92 cells, which lack CD16 and are dependent on IL-15 or IL-2 for
   survival, as described by Kennedy et al. ([284]73). To match the low
   dose of cytokine used throughout culture, during the 7-day culture, the
   dose was set to IL-15 (1 ng/ml) or an equifunctional dose of the TriKE
   (3.4 nM). When TriKE was added at the end of the culture period, the
   dose was set to 30 nM TriKE, 0.67 nM IL-15 as this had previously been
   shown to maximize cytotoxicity in this assay ([285]73).

Cytotoxicity of adherent cells by impedance (xCELLigence)

   This assay was adopted in addition to the live cell imaging assay
   because it allows efficient readout of adherent tumor cell survival
   without the need for a fluorescent reporter, so it can be used to
   evaluate a wide range of different tumor types.

   Briefly, enriched NK cells were cultured in various oxygen and pressure
   conditions described above for up to 7 days before challenging them
   with target cells (HT29, LN229, Cal33, or OVCAR8) with or without
   cetuximab in incubators capable of impedance-based cytotoxicity
   monitoring (xCELLigence RTCA, Agilent Technologies Inc.; humidified,
   20% O[2] 5% CO[2], 37°C). Seeding cell density and dose of cetuximab
   (0.1 μg/ml for cetuximab-sensitive Cal33 or 1 μg/ml for the other
   lines) were determined for optimal assay sensitivity to tumor survival
   while limiting tumor-intrinsic effects of cetuximab. Cal33 and OVCAR8
   cells were seeded at 12,500 cells per well; LN229 cells were seeded at
   25,000 cells per well; HT29 were seeded at 50,000 cells per well onto a
   96-well gold-coated plate. Impedance readings were taken every 15 min
   for 24 hours. NK cells and drugs were added at the 24-hour mark, and
   impedance readings continued for 60 hours. Impedance measurements were
   used to infer adherent tumor cell survival at each time point.
   Cytolysis was estimated within xCELLigence software (the impedance of
   treatment wells, normalized to the impedance reading when NK cells were
   added and normalized to the impedance of wells containing tumor cells
   alone) and inverted to obtain tumor survival graphs. Graphs are plotted
   from the time of NK cell addition.

CyTOF (Cytobank and Astrolabe)

   Enriched NK cells and expanded NK cells cultured for 1, 3, or 7 days in
   various oxygen conditions were stained with Cisplatin (catalog no.
   201064, Fluidigm) followed by barcoding using the Cell-ID 20-Plex Pd
   Barcoding Kit (catalog no. 201060, Fluidigm). Barcoded samples were
   combined into a single 5-ml polystyrene U-bottom tube and incubated in
   the surface marker antibody cocktail for 30 min at 4°C. Cells were then
   fixed using 2% paraformaldehyde, for 20 min. Cells were permeabilized
   with 0.1% Triton X-100 for 5 min at room temperature, followed by
   incubation with the intracellular antibody cocktail for 30 min at 4°C
   (listed in table S1). Stained cells were incubated overnight with
   Cell-ID Intercalator (catalog no. 201192A, Fluidigm). The following
   morning, cells were washed with Maxpar PBS (catalog no. 201058,
   Fluidigm). Samples were run on a time of flight mass cytometer (CyTOF
   2, Fluidigm). ViSNEs were generated of live, single NK cells
   (cisplatin^−, Cell-ID Intercalator single cells, CD45^+, CD56^+, CD3^−)
   using Cytobank software (Cytobank Inc., RRID:SCR_014043), where each
   plot was formed from data concatenated from three donors, organized on
   the basis of CD16, CD56, CD57, granzyme B, Ki67, NKG2A, and perforin.
   Differential expression analysis was performed in Astrolabe Diagnostics
   ([286]https://astrolabediagnostics.com/) where NK cells were considered
   to be CD56^+ CD3^− within live (cisplatin^−) single cells (Cell ID
   Intercalator).

   NK cell subsets were evaluated in Astrolabe Diagnostics software using
   the following definitions: the CD56^bright population (CD62L^+, CD16^−,
   CD57^−, KIR^−), CD56^dim immature population (CD62L^− CD57^− KIR^−
   NKG2A^+), CD56^dim mature population (CD62L^− CD57^− KIR^+ NKG2A^−),
   late-stage cells (CD57^+ NKG2C^−), and adaptive cells (CD57^+ NKG2C^+).
   The developmental subset definitions are not generally applied to NK
   cells activated by feeder cells, but KIR and NKG2A can dictate
   functional status ([287]57), and CD57^+ cells are known to have less
   proliferative potential ([288]97), so there is value in assessing the
   phenotypes of these populations. We acknowledge that some populations,
   such as the CD56^bright NK cells, were likely underestimated in these
   panels, for example, due to CD62L clipping. We included an “unassigned”
   cell population in plots to account for cells which fell outside these
   definitions (e.g., CD62L^+ CD16^+ cells).

RNA sequencing

   Briefly, NK cells cultured under different conditions were harvested,
   total RNA was extracted by the RNeasy Plus mini kit (Qiagen, Valencia,
   CA, USA), and RNA quality was evaluated by analysis of the RNA
   integrity number determined using a 2100 bioanalyzer (Agilent RNA 6000
   Nano Kit, Agilent Technologies, Santa Clara, CA, USA). Total RNA (0.5
   μg) was used for library prep with the TruSeq Stranded mRNA Library
   Prep Kit (Illumina) according to the manufacturer’s instructions.
   Validation of the library preparations was performed on an Agilent
   Bioanalyzer using the high sensitivity DNA kit (5067-4626, Agilent
   Technologies). Libraries were sequenced using the HiSeq sequencing
   system with a 2 × 150 bp paired-end read run (Illumina, San Diego, CA)
   by Novogene Corporation Inc. (Sacramento, CA). The overall quality of
   sequencing reads was evaluated using FastQC (v.0.11.9). Sequence
   alignments to the reference human genome (GRCh38) were performed using
   HISAT2 (v2.0.4). Read count was generated using featureCounts,
   differential expression analysis between normoxia and different O[2]
   groups was performed using DESeq2, and genes were considered
   significantly differentially expressed when the absolute log[2] fold
   change was >1 and Padj was <0.05. Gene Ontology and Kyoto Encyclopedia
   of Genes and Genomes (KEGG) pathway enrichment analysis were done using
   the g:Profiler—a web server for functional enrichment analysis
   ([289]https://biit.cs.ut.ee/gprofiler/). Data are deposited as
   GSE269629.

scRNA-seq analysis

   The peripheral blood dataset [290]GSE130430 and excised glioma dataset
   [291]GSE147275 were downloaded from the Gene Expression Omnibus
   database and reanalyzed by a commercial pipeline built by LatchBio Inc.
   (https://latch.bio/) with embedded Seurat package (v2.3.1). In the
   quality control step, we filtered out the cells that expressed <200
   genes or >2500 genes and removed cells with >5% mitochondrial
   transcripts content. These two datasets were then merged for
   differential gene expression analysis. Briefly, gene expression values
   for each cell were log-normalized and scaled on the basis of the number
   of Unique Molecular Identifiers in each cell and the cell mitochondrial
   transcript content. For differential gene expression, we used Wilcoxon
   rank-sum test and only selected the genes with adjusted P value of
   <0.05. A machine learning model using a ridge classifier was built on
   the scRNA-seq differentially expressed genes (DEGs) to evaluate the
   accuracy [i.e., (true positives + true negatives)/all] of separating NK
   cells from gliomas (considered hypoxic) from NK cells from blood
   (considered normoxic). We conducted two tests: using all DEGs
   identified from scRNA-seq analysis and using the DEGs that were only
   present in the top 50 DEGs identified from bulk RNA-seq analysis of NK
   cells exposed to hypoxia or normoxia in vitro. As a control test, we
   randomly chose six DEGs from all scRNA-seq DEGs and repeated them three
   times. For each classification test, we ran 10-fold cross-validation
   tests, and the error bar represents the SD from these 10 validation
   tests.

Oxidative phosphorylation and glycolytic rate assays

   To assess metabolic activity, enriched NK cells were harvested after 7
   days of culture and resuspended in Seahorse XF Assay Medium (Agilent
   Technologies). Plates were precoated with 0.01% poly-l-lysine solution
   (catalog no. P4707, 50 ml; MilliporeSigma) according to the
   manufacturer’s instructions, and 500,000 NK cells per well were
   immobilized in an even monolayer. The extracellular acidification rate
   and the oxygen consumption rate were measured (picomoles per minute) in
   real time on an XFe96 Extracellular Flux Analyzer (Agilent
   Technologies) after addition of drugs from the Mito Stress Test kit [1
   μM oligomycin, then 1 μM carbonyl cyanide
   p-trifluoromethoxyphenylhydrazone (FCCP), and then 0.5 μM
   rotenone/antimycin A; catalog no. 103015-100], the glycolytic rate
   assay kit (0.5 μM rotenone/antimycin A and then 50 mM
   2-deoxy-D-glucose; catalog no. 103344-100), and the ATP Rate Assay kit
   (1 μM oligomycin and then 0.5 μM rotenone/antimycin A; catalog no.
   103592-100; all Agilent Technologies). Oxygen consumption rates are
   reported from the Mito stress test and ATP rate assay; extracellular
   acidification rate is reported from the glycolytic rate assay and ATP
   rate assay.

ROS detection and ATP abundance

   To assess levels of ROS, 500,000 enriched NK cells per condition were
   stained with the 5 μM MitoSOX Red Mitochondrial Superoxide Indicator
   (catalog no. [292]M36008, Thermo Fisher Scientific). Live cells were
   identified by exclusion of TO-PRO-3 Iodide (catalog no. T3605, Thermo
   Fisher Scientific). Cells were kept on ice and evaluated on a LSR II
   flow cytometer (BD Biosciences) immediately. The levels of ATP in
   enriched NK cells were evaluated using the ATP Bioluminescence Assay
   Kit HS II (catalog no. 11699709001, Sigma-Aldrich) according to the
   manufacturer’s instructions. For this assay, 100,000 enriched NK cells
   were measured in triplicate per condition.

Quantification of mitochondrial surface area and volume

   For visualization of mitochondria in NK cells, chamber slides (μ-Slide
   eight-well polymer bottom 1.5, Ibidi, catalog no. 80826) were initially
   coated with fibronectin overnight at 4°C. After 7 days in 20 or 1%
   oxygen, NK cells were immobilized on the slides and stained with 200 nM
   MitoTracker deep red FM (Thermo Fisher Scientific, catalog no.
   [293]M22426) for 30 min at 37°C. After washing, cells were then fixed
   with 4% paraformaldehyde (Electron Microscopy Sciences, catalog no.
   15710) and permeabilized with 0.1% Triton X-100/PBS (Sigma-Aldrich,
   catalog no. 100261-4890) before a 4′,6-diamidino-2-phenylindole
   counterstain was added to visualize the nucleus (NucBlue fixed cell
   stain ReadyProbes; Thermo Fisher Scientific, catalog no. [294]R37606).
   NK cells were imaged immediately after staining on a Nikon A1Rsi HD
   confocal microscope, using 60× 1.27 numerical aperture water immersion
   objective with 405- and 640-nm laser excitation, detected using two
   GaAsP detectors and a transmitted light detector. Z-stack images were
   denoised in Nikon Elements using a Fourier transform. The mitochondrial
   structure was modeled as three-dimensional spots within Imaris
   software, and the dimensions of these spots were quantified for
   mitochondrial surface area and volume.

pSTAT5 flow cytometry

   Flow cytometry was used to examine the levels of STAT5 phosphorylation
   at Y694. After 4 days in culture, NK cells were restimulated or not
   with 3 nM SS1 TriKE or an equifunctional dose of IL-15 (0.2 nM) for 30
   min in the same oxygen conditions that they had been cultured in. See
   live cell cytotoxicity protocol for explanation of equifunctional IL-15
   dose. Cells were fixed in prewarmed BD Phosflow Fix Buffer I (catalog
   no. 557870; contains 4.2% paraformaldehyde) for 10 min. Cells were
   washed with flow buffer: 1× PBS, 0.5% human antibody serum (catalog no.
   HP-1022HI, Valley Biomedical), 2.5 mM EDTA; and then permeabilized
   using ice-cold BD Phosflow Perm Buffer III (catalog no. 558050;
   contains 87% methanol). Cells were stained with Alexa Fluor
   647–conjugated anti-STAT5 pY694 antibody (BD Biosciences, catalog no.
   612599, RRID:AB_399882) for 30 min at 4°C. Levels were evaluated using
   a BD LSR II flow cytometer.

Statistical analysis

   Statistical analysis was performed in GraphPad Prism, unless otherwise
   indicated. Details of statistical tests are given in figure legends.
   Data were tested for normality with the Shapiro-Wilk test. Where two
   variables appear in panel, e.g., time and oxygen or time and tumor
   target cells, a two-way analysis of variance (ANOVA) was first
   performed with repeated measures if donors appeared across time points
   and conditions. If the variables of interest were changing (P ≤ 0.05)
   according to the two-way ANOVA, then post hoc tests were performed as
   recommended by GraphPad Prism software. Care was taken to avoid
   unnecessary multiple testing and focus on the comparisons of interest.
   To differentiate between a post hoc test not being performed or
   returning a nonsignificant result (P > 0.05), we have marked the latter
   with gray bars on appropriate panels. Where data were missing at
   random, e.g., an autofocus failure on a time point in live cell
   imaging, a mixed-effects model was performed in preference to a two-way
   ANOVA.

Acknowledgments