(A) HeLa cells, treated with recombinant human GzmA with (right) or without (left) PFN, were analyzed by immunoelectron microscopy for the mitochondrial matrix protein Hsp60 (10 nm particles, black arrows) and GzmA (15 nm, blue arrows). Without PFN, there is no staining for GzmA. GzmA colocalizes with Hsp60 in the matrix in cells treated with both GzmA and PFN.
(B) Morphometry quantification of the number of matrix Hsp60 (black bars) or GzmA (white bars) gold particles per mitochondrial surface area. Means ± SD are indicated.

(A) HeLa cells were loaded with MitoTracker Deep Red and MitoSOX, a mitochondrial superoxide dye, and then treated with PFN alone (top) or PFN plus recombinant human GzmA (bottom) and analyzed by live-fluorescence microscopy. Superoxide anion is detected within 2 min in GzmA-exposed mitochondria, and mitochondria lose their characteristic elongated shape (insets).
(B) Electron micrographs of HeLa cells treated with GzmA alone (left) or PFN + GzmA (right). Higher magnification images are below. Mitochondria round up and lose their cristae when GzmA is delivered by PFN. Plasma membrane (P), nucleus (N), and mitochondrion (m). Results are representative of at least three experiments.

(A–C) Pseudo ρ⁰ K562 cells with reduced mtDNA, assessed by quantitative PCR (A), are resistant to GzmA-induced ROS (B) and cell death (C), assayed by annexin V and PI staining. In (C), black bars indicate K562 cells; gray bars, pseudo ρ⁰ K562 cells.
(D and E) The complex I inhibitor rotenone (Rot) and the superoxide anion scavenger Tiron (80 mM) inhibit GzmA-induced ROS (D) and cell death (E). K562 cells were pretreated with rote-none or Tiron and then incubated with PFN and/or GzmA. Light and dark gray bars indicate cells pretreated for 2 hr with 0.25 μM and 1 μM of rote-none, respectively.
(F) Scavenging superoxide anion does not affect upstream cleavage of NDUFS3 but inhibits downstream cleavage of SET when K562 cells are exposed to GzmA and PFN.
For all panels, results are means ± SD for representative experiments performed at least three times.

(A) Proteomic analysis. Freshly purified mouse liver mitochondria were treated with buffer or human GzmA for 10 min and resolved by 2D gel electrophoresis and silver staining. Most spots were unchanged. Spots 1 and 2 on the control gel, corresponding to missing proteins in the GzmA-treated sample, were excised and characterized by mass spectrometry.
(B) Magnification of the region corresponding to spots 1 and 2. Excised spots are indicated with arrows. Mass spectroscopy identified spot 1 as Ndufs3 and spot 2 as Atp5h. Spot 1 partially overlaps with another unchanged spot of higher apparent molecular mass and more acidic isoelectric point.
(C and D) Human K562 cells treated with human GzmA and PFN confirmed that GzmA specifically cleaves NDUFS3, but not ATP5H, in a dose- (C) and time-dependent (D) manner. SET is a previously described GzmA target. NDUFS3 was cleaved by GzmA but not GzmB. However, ATP5H was not a substrate for either Gzm. NDUFA9 is another complex I protein and Hsp60 is a mitochondrial loading control. Results are representative of at least four experiments.

(A) Fluorescently labeled GzmA (GzmA 488) is internalized by purified mouse liver mitochondria, but the control fluorescent recombinant SET protein (SET 488) is not. Purified mouse liver mitochondria were incubated with SET 488 or GzmA 488, gently trypsinized to remove surface proteins, and then analyzed by flow cytometry. Valinomycin, which disrupts ΔΨ_m, inhibits GzmA entry into mitochondria. Results are representative of at least three independent experiments.
(B) GzmA interacts with Hsp70 and Hsp90, two chaperones implicated in mitochondrial import. K562 cells treated with buffer or freeze-thawed human CTL granules were lysed and immunoprecipitated (IP) using GzmA antibody in the presence or absence of 10 mM of ATP and blotted for Hsp70, Hsp90, GzmA, and GAPDH. GzmA interacts with Hsp70 and Hsp90, but not with GAPDH, but only in the absence of 10 mM ATP, which inhibits Hsp chaperone function. GzmA was not detected in the input, which was loaded with 10% of the reaction mixture.
(C) Valinomycin also inhibits GzmA cleavage of Ndufs3 in isolated mouse liver mitochondria. (D and E) Valinomycin at nanomolar concentrations that disrupt ΔΨ_m (D) but are not toxic interferes with ROS production in K562 cells treated with GzmA and PFN (E). Cells treated with buffer or GzmA or PFN alone are background controls.

Mouse EL4 cells were stably transfected with vector, HA-tagged wild-type (WT) NDUFS3, or K56A-NDUFS3 expression vectors (Figure 3E).
(A) EL4 cells overexpressing K56A-NDUFS3 generate reduced ROS in response to GzmA and PFN, compared to cells transduced with WT NDUFS3 or vector. (Adding GzmA without PFN tended to reduce background ROS detection in cells for all conditions for unclear reasons, suggesting that GzmA may scavenge ROS.)
(B and C) Overexpression of K56A-NDUFS3 also protects EL4 cells from GzmA- and PFN-induced cell death measured by annexin V-PI staining (B) but does not substantially affect UV-induced ROS generation (C) or death (data not shown). Cells were treated with buffer (−), 0.5 μM GzmA (A), PFN (P), PFN plus 0.25 μM GzmA (gray bars), or PFN plus 0.5 μM GzmA (black bars).
(D) Mutant NDUFS3 also inhibits cleavage of the SET homolog I2PP2A in overexpressing EL4 cells. (E and F) These stably transfected EL4 cell lines were coated with gp33 peptide and used as targets for CTL effector cells derived from GzmB^−/− (E) or GzmA^−/− (F) mice in a 6 hr ⁵¹Cr-release assay. E:T ratio, effector:target cell ratio. Black bars represent EL4 cells transduced with vector; white bars, WT NDUFS3; gray bars, K56A-NDUFS3. Overexpression of K56A-NDUFS3 protects cells from GzmA-expressing GzmB^−/− CTLs but not from GzmA^−/− CTLs.
For all panels, results shown are means ± SD of at least 4–6 independent experiments. *p ≤ 0.02 compared with vector-transduced target cells and p < 0.03 compared with WT NDUFS3-transduced target cells.

(A) rhNDUFS3 was specifically and directly cleaved by human GzmA, but not mouse GzmB, to a 26.3kDa C-terminal fragment detected by immunoblot with His₆ antibody. Mass spectroscopy of this fragment identified Lys56 as the NDUFS3 cleavage site.
(B–D) To confirm the cleavage site and its relevance in intact cells, WT or K56A-NDUFS3 or control vector were expressed with a HA tag in K562 cells (B, immunoblot with HA antibody) and verified to be incorporated into complex I by HA antibody immunoprecipitation (C). NDUFA9, MTND6, and NDUFB6 are other complex I proteins (D). When transduced K562 cells were treated with GzmA and PFN for 1 hr, WT NDUFS3 was degraded, verifying Lys56 as the in vivo GzmA cleavage site. K56A-NDUFS3 is cleaved somewhat by GzmA but is more resistant than WT NDUFS3. Cleavage of the GzmA target SET serves as a positive control and Hsp60 is a negative loading control.
(E) Similarly K56A-NDUFS3 overexpressed in mouse EL4 cells (immunoblot, left) was relatively resistant to cleavage by GzmA. EL4 cells transduced with WT or K56A-NDUFS3 were lysed and treated for 20 min with the indicated GzmA concentration (right) or treated for indicated times with 250 nM GzmA (bottom). In (B)–(E) transduced NDUFS3 was detected by HA antibody.
(F–H) GzmA also cleaves mouse Ndufs3 in purified mouse complex I and inhibits complex I activity. Complex I was purified either by immunoprecipitation using a cocktail of anticomplex I subunit antibodies (F) or by anion exchange and size exclusion chromatography (G). Ndufs3 in complex I was cleaved to a 24 kDa C-terminal fragment. Treatment with GzmB or inactive S-AGzmA had no effect on Ndufs3. Another complex I protein, Ndufb6, was unchanged. (H) GzmA disrupts complex I activity. Complex I activity was assayed 15 min after mitochondria were treated for 15 min with S-AGzmA or GzmA by measuring NADH oxidation by absorbance at 340 nm. Inhibition of NADH oxidation required active GzmA and increased in both a dose- and time-dependent manner (data not shown). Means±standard deviation (SD) are indicated. *p ≤ 0.05 by 2-sided t test.