Abstract
Preclinical studies demonstrated that complement promotes tumor growth. Therefore, we sought to determine the best target for complement-based therapy among common human malignancies. High expression of 11 complement genes was linked to unfavorable prognosis in renal cell carcinoma. Complement protein expression or deposition was observed mainly in stroma, leukocytes, and tumor vasculature, corresponding to a role of complement in regulating the tumor microenvironment. Complement abundance in tumors correlated with a high nuclear grade. Complement genes clustered within an aggressive inflammatory subtype of renal cancer characterized by poor prognosis, markers of T cell dysfunction, and alternatively activated macrophages. Plasma levels of complement proteins correlated with response to immune checkpoint inhibitors. Corroborating human data, complement deficiencies and blockade reduced tumor growth by enhancing antitumor immunity and seemingly reducing angiogenesis in a mouse model of kidney cancer resistant to PD-1 blockade. Overall, this study implicates complement in the immune landscape of renal cell carcinoma, and notwithstanding cohort size and preclinical model limitations, the data suggest that tumors resistant to immune checkpoint inhibitors might be suitable targets for complement-based therapy.
Introduction
Early studies demonstrated deposition of complement proteins in several human malignancies (1). Therefore, complement was thought to contribute to immune surveillance through complement-dependent cytotoxicity and tumor cell lysis (2). These roles for complement are particularly clear in the context of anticancer mAb therapies in which complement-dependent cytotoxicity assists in tumor cell killing, especially in hematologic malignancies (3). However, because of overexpression of membrane and soluble complement regulatory (i.e., inhibitory) proteins, solid tumors are protected from complement-mediated lysis (2). Consequently, the functional significance of complement activation in tumors in the absence of therapeutic Abs has remained unclear.
In contrast to these beneficial antitumor complement roles, several studies demonstrated that the complement system promotes tumor growth by inhibiting antitumor immunity (4–6). In fact, complement is currently perceived as an important immunosuppressive mechanism in primary tumors (6, 7) and metastasis-targeted organs (8, 9). Complement proteins activate and recruit immunosuppressive cells, including myeloid-derived suppressor cells (MDSC), tumor-associated macrophages, and regulatory T cells, to tumors and premetastatic niches (7, 10). Recent work also demonstrated synergism between programmed cell death protein 1 (PD-1) blockade and complement inhibition to reduce tumor growth (11). Interestingly, the C5a/C5a receptor 1 (C5aR1) axis was shown to have prognostic value in human renal cell carcinoma (RCC) (12). In addition, the C1q and the classical pathway appear to contribute to RCC progression (13). However, despite this foundational knowledge, complement-based anticancer therapies have not yet advanced to the clinic (14). This may be due to the following two factors: 1) lack of understanding of best therapeutic targets within the complement cascade and 2) lack of knowledge about which cancer patients might benefit from complement-based therapies.
We performed a systematic analysis of expression of several complement genes in human solid tumors to identify: 1) cancer patients with deregulated complement that might potentially benefit from complement-based interventions and 2) complement-dependent mechanisms regulating tumor growth that can be targeted for therapy. We complemented these studies with analyses of complement proteins in plasma and investigated their predictive potential for the response to immune checkpoint inhibitors (ICI). Findings in patients were corroborated in a mouse model of RCC.
Materials and Methods
Human samples and data
Data on complement genes’ RNA expression and survival were obtained from Human Protein Atlas (https://www.proteinatlas.org/) and The Cancer Genome Atlas (https://cancergenome.nih.gov/) or from University of Texas Southwestern Medical Center Kidney Cancer Program (UTSW KCP) as previously reported (15). The p values included in Table I and Fig. 1 are based on the Kaplan–Meier survival analysis available through the Human Protein Atlas. Cutoffs are based on the lowest p value for separation corresponding to a maximum difference in survival. Deidentified blood samples from RCC patients were obtained from UTSW KCP or from healthy donors from Oklahoma Blood Institute (Oklahoma City, OK) following informed consent under the Institutional Review Board approved protocols (STU 012011-190 and STU 102010-051). Blood from UTSW KCP was collected with EDTA or with sodium heparin (CPT Cell Preparation Tubes) as anticoagulants. Blood from healthy donors was collected with EDTA. Blinded analysis of immunohistochemical staining of complement proteins in RCC and RCC grade, based on images available through Human Protein Atlas (https://www.proteinatlas.org/), was performed by a board-certified pathologist (M.M.M.). Scores from 1 to 5 were assigned based on semiquantitative evaluation of intensity and pattern of staining and correlated with nuclear grade as established based on the International Society of Urological Pathology recommendations (16).
Data availability
Sequencing data from UTSW KCP patients specifically consenting to placement of their raw genomic data in a protected publicly accessible database are deposited in the European Genome–Phenome Archive (https://www.ebi.ac.uk/ega/home) with accession numbers EGAS00001002786 and EGAS00001000926.
Mice, cell lines, and treatments
Mouse studies were approved by the Institutional Animal Care and Use Committee of the Texas Tech University Health Sciences Center. Eight- to twelve-week-old BALB/c, C3ar1, and C5ar1 knockout (KO) mice from The Jackson Laboratory were injected s.c. with 1 × 106 Renca cells (American Type Culture Collection CRL2947). When tumors reached ∼5 mm in diameter, mice were assigned to treatment cohorts and treated with C3aR1 inhibitor SB290157 (10 mg/kg i.p. twice a day; Sigma-Aldrich), C5aR1 inhibitor PMX53 (17) (1 mg/kg, i.p. every other day), or PD-1 Ab (RMP1-14, BE-0146, 250 μg per mouse i.p. every 4 d; Bio X Cell, Lebanon, NH) or RaIgG2a isotype control (BE-0089, 250 μg i.p. every 4 d; Bio X Cell), as previously described (8, 18, 19). CD8+ T cells were depleted by i.p. injection of 200 μg of CD8α-neutralizing Ab (2.43,; Bio X Cell) per mouse each day for 3 consecutive d prior to injecting tumor cells. To maintain CD8+ T cell depletion, mice were injected with 200 μg of Ab every third day beginning at day 3 after tumor cell injection. BALB/c control mice were treated in the same manner with rat IgG2b (LTF-2; Bio X Cell).
Immunofluorescence
Five-micrometer-thick frozen tissue sections were stained with CD31 (clone 390; BD Pharmingen); CD8a (53–6.7; BD Pharmingen); CD11b (550282; BD Biosciences), CD88 (C5aR1) (sc-31240; Santa Cruz Biotechnology); C3b/iC3b/C3c, which binds only to C3 cleavage fragments, but not to intact C3 (20) (HM 1065; Hycult Biotech); C1q (HM 1096BT; Hycult Biotech); mannan binding lectin (NB100-1502; Novus Biologicals); IgM Abs (14-5790-81; eBioscience); and Annexin V (sc-1929; Santa Cruz Biotechnology). Secondary Abs included goat anti-rat Abs (Invitrogen): streptavidin-Cy2, Texas Red, and Alexa Fluor (AF) 633–conjugated; and donkey anti-goat AF 488–conjugated Abs. Stainings were quantified with Nikon Elements Advanced Research Image-Analysis software based on analysis of at least 10 fields per section. Data are expressed as the binary area occupied by positive cells.
Flow cytometry
Cells from tumors were preincubated with CD16/32 Ab (Fc block; 2.4G2; BD Pharmingen) and stained with fluorochrome-conjugated Abs from BioLegend: BV605-CD45 (30-F11), AF 700–CD3 (17A2), PerCP/Cy5.5-CD4 (GK1.5), and PE/Cy7-CD8a (53-6.7) as recommended by the manufacturer. To quantify IFN-γ expression, cells were stained with surface markers and permeabilized with Cytofix/Cytoperm (554714; BD Biosciences), and washed with 1× Perm/Wash buffer (554714; BD Biosciences), followed by incubation with PE–IFN-γ (XMG1.2; BioLegend). Prior to intracellular staining, cells were incubated with brefeldin A and monensin (BD Biosciences) in the presence of CD3 and CD28 Abs (17A2 and 37.51; eBioscience) adsorbed to the 96-well plates for 6–8 h. Data were acquired on BD Fortessa and analyzed with FlowJo software (Tree Star).
Real-time quantitative PCR
RNA was extracted from frozen tissue using RNeasy Plus Mini Kit (QIAGEN), and cDNA was generated using High-Capacity RNA-to-cDNA Kit (Applied Biosystems). The real-time quantitative PCR was performed using High-Capacity cDNA Synthesis Kit, Fast SYBR Green, and StepOnePlus (Applied Biosystems). Relative expression was calculated using the 2−ΔΔCt method and RT2 profiler PCR Array Data Analysis (SAB Biosciences) and normalized to GAPDH. The primer sequences are as follows: Vegfa, 5′-CTGCTGTAACGATGAAGCCCTG-3′ and 5′-GCTGTAGGAAGCTCATCTCTCC-3′; Vegfb, 5′-ACTGGGCAACACCAAGTCCGAA-3′ and 5′-CACATTGGCTGTGTTCTTCCAGG-3′; Vegfc, 5′-CCTGAATCCTGGGAAATGTGCC-3′ and 5′-CGATTCGCACACGGTCTTCTGT-3′; Angpt1, 5′-AACCGAGCCTACTCACAGTACG-3′ and 5′-GCATCCTTCGTGCTGAAATCGG-3′; Tek, 5′-GAACTGAGGACGCTTCCACATTC-3′ and 5′-TCAGAAACGCCAACAGCACGGT-3′; Il1b, 5′-TGGACCTTCCAGGATGAGGACA-3′ and 5′-GTTCATCTCGGAGCCTGTAGTG-3′; Ctla4, 5′-GTACCTCTGCAAGGTGGAACTC-3′ and 5′-CCAAAGGAGGAAGTCAGAATCCG-3′; Pdcd1, 5′-ACCCTGGTCATTCACTTGGG-3′ and 5′-CATTTGCTCCCTCTGACACTG-3′; Btla, 5′-CTTCTGCTCCTTGCCTGTGTCT-3′ and 5′-GGTTAGTGTCCCTTCCTGCCAA-3′; Fas, 5′-CTGCGATTCTCCTGGCTGTGAA-3′ and 5′-CAACAACCATAGGCGATTTCTGG-3′; and Stat3, 5′-AGGAGTCTAACAACGGCAGCCT-3′ and 5′-GTGGTACACCTCAGTCTCGAAG-3′.
ELISA of human and mouse plasma
Mouse C5a ELISA was performed according to the manufacturer’s instructions (DY2150; R&D Systems). Human complement ELISA kits C1q (HK356-02), C3 (HK366-02), C5 (HK390-02), factor B (FB; HK367-02), factor H (FH; HK342-02), factor D (FD; HK-343-02), factor I (FI; HK355-02), C3c (HK-368), soluble CD59 (sCD59; HK374-02), and soluble C5b-9, known as membrane attack complex or complement terminal complex (TCC) (HK328-01), were obtained from Hycult Biotech (Uden, the Netherlands) and were used according to the manufacturer’s recommendations.
Statistics
Data were analyzed with t test or one-way ANOVA (more than two mean-value comparison). Impact of treatments on growth of mouse tumors over time was evaluated by two-way ANOVA. The log-rank test was used for patient survival analysis and survival data were visualized by the Kaplan–Meier estimator (the Human Protein Atlas). To determine the predictive value of complement proteins for the response to ICI, we used the time to next treatment (TNT) as a surrogate of response. We divided patients into cohorts with protein concentration above or below a set threshold. Then, we compared the distribution of TNT and calculated the p value of a difference between the cohorts. The optimum threshold was set at the minimum p value of separation, corresponding to a maximal difference in TNT. Multiple-protein analysis used the scikit-learn machine learning library version 0.22 (21). A p value < 0.05 was considered significant. Bar graphs indicate mean ± SEM. GraphPad Prism 6 was used for analyses.
Results
High expression of complement genes is associated with unfavorable prognosis in RCC
Using data available through the Human Protein Atlas (https://www.proteinatlas.org/) and The Cancer Genome Atlas (https://cancergenome.nih.gov/), we analyzed expression of complement genes in human solid tumors (Table I). We found 11 soluble complement proteins, receptors, and regulators that were associated with poor prognosis in RCC (Fig. 1A–J [10 genes are shown], Table I). One complement protein, CD59, was associated with improved outcomes (Fig. 1K), but CD59 is a negative regulator of the complement system, also known as C5b-9/membrane attack complex/TCC-inhibitory protein (22). In contrast to RCC, several complement genes were linked to favorable prognosis in other common human tumors, including liver, pancreatic, breast, and cervical carcinomas (Table I).
Complement Gene . | Cancer Type . | Prognosis . | p Value . |
---|---|---|---|
C1QA | Renal | Unfavorable | 1.58 × 10−6 |
C1QB | Renal | Unfavorable | 9.58 × 10−6 |
C1S | Renal | Unfavorable | 7.44 × 10−15 |
Liver | Favorable | 7.73 × 10−4 | |
C1R | Renal | Unfavorable | 1.94 × 10−14 |
C2 | Renal | Unfavorable | 2.64 × 10−7 |
C3 | Renal | Unfavorable | 1.09 × 10−5 |
Liver | Favorable | 8.11 × 10−4 | |
C5 | Liver | Favorable | 9.43 × 10−4 |
C6 | Liver | Favorable | 4.31 × 10−4 |
C7 | Liver | Favorable | 5.89 × 10−4 |
C8B | Liver | Favorable | 9.30 × 10−5 |
CFB | Renal | Unfavorable | 1.53 × 10−5 |
Breast | Favorable | 2.76 × 10−5 | |
CFD | Pancreatic | Favorable | 1.77 × 10−4 |
Renal | Unfavorable | 7.09 × 10−4 | |
CFH | Renal | Unfavorable | 1.92 × 10−6 |
CFI | Urothelial | Unfavorable | 6.06 × 10−4 |
CD21/CR2 | Breast | Favorable | 6.90 × 10−4 |
CD46 | Cervical | Unfavorable | 8.54 × 10−5 |
Stomach | Favorable | 2.38 × 10−4 | |
CD55 | Renal | Unfavorable | 9.96 × 10−4 |
CD59 | Renal | Favorable | 1.30 × 10−9 |
Pancreatic | Unfavorable | 2.92 × 10−5 | |
Head/Neck | Unfavorable | 3.10 × 10−5 | |
Cervical | Unfavorable | 3.81 × 10−5 | |
C5AR1 | Renal | Unfavorable | 1.16 × 10−4 |
Testis | Unfavorable | 7.99 × 10−4 | |
Ovarian | Unfavorable | 9.55 × 10−4 | |
Cervical | Favorable | 4.71 × 10−4 |
Complement Gene . | Cancer Type . | Prognosis . | p Value . |
---|---|---|---|
C1QA | Renal | Unfavorable | 1.58 × 10−6 |
C1QB | Renal | Unfavorable | 9.58 × 10−6 |
C1S | Renal | Unfavorable | 7.44 × 10−15 |
Liver | Favorable | 7.73 × 10−4 | |
C1R | Renal | Unfavorable | 1.94 × 10−14 |
C2 | Renal | Unfavorable | 2.64 × 10−7 |
C3 | Renal | Unfavorable | 1.09 × 10−5 |
Liver | Favorable | 8.11 × 10−4 | |
C5 | Liver | Favorable | 9.43 × 10−4 |
C6 | Liver | Favorable | 4.31 × 10−4 |
C7 | Liver | Favorable | 5.89 × 10−4 |
C8B | Liver | Favorable | 9.30 × 10−5 |
CFB | Renal | Unfavorable | 1.53 × 10−5 |
Breast | Favorable | 2.76 × 10−5 | |
CFD | Pancreatic | Favorable | 1.77 × 10−4 |
Renal | Unfavorable | 7.09 × 10−4 | |
CFH | Renal | Unfavorable | 1.92 × 10−6 |
CFI | Urothelial | Unfavorable | 6.06 × 10−4 |
CD21/CR2 | Breast | Favorable | 6.90 × 10−4 |
CD46 | Cervical | Unfavorable | 8.54 × 10−5 |
Stomach | Favorable | 2.38 × 10−4 | |
CD55 | Renal | Unfavorable | 9.96 × 10−4 |
CD59 | Renal | Favorable | 1.30 × 10−9 |
Pancreatic | Unfavorable | 2.92 × 10−5 | |
Head/Neck | Unfavorable | 3.10 × 10−5 | |
Cervical | Unfavorable | 3.81 × 10−5 | |
C5AR1 | Renal | Unfavorable | 1.16 × 10−4 |
Testis | Unfavorable | 7.99 × 10−4 | |
Ovarian | Unfavorable | 9.55 × 10−4 | |
Cervical | Favorable | 4.71 × 10−4 |
Expression of complement genes in RCC and prognosis based on data from the Human Protein Atlas. Survival probabilities in patients with high versus low expression of complement genes: (A) C1QA, cutoff = 197.13 fragments per kilobase of transcript per million mapped reads (FPKM); (B) C1QB, cutoff = 184.01 FPKM; (C) C1S, cutoff = 37.42 FPKM; (D) C1R, cutoff = 38.62 FPKM; (E) C2, cutoff = 1.99 FPKM; (F) C3, cutoff = 97.97 FPKM; (G) C5AR1, cutoff = 4.36 FPKM; (H) CFB, cutoff = 4.86; (I) CFD, cutoff = 5.09 FPKM; (J) CFH, cutoff = 8.72 FPKM; and (K) CD59, cutoff = 79.49 FPKM.
Expression of complement genes in RCC and prognosis based on data from the Human Protein Atlas. Survival probabilities in patients with high versus low expression of complement genes: (A) C1QA, cutoff = 197.13 fragments per kilobase of transcript per million mapped reads (FPKM); (B) C1QB, cutoff = 184.01 FPKM; (C) C1S, cutoff = 37.42 FPKM; (D) C1R, cutoff = 38.62 FPKM; (E) C2, cutoff = 1.99 FPKM; (F) C3, cutoff = 97.97 FPKM; (G) C5AR1, cutoff = 4.36 FPKM; (H) CFB, cutoff = 4.86; (I) CFD, cutoff = 5.09 FPKM; (J) CFH, cutoff = 8.72 FPKM; and (K) CD59, cutoff = 79.49 FPKM.
Complement proteins are deposited in tumor stroma and their abundance correlate with histological grade
Tumor-promoting functions of complement proteins are linked to an immunosuppressive tumor microenvironment (TME) (23). However, some studies demonstrated that complement promotes tumor growth via direct autocrine effect on tumor cells that is independent from inhibiting antitumor T cells (24). Therefore, it is important to determine localization of complement proteins in tumors. We focused our analyses on C1qA (n = 22), C1qB (n = 11), C3 (n = 29), and C5aR1 (n = 22) because of the prognostic value of these proteins in RCC, their strategic positions in the complement cascade, and their roles in regulating tumor growth in mouse models (6). Immunohistochemistry slides (84 samples) from 43 RCC patients, available through the Human Protein Atlas, were analyzed. C1qA was present as extracellular deposits in stroma (Fig. 2A) or was associated with the vasculature (Fig. 2B) and/or infiltrating cells/leukocytes (Fig. 2C) in all 22 tumor samples. Of nine samples with cytoplasmic staining of C1qA in tumor cells, six were clear-cell RCC (ccRCC), and three were papillary RCC (pRCC) (Fig. 2D, 2E). C1qB staining was limited to stromal deposits, scarce infiltrating cells, and vasculature (Fig. 2F) in 10 out of 11 ccRCC samples. Weak and focal staining of tumor cells was found only in one high-grade ccRCC (Fig. 2G). The most consistent C3 staining pattern was stromal and vascular deposition in all 29 samples, regardless of histologic subtype (Fig. 2H, 2I). C3-positive infiltrating cells were observed in 10 of 29 tumor samples (Fig. 2J). C3 cytoplasmic and membrane staining of tumor cells was observed in 12 out of 24 ccRCC sections (Fig. 2K) and in two out of five pRCC sections (Fig. 2L). C5aR1 expression was limited to infiltrating cells (Fig. 2M) and the vasculature (Fig. 2N) in 17 out of 22 ccRCC samples. Membrane staining of tumor cells was observed in only five high-grade ccRCC (Fig. 2O). A semiquantitative analysis of C1qA and C3 staining demonstrated that high-grade tumors (nuclear grades 3–4) had more widespread staining and higher intensity than low-grade (1, 2) tumors (Fig. 2P–S). Overall, these studies show that complement proteins are largely associated with the stroma in RCC.
Spatial distribution of complement proteins in RCC tumors and their association with grade. Detection of complement proteins in RCC tumors by immunohistochemistry: (A–E) C1qA, (F and G) C1qB, (H–L) C3, and (M–O) C5aR1. Scale bar, 50 μm. For (A–O) n1 = 22 males, average age 64 ± 9.25, and n2 = 24 females, average age 64.6 ± 10. (P) C1qA expression by immunohistochemistry in low (grade 1–2)– versus high (grade 3–4)–grade tumors. (Q) Semiquantitative analysis of C1qA expression. (P) n1 low grade = 8 and n2 high grade = 9. *p < 0.0001, by t test. (R) C3 expression by immunohistochemistry in low- versus high-grade tumors. (S) Semiquantitative analysis of C3 expression (C), n1 low grade = 10 and n2 high grade = 13. *p < 0.0001 by t test. Inf. Cells, expression in infiltrating cells; Stroma, stromal deposition; Tumor, expression in tumor cells; Vasculature, vascular presence.
Spatial distribution of complement proteins in RCC tumors and their association with grade. Detection of complement proteins in RCC tumors by immunohistochemistry: (A–E) C1qA, (F and G) C1qB, (H–L) C3, and (M–O) C5aR1. Scale bar, 50 μm. For (A–O) n1 = 22 males, average age 64 ± 9.25, and n2 = 24 females, average age 64.6 ± 10. (P) C1qA expression by immunohistochemistry in low (grade 1–2)– versus high (grade 3–4)–grade tumors. (Q) Semiquantitative analysis of C1qA expression. (P) n1 low grade = 8 and n2 high grade = 9. *p < 0.0001, by t test. (R) C3 expression by immunohistochemistry in low- versus high-grade tumors. (S) Semiquantitative analysis of C3 expression (C), n1 low grade = 10 and n2 high grade = 13. *p < 0.0001 by t test. Inf. Cells, expression in infiltrating cells; Stroma, stromal deposition; Tumor, expression in tumor cells; Vasculature, vascular presence.
Complement gene expression is associated with an aggressive inflammatory subtype of RCC
We previously reported the discovery of an inflammatory subtype (IS) of RCC characterized by local immune cell infiltration, systemic inflammation, poor prognosis, and BAP1 mutations (15). In that report, we used publicly available RNA-sequencing datasets from The Cancer Genome Atlas (n = 529 ccRCC) as well as from UTSW KCP (n = 181, including 39, 24, 15, and 22% patients with ccRCC, pRCC, chromophobe, and other subtypes, respectively). The IS cluster was enriched for gene signatures of regulatory T cells, NK cells, Th1 cells, neutrophils, macrophages, B cells, CD8+ T cells, and C1q (15). The identification of CD8+ T cells (tumor-infiltrating lymphocytes [TIL]) in this cluster was not surprising, as CD8+ T cells have been previously associated with poor prognosis in RCC, unlike in other tumor types (25).
We reanalyzed the UTSW KCP data for complement-related genes. CFB, C5AR1, CFH, C3, C1R, C1S C1QA, and C1QB were enriched in the IS in comparison with the non-IS subtype (NIS), especially for ccRCC patients (Fig. 3A, blue gene symbols in boxes). Conversely, complement regulatory genes (CD46, CD55, and CD59) encoding proteins inhibiting/controlling complement activation were enriched in NIS RCC (Fig. 3A,green gene symbols). These data extend the findings in Fig. 1 by showing that complement protein expression is associated with an IS of RCC, which we previously showed is characterized by poor prognosis (15). The findings were particularly striking for the ccRCC subtype.
Expression of complement genes in IS and NIS of RCC. (A) Hierarchical clustering of the UTSW KCP RCC tumors based on TME genes. Gene or single-sample Gene Set Enrichment Analysis scores for immune cells, angiogenesis, and complement genes (blue and green gene symbols for genes associated with unfavorable and favorable prognosis, respectively); red meaning higher activation/expression scores. BAP1 and PBRM1 mutations are shown with black bars. The red bar denotes ccRCC (n = 70), and white bar denotes all other RCC (n = 111). Expression of genes corresponding to (B) complement, (C) chemokine, (D) T cell regulation and exhaustion, (E) myeloid cell regulation and function, and (F) enzymes in IS versus NIS of the same patients as in (A). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by t test. (G) Correlative analysis of complement gene expression with genes associated with immunosuppression.
Expression of complement genes in IS and NIS of RCC. (A) Hierarchical clustering of the UTSW KCP RCC tumors based on TME genes. Gene or single-sample Gene Set Enrichment Analysis scores for immune cells, angiogenesis, and complement genes (blue and green gene symbols for genes associated with unfavorable and favorable prognosis, respectively); red meaning higher activation/expression scores. BAP1 and PBRM1 mutations are shown with black bars. The red bar denotes ccRCC (n = 70), and white bar denotes all other RCC (n = 111). Expression of genes corresponding to (B) complement, (C) chemokine, (D) T cell regulation and exhaustion, (E) myeloid cell regulation and function, and (F) enzymes in IS versus NIS of the same patients as in (A). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by t test. (G) Correlative analysis of complement gene expression with genes associated with immunosuppression.
The direct comparison of complement genes in IS versus NIS demonstrated higher expression of genes associated with poor prognosis in IS. In contrast, CD59, which is associated with favorable prognosis, had high expression in NIS together with another complement regulator CD46 (Fig. 3B). In addition, several chemokine encoding genes had relatively higher expression in the IS versus NIS (Fig. 3C). These chemokines are implicated in regulating growth of several cancers and in recruiting immune cells to tumors. Further, their expression is associated with T cell exhaustion (26, 27). Importantly, high expression of CCL4, CCL5, CXCL10, and CXCL11 was associated with worse outcomes in RCC (Supplemental Fig. 1A–D).
Poor prognosis associated with TIL in RCC and high number of TIL in IS suggest that these T cells are dysfunctional. We evaluated genes associated with T cell exhaustion/dysfunction (28) in IS versus NIS and found that they were upregulated in IS (Fig. 3D). Several of these genes (highlighted by pink background in Fig. 3D) were also associated with reduced patient survival (Supplemental Fig. 1E–G). Because macrophage genes were enriched in IS (15) and tumor-associated macrophages and other cells of myeloid origin in tumors are immunosuppressive (29), we evaluated genes linked to myeloid cell regulation and function. Several alternatively activated macrophage markers and macrophage regulators (30) had higher expression in IS versus NIS (Fig. 3E). Among those, high expression of CD86, IRF1, STAB1, TGFB1, F13A1, IL-6, and CD40 was associated with an unfavorable prognosis (Supplemental Fig. 1H–N). Genes encoding enzymes involved in extracellular matrix remodeling and immunosuppression also had higher expression in IS versus NIS (Fig. 3F). In this category, TGM2 and IDO1 were associated with lower survival (Supplemental Fig. 1O, 1P). Importantly, several genes potentially involved in suppression of antitumor immunity (see Fig. 3C–F) positively correlated with complement genes (Fig. 3G, gated area), suggesting that complement pathways may be intertwined with immunosuppressive mechanisms supporting immune escape of tumor cells in RCC.
Complement in plasma and the response to ICI
In general, the contribution of non–PD-1/CTLA-4 pathways to tumor immunosuppression predicts lack of or limited response to ICI (31). Because the high expression of complement genes is associated with 1) T cell exhaustion and 2) activated macrophage markers, we hypothesized that complement may contribute to these immunosuppressive pathways and that increased complement activity may predict resistance to ICI. Several complement proteins are secreted from cells and circulate in plasma. Complement function is routinely evaluated in plasma in the clinic (32). Furthermore, we previously showed in mouse models that activation of complement in tumors or in metastatic sites is reflected by changes in concentrations of complement effectors in plasma (8). Thus, we evaluated complement function in plasma and sought to establish whether plasma levels of complement proteins were associated with response to ICI.
We measured the concentrations of C1q, C3, C5, FB, FD, FH, FI, C3c, sCD59, and s5b-9 (TCC) in plasma collected from 24 RCC patients treated at the UTSW KCP prior to initiation of ICI (nivolumab monotherapy [n = 19] or combination ipilimumab and nivolumab [n = 5]) (Supplemental Table I) and from healthy donors (Oklahoma Blood Institute). The concentration of plasma complement proteins were correlated with TNT, a surrogate of response to ICI. TNT is defined as a time from starting of ICI until the next line of therapy, which is usually administrated because of disease progression. We used TNT, as it captures patient benefit from ICI beyond the treatment course, and it is less subjective than retrospective interpretation of imaging studies, which is also influenced by the criteria used and pseudoprogression. Thus, a shorter TNT indicates a limited response to therapy. For measurements of complement proteins, we used commercially available ELISA kits that are designed to work with plasma collected with different anticoagulants. However, as with other laboratory assays, the sample collection method may impact results. Therefore, we evaluated samples collected with EDTA versus heparin (CPT Cell Preparation Tube). For all measured complement fragments, except C1q, data distributions were similar with both anticoagulants (Supplemental Fig. 2). C1q concentration was affected by heparin (Supplemental Fig. 2A); therefore, only EDTA C1q samples were included in the subsequent analyses.
We found that the concentration of several complement proteins in RCC patients’ plasma was significantly higher than in control plasma from healthy donors, with the exception of FB (Fig. 4A–F). C1q and C3 in most RCC patients were within what is regarded as the normal range (Fig. 4A, 4B). For C5 and FI reference ranges vary significantly between laboratories; therefore, they were not included. FI and FB in RCC patients were higher and lower than in controls, respectively (Fig. 4E, 4F). Moreover, FB levels in RCC patients were below normal range (Fig. 4F). To separate patients into cohorts with high versus low concentrations of complement proteins, we used a cutoff based on the lowest p value for separation corresponding to a maximum difference in TNT (Fig. 4G–J). The patients with FH and FD below the cutoff had worse response to ICI as indicated by significantly shorter TNT (Fig. 4K, 4L). Conversely, low FI and TCC were associated with better response and longer TNT (Fig. 4M, 4N).
Complement proteins in RCC patients plasma and response to ICI. (A–F) Concentrations of complement proteins in plasma of RCC patients that were significantly different from concentrations in healthy donors. p < 0.05, by t test. (G–J) Graphs illustrating p values of separation as a function of analyte concentration. The optimal threshold is marked by blue triangle. (K–N) TNT in patients assigned to the cohorts based on cutoffs shown in (G); (K) p = 0.022, (L) p = 0.029, (N) p = 0.017, and (M) p = 0.042 (all p values by t test). (O) Multifactor decision tree algorithm to select patients responding to ICI. (P) Distribution of TNT: TNT < 12-mo nonresponders, red, and TNT > 12-mo responders, green. (Q) Distribution of responders (green circles) and nonresponders (red crosses) based on decision tree recommended thresholds for C5 >4.144 mg/dl and TCC <0.71973 AU/ml. (P) TNT in multifactor responders cohort [shown outside a yellow gate of (Q)] versus other patients. (R) TNT in multifactor non-responders vs. responders; p = 0.00000017 by t test.
Complement proteins in RCC patients plasma and response to ICI. (A–F) Concentrations of complement proteins in plasma of RCC patients that were significantly different from concentrations in healthy donors. p < 0.05, by t test. (G–J) Graphs illustrating p values of separation as a function of analyte concentration. The optimal threshold is marked by blue triangle. (K–N) TNT in patients assigned to the cohorts based on cutoffs shown in (G); (K) p = 0.022, (L) p = 0.029, (N) p = 0.017, and (M) p = 0.042 (all p values by t test). (O) Multifactor decision tree algorithm to select patients responding to ICI. (P) Distribution of TNT: TNT < 12-mo nonresponders, red, and TNT > 12-mo responders, green. (Q) Distribution of responders (green circles) and nonresponders (red crosses) based on decision tree recommended thresholds for C5 >4.144 mg/dl and TCC <0.71973 AU/ml. (P) TNT in multifactor responders cohort [shown outside a yellow gate of (Q)] versus other patients. (R) TNT in multifactor non-responders vs. responders; p = 0.00000017 by t test.
Next, we sought to determine if using plasma concentrations of several complement proteins simultaneously would provide better prediction of response to ICI than the concentration of any single protein. To identify which combination of complement protein concentrations was the best predictor, we used the scikit-learn machine learning library (21) (http://scikit-learn.org). This allowed us to establish an optimal decision tree (Fig. 4O). Based on the distribution of TNT (Fig. 4P), patients were separated into two cohorts using 12 mo as the TNT cutoff (TNT < 12-mo red color code, worse responders, and TNT > 12-mo green color code, better responders). We trained a decision tree using C1q, C3, C5, FB, FD, FH, FI, C3c, sCD59, and s5b-9 (TCC) concentrations as input features to find optimal separation with constrained tree depth. Decision trees are grown iteratively by finding the single variable split that best subdivides the group into cohorts (as measured by the greatest decrease in the node’s Gini impurity) (33). The algorithm searched through all proteins looking for a single split that best subdivided all patients (n = 24). The optimal transition was to separate patients based on whether TCC was above or below 0.71973 AU/ml (Fig. 4O), as three patients with concentrations below this threshold were responders (Fig. 4O). The algorithm was repeated on the remaining patients (n = 21), determining that a C5 concentration threshold of 4.144 mg/dl optimally split the remainder group as two patients with concentrations of C5 >4.144 mg/dl were responders (Fig. 4O). To reduce chances of data overfitting, we stopped the algorithm at this point. Thus, of the original patients (n = 24), the decision tree identified a cohort of five patients with C5 >4.144 mg/dl or TCC <0.71973 AU/ml that were all responders (Fig. 4Q, green circles outside the yellow gate). The average TNT (mean TNT = 22.06 ± 2.10 mo) in this group was significantly higher than in the alternative group (mean TNT = 7.99 ± 3.70 mo) containing 16 nonresponders and three responders (Fig. 4Q, red crosses and green circles in a yellow gate; Fig. 4R). Thus, highest benefit from ICI is observed among patients with low TCC and high C5. To extend these conclusions to the general RCC patient population, larger patient cohorts will be required.
Complement inhibition reduces growth of anti–PD-1–resistant renal tumors in mice by improving TIL function and inhibits angiogenesis
To test in vivo the role of complement proteins, we used RCC-Renca (American Type Culture Collection CRL-2947), a murine RCC model that although differing from human RCC, has been extensively evaluated in immunotherapy studies and is known to be recalcitrant to ICI (34) and, therefore, potentially resembles ICI-resistant RCC. To evaluate the suitability of this model to study the role of complement, we stained mouse tumors for complement proteins. C3 fragments were found in the vicinity of vasculature (Fig. 5A). C3 deposits colocalized with C1q (Fig. 5B) indicating a possible contribution of the classical pathway to the activation of complement (13). C3 deposition without association with C1q likely indicates involvement of the alternative pathway because this pathway is initiated by spontaneous C3 hydrolysis, followed by C3b deposition (35). Mannose-binding lectin, which initiates the lectin complement pathway, was not found in these tumors (data not shown). Of note, the alternative pathway amplification loop was demonstrated to contribute to 80–90% of C5a generation when complement is initiated through the classical pathway (36), and we found higher amounts of C5a in plasma of tumor-bearing mice versus tumor-free mice (Fig. 5C). Because the classical pathway is initiated by C1q binding to immune complexes containing either IgG or IgM, with the latter having much greater capacity to activate complement, we examined deposition of IgG and IgM along with C1q in mouse tumors. C1q colocalized with IgM (Fig. 5D), but not with IgG (data not shown), suggesting that IgM initiates complement activation. IgM deposits colocalized with Annexin V bound to apoptotic cells in mouse tumors (Fig. 5E). We found numerous CD11b+ (myeloid) cells and CD8+ T cells expressing C5aR1 (Fig. 5F, 5G), which is consistent with the role of C5aR1 in regulating MDSC (4) and TIL (37).
Complement in mouse Renca model of RCC. (A) C3 deposition along CD31 Ab-stained vasculature, (B) C1q and C3, (C) C5a concentration in plasma from tumor-free (TF) and tumor-bearing (TB) mice, *p < 0.0001 by t test. (D) C1q and IgM. (E) Annexin V and IgM. (F) C5aR1 and CD11b. (G) C5aR1 and CD8+. Arrows denote areas of colocalization. Scale bar, 50 μm. (A, B, and D–G) immunofluorescence and (C) ELISA.
Complement in mouse Renca model of RCC. (A) C3 deposition along CD31 Ab-stained vasculature, (B) C1q and C3, (C) C5a concentration in plasma from tumor-free (TF) and tumor-bearing (TB) mice, *p < 0.0001 by t test. (D) C1q and IgM. (E) Annexin V and IgM. (F) C5aR1 and CD11b. (G) C5aR1 and CD8+. Arrows denote areas of colocalization. Scale bar, 50 μm. (A, B, and D–G) immunofluorescence and (C) ELISA.
As previously reported (34), PD-1 blockade was ineffective in this model (Fig. 6A). To assess the role of complement proteins, we evaluated Renca growth in complement KO mice. Tumor growth was significantly impaired in C3aR1 KO (Fig. 6B). C5aR1 KO also showed a trend for reduced tumor growth, although differences did not reach statistical significance. Consistent with the phenotypes in KO mice, WT mice treated with C3aR1 (SB290157) and C5aR1 (PMX53) inhibitors had reduced tumor growth (Fig. 6C). Because both C3aR1 and C5aR1 are implicated in suppressing antitumor T cell responses (23, 37), we investigated the impact of complement deficiencies/blockade on T cells in tumors. Although we did not observe an effect from C3aR1 or C5aR1 genetic deficiencies on recruitment of CD8+ T cells to tumors, which is similar to anti–PD-1 therapy (Fig. 6D, 6E), pharmacological inhibition of C5aR1 resulted in an influx of CD8+ T cells to tumors (Fig. 6F), consistent with previous studies (4). Unlike anti–PD-1 (Fig. 6G, 6H), genetic deficiencies and pharmacological blockade of complement receptors improved TIL function, as demonstrated by increased production of IFN-γ upon ex vivo restimulation (Fig. 6G, 6I, 6J). The highest production of IFN-γ was associated with lack of or blockade of C3aR1. The increase in IFN-γ production suggests improved functionality and could indicate a reversal of T cell exhaustion (27). Therefore, we next evaluated expression of genes associated with T cell exhaustion in mouse cohorts that exhibited significant phenotypes (C3aR1 KO, SB290157, and PMX53). We found downregulation of T cell inhibitory pathways indicated by a reduction in the expression of genes for T cell inhibitory receptors (Pdcd1, Ctla4, and Btla), cellular receptor Fas, and transcription factor Stat3 (Fig. 6K). Btla and Fas were upregulated in SB290157-treated mice, reflecting multifaceted impact of these genes on T cell function. Although BTLA and Fas are well-recognized markers of T cell exhaustion (27), BTLA may also provide costimulatory signals to CD8+ T cells (38), and Fas/FasL signaling is critical for the survival of exhausted CD8+ T cells during immune response to tumors (39). In accordance with our recent work demonstrating contributions of complement to angiogenesis (40), we found that complement deficiency/blockade was associated with reduced vascular density in tumors (Fig. 6L) and reduced expression of proangiogenic factors (Fig. 6M), except Tie2 in a C3aR1 KO. To confirm that the therapeutic effect of complement deficiencies is T cell dependent, we chose to deplete CD8+ T cells in C3aR1 KO, as these mice had most profoundly reduced tumor growth (Fig. 6B). The treatment of mice with neutralizing anti-CD8α led to CD8+ T cell depletion, as demonstrated by lack of these cells in the spleen and peripheral blood (Fig. 6N, 6O). As expected, injections of isotype-matched control IgG did not affect CD8+ T cell subsets (Fig. 6N, 6O). We challenged these mice and controls with Renca cells 3 d after the first dose of anti-CD8α (or IgG). C3aR1 KO receiving control IgG had significantly reduced tumor growth when compared with IgG-treated WT controls, as expected (Fig. 6P). When CD8+ T cells were depleted from C3aR1 KO, the beneficial impact of C3aR1 deficiency on tumor growth disappeared (Fig. 6P). The finding that tumor growth in WT mice with or without CD8+ T cells is similar suggests that C3aR1 renders T cells dysfunctional. Overall, these data show that CD8+ T cells are required for reduced tumor growth in C3aR1 KO, suggesting contributions of this receptor to T cell dysfunction in this RCC model.
Complement inhibition in an anti–PD-1–recalcitrant model of RCC. Tumor volumes in (A) isotype-IgG or anti–PD-1–treated mice; (B) placebo (PBS)–treated WT, C3aR1 KO, or C5aR1 KO; and (C) placebo, SB290157 (C3aR1 inhibitor), or PMX53 (C5aR1 inhibitor)–treated mice [the same placebo cohort is shown in (B) and (C)]. (D–F) CD8+ T cells in tumors (TIL) from cohorts (A)–(C) by FACS. (G–J) IFN-γ expressing TIL in cohorts (A)–(C). (G) Representative FACS dot plots and (H–J) quantification of FACS data. (K) Expression of genes associated with T cell exhaustion in tumors from C3aR1 KO and SB290157- and PMX53-treated mice relative to placebo (horizontal lines). (L) CD31 immunofluorescence (vascular density) in tumors from cohorts in (K) and quantification of vascular density based on CD31 immunofluorescence. Scale bar, 50 μm. (M) Expression of proangiogenic genes in tumors form cohorts as in (K) relative to placebo (dashed line). (N and O) Representative FACS dot plots (with quantification) of spleen (N) and blood (O) from C3aR1 KO and WT mice treated with anti-CD8α or isotype-matched IgG. (P) Tumor growth in cohorts (N) and (O). Data are representative of one experiment with n = 5–15 mice. *p < 0.05, **p < 0.01 in (B) C3aR1 versus placebo and in (C) SB290157 or PMX53 versus placebo, ***p < 0.001 in (P) C3aR1-CD8+ T cells present versus WT-CD8+ T cells present, ****p < 0.0001, by two-way ANOVA for (A)–(C) and (P), one-way ANOVA for (E), (F), and (I)–(M), and t test for (D), (H), (N), and (O). ns, not significant.
Complement inhibition in an anti–PD-1–recalcitrant model of RCC. Tumor volumes in (A) isotype-IgG or anti–PD-1–treated mice; (B) placebo (PBS)–treated WT, C3aR1 KO, or C5aR1 KO; and (C) placebo, SB290157 (C3aR1 inhibitor), or PMX53 (C5aR1 inhibitor)–treated mice [the same placebo cohort is shown in (B) and (C)]. (D–F) CD8+ T cells in tumors (TIL) from cohorts (A)–(C) by FACS. (G–J) IFN-γ expressing TIL in cohorts (A)–(C). (G) Representative FACS dot plots and (H–J) quantification of FACS data. (K) Expression of genes associated with T cell exhaustion in tumors from C3aR1 KO and SB290157- and PMX53-treated mice relative to placebo (horizontal lines). (L) CD31 immunofluorescence (vascular density) in tumors from cohorts in (K) and quantification of vascular density based on CD31 immunofluorescence. Scale bar, 50 μm. (M) Expression of proangiogenic genes in tumors form cohorts as in (K) relative to placebo (dashed line). (N and O) Representative FACS dot plots (with quantification) of spleen (N) and blood (O) from C3aR1 KO and WT mice treated with anti-CD8α or isotype-matched IgG. (P) Tumor growth in cohorts (N) and (O). Data are representative of one experiment with n = 5–15 mice. *p < 0.05, **p < 0.01 in (B) C3aR1 versus placebo and in (C) SB290157 or PMX53 versus placebo, ***p < 0.001 in (P) C3aR1-CD8+ T cells present versus WT-CD8+ T cells present, ****p < 0.0001, by two-way ANOVA for (A)–(C) and (P), one-way ANOVA for (E), (F), and (I)–(M), and t test for (D), (H), (N), and (O). ns, not significant.
Discussion
For several decades, complement was thought to contribute to cancer immune surveillance through complement-mediated lysis of tumor cells (1). In contrast, we found that complement promotes tumor growth through the inhibition of antitumor immunity, mediated via activation and recruitment of MDSC to tumors (4). Follow-up studies documented a critical role of complement in immunosuppression in several mouse models and discovered other mechanisms involved in this process (6). In addition to their effects on the TME, complement proteins and receptors directly impact tumor cells (24). The most recent studies established a link between complement and cancer metastasis (8, 9) and demonstrated synergism between PD-1 blockade and complement inhibition (11). Thus, there is substantial evidence from preclinical studies pointing to complement as a potential therapeutic target in cancer. Furthermore, studies using human samples indicate that early complement components (C1q, C2, and C4) are prognostic biomarkers in lung and kidney cancer (13, 41). However, there is limited understanding of complement in human malignancies.
We found that high expression of 11 complement genes was associated with unfavorable prognosis in RCC. In contrast to recent studies (13), our analysis of large datasets from The Cancer Genome Atlas and the Human Protein Atlas failed to demonstrate a prognostic role for C4 in RCC. No other solid tumor evaluated had such striking correlation between complement gene expression and prognosis. Supporting the same notion, we found that the MAC/TCC inhibitor CD59, which inhibits complement, was associated with good prognosis. These data also indicate that limiting the final stage of complement activation may be beneficial for RCC patients. In contrast to CD59, expression of another complement regulatory protein, CD55, was not associated with improved prognosis (https://www.proteinatlas.org/ENSG00000196352-CD55/pathology). In contrast to RCC, high expression of complement genes was associated with favorable prognosis in liver, breast, pancreatic, and cervical carcinomas. Therefore, inhibiting complement may not be universally beneficial for all cancer patients, and RCC seems to be an optimal target for complement-based therapy. Immunohistochemistry data confirmed the presence of complement proteins in RCC tumors. However, where these proteins are produced is unclear. For example, they may be synthesized in the liver and deposited in the tumor stroma and vasculature.
Gene expression analyses from the UTSW KCP focusing on the TME identified an association between complement gene expression and an aggressive IS of RCC, which is consistent with key roles of several complement fragments in inflammation (22). The correlation between complement and markers of T cell exhaustion/dysfunction and alternatively activated macrophages implicates complement in regulating immunosuppression in human RCC. This is supported by reduced expression of genes associated with T cell exhaustion as a result of complement deficiency/inhibition in a mouse model. The contribution of C3aR1 to T cell dysfunction is further corroborated by studies of T cell depletion showing that intact CD8+ T cells are required to reduce tumor growth in C3aR1 KO mice (i.e., C3aR1 loss improves T cell function). Based on these data, complement appears to act as an additional checkpoint in RCC, which corresponds to studies indicating roles of C3aR1 and C5aR1 in the regulation of cytolytic activity of TIL in other mouse models (37). Thus, in the presence of complement-imposed immunosuppression, therapeutic inhibition of the PD-1/CTLA-4 pathways may not reach its full potential. The resistance to ICI in patients with high TCC levels and low C5 levels suggest that complement activation and possibly subsequent complement consumption plays a role in RCC pathogenesis. This consumption is best characterized in autoimmune diseases such as systemic lupus erythematosus and urticarial vasculitides, in which hypocomplementemia supports the diagnosis and is used to monitor disease activity (42). Because the C1q gene signature is strongly associated with poor prognosis and aggressive IS of RCC and C1q is deposited in human and mouse tumors, the contribution of the classical pathway of complement activation to RCC should be considered. C1q colocalized with IgM in mouse tumors; therefore, we theorize that IgM may trigger the classical pathway. These IgM may represent polyreactive natural Abs constitutively present in high quantities in the body fluids that bind endogenous Ags in dying, damaged, or otherwise stressed cells (43, 44). Of note, IgM decorated apoptotic cells marked by Annexin V in the mouse RCC model.
In conclusion, complement appears to be involved in RCC pathogenesis and may impact antitumor immunity, which is reflected by its association with inflammation and poor prognosis. Targeting complement might be a therapeutic option for RCC patients.
Acknowledgements
We thank Nardelio DaSilva and Robin Rajan for technical assistance with experiments included in this study. We acknowledge UTSW KCP patients for donated samples used in this study as well as the healthy volunteers.
Footnotes
This work was supported by the National Cancer Institute, National Institutes of Health (R01CA190209 to M.M.M., P50CA196516 to J.B., and CCSG 5P30CA142543 to T.W.) and the Cancer Prevention and Research Institute of Texas (RP190208 to T.W.).
The online version of this article contains supplemental material.
Abbreviations used in this article:
- AF
Alexa Fluor
- C5aR1
C5a receptor 1
- ccRCC
clear-cell RCC
- FB
factor B
- FD
factor D
- FH
factor H
- FI
factor I
- ICI
immune checkpoint inhibitor
- IS
inflammatory subtype
- KO
knockout
- MDSC
myeloid-derived suppressor cell
- NIS
non-IS subtype
- PD-1
programmed cell death protein 1
- pRCC
papillary RCC
- RCC
renal cell carcinoma
- sCD59
soluble CD59
- TCC
complement terminal complex
- TIL
tumor-infiltrating lymphocyte
- TME
tumor microenvironment
- TNT
time to next treatment
- UTSW KCP
University of Texas Southwestern Medical Center Kidney Cancer Program
- WT
wild-type.
References
Disclosures
J.B. is a paid consultant for Exelixis and Arrowhead Pharmaceuticals. The other authors have no financial conflicts of interest.